Epidemiological Research

O.S. Miettinen I. Karp

Epidemiological Research:
An Introduction

123
O.S. Miettinen I. Karp
McGill University Universite de Montreal
Universite de Montreal Montreal, QC
Montreal, QC Canada
Canada
Cornell University
New York, NY
USA

ISBN 978-94-007-4536-0 ISBN 978-94-007-4537-7 (eBook)

DOI 10.1007/978-94-007-4537-7
Springer Dordrecht Heidelberg New York London

Library of Congress Control Number: 2012942327

Springer Science+Business Media Dordrecht 2012

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publishers location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Foreword

In my life ever since medical-school graduation half-a-century ago, Ive had the
dream of reaching true understanding of the theory of the research that would best
serve to advance the knowledge-base of medicine, of genuinely scientific medicine.
Olli Miettinen wrote this a year or so ago in his Epidemiological Research: Terms
and Concepts. For those who have had the privilege to witness his odyssey since
the 1970s, the current book comes as both a wonderful revisit of the past and
a great leap into the future. Miettinens text accompanying his course in the
1970s at Harvard was the first systematic introduction to theoretical epidemiology
(Greenland) and can be viewed as the start of modern epidemiology (Morabia).
But he never published it and therefore the book that you read now is the first
published introduction to epidemiological research by the father and grand master
of modern epidemiology.
Epidemiological Research: An Introduction, the current book of Miettinen,
in collaboration with his junior colleague Igor Karp, is a true milestone for
epidemiology, but a cautionary word may be in place about the introduction. I
remember Miettinen referring to his courses as basic, but not basal. The current
book, similarly, is basic, but not basal. It is introductory in that it develops the objects
and methods of epidemiological research from first principles, but it does so in a
breathtakingly sophisticated way, alternatingly grand and subtle in argumentation,
visionary and down-to-earth, broad and deep. For this reader one thing is particularly
clear: Miettinens discussion is still unparalleled in our field, the logic and coherence
is as spellbinding as ever (and I need to think a bit more, and better, if and when I
do not fully understand what he writes).
The structure of this book must be a treat for all epidemiologists. From epidemi-
ology: grappling with the concept through etiology as a pragmatic concern and
the object of study to the books core on objects design and methods design,
it is like travelling to familiar destinations along new roads. Although Miettinen
has always stressed the importance of objects design from first principles, I think
this book is the first to treat this topic systematically and somewhat extensively.
And although Miettinen has published quite comprehensively on the fallacies in the

v
vi Foreword

design of epidemiological studies and their remedies, I find his discussion of the
etiologic study fresh and summarized aptly and succinctly in e pluribus unum and
e unum pluribus.
What will be the effect on the practice of epidemiological research? I have no
doubt that it will be vast, but also that it will be slow to come. In the long run,
his arguments will turn out to be irresistible, although most likely modified and
expanded. It is like the effect of epidemiological research on medical practice: it is
hardly ever direct, it nearly always takes a long time, but in the end it makes a true
core contribution.
I pay tribute to the father of modern epidemiology, and recommend this volume
to assist in deep epidemiological introspection. It will benefit epidemiology and
epidemiologists greatly.

Albert Hofman, M.D., Ph.D., Professor and Chair, Department of Epidemiology,

Erasmus Medical Center, Rotterdam, The Netherlands.
Preface

Anyone conducting epidemiological research is prone to encounter obvious major

challenges of a conceptual nature, sometimes seeing them to be tauntingly complex,
at other times subtle beyond concrete grasp. But the challenges can also remain
unrecognized and thereby unmet, ones of major consequence included.
Even when no longer a beginner in the research, the investigator may wonder
about the adequacy of the copings with these challenges, notably when considering
how controversial many of even the much-researched substantive issues remain and,
thus, how little consequence the research his/her own and that of others too
is having in the evolution of knowledge-based societal policies about healthcare
and in the advancement of public-health practices within their respective policy
frameworks.
The key to attaining, and maintaining, the conceptual understandings that form
the basis for maximally consequential careers in epidemiological research we take to
be suitable introduction to and thereby the attainment of a wholesome outlook in
such research. To this end, authors of introductory textbooks on the research need
to try to present basic ideas that are so obviously well-focused and so obviously
tenable that they thereby get to be even where they arent yet commonly agreed
upon by the teachers as properly constituting the core content of an introductory
course on the research.
In our view, an introductory course on epidemiological research should bring
to focus, and give tenable answers to, such orientational, normative questions as:
To what pragmatic ends should the research be conducted? What, as for both
substance and form, should the population-level research be about? What should
be understood to be the necessary, logical nature of those studies themselves on
the principal generic types of object of study? What are the main concerns and
principles in the optimization of the objects and methods of those studies? and
How should the evidence from the studies be transmuted into knowledge about the
respective objects of study?
We here make a serious effort, our first, to formulate answers to these, and
related, questions for possible incorporation into teachers efforts to properly
introduce their students to the research specifically, as insinuated above, to

vii
viii Preface

epidemiological research that would be maximally consequential and hence per-

sonally most gratifying to them, with society at large not only the sponsor but also
the correspondingly, if not even more richly, rewarded beneficiary of it.
In this effort we are guided by our belief that a proper introductory course on
epidemiological research, like its counterpart on physics for example, conveys the
most advanced insights into the most elementary the most appropriately chosen to
be the most elementary component topics within the overall topic; and our aim is
to introduce them in a logical sequence, for most natural and effective study by the
students. A contemporary introductory course on physics teaches, for example, that
the formerly common idea of ether as a ubiquitous medium for electric and magnetic
forces (`a la Maxwells equations) is now seen to have been a misunderstanding; and
a suitably advanced introductory course on epidemiological research now teaches,
for example, that the still-common concepts of cohort study and case-control study
should already be passe.
There is a story (apocryphal) about the physicist Niels Bohr and the philosopher
Bertrand Russell concerning their respective decisions not to study psychology in
preparation for a career in it, about their respective decisions to study mathematics-
cum-physics and mathematics-cum-philosophy instead. Bohr is said to have rejected
the psychology option on the ground that this field is too easy, and Russell on the
ground that it is too difficult, to gain mastery of. We are of the view that preparation
for a productive and thereby gratifying career in epidemiological research different
from a stellar career in quantum physics or theoretical philosophy does not require
any extraordinary talent. But we also are keenly aware, from our personal struggles,
that it requires much effort and to say it again a proper introduction and its
consequent proper orientation as important prerequisites. The student needs to make
the investment of the effort, upon us having endeavored to help the teacher to
provide the latter.
Much of an introductory course on epidemiological research necessarily is about
concepts and the corresponding terms germane to such research (refs. 1, 2
below). It thus likely would materially enhance the teaching, and the learning, to
supplement this textbook (or any other, for that matter) by a compendium providing
suitable introductory exposition and discussion of those concepts and terms (ref. 2),
for consultation as needed. For, dwelling on the concepts and terms in the flow of
a course like this would tend to take away from the students grasp of the logic
underpinning the sequence of concepts-based ideas being introduced, many of them
quite original.

References
1. Miettinen OS. Important concepts in epidemiology. In: Olsen J, Saracci R,
Trichopoulos D. Teaching Epidemiology. Third edition. Oxford: Oxford University Press,
2010.
2. Miettinen OS. Epidemiological Research: Terms and Concepts. Dordrecht: Springer, 2011.
Acknowledgements

Miettinen fils a polymath, even if thus far without the evidentiary portfolio of
writings studied a late-version manuscript of this book. We asked him to imagine
the youngest one of his children, the only one still young enough to realistically
consider, as a potential student of one of this youngsters paternal grandfathers
fields (theory of epidemiological research, the other one being theory of meta-
epidemiological clinical research) and to consider critically! this text as the
textbook in that youngsters introduction into this field. And we asked him to
comment from the general industrial perspective on what we wrote (sect. 14.5) about
epidemiologists role in quality assurance economic as well as medical in the
hospital-based segment of healthcare, which in Canada now absorbs about half of
the ever less sustainable fiscal burden of the countrys public-health industry.
He delighted us, for one, with a learned commentary on introductory teaching of
scholarly subjects at large and on how this text conforms with the most notable ideas
about it; and for another, he provided us with an equally insightful commentary on
industrial quality assurance in general and on its implications for modern public-
health practice. And for good measure, he permitted us to incorporate both of these
commentaries in this book (Apps. 4, 5).
We also asked Albert Hofman, of Erasmus University Medical Centre, Rotter-
dam todays preeminent leader of teaching on all aspects of epidemiological
research, from introductory courses on up, in the Erasmus Summer Programme in
particular to critically review a near-final draft of this book, with a view to his
possibly writing the Foreword to it.
Hofman indeed was kind enough to read the draft, and he had quite gratifying
words to say about it. What is much more, he did agree to write the Foreword. We
are much obliged to him.

ix
Contents

1 Epidemiology: Grappling with the Concept . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1

Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.1 Epidemiology as Practice of Healthcare . . . . . . . .. . . . . . . . . . . . . . . . . . . . 2
1.2 Epidemiology as Health Research . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 5
1.3 Epidemiology as a Research Discipline . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 8
1.4 Epidemiology as a Subject of Study.. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 10
2 Epidemiological Knowledge: Examples . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 15
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 15
2.1 Nutrition and Atherogenesis . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 16
2.2 Nutrition and Oncogenesis .. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 18
2.3 Nutrition and Ageing .. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 21
2.4 Nutrition and Obesity . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 22
2.5 Nutrition and Hypertension . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 24
2.6 Epidemiologic Reviews in Review . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
2.7 Overview of the Achievements . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 31
3 Etiology as a Pragmatic Concern . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 35
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 35
3.1 The Two Genera of Causality in Medicine .. . . . .. . . . . . . . . . . . . . . . . . . . 36
3.2 Etiology as a Community-medicine Concern .. .. . . . . . . . . . . . . . . . . . . . 38
3.3 Etiology as a Clinical-medicine Concern . . . . . . .. . . . . . . . . . . . . . . . . . . . 39
3.4 Etiology as a Legal Concern .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 40
4 Etiology as the Object of Study . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
4.1 Some Introductory Examples .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 44
4.2 From Case Series to Rate Comparison . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47
4.3 Standardized RR as the Result . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 49
4.4 The Object Imbedded in a Model.. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 51

xi
xii Contents

5 Etiologic Studies Essentials . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 55

Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 55
5.1 Source Base, Study Base . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 56
5.2 Case Series, Base Series . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 57
5.3 Model Fitting, Study Result . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 58
5.4 Demystifying the Modeling.. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 60
6 Etiologic Studies Typology .. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 63
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 63
6.1 Cohort Study: Its Essence Corrected . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 63
6.2 Case-control Study: Its Essence Corrected.. . .. . . . . . . . . . . . . . . . . . . . 65
6.3 Cross-sectional Study: Its Essence Redefined . . . . . . . . . . . . . . . . . . . . 67
6.4 Semi-cohort Study: Its Essence Corrected . . . .. . . . . . . . . . . . . . . . . . . . 68
6.5 E Pluribus Unum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 69
6.6 E Unum Pluribus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 70
7 Etiologic Studies Objects Design . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 73
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 73
7.1 Some Considerations in the Design . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 74
7.2 Some Principles of the Design .. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 75
7.3 The Result of the Design .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 83
7.4 Implications of the Design .. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 84
8 Etiologic Studies Methods Design. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 87
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 87
8.1 Some Considerations in the Design . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 88
8.2 Some Principles of the Design .. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 89
8.3 Implications of the Design .. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 94
9 Etiologic Studies Intervention Counterparts . . . . . . .. . . . . . . . . . . . . . . . . . . . 97
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 97
9.1 The Eminence of Experiments.. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 98
9.2 The Objects of the Studies . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 99
9.3 The Methods in the Experiments . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 100
9.4 The Ethics in the Experiments .. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 103
9.5 Quasi-experimental Intervention Studies. . . . . . . .. . . . . . . . . . . . . . . . . . . . 104
10 Causal Studies Acausal Counterparts .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 107
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 107
10.1 The Focus on Prevention-oriented Research .. . .. . . . . . . . . . . . . . . . . . . . 108
10.2 The Acausal Knowledge-base of Prevention .. . .. . . . . . . . . . . . . . . . . . . . 109
10.3 The Objects of the Acausal Studies . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 110
10.4 The Acausal Studies Proper.. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 110
11 Studies on Screening for a Cancer .. . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 113
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 113
11.1 Introductory Example . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 114
11.2 Epidemiologists Outlook .. . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 115
Contents xiii

11.3 The Gold Standard Trial . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 116

11.4 Meaningful Types of Trial . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 117
11.5 Meaningful Etiologic-type Study .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 120
12 Some Paradigmatic Studies. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 123
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 123
12.1 The Role of Paradigmatic Studies . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 124
12.2 Paradigmatic Causal Studies .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 125
12.3 Successful Extraparadigmatic Studies. . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 127
13 From Studies to Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 131
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 131
13.1 Original Study versus Derivative Study .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 132
13.2 The Evidence from a Study .. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 133
13.3 Interpretation of the Evidence . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 133
13.4 Inference from the Evidence .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 134
13.5 Knowledge from the Evidence.. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 135
13.6 The Needed Innovations . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 142
14 Fact-finding in Epidemiological Practice . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 147
Abstract.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 147
14.1 Fact-finding vis-`a-vis Research . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 148
14.2 Fact-finding by Use of Routine Sources . . . . . . . .. . . . . . . . . . . . . . . . . . . . 148
14.3 Fact-finding by Ad-hoc Surveys and Monitorings .. . . . . . . . . . . . . . . . . 149
14.4 Investigations of Epidemics.. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 150
14.5 Quality Assessment of Hospital Care . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 150

Appendix 1: Canadian Task Force on Preventive Health Care . . . . . . . . . . . . . 157

Appendix 2: CTFPHC on Screening for Breast Cancer .. . . . . . . . . . . . . . . . . . . . 161

Appendix 3: Needed Task Forces on Preventive Healthcare . . . . . . . . . . . . . . . . 169

Appendix 4: On Introductory Teaching of Scholars . . . . .. . . . . . . . . . . . . . . . . . . . 171

Appendix 5: Quality Assurance for Modern Public-health Practice . . . . . . . 175

Author Index.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 181

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 183

Chapter 1
Epidemiology: Grappling with the Concept

Abstract Public health as a segment of healthcare naturally is healthcare in the

public domain, as distinct from healthcare outside of societys purview; and the care
naturally is paramedical care together with medical care hence the term health,
rather than medicine, in public health.
Medicine encompasses community medicine in addition to clinical medicine.
This segment of medicine inherently is public-health medicine, whereas clinical
medicine falls under public health only to the extent that national health insurance
has been introduced.
Community medicine used to be focused on epidemics of communicable dis-
eases; and a natural term for this segment of medicine thus was epidemiology. As the
concerns in community medicine were extended to encompass endemic occurrence
of communicable and also of non-communicable illnesses, endemiology could
have been introduced as a term for this extension; but instead, the concept of
epidemiology was expanded to community medicine in the thus-expanded meaning
of it.
Epidemiological research naturally is research to advance (the practice of)
community medicine of epidemiology, that is. This research includes bench or
basic research aimed at the development of vaccines, for example; and it falls
under various medical sciences instead of constituting a science unto itself.
All of this presumably is natural and quite obviously true in the judgements of
beginning students of epidemiological research, but it is here said for the troubling
reason that concepts of epidemiology and epidemiological research different from
these are endemic in todays epidemiologic academia.
The purpose of this chapter is to help the beginning student to find proper
orientation in this academia.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 1

DOI 10.1007/978-94-007-4537-7 1, Springer ScienceCBusiness Media Dordrecht 2012
2 1 Epidemiology: Grappling with the Concept

1.1 Epidemiology as Practice of Healthcare

The Black Death of 1348 and 1349, and the recurrent epidemics of the fourteenth
and fifteenth centuries, were the most devastating natural disasters ever to strike
Europe [ref.]. : : : Europe about 1420 could have counted barely more than a third
of the people it contained one hundred years before. So writes David Herlihy in his
The Black Death and the Transformation of the West (1997; p. 17).
Herlihy continues: The devastating plagues elicited a social response that
protected the European community from comparable disasters until the present
(p. 17). And: One hundred years ago, the great bacteriologist Louis Pasteur
declared: It is now in the power of man to cause all parasitic diseases to disappear
from the world [ref.] (p. 18).
Regarding Pasteurs grand vision, it is instructive to consider the history of
epidemics of smallpox, also known as variola.
As there never was any effective treatment for this commonly-fatal communica-
ble disease, healthcare directed to it always focused on prevention. Pre-scientifically
this was a matter of variolation: immunization against variola by means of
inoculation (i.e., injection of attenuated matter from variola patients pustules,
to produce a mild case of the disease). Records of this practice in China date
back to the sixth century BCE, but in Britain and its American colonies it became
rather commonplace only in the early 1700s. While fully effective, it remained
controversial on account of concerns about its safety (unintended causation of
severe, commonly fatal cases of the disease).
The dawn of science relevant to the practice of variolation was heralded by
Benjamin Franklins work on this. His goal was simple and straightforward. He
wanted to give anxious patients evidence that it was safe to have their children
inoculated. The data he assembled were most impressive. : : : This, and more, on
Franklins work on variolation can be found in I. Bernard Cohens The Triumph of
Numbers: How Counting Changed Modern Life (2005; p. 90 ff.).
Then came what has been, arguably at least, the most spectacular scientific
breakthrough in medicine, resulting in the introduction and adoption of vaccination
as the replacement of inoculation in the prevention of smallpox. (Vaccinia was a
synonym for cowpox, the smallpox vaccine being based on matter from pustules of
cowpox a disease similar to but milder than smallpox.) The scientific breakthrough
introducing vaccination was not a triumph of numbers (cf. above): it resulted
from Edward Jenners work on a single (young) person. But the resulting triumph
in medicine was, ultimately, complete: a concerted global program of vaccination,
launched by the World Health Organization, resulted in eradication complete
elimination of the smallpox virus from human populations.

In his Civilization: The West and the Rest (2011), Niall Ferguson asserts that what
distinguished the West from the Rest the mainsprings of global power were six
identifiably novel complexes of institutions and associated ideas and behaviours
(p. 12). One of these he specifies (p. 13) as:
1.1 Epidemiology as Practice of Healthcare 3

Medicine a branch of science that allowed a major improvement in health and life
expectancy, beginning in Western societies, but also in their colonies

A historiographer as he is, Ferguson is to be forgiven for this excessive deference

to medical sources, which commonly (and intentionally?) confuse physicians
practice of healthcare which we take to be the true concept of medicine
with medical science. Major improvement in Western and other societies from
practices based on medical science is, however, an incontrovertible fact of history.
Ferguson addresses both medical science and the practice of medicine only
in reference to that segment of medicine which produced the historically so
consequential, major improvements in human health, namely community medicine.
For this segment of medicine he uses its umbrella term public health, which means
healthcare in the public domain, and subsumes also paramedical care. He sketches
the enormous accomplishments of European medical science in providing the basis
for highly effective community programs to prevent infectious diseases, tropical and
other, and the startling gains in population health that have been achieved by means
of those programs.
Neither that science nor those practices does Ferguson characterize as epidemi-
ological, nor as epidemiology; in fact, neither one of these terms can be found
anywhere in that admirably erudite book. Once aware of this, a student in an
introductory course on epidemiological research might be driven to ask: Is this
an oversight? or, Is it that there indeed is no epidemiology in the meaning of
physicians epidemiological community-level practice of healthcare, served
by epidemiological research, while clinical research obviously is in the service of
physicians practice of clinical healthcare?

Looking for the answer, the student might explore, for example, the websites
concerning Ottawa Public Health, the agency providing medical and paramedical
healthcare for the community/population of Canadas capital city. This agency is
constituted and functioning in accordance with the Health Protection and Promotion
Act of the province of Ontario, in which Ottawa is located. The purpose of this
Act is to provide for the organization and delivery of public health programs
and services, the prevention of the spread of disease and the promotion and
protection of the health of the people of Ontario. Among the Mandatory health
programs and services this Act specifies, quite notably, Collection and analysis of
epidemiological data (italics ours).
Pursuant to this Act, Ottawa Public Health is headed by a Medical Officer of
Health, who is a physician with provincially legislated powers to promote and
protect the publics health (italics ours). The other personnel of this agency is
composed of specialized teams of health professionals and support staff, including
: : : physicians, epidemiologists, : : : (italics ours).
This agency periodically publishes the City of Ottawa Health Status Report. The
latest one, from 2006 (as of Dec. 2011), presents a wide range of health-related in-
formation on mortality, morbidity, communicable diseases, reproductive outcomes,
4 1 Epidemiology: Grappling with the Concept

environmental indicators, and behavioral risk factors. It deserves note, in the

present context, that of this reports four authors (under the direction and guidance
of the Medical Officer of Health), three are specified to be epidemiologists as
might be wholly unsurprising, given what the report is about. And lets be clear:
they are not reporting on epidemiological research (scientific) but on their practice
as epidemiologists.
It thus is evident that epidemiology in at least one meaning of the term is
community medicine, practice of this. In fact, this could well be the only meaning of
the term: Research for clinical medicine is termed clinical research; but this research
is not thought of as constituting one of the meanings much less the only meaning
of clinical medicine. By analogy, it would seem natural to think of epidemiological
research as being extrinsic to that which it serves, namely epidemiology in the
meaning of community medicine, its practice; it would seem natural to think that the
research is not epidemiology by virtue of being in service of epidemiology. In these
terms, just as a clinical researcher is not inherently a clinician, so an epidemiological
researcher is not inherently an epidemiologist.
The work of Ottawa Public Health illustrates the essence of epidemiology as a
genre of the practice of healthcare: not only is it community medicine but, as such, it
inherently is (a physicians practice of) public health as a matter of community-level
preventive medicine. More on this in the last chapter (ch. 14) of this book.

Eminent in public-health practices in respect to non-communicable diseases have

been those centering on occupational populations. As Robert Proctor points out
in his The Nazi War on Cancer (1999), German physicians had a long tradition
in industrial hygiene, and Nazi physicians continued this practice; and the very
major anti-cancer program of that era (19331945) in Germany had an emphasis
on occupational carcinogenesis (p. 73).
In a 1973 book entitled Occupational Health Practice, M.L. Newhouse and
R.S.F. Schilling (the Editor) have a chapter on Uses and methods of epidemiology.
They say that for an occupational health service to achieve higher standards of
health in a working community, it is essential to observe people as groups by
using the methods of epidemiology, a branch of medical science : : : (p. 169).
To them and by no means alone occupational-health practice is not practice of
epidemiology but, instead, involves use of it, of the methods of that science. They
refer to a textbook source with a telling title: Uses of Epidemiology (Morris JN,
1964).

The physicians who, in Ottawa and elsewhere, produced the community diagnoses
about, say, the SARS and H1N1 (swine flu) epidemics and directed the population-
level programs to control these, and physicians who are caring for, say, occupational
populations as populations (rather than individually), we really think of as epi-
demiologists in the practice of epidemiology of community-medicine, that is. We
really do not think of them as users of epidemiology in something that is not
epidemiology.
1.2 Epidemiology as Health Research 5

1.2 Epidemiology as Health Research

Oncology, in one meaning of the term, is a discipline (specialty) of medicine;

it is the discipline of a physicians practice of cancer-related healthcare, clinical
or community-level practice. Oncology also is a science; it is the medical science
aimed at advancing the practice of oncology. Oncological research is applied
science in this meaning; and this it is regardless of how basic it is as when aimed
at, say, potential development of a proteomic biomarker for latent-stage diagnoses
about a cancer or a protein to turn off an activated proto-oncogene in the treatment
of a cancer.
By analogy, this: Insofar as the practice of community medicine is viewed as
being epidemiology, epidemiological research is naturally taken to be research
any research, however basic aimed at advancement of the practice of community
medicine. We hold to the predicate in this (sect. 1.1 above), and we therefore hold
this inclusive conception of epidemiological research.
The earliest, and arguably the most spectacular, example of epidemiological
research, in these terms, is the study for which Edward Jenner is so famous
(sect. 1.1). He injected into a little boy, who had not experienced smallpox,
matter from a cowpox lesion of someone else; and when Jenner a few weeks
later inoculated this boy with smallpox matter, the boy did not come down with
smallpox the boy evidently had been immunized against this much more serious
disease. (Jenner had developed, as a practitioner, the impression that those having
undergone cowpox had been resistant to smallpox.) And recent examples include,
among others, the research culminating in the availability of a vaccine for the
prevention of HPV infection and, thereby, prevention of cervical cancer.
The concepts of oncological and epidemiological research are not fully analo-
gous, however. Oncological research in the aggregate amounts to a science (called
oncology), in the research (rather than knowledge) meaning of that science. This
is because oncological research has a coherent, singular subject-matter or material
object, namely malignant neoplasm. Epidemiological research, very different from
this, can address a material object from any one of a large number of health
sciences oncology, immunology (as in Jenners case), etc. As a consequence,
epidemiological research is not definitional to a science (called epidemiology); it
like morphological research, for example is subsumed under a variety of sciences.

Students of epidemiological research are prone to be left confused about the essence
of this line of research, at least early in their studies. To wit, one of us wrote his thesis
for the degree of Master of Public Health, in 1964, on Epidemiology and Its Method,
which at the time were topics of vigorous debate in the American Journal of Public
Health. Drawing from the various then-prevailing definitions, he merely synthesized
them in the framework of explicit principles as amounting to this: Epidemiol-
ogy is the science of the occurrence of health and illness; the scope of epidemiology
is the entirety of the occurrence of health and illness; and the method of epidemiol-
ogy is essentially the scientific method but may have some distinctive characteris-
6 1 Epidemiology: Grappling with the Concept

tics. He thus found, most notably, that to his elders epidemiology was a science, and
only this, rather than the practice of community medicine, this alone (cf. Schilling
in sect. 1.1). He refrained from expressing his own views about these ideas of his
elders (even though he hadnt been brought up in the Presbyterian culture).
David Lilienfeld son of the eminent epidemiologist Abraham Lilienfeld when
he still was a college student in the 1970s, took a more-than-casual interest in what
his fathers field really was. He delved deep into the literature on the matter; and
he brought to view 23 published definitions of epidemiology, the first one from
1927. Dissatisfied with them all, he added one of his own: epidemiology is a
method of reasoning about disease that deals with biological inferences derived from
observations of disease phenomena in population groups. (The culture surrounding
his youthful development different from that of the youngster alluded to above
presumably was the one which, quite uniquely, encourages critical study of the
Scripture, even.)
We present, in Table 1.1 below, seven notable definitions of epidemiology, from
1956 to the present. The strong impression from these definitions, disappointing to
us, is that epidemiology is seen to be a matter of research alone, and this, even,
without any inherent service relation to the practice of healthcare different from,
say, that of oncological research in relation to the practice of oncology (cf. above).
It seems that the research is seen to be conducted for the sake of the research itself,
as though there were no practice of epidemiology being served by the research (cf.
sect. 1.1).
It really does seem, on the basis of those definitions, that among epidemiological
researchers of late there hasnt been any conception of epidemiology as practice
of healthcare. Sight seems to have been lost of the field pioneered by Hippocrates
pre-scientifically, as most notably described in his Of the Epidemics; this field
made vastly more consequential by the spectacular scientific contributions to it by
Edward Jenner, John Snow, Louis Pasteur, and Robert Koch, among others; this
field later extended to combating also epidemics of non-communicable illnesses
such as congenital malformations, cancers, degenerative cardiovascular diseases,
and diabetes; this field now combating even endemic rates of population occurrence
of whatever type of illness. These practices do exist (sect. 1.1), and they are very
important. But these practices of community medicine, according to the definitions
in Table 1.1, are not epidemiology; and Ottawa Public Health, when it periodically
reports on the health of the citys residents, as determined in its practices there, is
not, according to those definitions, reporting on epidemiology. (Cf. sect. 1.1.)
According to most of those definitions in Table 1.1, epidemiology is merely the
study of something, sometimes alternatively (and unjustifiably; cf. above) said to
be the science of that something; and to this has recently and only exceptionally
been added application of the results of that study. We take those definitions to be
intended to mean that epidemiology is, in the only meaning of the term, a line of
research population-level research on the rates of occurrence of phenomena of
human health.
In those definitions there is no expression of the conception of a certain
type of practice of healthcare as inherently being epidemiology, even when its
knowledge-base does not derive from epidemiology-the-study. The knowledge-base
1.2 Epidemiology as Health Research 7

Table 1.1 Select definitions of epidemiology

Maxcy KF. Epidemiology. In: Maxcy KF (Editor). Rosenau Preventive Medicine and
Public Health. 8th edition. New York: Appleton-Century-Crofts, 1956:
that field of medical science which is concerned with the relationships of the
various factors and conditions that determine the frequencies and distributions of
an infectious process, a disease, or a physiological state in a human community
(p.1289)
MacMahon B, Pugh TF, Ipsen J. Epidemiologic Methods. Boston: Little, Brown and
Company, 1960:
the study of the distribution and determinants of disease prevalence in man (p.3)
Lilienfeld AM. Foundations of Epidemiology. New York: Oxford University Press, 1976:
the study of the distribution of a disease or a physiological condition in human
populations and of the factors that influence this distribution (p. 3)
Sartwell PE, Last JM. Epidemiology. In: Last JM (Editor). Maxcy-Rosenau Preventive
Medicine and Public Health. 11th edition. New York: Appleton-Century-Crofts, 1980:
the study of the distribution and dynamics of disease in populations (p. 9)
Last JM (Editor). A Dictionary of Epidemiology. 1st edition. Oxford: Oxford University
Press, 1983:
the study of the distribution and determinants of health-related states and events in
populations, and the application of this study to control of health problems
Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. 3rd edition. Philadelphia:
Lippincott Williams & Wilkins, 2008:
the study of the distribution of health-related states and events in populations
(p. 32)
Porta M (Editor), Greenland S, Last JM (Associate Editors). A Dictionary of
Epidemiology. 5th edition. Oxford: Oxford University Press, 2008:
the study of the occurrence and distribution of health-related states or events in
specified populations, including the study of the determinants influencing such
states, and the application of this knowledge to control the health problems

of mass screening for tuberculosis has been entirely clinical and, thus, not even in
part the result of epidemiological study. Such screening therefore has not been
epidemiology according to any of those definitions. Nor were the recent population-
level efforts to control the SARS and H1N1 (swine flu) epidemics epidemiology,
as they were programs of the practice, not ones of mere study.

With academics having lost sight of epidemiology as practice of healthcare (of

community-level preventive medicine), epidemiological research does not get to be
defined according to what it is intended to serve. It used to be commonly defined
by the use of the epidemiological method (cf. above), but more recently it is
commonly defined in terms of something about its objects of study. And whereas all
phenomena of health are in its purview, as potential material objects of the research,
the felt need now is to define epidemiological research by its formal objects, the
(characteristic) form of its objects (on whatever phenomena of health). This is what
the definitions in that Table 1.1 are about.
8 1 Epidemiology: Grappling with the Concept

However well this is done, the implication is that excluded from epidemiological
research is the basic or bench research that underpins, for example, the develop-
ment of vaccines. In these terms by the form of its objects one can define only
a part of what we take to be epidemiological research (cf. above). This, we hold,
should be the idea. For by no means is all of the research relevant to dealing with
communicable-disease epidemics, for example, singular in the form of its objects,
just as, say, oncological research isnt.
Insofar as there is taken to be a definitional form of the objects of epidemiological
research, this has to do with research on phenomena of health on the population
mass level, the level of community medicine. Relative to the clinical level,
phenomena of health on this higher level have new, emergent properties. On
the clinical level a given phenomenon of health characterizes an individual at a
particular moment in time; the phenomenon either is or is not associated with
that person-moment; it either is or is not occurring in it; it has no frequency of
occurrence at that person-moment. On the population level, by contrast, any given
phenomenon of health inherently occurs, and doesnt occur, in association with a
multitude of person-moments. It has, thus, the emergent quality of the frequency
the rate of this occurrence. This is the central element in the form of the objects
in epidemiological research on the population (rather than bench) level.
A health phenomenons rate of occurrence (on the population level) is, inher-
ently, a quantitative phenomenon; but the rate quite generally is non-singular in
magnitude. Its level varies according to various determinants of this level; it is a
function of its determinants. Therefore, epidemiological research, when addressing
rates of occurrence of phenomena of health, necessarily has occurrence relations as
its objects: a health phenomenons rate of occurrence in relation to as a function
of its determinants, this in a defined domain of people.
Given those definitions in Table 1.1 above, it should be noted that population-
level epidemiological research is not about the distributions of the phenomena of
health; and that the determinants of the rates of occurrence of phenomena of health
are not objects of epidemiological research only rates relations to these are. On
the clinical level, cases of a phenomenon of health inherently have a distribution by
gender and age, for example, while on the population level the phenomenon has
a rate of occurrence that may, or may not, depend on gender and age. And, for
example, behavioral determinants of the occurrence of phenomena of health are
objects of study in behavioral and social sciences and not in epidemiological
research, which is biomedical science (though not constituting a science; cf. above).

1.3 Epidemiology as a Research Discipline

There is no single discipline of how to conduct oncological research no textbooks

nor any training courses on what, as to form at least, to study and how. In basic
oncological research in particular, progress is not a consequence of professionalism
in the research. Instead, progress derives from the process of creative conjectures
1.3 Epidemiology as a Research Discipline 9

and their critical testings, as the philosopher Karl Popper explains in his venerable
Conjectures and Refutations (1963). In these terms there may be no limits to
scientific progress, as the polymath David Deutsch argues in his The Beginning of
Infinity (2011). A wonderful biography of oncological research is The Emperor of
All Maladies (2010) by Siddharta Mukherjee.
When a line of health research is defined by the form of its objects, without
coherence of the material objects, as epidemiological research now commonly is
(sect. 1.2 above), there is a corresponding opportunity together with, as usual, a
corresponding risk. That the focus in this research is on objects of a singular form
that of occurrence relations (sect. 1.2) provides for the development of a more-or-
less coherent body of theory to guide research on objects of that form, a learnable
research discipline for professional study of phenomena of health on the population
level and of non-health phenomena just the same. An associated risk is a certain
formalism in the research replacement of creative and critical thought by mere
conformity with the prevailing professional norms, unjustifiable ones included.
The discipline of how to conduct population-level health research is the sub-
ject of several contemporary textbooks of Epidemiology (e.g., MacMahon and
Trichopoulos, 1996, and Rothman et alii, 2008), implying that this discipline is
the denotation the only one of epidemiology. To us this is, however, only one
the most recent, a third meaning of the term (cf. sects. 1.1, 1.2), and a questionable
one at that.
The Enlightenment maxim Sapere aude! (Dare to reason!), eminently prop-
agated by the philosopher Immanuel Kant among others, should be understood
to apply to the prevailing teachings about epidemiological occurrence research as
well. After all, questioning received knowledge (scholastic) was at the very core
of Enlightenment; and the Enlightenment outlook in turn was the springboard of
the enormous progress in science subsequent to the advent of this outlook in the
seventeenth century, as Ferguson explains (sect. 1.1) even if, arguably at least,
its antecedent scientific progress was at the root of Enlightenment. (In the pre-
Enlightenment era, Copernicus dared publish only post-humously, while Galileo
was more courageous than that and suffered for it.)

When people in our post-industrial societies concern themselves with that which
is the concern in community medicine in epidemiology in this meaning of
population-level practice of preventive healthcare (sect. 1.1) they think of their
nutrition diet and dietary supplements first and foremost. It therefore is
particularly illustrative of epidemiological research in this disciplinary sense of
it, to examine its accomplishments with focus on nutrition: what is known? what
is the basis of this knowledge from epidemiological research? and what does this
say about the scientific prowess of our contemporary discipline of epidemiological
research?
We devote much of chapter 2 of this book to this topic. It illustrates the stagnation
that in research tends to flow from formalism and from its associated academic
routine of counting publications instead of contributions. The central flaw in todays
disciplinary framework of epidemiological research is, as we will illustrate, its
10 1 Epidemiology: Grappling with the Concept

focus on the methodology of the research, without regard for disciplined objects
design before methods design. (Stagnation consequent to formalism in societal
development was a major concern of the Soviet leader Michael Gorbachev when he
launched his famous programs of openness/glasnost and restructuring/perestroika.)

1.4 Epidemiology as a Subject of Study

As set forth in the three sections above, there now is epidemiology in three different
meanings of the term; and there could just as well be a fourth one: given that
there is epidemiology in the meaning of the theory concepts and principles
(and terminology) of epidemiological research (sect. 1.3 above), a related, more
proximal but very different meaning of epidemiology would be that of the theory
of epidemiological practice, a theoretical discipline yet to be introduced, even
though that of clinical practice has been in Up from Clinical Epidemiology &
EBM (2011) by one of us.
Given the prevailing triad of meanings of epidemiology (sects. 1.1, 1.2, 1.3
above), a student now pursuing an academic degree in epidemiology unspecified
(as now is usual) needs to be clear on which one of the prevailing three meanings
of the term is the one of his/her chosen future career. For this is a prerequisite for
understanding how the studies preparatory to the career should be composed from
the three epidemiologies that now may be available for study.
As for epidemiology in the meaning of a discipline of research, the most
important orientational questions are these: Does a future practitioner of community
medicine need to study this? and, To what extent does study of this discipline
constitute the necessary preparation for a career in epidemiological research? To
that first question our answer is simple: Not really, but an introductory course might
be a justifiable investment of time and effort, with merely impressionistic education
the aim of this. (The future practitioner really would need to study the theory
concepts and principles of the practice, were such study to be available, and then
the current knowledge-base of this practice and how to update this knowledge.) And
to that second question our answer, below, is rather involved. Preparation for a career
on the theory of population-level epidemiological research the development and
teaching of this is a very uncommon concern of the student, and it thus is not our
express concern in this book, even though an introductory course on epidemiological
research is very useful for the future theoretician as well.
The student whose future career will be one of population-level epidemiological
research needs to make good choices in relevant areas. One of these now receiving
much too little attention has to do with the substantive focus in the studies,
including in any thesis work (of which the subsequent research commonly is an
unwitting extension). For (s)he needs to understand that society does not sponsor the
research in order that the researchers can indulge in their interests; there needs to be
a societally justifiable purpose in the research (cf. Preface). Consider two examples:
1.4 Epidemiology as a Subject of Study 11

One of us, as a doctoral student of epidemiology, half-a-century ago, thought

that a good topic for his PhD thesis in epidemiology would be the rate of biological
ageing, this in causal relation to various aspects of lifestyle. In fact, he thought
that this would be a good line of research for his entire career, and for a number
of other researchers as well. These judgements sprang from his consideration of
the publics great concern to retard the process together with his vision of the
feasibility of studying it feasibility critically in terms of how an operational scale
for the rate of ageing could be developed. He therefore studied, on his own and
in depth, the biology of ageing. But in the end he did not pursue this research,
because he judged that his preordained career line, one in that prevailing third
meaning of epidemiology before extending it to the theory of clinical research
is better served by further studies in biostatistics and thesis work in this area. (This
judgement he did implement.)
Another example to consider is the entire realm of social epidemiology, now
in notable ascendancy. In it, the concern is to study population health in relation
causal relation to such social characterizers of people as their race and level of
income, for example. But for health effects of race to represent even a theoretically
admissible choice let alone a good one for epidemiological research, it should
be that race effects on health (or whatever else) are imaginable; but they arent,
given that for a persons actual race there was no alternative: a given, real person
could not have been conceived and born as a representative of some alternative
race; for a given person the racial implication of the pair of parental gametes is an
immutable given. And if it be granted even if only for the sake of argument
that the relative effects of alterative distributions of income on population health
are meaningfully studyable, meaningful questions would arise. For example: Would
this research possibly help justify societal policies concerning, say, minimum wage?
Would it even be considered? Or is it, instead, that poverty is, and is to be viewed as,
a problem as such, and not because of what it causes as health effects, for example.
More on this in section 2.6.
A student preparing for a career in epidemiological research should enter into a
career-long program of self-development with a view to maximizing good choices
for the substantive topics of his/her research. Central to this pursuit is understanding
what the best source of advice is, namely, the practitioners out there. So, if the
student would contemplate, and have difficulty perceiving, the relative merits of the
two lines of research in the examples above, (s)he should ask epidemiologists of
the first practicing type (sect. 1.1) this: For you to better serve the protection of
your client populations health, which one of these two would you rather learn from
epidemiological research: the effects of carbohydrate-rich diets on individuals rates
of biological ageing, or the health effects of low wages?
Thus, a student preparing for a career in community-level health research
(biomedical; sects. 1.2, 2.6) needs to get to the habit career-long habit of
maintaining up-to-date familiarity with the goings-on in the practice of community
medicine, notably with a view to familiarity with the gaps in the knowledge-base of
the practices as seen by the practitioners themselves. The academic program should
include arrangements for the students inquisitive mingling with practitioners.
12 1 Epidemiology: Grappling with the Concept

Parallel with this program of maintaining awareness of the goings-on in the field
there is, of course, the need for a counterpart of this in respect to epidemiological
research, which, ideally at least, is in the service of the practice. The research at
large deserves only superficial following, to remain up-to-date on what the topics
are. On the other hand, research in ones own area of subject-matter needs to be
followed both inclusively and in depth, the relevant basic research included.
No good purpose in the education of epidemiological researchers, we think, is
served by a general course addressing research on select topics in substantive epi-
demiology, given the great diversity of subject-matter in the students future lines of
epidemiological research or, for that matter, of epidemiological-type research that
actually is extraepidemiological. Equally meaningless are seminar presentations of
studies on a limitless melange of subject-matter, empty of content that the audience
at large should consider for self-development as epidemiological researchers.

Last, but by no means least, the most obvious point here: For future researchers
on population-level topics of the epidemiological form (as superficially specified
in sect. 1.3 above) the generally directly-relevant preparatory studies toward good
choices are ones of the theory of that research. Epidemiology in this research-
discipline sense is what the studies principally are about. Study of it should be
materially indeed, critically helpful in efforts to optimize decisions about what,
exactly, to study a matter of the studies objects design and also how to study
it a matter of the studies methods design so long as at issue is population-level
(rather than basic) research.
It may not be obvious, however, what these theory studies, even in their broadest
outlines, are or should be about. But even before this question there is the one about
the necessary prerequisites for these studies. The student should master certain
general concepts of medicine and science, but the availability of a suitable reference
text should allow inquiry into these as needed, rather than necessarily in advance
of the studies on the theory specific to population-level epidemiological research.
(Cf. Preface.)
The student absolutely needs a background in (probability theory and then)
statistics. For as mathematics is to physics, so statistics is to population-level
epidemiological research. Major background study of mathematics is an absolute
prerequisite for any serious study for a career in physics; but: mathematics is not
the theory the embodiment of the concepts and principles of physics, nor is
physics an outgrowth of mathematics (a field of meta-mathematics, as it were). The
objects of population-level epidemiological research are generally about frequencies
(sect. 1.2) and hence statistical in form; such epidemiological research therefore is
statistical research in form (while medical in substance). But as with mathematics in
relation physics, statistics is not, nor does it subsume, the theory of population-level
epidemiological research.
Students of epidemiological research thus need to have suitable preparatory
education in (probability theory and) statistics, and of course the necessary math-
ematical preparedness for this. The suitable education in statistics would mainly
focus on topics of multiple regression; and it would not confuse and mislead
1.4 Epidemiology as a Subject of Study 13

the students with endogenously statistical ideas about research with survey
research, sample size determination, multiple hypothesis testing, and data-
suggested hypotheses, among others. For, to say it again, statistics is not, nor does
it subsume, the theory of statistical science.
The rest of this book is, in the main, quite extensive an exposition of what
we think these theory studies should be about, on the introductory level. But in
utterly succinct terms, as insinuated above, it can be said that at issue really are
only two broad topics: objects design and methods design for population-level
epidemiological research. With these broad topics suitably construed, theory of
epidemiological data analysis (really, synthesis of the data to the study result)
becomes essentially a non-topic.
As a student setting out to prepare for a career in epidemiological research
grapples with the concept of epidemiology per se and that of epidemiological
research in relation to this these frustrating topics at the very outset of the studies
(s)he would do well, we think, being mindful of two principles bearing on this.
1. Critical for progress in any field of science generally is propitious definition of
the field in question. Highly illustrative of this is the historical fact that chemistry
started its spectacular progress only once Robert Boyle distinguished between
mixtures and compounds; defined elements as constituents of compounds,
introducing the term analysis for the study of these compositions (cf. data
analysis as a term for data synthesis, above); and defined chemistry itself by its
being not about mixtures but about compounds. These revolutionary ideas he
set forth in his book fittingly entitled The Sceptical Chymist (1661).
The student of epidemiological research should reflect on the relative merits
of defining epidemiological research as (a) the study of the distribution and
determinants of phenomena of health in human populations (or some variant of
this), or as (b) research including basic research for advancement of the
practice of community medicine, of epidemiology in this meaning.
2. Read not to contradict, nor to believe, but to weigh and consider (ref. 1
below). In his/her understanding of the theory of epidemiological research, the
student of this will ultimately be alone, solely responsible for his/her own (sic)
understanding. For the vision of one man lends not its wings to another man
(ref. 2 below). (A practitioner of epidemiology, by contrast, needs to defer to
experts shared beliefs knowledge in this meaning on the substantive topics
that are relevant.)

References
1. Bacon F. Of studies. In: Bacon F. The Essays or Counsels Civil and Moral. Edited with an
Introduction and Notes by Brian Vickers. Oxford: Oxford University Press, 1999; p. 134.
2. Gibran K. On teaching. In: Gibran K. The Prophet. New York: Alfred A Knopf, 1970.
Chapter 2
Epidemiological Knowledge: Examples

Abstract Whereas the natural the only justifiable purpose of epidemiological

research is the advancement of epidemiology in the meaning of (the practice of)
community medicine, a student starting to prepare for epidemiological research
naturally wishes to gain some orientational sense of the status of knowledge
and then, especially, of lack thereof relevant to community medicine, despite the
vastness of the number of issues involved in this. In respect to the status of the
knowledge there thus is a need for suitable focusing of the exploration.
One guide to the selection of a suitable focus in the exploration of the present
state of epidemiological knowledge is the fact that community medicine is preven-
tive rather than curative/therapeutic or rehabilitative medicine. The knowledge-
base of community medicine thus is naturally organized by determinants (causal)
of population health, with a view to all of the various health implications of these
rather than by health outcomes, with prevention of each treated as separate topics.
Thus, an encyclopedia of preventive medicine would address the entirety of the
health effects of smoking under the rubric of smoking not the role of smoking in
the causation of various illnesses, with these outcomes as the organizing principle.
In this spirit, we focus on the eminent topic of nutrition as a determinant
of select aspects of health. More specifically, we mainly address the fats-versus-
carbohydrates contrast in the energy content of the diet, but to an extent we also
address omega-3 fatty acids, antioxidants, and salt in the diet.
As another, very different depiction of the state of epidemiological knowledge
(concerning the prevention of illness) we examine the annual issues of Epidemio-
logic Reviews, representing the state of the art these reviews being comprehensive
and critical and produced under the auspices of an eminent school of public health
together with an eminent journal of epidemiology affiliated with it.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 15

DOI 10.1007/978-94-007-4537-7 2, Springer ScienceCBusiness Media Dordrecht 2012
16 2 Epidemiological Knowledge: Examples

2.1 Nutrition and Atherogenesis

In epidemiological practice (of community medicine; sect. 1.1), an eminent line of

action is population-level health education, and especially that about lifestyles
education that presumedly is based on professional knowledge about the health
consequences of various features of lifestyle when maintained in lieu of particular
alternatives to these. The regulatory and service approaches to improvement of
population health are, of course, equally dependent on the availability of knowledge
of what in peoples lives bears on the preservation of their health, and to what extent.
The relevant knowledge is about health in relation (causal) to lifestyles in respect to
nutrition first and foremost.
Among peoples common nutrition-related health concerns are the risks of
degenerative vascular diseases (mostly coronary and cerebro-vascular disease)
consequent to relatively advanced stages of atheromatosis (fatty deposits in arteries;
Gr. ather
N e,
N gruel). And nutrition-based prevention of these diseases therefore is
principally directed to slowing down the process (life-long) of the progression of
atheromatosis of further atherogenesis, that is. (Thrombogenesis is a related other
concern, though much less eminent; Gr. thrombos, clot.)
Concerning the present state of knowledge about the role of nutrition in atheroge-
nesis, there scarcely is an exposition superior to Good Calories, Bad Calories: Fats,
Carbs, and the Controversial Science of Diet and Health (2007), by Gary Taubes.
The title is telling: the science concerned with something that is of central concern in
community-level preventive medicine and in its clinical counterpart too remains
a matter of eminent controversies. In fact, so Taubes argues, the now-dominant
central tenets about nutritional atherogenesis are likely wrong. In developing his
ideas he draws from epidemiological research in the inclusive meaning we give to
this, basic research very definitely included (sect. 1.2).
Taubes focuses on the relative proportions of fats and carbohydrates in the overall
energy composition of diet, fully appreciating that a low-fat diet tends to be, in
isocaloric terms, a diet correspondingly high in carbohydrates, and conversely.
Related to this, in his book, Part I is entitled The Fat-Cholesterol Hypothesis
and describes how we came to believe that heart disease is caused by the effect of
dietary fat and particularly saturated fat on the cholesterol in our blood. It evaluates
the evidence to support that hypothesis. Part II is entitled The Carbohydrate
Hypothesis. It describes the history of the carbohydrate hypothesis of chronic
disease, beginning in the nineteenth century. It then discusses in some detail the
science that has evolved since the 1960s to support this hypothesis, and how
this evidence was interpreted once public-health authorities established the fat-
cholesterol hypothesis as conventional wisdom (p. xxiii).
While Parts I and II indeed evaluate those profoundly different hypotheses,
a much larger picture actually is painted by the two: how the purported knowledge
about the role of diet in atherogenesis, as formulated by health agencies both non-
governmental and governmental and even by a committee of the U.S. Senate, is a
consequence of the superficial appeal of the now-dominant hypothesis (postulating
2.1 Nutrition and Atherogenesis 17

an atherogenetic effect of dietary fat, mediated by level of serum cholesterol);

the zealous and persistent campaign of the originator of that hypothesis (Ancel
Keys) and of his adherents (Jeremiah Stamler, i.a.) to keep that hypothesis alive;
and various improprieties that have distorted the translation of the evidence into
purported knowledge.
Following some laboratory experiments on animals of arguably misleading (her-
bivorous) species, the initial human-level evidence for the fat-cholesterol conception
of atherogenesis was derived from geographic comparisons (by Keys); but it was,
as Taubes explains, biased by design. On the other hand, evidence against it
commonly ignored came from the venerable Framingham Heart Study even
if an important part of this the proponents of the hypothesis kept unpublished for
a long time. Countervailing evidence from a randomized trial by partisans of that
conception also was substantially delayed in its publication.
Regarding the carbohydrate hypothesis, Taubes describes, extensively, the
science that was pushed aside as investigators and public-health authorities tried
to convince first themselves and then the rest of us that dietary fat was the root
of all nutritional evils (p. 152). The pushed-aside science at issue here concerns,
most eminently, two topics: blood levels of triglycerides and lipoproteins and their
dependence on carbohydrates rather than saturated fats in the diet; and the sac-
charine disease or syndrome x or metabolic syndrome conception of diseases
of civilization, coronary heart disease among these, with dietary carbohydrates
refined starches and sugar in particular the civilized cause of this aggregate of
diseases.
This pushed-aside science is, as Taubes describes it, complex but quite com-
pelling. All of it is epidemiological in our inclusive meaning of epidemiological
research, though largely not in the now-common, restricted meaning (sect. 1.2).
As public-health science, when all of its chronic-disease implications are con-
sidered, the importance of this nutritional science is second only to the science
epidemiological laboratory science that provided for the ascent of man (Jacob
Bronowski) over the scourges of communicable-disease epidemics (sects. 1.1, 1.2).
(Chronic disease, while a term in common use by epidemiologists, is not apposite
as an antonym of communicable disease.)
In the face of all of this, the young student of epidemiological research would
do well, we suggest, sharing the optimistic view of the eminent physicist Richard
Feynman, whom Taubes quotes as having said that, whatever be ones bias, there
comes a time when the perpetual accumulation of [evidence] : : : cannot be
discarded any longer (p. 83).

Another eminent topic in dietary causation etiology/etiogenesis of atheromatotic

disease now has to do with paucity of omega-3 fatty acids in the diet. Taubes does
not deal with this, but the two Scientific Statements on this topic by the American
Heart Association are of note.
In the first one of these two Statements Stone JF, Circulation 1996; 94: 2337-
40 the up-front content is this: Reducing intake of saturated fat and dietary
cholesterol and avoiding excess calories, which can lead to obesity, remain the
18 2 Epidemiological Knowledge: Examples

cornerstone of the dietary approach to decreasing risk of atherosclerotic vascular

disease [cf. above]. During the past 20 years, however, there has been renewed
interest in other dietary components that might favorably improve [sic] lipid profiles
and reduce risk of coronary heart disease (CHD). Fish and fish oil, rich sources of
omega-3 fatty acids, have sparked intense interest in both epidemiological studies,
which suggest a favorable effect on CHD, and metabolic ward studies, which
show a striking improvement in lipid profiles in hyperlipidemic patients. Confusion
has resulted from clinical trials of fish oil in patients with CHD, which did not
corroborate early observational findings, and newer results, which suggest clinical
benefit due to a mechanism independent of lipid effects.
The more recent, second Statement Krist-Etherton PM et alii, Circulation 2002;
106: 2747-57 has a summary which opens thus: Omega-3 fatty acids have been
shown in epidemiological and clinical trials to reduce the incidence of CVD. Large-
scale epidemiological studies suggest that individuals at risk for CHD benefit from
the consumption of plant- and marine-derived omega-3 fatty acids, although the
ideal intakes presently are unclear. Said in the body of this Statement is that in a
recent meta-analysis of the results of 11 randomized trials with 7951 patients in
the intervention groups, : : : the risk ratio of nonfatal MI was 0.8, for fatal MI it was
0.7, and for sudden death (in 5 trials) it was 0.7 as though these numbers had
meaning without specificity in respect to the actual contrast and its duration, and the
domain of this.
So, while popular interest in this nutritional topic also is broadly-based and
keen, available from epidemiological research on it seems to be conflictual ideas
and meaningless empirical numbers rather than secure knowledge derived from
meaningful evidence.

2.2 Nutrition and Oncogenesis

Taubes (ref. in sect. 2.1 above) makes a quite compelling case for cancer, too,
being a disease of civilization, emerging in all essence only with the advent of the
civilized diet low in fats and high in carbohydrates as a feature of agricultural
societies, up to 10,000 years already.
He describes as a piece of seminal work a 120-page report by Richard Doll
and Richard Peto, in 1981, that reviewed the existing evidence on changes in
cancer incidence over time, changes upon migration from one region of the world to
another, and differences in cancer rates between communities and nations (p. 209).
They concluded, Taubes says, that diet was causing 35 percent of all cancers,
though the uncertainties they considered to be so vast that the number could be
as low as 10 percent or as high as 70 percent (pp. 209-10).
By the end of the 1990s, clinical trials and large-scale prospective studies
had demonstrated that the dietary fat and fiber hypotheses of cancer were almost
assuredly wrong, and similar investigations had repeatedly failed to confirm that red
meat played any role (p. 211).
2.2 Nutrition and Oncogenesis 19

The saccharine disease or metabolic syndrome concept having been advanced

decades earlier (cf. sect. 2.1 above), Taubes points out that Richard Doll and Bruce
Armstrong had found sugar intake in international comparisons to be positively
correlated with both the incidence and mortality from colon, rectal, breast, ovarian,
prostate, kidney, nervous-system, and testicular cancer, and that other investigators
have produced similar findings (pp. 211-2).
Taubes proceeds to an extensive and deep examination of the role, in oncogenesis
(Gr. oncos, mass, bulk), of two hormones whose blood levels are closely connected
to carbohydrates in the diet, namely insulin and insulin-like growth factor (IGF).
That these hormones play a substantial role in both the genesis and growth of
cancers, and that therefore dietary carbohydrates do, he makes quite plausible even
if these insights, too, tend to be pushed aside (cf. sect. 2.1 above) by authorities on
nutrition and (public) health.
Taubes refers twice to the 700-page report Food, Nutrition and Prevention of
Cancer, published in 1997 by the World Cancer Research Fund together with the
American Institute for Cancer Research. He writes that the assembled experts could
find neither convincing nor even probable reason to believe that fat-rich diets
increased the risk of cancer (p. 74). And as for carbohydrates, he quotes (p. 118)
from the report this:
The degree to which starch is refined in diets, particularly when the intake of starch is high,
may itself be an important factor in cancer risk, as may be the volume of refined starches
and sugars in diets. Epidemiological studies have not, however, generally distinguished
between degrees of refining or processing of starches, and there are, as yet, no reliable
epidemiological data on the effects of refining on cancer risk.

The Cancer Prevention Overview (PDQ ) of the U.S. National Cancer Institute,
the Health Professional Version of this (consulted in December 2011, last modified
04/29/2011), gives the following synopsis of that massive WCRF/AICR review:
With respect to dietary factors that may protect against cancer, the greatest consistency
was seen for fruits and non-starchy vegetables. In the WCRF/AICR report, conclusions
were reached that both fruits and non-starchy vegetables were associated with probable
decreased risk of lung cancer. Thus, even for the two dietary exposure classes that the
current evidence suggests may have the greatest cancer prevention potential, the evidence
was judged to be less than convincing and was applicable to only a few malignancies.

From this it may appear that the WCRF/AICR review had nothing to say about
cancer prevention in accordance with either one of the two hypotheses addressed
by Taubes not by avoidance of fats nor by avoidance of carbohydrates, in the
diet. These topics are, however, among the 61 (sic) items of food and drink
the WCRT/AICR report addressed in terms of total fat and foods containing
sugars, and neither one of these two did the WCRF/AICR experts judge to confer
convincing increased risk or probably increased risk for any of the 13 cancers
(by site; table in the foldout; cf. Taubes quote from the review above). Those
experts judgements were based on review of studies on humans, non-experimental
(cohort and case-control studies) as well as experimental.
20 2 Epidemiological Knowledge: Examples

Based on its own reviews, the NCI places no dietary factors under Risk factors
causally associated with cancer nor under Risk factors with uncertain associations
with cancer. It says only this:
Observational epidemiologic studies (case-control and cohort studies) have suggested
associations between diet and cancer development, but randomized trials of interventions
provided little or no support. : : : Ecologic, cohort, and case-control studies found an asso-
ciation between fat and red meat intake and colon cancer risk, but a randomized controlled
trial of low-fat diet in postmenopausal women showed no reduction in colon cancer. The
low-fat diet did not affect all-cancer mortality, overall mortality, or cardiovascular disease.

Evidently, the NCI places more reliance on randomized trials than ecologic,
cohort, and case-control studies. But it leaves open the possibility that the trials
have been about relatively short-term effects, which is not what the fat and
carbohydrate hypotheses are about.

Apart from fats and carbohydrates in the macro-composition of diet, an eminent

topic in respect to prevention/retardation of oncogenesis has been micronutrients in
the diet, in a particular theoretical framework.
As Taubes explains, a normal feature of metabolism is the production of : : :
oxidants, [which] oxidize other molecules : : : The object of oxidation slowly
deteriorates. : : : Antioxidants neutralize [oxidants], which is why antioxidants have
become a popular buzzword among nutritionists (p. 191; italics in the original).
Part of that deterioration occurs in the genome, as a matter of mutations possibly
leading to turned-on oncogenes and turned-off tumor-supressor genes, which jointly
constitute the deepest essence of cancer. Retardation of this genetic deterioration
is the direct aim of recommending foods rich in antioxidants, and antioxidant
supplementation of diet, in preventive oncology. The extent to which dietary
antioxidants or their supplementations actually serve to slow down oncogenesis, or
to delay/prevent overt cancer, Taubes does not discuss, as his interest focuses on the
fact that increasing level of blood sugar due to dietary carbohydrates increases
the production of those oxidants.
The WCRF/AICR report addresses mainly foods containing antioxidants
folate, carotenoids, beta-carotene, lycopene, vitamin C, selenium, pyridoxine,
vitamin E, and quercetin. It associates the judgement of convincing or probably
decreased risk of cancer with foods containing folate (oesophagus, pancreas, and
colorectum), carotenoids (mouth, pharynx, larynx, and lung), beta-carotene (oe-
sophagus), vitamin C (oesophagus), selenium (lung, stomach, colorectum, prostate),
pyridoxine (oesophagus), vitamin E (oesophagus, prostate), and quercetin (lung).
That is, it associates at least probable decrease in risk of some cancer or cancers with
each of the foods containing antioxidants, except the dietary sources of lycopene.
In respect to antioxidant supplementation of diet this report addresses beta-
carotene, selenium, retinol, and alpha-tocopherol. It associates convincing or proba-
ble increase (sic) of risk with beta-carotene (lung), neither convincing nor probable
decrease of risk with retinol (any cancer), and either convincing or probable
decrease of risk with selenium (prostate, lung, colorectal) and alpha-tocopherol
(prostate).
2.3 Nutrition and Ageing 21

The NCI site addresses vitamin and dietary supplement use, under the rubric
of Interventions. The antioxidants among these have no representation at all
under those with proven benefits. Under the other rubric, those with no proven
benefits, the NCI paints the big picture thus:
Some have advocated vitamin and mineral supplements for cancer prevention. Many
different pathways for anticancer effects have been invoked. A commonly tested hypothesis
is that antioxidant vitamins may protect against cancer, based on the premise that oxidative
damage to DNA leads to cancer. However, the evidence is insufficient to support the use of
multivitamin and mineral supplements or single vitamins or minerals to prevent cancer [ref.]
: : : Research into the potential anticancer properties of vitamin and mineral supplements is
ongoing, and the results continue to reinforce the lack of efficacy of vitamin supplements
in preventing cancer. Results from several large-scale randomized trials were published in
2009. [The supplementation] : : : did not reduce : : : had no benefit : : : was not efficacious
: : : was not efficacious : : : [produced] an increase : : :
But again, were those randomized trials addressing the much-delayed effect of
reduced DNA damage, in early life, on occurrence of overt cancer in middle age or
later? See sections 7.1 and 7.2.

2.3 Nutrition and Ageing

For reasons of symmetry, as the saying goes, we could have invoked the neologism
presbogenesis a` la atherogenesis and oncogenesis to denote the process of
biological ageing (of tissues; Gr. presbus, old man).
From the vantage of nutrition and health, the ultimate concern is not the
occurrence nor the nature of this process but the rate of it, its rate in causal relation to
nutritional determinants of it, micronutrient supplementation of diet included. The
slower is this rate, the longer tends to be (by a reasonable presumption) the duration
of life, including life free of (overt) cardiovascular disease from atheromatosis
and free of (overt, adult-onset) cancer. These degenerative diseases (along with
arthrosis and dementia, i.a.) can be viewed as largely being manifestations of
relatively advanced stages of this (cumulative) presbogenesis.
Taubes (ref. in sect. 2.1) reviews the spectacular benefits of semi-starvation on
the health and longevity of laboratory animals (p. 218 ff.). The calorie-restricted
animals live longer because of some metabolic or hormonal consequences of semi-
starvation, not because they are necessarily leaner or lighter (p. 219). And, back to
his leit motif, he writes (pp. 219-20) this:
The characteristics that all these long-lived organisms seem to share definitively are reduced
insulin resistance, and abnormally low levels of blood sugar, insulin, and insulin-like growth
factor. As a result, the current thinking is that a lifelong reduction in blood sugar, insulin,
and IGF bestows a longer and healthier life. The reduction in blood sugar also leads to
reduced oxidative stress : : : [Italics ours.]

Remarkably, Taubes makes no reference to population-level epidemiological

research on human nutrition and ageing, despite the great interest of people in this
topic. It seems that this research has not yet really begun (cf. sect. 1.2). Alchemy,
22 2 Epidemiological Knowledge: Examples

by contrast, represented a very determined long-term effort to develop an antidote

to human ageing. As Jacob Bronowski explains in his The Ascent of Man (1972),
alchemists search to make gold and to find the elixir of life are one and the same
endeavor. : : : every medicine to fight old age contained gold : : : and the alchemists
urged their patrons to drink from gold cups to prolong life (p. 138).

2.4 Nutrition and Obesity

Peoples concern to preserve their health is most pronounced in such truly modern
countries as the U.S.; and, as weve noted, this concern translates into preoccupation
with diet first and foremost. In these countries, peoples main dietary health-concern
now most commonly is prevention and/or control of obesity.
Given the major epidemic of obesity that has emerged in the U.S. and elsewhere
in the last few decades, public-health officials, too, now view obesity as the
main diet-related health problem in countries such as the U.S. Their concern is
accentuated by the common conception of obesity as a causal precursor of diabetes
and cardiovascular disease.
In the context of this major epidemic there has been no commensurate surge in
research into the etiology/etiogenesis of obesity barogenesis could be the term for
weight gain (Gr. baros, weight) as the causation of weight gain has been taken
to be obvious and well known, on a-priori grounds. Gary Taubes (ref. in sect. 2.1)
addresses this situation quite critically.
Everyone concerned with this matter, Taubes included, is committed to the first
law of thermodynamics that of the conservation of energy as it bears on the
storage of energy (as fat) in adipose (fat) tissues: If over a given interval of time
there is a given change in the amount of energy a person has stored as fat, this
is associated with a corresponding, energy-equivalent cumulative difference, over
that interval of time, between the persons energy intake (from diet) and energy
expenditure (in physical activity, i.a.) and conversely, of course.
The ordinary reading of this equivalence focuses on the idea quite correct
that any induced imbalance/difference in the rates of energy intake and energy
expenditure, cumulatively over an interval of time, produces its energy-equivalent
change in the amount of adipose tissue. And from this flows, first, the idea that to
avoid weight gain, the need is to see to it that the intake of energy does not exceed
the expenditure of it, and, second, the idea that for weight reduction the need is to
decrease the (weekly or so) quantity of ones energy intake and/or to increase that
of energy expenditure (by increasing physical activity).

Taubes makes a strong case for believing, or at least seriously entertaining, an

alternative idea: By suitably changing the quality of the intake of energy, the
consequence is a metabolic change away from the propensity for excessive storage
of energy in adipose tissues even without any restriction in the quantity of the
intake of energy. The qualitative change at issue is, again, lesser consumption of
2.4 Nutrition and Obesity 23

carbohydrates, refined ones and sugars in particular, naturally compensated for by

greater consumption of fats and/or proteins (even ad libitum).
Taubes describes the extensive research in direct support of this idea, and how
even the mechanism of the effect is understandable: Consumption of carbohydrates
stimulates the pancreatic excretion of insulin, which promotes the entry of fatty
acids derived from carbohydrates in the liver into adipose tissues; and insulin
also inhibits their release from this storage.
As a matter of practicalities, sustained weight reduction by means of restricted
intake of energy is generally unattainable because of inability to adhere to such a
diet, while avoidance of carbohydrates can be maintained as a feature of lifestyle;
and exercise (physical activity) too is overadvertised: it increases (appetite and
its consequent) energy intake in proportion to the energy expenditure (quite minor)
in it. Taubes reviews the studies behind these understandings.

Taubes relates the appearance, in 1976, of Theodore Cooper, then U.S. Assistant
Secretary of Health, as an expert witness before U.S. Senator George McGoverns
committee which was developing its Dietary Goals for the United States. Cooper
first testified that the consumption of high carbohydrate sources with the induction
of obesity constitutes a very serious public health problem in the underprivileged
and economically disadvantaged (p. 404).
But in response to the Senators looking for a rule of thumb, Cooper reversed
that first statement: I think in order to have an effective reduction in weight and
a realignment of our composition we have to focus on fat intake (p. 406). Taubes
sees this as a turning point in both public policy and research on obesity:
The shift in the nutritional wisdom was now taking place, driven by the
contagious effect of Ancel Keyss dietary-fat/heart-disease hypothesis on the closely
related field of obesity. Any diet that allowed liberal fat consumption was to be
considered unhealthy (p. 410). The small contingent of influential nutritionists
from Fred Stares department at Harvard provide an example of how this process
of entrenchment evolved. : : : (p. 411). The result is an enormous enterprise
dedicated in theory to determining the relationship between diet, obesity, and
disease, while dedicated in practice to convincing everyone involved, and the lay
public, most of all, that the answers are already known and always have been
an enterprise, in other words, that purports to be a science and yet functions like
a religion (pp. 451-2).

In closing here, an instructive vignette from earlier history. In his Bismarck (2011),
Jonathan Steinberg describes one of that great statesmans sons, suffering from
obesity, as having been helped to lose weight by following instructions from his
doctor, Ernst Schweninger (p. 413). And then, with the Reich Chancellor himself,
in 1883, again desperately sick, this son of his brought Schweninger to see him (p.
414) also obese, and famously gluttonous to boot.
Schweninger saved Bismarcks life by : : : [prescribing] a new diet for the
entire family tea or milk with eggs for breakfast, a little fish and roast meat
(no vegetables) at noon, a small jug of milk at 4:00 and yet another in the evening
24 2 Epidemiological Knowledge: Examples

(p. 416). The pains, the facial neuralgia, and the headaches vanished. Bismarck
was able to ride again. His weight began to go down, : : : From 1886 on [till his
death in 1898] he never went above 227 pounds, a perfectly reasonable weight for a
man six feet four (p. 415).
Bismarck retained Schweninger as his personal physician for the rest of his
life, and in gratitude Bismarck imposed his House Doctor on the Berlin medical
faculty, which regarded him as a charlatan and refused to talk with him (p. 416).

2.5 Nutrition and Hypertension

In public education about healthy diet, a very eminent recent governmental message
in Canada has emphasized the need to change the typical salt (sodium chloride)
consumption from the prevailing mean of over 8 g/day to less than 4 g/d for
persons of all ages and with even lower target levels for those under 10 or over
50 years of age. In this aspect of diet the proximal concern is the level of blood
pressure: prevention or amelioration of hypertension. (This term is a common
medical misnomer for relatively high blood pressure, which might better be termed
hemohyperbarism; Gr. haima, blood, Gr. baros, weight, pressure). Hypertension,
in turn, is of concern as a major cause of vascular disease (cardiac and cerebral).
Taubes (ref. in sect. 2.1) makes the point that Systematic reviews of the evidence
: : : have inevitably concluded that significant reductions in salt consumption
cutting our salt intake in half, for instance, which is difficult to accomplish in the
real world will drop blood pressure by perhaps 4 to 5 mm Hg in hypertensives
and 2 mm Hg in the rest of us. : : : So cutting our salt intake in half : : : makes
[in proportional terms] little difference (p. 146). Available not yet to Taubes
but to the Sodium Working Group as background for its 2010 recommendations
to Health Canada also was A comprehensive review on salt and health and
current experience of worldwide salt reduction programmes by F.J. He and G.A.
MacGregor, published in Journal of Human Hypertension 2009; 23: 363-84.
Until the recent salt hypothesis began receiving serious attention in the 1960s,
Taubes explains, the investigators paid little attention to nutritional explanations
for the rise in blood pressure that accompanied Western diets and lifestyles. : : :
Of course, the same societies that ate little or no salt ate little or no sugar or white
flour, so the evidence supported both hypotheses, although the investigators were
interested only in one (p. 146).
The laboratory evidence that carbohydrate-rich diets can cause the body to
retain water and to raise blood pressure, just as salt consumption is supposed
to do, dates back over a century. : : : In the late 1950s, a new generation of
investigators rediscovered the phenomenon, : : : By the early 1970s, researchers
had demonstrated that the water-retaining effect of carbohydrates was due to the
insulin secreted, which in turn induced the kidneys to reabsorb sodium rather than
excrete it, and that insulin levels were indeed higher, on average, in hypertensives
than in normal individuals. (Pp. 148-9).
2.6 Epidemiologic Reviews in Review 25

None of this accompanies the now-typical teaching about dietary sodium in

Canada (or elsewhere). Similarly, the adverse effects of severely restricted intake
of salt are left out of this health education.

2.6 Epidemiologic Reviews in Review

Epidemiologic Reviews, by its self-description, is a yearly journal published

by Oxford University Press : : : for the Johns Hopkins Bloomberg School of
Public Health and is sponsored by the Society of Epidemiologic Research and the
International Epidemiological Association. It is a journal in the usual meaning that
authors submit their articles to it. A sister publication of the American Journal of
Epidemiology, it is devoted to publishing comprehensive and critical reviews on
specific themes once a year.
This journal thus is a source from which to learn, on superficial review of
it already, what the topical concerns in the advancement of epidemiological
knowledge in recent years have been, while the respective comprehensive and
critical reviews reveal the corresponding states of the attained evidence and, one
hopes, knowledge.
The yearly specific themes in the 20012011 period were these:
2001 Prostate cancer
2002 Clinical trials
2003 Injury prevention and control
2004 Social epidemiology
2005 Epidemiologic approaches to disasters
2006 Vaccines and public health
2007 The obesity epidemic
2008 The epidemiology of mental disorders
2009 Epidemiologic approaches to health disparities
2010 Epidemiologic approaches to global health
2011 Screening
Some observations on select ones of these follow.

Given our fundamental tenet that epidemiology is (practice of) community medicine
(sect. 1.1), we first examine the 2005 issue, concerned with epidemiological
approaches to disasters, to determine whether population-level medical care for
disaster-afflicted populations was viewed as being epidemiology. Of main interest
in this respect is the Editorial in that issue, by the eminent epidemiologist Michel
Ibrahim.
As evidenced in this issue, Ibrahim wrote, epidemiology has an important
role to play in the study of characteristics of populations affected by disasters and
the deaths and injuries that follow. A seminal article : : : shows epidemiologic ap-
proaches to disaster assessment at its best. : : : As presented in several papers in this
26 2 Epidemiological Knowledge: Examples

issue, epidemiologic methods and public health strategies are important in assessing
and containing disasters. : : : The time-honored prevention and control measures
come into play: : : : vaccinating high-risk groups against infectious diseases : : :
The final paragraph is particularly telling: Now more than ever, public health
preparedness has become a high priority of governments that must have the ability
to predict, prevent, and control disasters and their consequences. Adequate funds are
crucial to achieve this purpose, but epidemiologic know-how also has a role to play.
Now, as a disaster an earthquake with or without a tsunami, say produces
public-health problems in terms of epidemics of injuries and other illnesses, it is, as
Ibrahim said, the responsibility, specifically, of the governments public-health offi-
cials to predict, prevent, and control these epidemics. The leading officials in this
naturally are (practicing) epidemiologists. As remains commonplace, Ibrahim wrote
about public-health practices when at issue were, specifically, the epidemiological
varieties of these (cf. sect. 1.1, i.a.). He wrote about epidemiologists in the practice
of community medicine in the practice of epidemiology, that is (cf. sect. 1.1).

Related to the essence of epidemiology, the point that social epidemiology now is
on the ascendancy came up in section 1.4. And it is of note that in the period from
2001 to 2011, the 2004 issue was expressly devoted to this, while related to this also
was the 2009 issue. Suffice it here to quote from the Introduction to the 2004 issue,
by Lisa Berkman:
Social epidemiology, disguised in other forms and known by other names, has been with
us for decades, if not centuries. : : : the field is not a new one, nor are epidemiologists
the only scientists contributing to a deepening understanding of the social determinants of
health. : : : I would argue that the important insights of social epidemiologists are based on
the training they have in the assessment of health, disease, and biological pathways and the
capacity to integrate upstream social dynamics into their modeling of disease causation.
[Italics ours.]

A fuller account of Berkmans and others conception of social epidemiology

can be found in Social Epidemiology (2000), of which Berkman and I. Kawachi are
the Editors.
Of added note, a propos, eminently is Epidemiology and the Peoples Health
(2011) by Nancy Krieger. She writes (p. 163) that,
Among the extant social epidemiologic perspectives, I would argue three distinct theoretical
trends exist. : : : Of these, the first, sociopolitical, focuses principally on power, politics,
economics, and rights as key societal determinants of health; the second, psychosocial,
emphasizes psychosocially-mediated social determinants of population health. : : : The
third trend : : : is more nascent and is best represented by ecosocial theory, which builds
on and extends these first two frameworks by solving both the embodied population
distributions of disease and health and epidemiologic theories of disease distribution, each
in relation to their societal, ecological, and historical context. [Italics in the original.]

Given what social epidemiology is said to be about, it is unsurprising that it

is not an outgrowth of epidemiology in the research meaning of epidemiology.
It arose from sociology, and only quite recently (cf. Berkman above). As Krieger
explains (p. 165),
2.6 Epidemiologic Reviews in Review 27

By 1969, enough familiarity with the field exists that Leo G. Reeder presents a major ad-
dress to the American Sociological Association called Social Epidemiology: An appraisal
[ref.]. Defining social epidemiology as the study of the role of social factors in the etiology
of disease [ref.], he asserts that social epidemiology : : : seeks to expand the scope of
investigation to include variables and concepts drawn from a theory [ref.] in effect, calling
for a marriage of sociological frameworks to epidemiologic inquiry. [Italics in the original.]

Recall, a propos, from section 1.4 a counsel of Francis Bacon: Read not
to contradict, nor to believe, but to weigh and consider. Something to here
weigh and consider in particular is this: Given that epidemiological research
without the social adjective addresses population occurrence of illness in relation
to characterizers of the populations individual members their constitutional,
environmental, and behavioral characterizers as a matter of biomedical research,
isnt it natural to think of study of the distribution of these determinants in
(human) populations/communities/societies as social and behavioral rather than
biomedical research, as a supplement to, rather than a type of, epidemiological
research? (Cf. sect. 1.4.)
And here is something more specific to weigh and consider. As epidemiologists
by their biomedical research have come to know how rates of various illnesses in
human populations are determined, causally, by the pattern of histories of, say,
smoking in the population, social epidemiologists have nothing to add to this,
whatever may be their claims about the training they have in the assessment
of health, disease, and biological pathways. By the same token, epidemiological
researchers of the established, biomedical variety are not experts on the patterns of
smoking in human populations and, especially, on how best to reduce these. They
defer to social epidemiologists a misnomer for medical sociologists on this,
and, ultimately, to policy-promulgating societal authorities.
And finally here, while social epidemiology has been with us for decades
(Berkman, above), it remains inchoate: the order of the day evidently still is
conceptualization of what it is, rather than review of its accomplishments in the
advancement of the knowledge-base of community-level preventive medicine.

Epidemiological research and knowledge typically focus on a particular illness. It

thus is instructive to consider the 2001 issue, focusing on prostate cancer. Following
a brief section on descriptive epidemiology and natural history, most of that issue,
by far, was about factors in the causal origin etiology/etiogenesis of the illness:
genetic factors, hormones and growth factors, and environmental factors
with diet classified as an environmental factor. There was, however, a short section
also on prevention and control, addressing chemoprevention and screening.
A single article addressed descriptive epidemiology, for which the term now is
ecological epidemiology: In this review, the authors said, we present patterns
and trends of prostate cancer in various countries to provide further insights for
future epidemiologic studies. A propos, we remind the student that studies of
this type have had a major role in nutritional epidemiology (sects. 2.1, 2.2).
And we add that it has been commonplace to distinguish between descriptive
and analytic epidemiology, with the idea that studies of the former type serve
28 2 Epidemiological Knowledge: Examples

hypothesis generation (regarding etiogenesis), while hypothesis testing requires

studies of the latter type, defined by the units of observation being individuals
(rather than populations). Epidemiology in the research-discipline meaning of the
term (sect. 1.3) focuses on analytic studies (which term is as gross a misnomer
in epidemiological research its logic being synthetic as its semi-cognate data
analysis meaning data synthesis is in statistics).
There was a single article on natural history of prostate cancer but it was
not about the natural (untreated) course of the disease (for which the term is,
in medicine, a commonly-used misnomer). Instead, surprisingly, The intention
of this brief review is to present some of the most pertinent facts regarding the
pathobiology of prostate and its malignancies, and thereby facilitate the future work
of epidemiologists studying this important disease. We very much agree with
this intention: even if the conception of epidemiological research be limited to
population-level occurrence research (sect. 1.2), knowledge about the anatomic
site, pathology, and clinical aspects of the illness at issue may well facilitate the
research; and it generally is essential for its scientific quality assurance.

A propos of section 2.2 above, it is of interest that one article did address fat intake in
the etiogenesis of prostate cancer, and that none of them dealt with the etiogenetic
role of the carbohydrate composition of diet. Illustrative of that particular review
is this: Data from several ecologic studies have shown positive correlations (r
 0.6) between per capita intake of total, saturated, or animal fat and prostate
cancer incidence or mortality [refs.]. : : : Although several case-control studies
found positive associations (odds ratio (OR)  1.3) between total fat intake and
the risk of prostate cancer [refs.], only slightly fewer failed to find this relation
[refs.]. : : : Because cohort studies of diet and cancer avoid the problem of recall
bias and are less prone than case-control studies to other [sic] selection biases, they
are generally given more weight in overall assessments of the literature. Three such
studies : : : examined total fat and prostate cancer; two of these studies [refs.] found
a positive association (relative risk  1. 3).
So, with descriptive/ecologic studies again the point of departure, the empha-
sis was on the now-perceived two principal types of analytic study: case-control
and cohort studies, the former producing findings in terms of odds ratio, the
latter in terms of relative risk and both with no indication of what contrast (based
on the quantitative determinant) was at issue. Review of the findings of a given one
of the latter two types of study were taken to be a matter of noting what proportion
of the studies found a positive association and what proportion failed to find this
relation. And the studies none of them of basic science were seen to be marred
by potential biases.
All of this is, at present, rather typical of epidemiological concepts, terminology,
thinking, and writing; and we find it to be, in many ways, disagreeable. For a
student these excerpts from this review are indicative of what now is out there, even
if it remains far from being understood correctly interpreted and evaluated. We
present our critical understandings in the ensuing chapters, and we suggest that the
2.6 Epidemiologic Reviews in Review 29

student return to these passages for critical examination of them after having studied
chapter 6 at least.
This review of dietary fat in the etiogenesis of prostate cancer was placed
under the rubric of environmental factors as distinct from genetic factors and
hormones and growth factors, but we think in terms of a triad of categories for
illness-causing factors: constitutional, environmental, and behavioral. All somatic
factors, whether congenital (e.g., genetic) or acquired (post-natally), we think
of as constitutional (with hormones and growth factors among these). And we
decisively separate behavioral factors active ones of lifestyle, such as dietary
habits from factors that are, passively, environmental (from the vantage of a given
person).
The last, brief section in this issue of the Reviews is noteworthy by its title:
Prevention and control. So, prevention and control of the occurrence of prostate
cancer these matters of practice apparently were taken to be matters of
epidemiology (rather than extra-epidemiological practice of public health). But this
is not very clear from the three reviews under this rubric, one of them addressing
chemoprevention of prostate cancer, the other two screening for it. However, one
of the two screening reviews clearly went to matters beyond research, to matters of
practice of screening:
In summary, patients must be informed about the risks and benefits of prostate cancer
screening before it is carried out. : : : If screening (and eventual treatment) is to be offered
to asymptomatic patients, it should be offered to those in whom age and health status are
such that they may benefit from early detection of a disease which may have a protracted
natural history [meaning course in the absence of treatment; cf. above].

The current, major epidemic of obesity, in the U.S. and elsewhere (sect. 2.4), makes
the 2007 issue both timely and interesting.
The lead article, by B. Caballero, provides an introduction to this issue of the
Reviews. It pointed out that,
In spite of extensive research over the past decades, the mechanisms by which people attain
excessive body weight and adiposity are still only partially understood : : : As discussed
in this overview, there is an increasing consensus among obesity experts that changing the
obesogenic environment is a critical step toward reducing obesity. Reversing the factors
: : : that lead to increased caloric consumption and reduced physical activity would require
major changes in urban planning, transportation, public safety, and food production and
marketing. [Italics ours.]

Caballeros overview is illustrative of the now-common thinking about diet as

bearing on obesity through the quantity of its energy content, with no thought about
the role of the quality of the diet in causing the metabolism to become obesogenic
at whatever level (quantitative) of energy intake (cf. sect. 2.4). Curiously, diet and
physical activity were here seen, a` la obesity experts, to be environmentally
determined aspects of behavior.
Of the other articles in this issue of the Reviews, of particular note, a propos
of the sections above, is that by R. Kelishadi: In a systematic review : : : of the
literature from 1950-2007, [he] compared data from surveys on the prevalence
30 2 Epidemiological Knowledge: Examples

of overweight, obesity, and the metabolic syndrome among children living in

developing countries. : : : Time trends in childhood obesity and its metabolic
consequences : : : should be monitored in developing countries in order to obtain
useful insights for primordial and primary prevention of the upcoming chronic
disease epidemic in such communities.
The definition of the metabolic syndrome in childhood and adolescence is
variable, an example of the definitions being that by Lambert et alii in a Canadian
study, as presented by Kelishadi: Hyperinsulinemia combined with 2 risk factors
including overweight, high systolic blood pressure (>90th percentile), impaired
fasting blood glucose, high triglycerides (>75th percentile), and low HDL-c (<25th
percentile). In this, non-metabolic elements are intermixed with metabolic ones.
Kelishadi made these remarks about actual risk factors of the metabolic syn-
drome:
unhealthy dietary habits, particularly high carbohydrate and fat intakes, are associated
with the metabolic syndrome in adults [ref.]. : : : we found that the risk of the metabolic
syndrome among children and adolescents rose with the consumption of solid hydrogenated
fat and whole-flour bread [ref.]. While the frequency of intake of sweets (candies) increased
the risk of the metabolic syndrome in both sexes, the frequency of eating fast foods and the
frequency of eating carbohydrates increased the risk in boys and girls, respectively. In both
sexes, the frequency of consumption of fruits and vegetables, as well as dairy products,
decreased the risk of having the metabolic syndrome [ref.].

In this he repeatedly makes the common mistake of equating the finding of an

outcome-antecedent association with documentation of an effect-cause relation, as
though the latter were a phenomenon, subject to observation: X increased the risk
of Y : : : See section 4.2.
In the Conclusion, very notably, there was nothing about dietary carbohydrates
in the etiogenesis of obesity and of its associated metabolic anomalies in the
metabolic syndrome. Instead, the review was said to have been about childhood
overweight and its metabolic consequences (italics ours). And as for the etiogenesis
of overweight/obesity, the conclusion was a simple affirmation of the prevailing
orthodoxy (sect. 2.4): Strategies aimed at reducing caloric intake and increasing
caloric expenditure through regular exercise, early and aggressively, are necessary
:::

The 2011 issue of the Reviews, on screening a topic in respect to prostate cancer
already (above) began with an Overview by R. Harris. Under the important
subheading of Where we are and where we may be heading he wrote this:
Overall, then, the present truths about screening are that it is sometimes effective in
improving health outcomes, that it is also associated with important harms, : : : With few
exceptions, the contribution of screening to improving the health of the public is small, : : :
Perhaps we should not think of screening as our primary prevention strategy but rather use
screening to make a real, but limited contribution to population health for a few conditions.
: : : Fletchers [ref.] thoughtful perspective on breast cancer screening is consistent with this
vision of the future of screening. : : :
2.7 Overview of the Achievements 31

S.W. Fletcher, whom Harris so put on a pedestal, wrote about Breast cancer
screening: A 35-year perspective. She made the familiar but ever-so-stunning
points that randomized trials of breast cancer screening have involved more than
650,000 women, and that even so, no other cancer screening has produced such
heated controversy. Multiple reviews have been published, : : : See Appendix 2.
Regarding the formulation of ideas and recommendations about screening for
a cancer, Fletcher described the pivotal roles that have been played by two task
forces, from the vantage of herself having been a founding member of both of
them. She gave no justification for the formation, membership, or functioning of
these task forces. See Apps. 1, 2, and 3.
Both the Canadian and U.S. task forces emphasized the importance of ran-
domized trials in their assessments of and recommendations for cancer screening
(italics ours). And illustrative of the results of these trials is this passage by Fletcher:
The most recent meta-analysis [by the U.S. Preventive Medicine Task Force, of the results
of the eight trials, having involved the total of over 650,000 participants, as noted above]
found that breast cancer mortality reduction among women invited to screening was 15%
for women aged 39-49 years, 14% for women aged 50-59 years, and 32% for women aged
60-69 years, with corresponding numbers needed to invite to screening to prevent 1 breast
cancer death of 1,904, 1,339, and 377, respectively [ref.].

While presenting these numbers concerning mortality reductions that

purportedly were found and purportedly implied with four-digit accuracy!
various numbers needed to invite to screening to prevent 1 breast cancer death
irrespective of the nature and duration of the screening! she had nothing to say
about them, thoughtful or otherwise. See Appendix 2.

2.7 Overview of the Achievements

The aim of epidemiological research should be understood to be advancement

of epidemiological practices in the improvement of population health in human
communities. We defined epidemiological research by this aim of it, regardless of
how basic the research is (sect. 1.2). And we outlined the enormous achievements
of epidemiological research, thus defined, in providing for the prevention and
control of epidemics of communicable diseases (sect. 1.1).

Given that the accomplishments of epidemiological research and practice have

been enormous in respect to communicable diseases and have more recently been
extended to population-level prevention of illnesses of whatever kind (sect. 1.1),
the question now is, specifically, about the accomplishments of the research in the
advancement of the relatively novel types of epidemiological practice. We sought a
partial answer to this question by focusing on nutrition as a determinant the health
of populations (through its effects on individuals), health as for various diseases
(incl. obesity) and longevity too (sects. 2.1, 2.2, 2.3, 2.4, 2.5); and we supplemented
this by a cursory review of Epidemiologic Reviews (sect. 2.6).
32 2 Epidemiological Knowledge: Examples

The alert student will have taken note of the subtitle of Taubes book (ref. in
sect. 2.1), the element of controversial science in it. Taubes quotes Henry Blackburn,
the eminent cardiovascular epidemiologist, as having written, in 1975, that two
strikingly polar attitudes persist on this subject [i.e., the etiology/etiogenesis of
coronary heart disease, as to whether it has to do with fats or carbohydrates in the
diet], with much talk from each and little listening between (p. 22).
At the root of the controversies are, as Taubes explains, two aberrations of
the culture surrounding epidemiological research: inattention to relevant basic
research and dogmatism, among the investigators and also the agencies sponsoring
the research.
One of us was an author in a paper indicating that alcohol consumption
indeed does prevent coronary heart disease. Puzzled why evidence of this had
not been published from the Framingham Heart Study, in the blue books of
which the association (negative) had been quite apparent, he asked one of the FHS
investigators about this. The answer was that this colleague himself actually wrote a
manuscript on it. But: its approval process in the sponsoring agency the National
Heart and Lung Institute of the U.S. went, quite exceptionally, all the way up to
the head of the Institute; and he blocked the articles publication on the ground that
knowledge of this salutary effect of alcohol consumption would promote alcoholism
among the American people.
Taubes gives several examples of censorship, doctrinaire investigators self-
censorship as well as censorship by their sponsoring agencies. Particularly in-
structive about the role of institutional dogmatism are his descriptions of how
institutional ideas specifically of what the truth about the matter at issue is play a
role in the selection of like-minded experts to committees with the charge to review
the evidence and to seek to determine what the truth is, thus making official the
conception of the truth that the agency already was holding.
In the framework of this culture surrounding the relevant scholarship, people
have been taught, for decades, to avoid fats in their diets; and so they have consumed
carbohydrates instead and have grown obese, quite possibly as a consequence of
this. Now, somewhat hesitantly, the reversal of this teaching is taking place; now
the idea increasingly is that to be avoided are those carbohydrates, the consequence
naturally being increase in the consumption of fats (and proteins too).
These matters should be seen to be at the very core of non-communicable,
chronic disease epidemiological research and knowledge. Yet the importance of
them is not manifest in the topics addressed in Epidemiologic Reviews in 2001-2011.
Moreover, these quite generally ignore the relevant basic science and have no
genuine knowledge to present. It took Taubes a free-lance science-writer with no
official credentials in epidemiological research to teach us about these matters,
with unprecedented range of and depth of insights in the coverage.
To be sure, epidemiological research of this more recent, chronic-disease genre
(the dichotomy communicable vs. non-communicable actually is nowhere near
coterminous with the acute-chronic dichotomy) has served to discover, and to an
extent quantify, the effects of a large number of health hazards, those of cigarette
smoking prime among these. But even in respect to this prime example, the history
2.7 Overview of the Achievements 33

of progress is less than glorious: Franz Muller, in his medical thesis in Germany in
1939, gave quite compelling evidence to the effect that the main cause, by far, of the
lung-cancer epidemic (which had been evolving for decades already) was smoking
of cigarettes; and very strict anti-smoking policies were adopted in Germany, in
part on the basis of this knowledge. Yet, as Robert Proctor in his The Nazi War
on Cancer (1999) points out, Wynder and Grahams famous paper from [1950]
characterized tobacco merely as a possible etiologic factor in the increase of the
disease (p. 196).
While Mullers study involved a mere 86 cases of death from lung cancer
and 86 living and healthy controls, 28 other case-control studies with a total
of 15,492 cases and 101,215 controls, and seven very large, expensive, and time-
consuming cohort studies one of them with a cohort of 1,085,000 persons were
conducted before it got to be a piece of official knowledge in the U.S. that smoking
causes lung cancer. This landmark event took place as late as in 1964, in the form
of the publication of the Surgeon Generals Smoking and Health report. To say that
which would go without saying, an enormous number of people have died of lung
cancer as a consequence of their smoking in the quarter-century before 1964.
Given that epidemiological research has as its raison detre the advancement of
community medicine, there should be continual transmutation of epidemiological
evidence into genuine epidemiological knowledge. The requisite innovations for
the advancement of epidemiological knowledge we discuss in section 13.6 and
in Appendix 3, while perfunctory programs to derive recommendations/guidelines
for practice from evidence, without attention to expert knowledge, we address in
Apps. 1 and 2.
While the Epidemiologic Reviews are supposed to be comprehensive and
critical, we find them to be strikingly uncritical loose in both concepts and
principles. The epitome of this lack of discipline (sect. 1.3) is the piece by an
eminent clinical epidemiologist a founding member of the Canadian and U.S.
task forces on preventive medicine on an enormous amount of research that
substantively has amounted only to heated controversy, this on a topic that in
critical conception is wholly extrinsic to preventive medicine! See Apps. 1 and 2.
Chapter 3
Etiology as a Pragmatic Concern

Abstract Epidemiological research is, almost exclusively, concerned with etiology

of illness; but the concept of this remains quite generally misunderstood. To wit,
according to A Dictionary of Epidemiology (2009) a handbook sponsored by
the International Epidemiological Association etiology is Literally, the science of
causes, causality; in common usage, cause. (Tautology is not, literally or otherwise,
the science of unnecessary repetition; nor is morphology, literally or otherwise, the
science of shape, even though an important aspect of many sciences.)
For a student in an introductory course on epidemiological research it is essential
to learn, for a start, that etiology we prefer to term it etiogenesis (in analogy
with pathogenesis for the closely related descriptive concept) is one of the two
fundamental concepts of causation in medicine. And then the need is to learn,
securely, the essence of this particular genre of causation in medical thought, to
learn to distinguish it from interventive causality.
Once the medical concept of etiology/etiogenesis has been correctly introduced
and internalized by the students, they are ready to be introduced to the role of
etiogenetic knowledge in (the practice of) community medicine in community
etiognosis, preparatory to any action (education, regulation, or service) aimed
at reduction of morbidity from some particular illness(es). In this etiognosis,
general/abstract knowledge about etiogenesis is brought to bear on ad-hoc facts,
to arrive at this particularistic type of knowing (about the causal origin of a given
rate of morbidity in the community/population being cared for).
Community etiognosis (about a morbidity rate) has its counterpart in clinical
medicine, focusing on a single case of an illness or sickness not due to illness and
thus differing in its particulars from community etiognosis. The requisite knowledge
is generally considerably more detailed/specific than in community etiognosis, and
hence distinctly less secure (in todays context at least).
Clinical etiognosis is quite commonly a concern in legal contexts as well, but
there it takes on a character different from what it is in clinical practice.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 35

DOI 10.1007/978-94-007-4537-7 3, Springer ScienceCBusiness Media Dordrecht 2012
36 3 Etiology as a Pragmatic Concern

3.1 The Two Genera of Causality in Medicine

Meteorological information merely descriptive of weather in a given place at a given

time, present or prospective, is of value to the people in that place at that time.
For, having such information people can make arrangements to adapt themselves
to the weather; and even if this isnt their concern, they may appreciate the
information out of sheer interest or curiosity. Of even greater, added value would
be causal information about weather, were weather commonly to be subject to
intentional change by human actions. In that case, information about existent
weather would call for information also about meteorologic actions or inactions
as causal explanations of it, and weather forecasts would be made conditional on
what actions might, or will, be taken and would thus be made relevant for decisions
about professional actions (prospective, communal) in the management of weather.
Like meteorological practice, medical practice too commonly is pursuit of
certain facts and insights; but different from meteorology, medicine has been, for
millennia already, much concerned with causality. In todays community medicine,
the doctor epidemiologist (sect. 1.1) may proceed from community diagnosis
about the existent level of morbidity from a particular illness in the cared-for
population to consideration of the causal origin etiology, etiogenesis of this.
And from morbidity diagnosis possibly supplemented by insight into the etiogenesis
of the level of the morbidity from community diagnosis possibly supplemented by
community etiognosis, that is (s)he proceeds to consider the future course of the
morbidity, aiming to arrive at prognosis about this and, in particular, at intervention-
prognosis to guide the decisions about the adoption of interventions. These two
generic types of causality in community medicine etiogenetic and interventive
have their obvious counterparts in clinical medicine, with presence/absence of a
particular illness in place of the level of morbidity from it.
These two types of causal concern in medicine are very different, one from the
other. Ad-hoc knowing about etiogenetic causality etiognosis, that is (cf. above)
is tantamount to having a causal explanation of an existent outcome (level of a
morbidity, or presence of an illness); it thus inherently is retrospective from the
vantage of an existent outcome. By contrast, ad-hoc knowing about interventive
causality is a decision-relevant input into prognosis; it is knowing about the causal
change in a future outcome resulting from an intervention, the adoption of which
(notionally or actually) is a given; it is prospective from the vantage of the decision
about adoption (or continuation) of the intervention.
Despite this profound difference, these two types of causality in medicine share
two important features. One of these is inherent in the very concept of cause.
A potentially etiogenetic history, and a potentially outcome-changing intervention
just the same, can be thought of as a potential cause only in the framework of
a defined causal contrast: the potential cause being present (retrospectively or
prospectively, according to which one of the two genera of causation is at issue)
in lieu of its defined alternative being present. For example, thus, a given level of
hypertension (systemic) is a potential cause of the occurrence of stroke not per
3.1 The Two Genera of Causality in Medicine 37

se but only if a lower level of blood pressure is taken to be the alternative (while
with a higher BP the alternative, the given BP is a potential preventive cause of
non-occurrence of stroke).
The other shared feature is this: Something that in general terms is a cause of the
outcome need not be so in a particular instance, even when present; in a particular
instance it is only a potential possible cause. Whether it in a given, individual
instance actually was, or will be, causal will generally remain unknowable. On
the individual, clinical level, only probabilistic knowing is realistic to pursue for
etiognosis as well as for intervention-prognosis knowing about the probability
(objective) that the potential cause actually was, or will be, causal to the outcomes
occurrence (its change from non-occurrence to actually occurring).
This said about the commonalities, another difference also is worth noting here.
It has to do with the causal contrast, specifically with the causal determinants index
category in the contrast, representing the presence of the cause, as distinct from
the reference category, representing the alternative to (rather than mere absence
of) the cause. An etiognostic potential cause is what it happens to be: a particular
pattern of the determinant status constitutional, environmental, or behavioral
over the entire span of etiogenetically relevant time (retrospective), deviating from
the corresponding, defined reference status throughout this period of time. In
intervention-prognostic causation, by contrast, the potential cause is what it is taken
to be, by choice, as for both its type and the duration of its implementation; it is
algorithmic for the duration of its implementation (notional or actually planned),
just as is its defined alternative.
Following these subtleties, it may be good to concentrate the mind by returning
attention to the difference profound between the etiognostic and intervention-
prognostic questions: In community medicine it is one thing to consider to what
extent the current level of morbidity from a particular illness is due to the
populations distribution by history in respect to a particular cause of it, this
distributions deviation from the reference distribution; and it is quite another thing
to consider what would be the prospective change in that morbidity if a particular
intervention were adopted (prospectively). The issue of the role of smoking in the
etiogenesis of the prevailing rate of lung-cancer incidence is profoundly different
from that of the change in this rate that would result from the adoption of a particular
regulation concerning smoking, including as for the attainability of gnosis about it.
A modality of a doctors action aimed at prevention of an illness in the client,
in clinical as well as community medicine, commonly is education of the client
(individual or population) about possible change in health-relevant lifestyle. This
is not actual intervention on the course of the clients health, not intervention by
the doctor nor by the client any more than, say, intention-to-treat based on
randomization is actual intervention in an intervention trial. In community medicine,
an actual intervention (preventive) is a change brought about by adopted public
policy into the peoples constitutions (as to, e.g., immunity), environments (e.g.,
pollution control), or behaviors (e.g., use of seatbelts). In clinical medicine an
intervention is a physician-effected change in the clients constitution (e.g., by
injection of a vaccine or a medication, or by tonsillectomy). Yet, even though
38 3 Etiology as a Pragmatic Concern

prevention-oriented education of the client (population or individual) isnt actual

intervention, it is justifiable for simplicity and convenience to subsume under
intervention-prognostic issues those that have to do with education about health-
relevant aspects of lifestyle.

3.2 Etiology as a Community-medicine Concern

Apart from surveys and surveillances directed to health hazards and general
promotion of healthy constitutions, environments, and behaviors in the cared-for
population, a community doctors concern is to get to know about the rates of
morbidity from various illnesses in the population that (s)he is caring for to achieve
community diagnosis in this multifaceted sense and then to devise specific ways
of reducing those rates, unduly high ones in particular.
Regarding a given one of those unduly high rates, the epidemiologists
development of a plan to reduce it begins with his/her answer to the question, Why
is this rate this high in this population at this time? What is causing this rate to be
this high? The beginning of the control of morbidity from a particular illness thus
is the answer or the set of answers to this question about etiology/etiogenesis,
to this specifically etiognostic question. The rate at issue must be either specific in
terms of its descriptive determinants or suitably adjusted for these, so that its high
level does not have a descriptive, acausal explanation but needs a causal one.
Etiogenetic explanation of the rate of morbidity from a particular illness in a
particular population at a particular time addresses the proportion of the rate that
is caused by the factor at issue (by the pattern of its retrospective presence, in
lieu of the defined alternative to this, in the members of the population). This is
the proportion of the existing rate that would not be there but for the populations
patterns of positive histories for the factor (ceteris paribus).
This explanation the perception of this overall etiogenetic proportion the
epidemiologist can derive from two types of input. One of these types of input
is, generally, peculiar to the cared-for population at the time in question. If the
etiogenetic history is simply binary, such as use/non-use or seatbelt at the time of
road accident, this first input is the proportion of the cases of the illness in which the
affected person has the history of the factor/cause (non-use of seatbelt, say) having
been present in the relevant past (i.e., in which the person had the index history and
the case thus could have been caused by the factor at issue). This is the proportion
of the cases of the illness and hence of the rate of morbidity from the illness
that would be attributable to the factor were its presence (retrospective) always
to be etiogenetic to cases of the illness occurring in association with its presence
(retrospective).
The other input into the overall etiogenetic proportion into the proportion now
purportedly (ref. below) commonly termed attributable fraction (population) can
be a matter of general epidemiological knowledge about cases of the illness that
are associated with the potentially etiogenetic antecedent. It is the proportion of
these antecedent-associated cases such that the antecedent actually is causal to the
3.3 Etiology as a Clinical-medicine Concern 39

outcome, to the illness occurring the factor-conditional etiogenetic proportion

now purportedly (ref. below) properly termed attributable fraction (exposed). The
overall etiogenetic proportion for the cause at issue, in the population at the time, is
the product of the two input proportions.
Reference: Porta M (Editor), Greenland S, Last JM (Associate Editors). A Dictionary of
Epidemiology. A Handbook sponsored by the I.E.A. 5th edition. Oxford: Oxford University
Press, 2008.
If the factor at issue is not of the all-or-none type but has, instead, more than a
single index category, this calculation needs to be carried out separately for each
of these. Then, the overall etiogenetic proportion is the sum of these component
proportions.
Once the community doctor has a perception of the proportion caused by a given
factor in the morbidity from a particular illness in the population (s)he is caring
for, (s)he thereby has a sense prognostic of the extent to which that morbidity
would be reduced if that factor were to be completely removed from the population;
that is, if the prevailing distribution of histories concerning the risk factor at issue
were to be replaced by their singularity representing the historys reference category.
A regulatory intervention can have the potential to completely eliminate the cause-
specific excess morbidity from the illness at issue. But the adoption of a program
of community-level health education, and the introduction of a prevention-oriented
communal service just the same, when directed to a particular cause of morbidity,
generally has a morbidity-reducing consequence that is an inestimable fraction of
the etiogenetic proportion for the factor at issue. The inestimability of that fraction
is mainly due to the unpredictability of the peoples behavioral responses to the
education about healthy lifestyles and of their actually making use of the preventive
service(s) made available.
Any community-level health education directed to prevention would ideally be
founded on knowledge that has evolved from etiognostic insights into expressly
prognostic ones. The knowledge would ideally be specific to particular risk profiles
of individuals in the population; and in reference to these it would address the
way the prospective risk of the illness depends on the individuals prospective
lifestyle in respect to the aspect of it that is at issue, or on the individuals use of
preventive services communal and/or clinical so as to maintain particular levels
of (patho)physiological risk factors for the illness. Presentation of this educational
content would be complex, but it would be quite feasible via the Internet. The
problem is the common unavailability, at present still, of such educational content.
(Cf. ch. 2.)

3.3 Etiology as a Clinical-medicine Concern

When a clinician has arrived at a high-probability diagnosis (rule-in dgn.) about

a particular illness, or strongly suspects a case of some sickness not to represent
manifestation of any illness (somatic anomaly) but to be an adverse reaction to some
40 3 Etiology as a Pragmatic Concern

exogenous agent (an ingested medication, say), (s)he is confronted with the clinical-
type etiognostic question: Why is this person having this illness at this time? or,
What agent is producing this adverse reaction in this person at this time? The former
type of question (s)he may be able to regard as inconsequential, as the particular
etiogenesis may not have implications for the management of the case at hand nor
for prevention of recurrence of the illness. The answer to the latter kind of question,
by contrast, is quite commonly consequential for the management of the case and
practically always for prevention of its recurrence.
When setting out to meet this challenge, the clinician in ideal practice at least
recalls the known causes of the outcome at hand, ascertains the patients history
in respect to each of these, and translates these histories into the corresponding
etiognostic probabilities. Relative to etiognosis about a particular cause in explain-
ing a particular rate of morbidity in a practice of community medicine, the clinical
situation is generally characterized by greater detail in the history about the potential
cause and also about the host of the outcome about various aspects of the general
etiogenetic triad (constitution/environment/behavior).

3.4 Etiology as a Legal Concern

In a court of law, the plaintiff in a class action suit may be alleging that something
industrial a particular practice or environment or product is causal to a particular
illness experienced by people exposed to it; and in particular, that this is so in each
case of the illness in a particular class of exposed people in general and, therefore,
in each case among the members of the plaintiff class of people. A case may also
be brought to a court of law by an individual against a company or against his/her
own doctor, alleging damage to his/her own health from an industrial environment
or product or from a medical action.
The legal issue can be (as in the U.S.) whether the existence of the causal
connection between the health outcome and its antecedent exposure is more
probable than its non-existence. In a class action suit this has to do with such
particulars of the exposure and the persons as are definitional to the class; and in
the individual suit the issue is analogous to this, only more specific in its particulars.
In courts of law, just as in the practices of community and clinical medicine, there
thus are issues about the etiology/etiogenesis of illness. But arguably different from
medical practices, the legal practices related to etiogenesis of illness should involve
express focus on the general question of what is known in the community of the
relevant experts about the causal connection at issue, about the probability of its
existence in such individual cases as the one(s) at issue. The relevant knowledge
is about the proportion of instances like the one(s) at issue in general such that
the antecedent is causal to the outcome, this proportion being the probability
in question. Given the amount of relevant detail on an individual plaintiff, the
knowledge-base for rational resolution of the legal cases tends not to derive
3.4 Etiology as a Legal Concern 41

satisfactorily from epidemiological (population-oriented) research on etiogenesis;

it tends to require clinical (individuals-oriented) research of this genre.
Judges and lawyers cannot be presumed to understand the causality issues
in these cases, notably as to the state of the scientific knowledge; they should
be deferential to experts on the topic in question. They need to appreciate that
science can, in principle at least, address such topics as the probability that a given
medication use was causal to a particular adverse health event that has occurred in
association with it, even though there can be no science to address the probability
that a given person was causal to guilty of a tort or crime that has occurred.
This deference involves, in the context of the present state of science-based
epidemiological and meta-epidemiological clinical knowledge (sect. 2.6), the
problem that experts are likely to have only widely-divergent personal opinions
about the level of the probability in question, in lieu of actual knowledge about it.
So, in matters of science, even, the courts actually are down to witness testi-
monies separately for and against the causality in the case(s) at issue, generally by
pseudo-experts. The only difference is that these witnesses present opinions about
the probability regarding which the decisive opinion still is that of the judge or
the jury instead of merely presenting evidence relevant to that ultimately-relevant
opinion.
Chapter 4
Etiology as the Object of Study

Abstract Research on etiogenesis of morbidity or of illness per se is, by

the very nature of this genre of causation in medicine, generally bound to be
non-experimental; but a much greater added challenge is that causation is not a
phenomenon, subject to observation; it is a conception a priori, a noumenon
(Kant), needing to be inferred from phenomenal patterns.
If etiogenesis were a phenomenon, study of it would be based on a mere
case series of the illness, the cases selected independently of their etiogenetic
backgrounds, and the concern in it would be the proportion of the cases with the
antecedent at issue such that the antecedent actually was (seen to be) causal to the
case this factor-conditional etiogenetic proportion.
But as etiogenesis is but an unobservable noumenon, the inescapably needed
case series is to be an element in documentation of a phenomenon: the cases rate
of occurrence in a defined study base, the necessary added element being a sample
of that study base, a base series that is.
The case and base series, considered jointly, allow for documentation of the
relative levels of the rates of the cases occurrence in segments of the study base,
and of interest here are the rates for the index and reference segments of the study
base the rates for those with a positive history for the etiogenetic factor and for its
alternative, respectively. The rate ratio for this contrast allows for calculation of a
first approximation to the factor-conditional etiogenetic proportion (cf. above).
When, as is usual, the index and reference segments of the study base have
different distributions by extraneous determinants of the cases occurrence, control
of this confounding is needed.
One option in this control is suitable standardization of the rate ratio. But this
intuitively appealing approach is impracticable when the number of confounders is
appreciable.
This limitation of the standardization approach to the control of confounding
is avoided by invocation of a suitable model for the cases rate of occurrence.
A suitable log-linear model implies the rate ratio as a function of its modifiers,

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 43

DOI 10.1007/978-94-007-4537-7 4, Springer ScienceCBusiness Media Dordrecht 2012
44 4 Etiology as the Object of Study

conditionally on the set of confounders and thus, free of confounding by them, and
calculation of this result for the factor-conditional etiogenetic proportion can be
based on this functions fitted counterpart.

4.1 Some Introductory Examples

1. Policies about DUI. Legal and other public-policy stipulations about DUI
(driving under the influence, of alcohol) are principally guided by information
about the excess rate of fatal injuries that in the population within the policy-
makers purview is attributable (causally) to peoples DUI, as defined, perhaps
distinguishing among some demographic strata (by categories of age, most notably).
This DUI-attributable excess rate of car-accident fatalities (in a given demo-
graphic stratum, say) is the overall rate of these fatalities in the population in
question, regardless of causation, multiplied by the proportion of the overall rate that
is attributable to DUI; and the latter, that overall etiologic/etiogenetic proportion,
in turn, is the product of two proportions: the proportion particularistic, in the
population at the time of the fatalities occurring in association with DUI, and
the proportion more-or-less general of DUI-associated fatalities that actually are
DUI-attributable (sect. 3.2). The latter proportion presumably is practically invariant
over whatever demographic strata of the population of public-policy concern (so
long as the distribution of the severity of the inebriation is practically the same over
the strata).
It thus is knowledge of the etiognostic, rather than intervention-prognostic, type
from etiologic/etiogenetic research that matters in this example; and that other type
of research-based knowledge wouldnt even be practical to pursue for the purpose
here (see below).
2. Policies about seatbelt use. The requisite knowledge-base for public policies
about seatbelt use is, in its nature and genesis, quite analogous to that concerning
DUI; but the other genre of causal research is now realistic and instructive to
consider.
In the spirit of intervention-prognostic research (sect. 3.1), one would think
of, and address, drivers staying in a given category of the risk factor (use/non-
use of seatbelt) over considerable amounts of accrued mileage; and the study
would address the incidence of fatal injury to the driver, from traffic accidents, for
the contrasted categories. For categories of appreciable degrees of inebriation the
capturing of this type of experience is quite unimaginable, but for driving with and
without the use of the seatbelt it isnt.
Even though rather practicable in respect to seatbelt use, that alternative to the
etiognostic outlook and approach is not the preferable one in this case, either. The
rate has a multitude of determinants, both causal and acausal, other than the use/non-
use of seatbelt. In the face of this complexity, sufficient for policies about the use
of seatbelt is knowing its consequent proportional reduction general in the rate,
this reduction in conjunction with what the rate local in the absence of seatbelt
4.1 Some Introductory Examples 45

use is or would be. The relevant absolute reduction in the rate is the product of these
two (cf. sect. 3.2), that general input into it naturally taken to be independent of that
ad-hoc input.
So, again, the etiognostic genus of causal research is sufficient while also
obviously simpler than its intervention-prognostic alternative.
3. Policies about smoking. Like DUI and non-use of seatbelt, smoking is a matter of
peoples behavior that is of concern to promulgators of societal policies directed to
promotion and preservation of public health (i.e., the health of the population in the
policy-makers purview; sect. 1.1). But the legal-regulatory stipulations restricting
the peoples admissible behaviors are in this context supplemented by regulation
of the peoples environments: the concern is not only to ban smoking in various
particular types of setting; the peoples environments are controlled as to whether,
or what types of, tobacco products can be available (where and to whom, according
to age). Another difference is that community-level health education about the
behaviors their health implications is promoted by the policies being that
the risky behavior in many settings remains admissible (different from the examples
above).
The policy-makers direct concern is to reduce morbidity from illnesses for which
smoking is etiogenetic, especially from illnesses through which smoking is a ma-
terial contributor to mortality in the population. They thus need to learn from the
epidemiologist(s) caring for the population in respect to illnesses such as lung can-
cer the prevailing rate of mortality in the population; or, more to the point, they need
to learn the total mortality from smoking-related illnesses in the aggregate. And they
need to learn the proportion of this mortality really of the underlying morbidity
that is attributable (etiogenetically) to the peoples (histories of) smoking (sect. 3.2).
In these, essential terms the scientific knowledge-base of public policy about
smoking is analogous to that of the two examples above, with the relevant
knowledge from epidemiological research again being of the etiognostic rather
than of the intervention-prognostic type. A difference is, however, that smoking
as the etiogenetic factor does not explain (causally) a proportion of the cases that is
anywhere near invariant across all demographic strata of the population of concern.
Regarding age in particular, the general (epidemiological) knowledge needs to be
stratum-specific, including in respect to the typical positive history of smoking in
materially relevant respects.
Once thus informed about the excess mortality from smoking-related illness (by
categories of age, at least), the regulators would like to have an added input into their
decisions about the various possible regulatory interventions under consideration,
namely the intervention-prognostic magnitudes of the effects of these (also by
categories of age, at least). Knowledge about the magnitudes of these effects is,
however, very impractical to acquire by epidemiological (or other) research. But it
also is rather immaterial for the regulatory decisions, as cost-effectiveness is not
really an issue in these decisions.
For community-level health education about smoking-related life-styles, it is
essential to have research-based knowledge sketchy knowledge at least about the
46 4 Etiology as the Object of Study

health notably survival implications of the principal choices that individuals in

the population may make for themselves. These are the choices of whether to take up
smoking and, insofar as one has taken it up, whether to quit the habit. For the devel-
opment of this knowledge-base for health education, etiognostic research is both rel-
evant and feasible, while prognostic research is quite problematic on account of the
instability of the smoking habit among those who ever take up the habit including
incompleteness of the follow-through with decisions to permanently quit smoking.
4. Policies about fluoridation of drinking water. When behavioral etiogenetic
factors such as DUI or non-use of seatbelt or smoking are removed from a
population, this scarcely has adverse consequences on public health in terms of
increase in morbidity from some illness(es) to which the behavior change was not
directed. The same quite obviously is true of removing some pathogenetic agent
from peoples environments airborne asbestos, for example.
Matters are different in respect to micronutrients, whether in food or drink or
taken in as dietary supplements. Whereas all types of tissue in the human body have
the same, known optimum (i.e., zero) for the concentration of the ingredients of
tobacco smoke, for instance, this may not be the case for a given micronutrient.
Fluorine appears to be a case in point.
Fluorine intake is essential for the health of bones in particular, freedom from
dental caries included; and while the sources of this micronutrient include sea fish
and tea, among others, the principal source generally is drinking water.
Etiogenetic research has produced the understanding that fluorine concentration
less than 1 ppm in drinking water is causal to dental caries (while high concentra-
tions, such as 10 ppm, cause pathologic fluorosis of bones, including teeth); and
this understanding has led to policies of fluoridation of the drinking water in various
municipalities with naturally low concentrations of fluorine. Predictably, this has
had the consequence of reduced rates of dental caries in those municipalities.
But, no biological principle is to the effect that a level of fluorine intake that
is optimal for dental health also is optimal for all other aspects of health for
minimizing the rate of oxidative genetic damage and of its resultant development
of oncogenes and then cancers, for example. Concerns about adverse side effects
of fluoridation of drinking water as for morbidity from illnesses other than dental
caries have indeed arisen, and some programs of such environmental adjustments
have been discontinued on this basis.
Intervention-prognostic studies on this topic, with attention to all outcomes
potentially affected, would have provided the knowledge-base for these interven-
tions, had they been practicable in the face of the need for a large number of
municipalities and a very long-term follow-up (in respect to potential carcinogenesis
in particular). But as they havent been and perhaps wont be, etiogenetic studies
based on naturally varying levels of fluorine in drinking water need to be extended
to health outcomes other than dental caries.
4.2 From Case Series to Rate Comparison 47

4.2 From Case Series to Rate Comparison

Very important to appreciate about causation is the fact that it is not a phenomenon,
that it thus is not subject to being observed. It is, instead, what Immanuel Kant
termed a conception a priori, a noumenon. In this sense, as Kant points out, causation
is akin to, say, space and time not formed by experience but innate.
The concept of causation, noumenal as it is, has been remarkably baffling to
many notable philosophers. Insofar as the Greek aitia (etymologic to the etio- prefix)
can justifiably be translated as cause, Aristotle envisioned four fundamental genera
of cause: material, formal, effective, and final; but of these, only effective aitia
actually is a cause and specifically etiogenetic/etiognostic rather than intervention-
prognostic cause (sect. 3.1) in our contemporary meaning of cause.
William of Ockham (1285?1349?) is now commonly (though falsely) taken to
have been the originator of the important principle of parsimony (L. parsimonia,
frugality), namely that concepts are not to be multiplied beyond necessity; and one
of his applications of this principle was advocacy (sharp, as usual, a` la Ockhams
razor) of seeing causation to be an unnecessary concept, an undeserving supple-
ment to that of regular succession. Similarly, the eminent empiricist philosopher
David Hume (17111776) argued that knowable only is constant conjunction, in
lieu of causation. And subsequent positivist philosophers, even of the twentieth
century, had no use for noumenal concepts, such as causation.
Philosophers, while much concerned to understand what is and what is not
real (ontologically admissible to consider), have thus been conspicuously out of
touch with the realities of medicine. Hippocrates (460? BCE370? BCE), already,
is taken to have brought forth the fundamental idea that illnesses are not divinely
instigated; that instead, the beginning of the disease process is an offending material
(materia peccans) bringing the bodily humors (blood, phlegm, yellow bile, and
black bile) into a state of bad mixture (dyskrasia). Akin to this, a subsequent
concern in medicine has always been to understand the occurrence of an illness
(individually or in a population) as a matter of causal explanation, and beyond this,
to intentionally cause a change in a persons or populations health for the better
(cf. sect. 3.1). Without the concept of causation medicine would be as passive about
peoples health as, say, philosophers commonly elect to be about everything in their
surrounding reality (insofar as they, perhaps only grudgingly, grant its existence)
and cosmologists necessarily are about the goings-on in the cosmos.
In this aloofness from reality, philosophers have been in the Platonic mode. To
his persona as the first academic [Plato] added or superimposed the complementary
persona of the first intellectual, by which I mean someone who thinks ideas matter
more than people. So writes the eminent historiographer Paul Johnson in his
Socrates: A Man of Our Times (2011; p. 11).

To gain the most fundamentally orientational understanding of what an etio-

logic/etiogenetic study is to be empirically about, it is helpful to first consider the
thinking that would be natural if causation were a phenomenon. In this situation
48 4 Etiology as the Object of Study

(counterfactual; cf. above) causation would be observable (and documentable) in

each instance in which an antecedent (identifiable) actually is causal to a subsequent
(also identifiable), serving to produce (sic) the subsequent this outcome while
its defined alternative would not have had this effect (cf. sect. 3.1).
In this hypothetical situation, any etiogenetic study would focus on cases of the
illness; and among these the narrower focus would be on the ones associated with
the antecedent in question, as non-causation in the absence of the potential cause at
issue is obvious a priori. In respect to these illness-antecedent instances, the object
of study would be the proportion of them such that the antecedent is causal
etiogenetic to the illness. The magnitude of this proportion would be directly
studyable on the present premise as a function of determinants of its magnitude
on the basis of its empirical counterpart in a case series of the illness.
An added object of study (scientific) would commonly be the proportion of
the cases that are associated with the antecedent in question, if this commonly
were a parameter constant of Nature. But it actually is prone to be an ad-hoc
matter, devoid of universality (cf. sect. 4.1). Scientific study of this proportion, when
meaningful on account of its universality, is not made more challenging by the fact
that causation is a noumenon rather than a phenomenon.
Implicitly, in these remarks, at issue has been a (potential) cause of the all-or-
none type, one without subtypes (by intensity or duration, say) to be distinguished.
Insofar as there are such subtypes, those remarks apply to each of these separately,
to each index category vis-`a-vis the shared reference category.
In the real world the need is to move away from consideration of that un-
observable etiogenetic proportion for a given index history in a series of cases
of the illness with the antecedent, to consideration of certain related, inherently
observable, phenomena: that proportions manifestation in rates into which the case
series supplies numerator inputs. To this end it is necessary to think of the case
series as constituting those occurring in a defined experience of the cases potential
occurrence, the experience the study base classified according to whether the
antecedent in question was present, or the alternative to this was, or neither one of
these was.
Focusing on the empirical rates Ri and R0 for the respective experiences with
the ith index category of the antecedent and the reference category (present
retrospectively as of the outcome; sect. 3.1), the factor-conditional etiogenetic
proportion addressed above is, to a first approximation,

.Ri  R0 / =Ri D .RRi  1/ =RRi ;

where RRi D Ri /R0 , the ith rate ratio, the index rate divided by the reference rate.
This measure of the etiogenetic proportion of interest, while based on phenomena
(rates), adduces a new challenge: the magnitude of this RRi , based on crude
(unadjusted) overall rates as it is, can have extraneous explanations in addition
to the etiogenetic proportion in question; therefore, the need is to consider the
corresponding RRi that is not subject to extraneous explanation.
4.3 Standardized RR as the Result 49

4.3 Standardized RR as the Result

When epidemiologists in their practice of community medicine (sect. 1.1) compare

two or more rates, there commonly is a need to adjust one or more of the
directly available, crude overall rates for the compared experiences to achieve
comparability most commonly so as to eliminate the role of the differential
distributions of the respective populations by gender and age. The common means
of this adjustment is mutual standardization of the compared rates.
Following this practice as the paradigm leads to replacement of a crude
comparative measure, the crude RRi in section 4.2 above, by the corresponding
standardized RRi , involving mutually standardized rates in lieu of the corresponding
crude ones. The magnitude of a suitably standardized RR cannot be explained,
even in part, by the difference in the compared populations distributions by the
determinants of the rate that were standardized for. In this way, confounding by
differential distributions of the compared experiences according to those extraneous
determinants of the outcomes rate of occurrence is controlled excluded
from being a possible explanation of the magnitude of the result, the empirical
comparative measure, that RRi , notably its deviation from the null value of RRi D 1.

The principle underlying this standardization of empirical rates is founded on the

latent structure of any crude overall rate. With C the total number of cases of an
illness occurring in PT amount of population-time, the crude overall rate, C/PT, is
a (latently) weighted average of the specific rates in the strata (by gender and
age, say, here indexed by j D 1, 2, : : : ) contributing to the crude overall rate:
X X
C=PT D Cj PTj
j j
X  X
D PTj Cj =PTj PTj
j j
X X
D PTj Rj PTj
j j
X   X  
D PTj =PT Rj PTj =PT
j j
X
D Wj Rj ;
j

P
where j denotes summation over the strata (j D 1, 2, : : : ) and WP j D PTj /PT is the
latent weight for the specific rate, Rj D Cj /PTj , in the jth stratum. ( j Wj D 1).
Given this structure of a crude rate, it follows that two or more rates are
mutually standardized for one or more stratification factors if they involve the
same (standard) set of weights for their stratum-specific component rates. And
a standardized rate ratio is based on two mutually standardized rates.
50 4 Etiology as the Object of Study

The implication of this standardization principle here is this: Of the crude index rate,
R1i , among those with the ith index history for the potentially etiogenetic factor in
the study base, the proportion actually caused by the factor the factor-conditional
etiogenetic proportion (sect. 3.2) is addressed by

b 

RRi  1 RRi ;

b
b 
where RRi is the standardized rate-ratio in a particular meaning. First, RRi in- b 

volves R1i , the crude unadjusted index rate (conditional on the ith index history).
Second, it involves R0i , the reference rate adjusted to the latent structure of the
index rate (of its denominator input). And third, this adjustment of R0i involves all
extraneous determinants (acausal as well as causal) of the reference rate in terms of
which the index and reference experiences differ.
While any choice of shared weights generally serves the purpose of standardiza-
tion (attainment of comparability in particular respects), here the weights need to be

b b
thought of as those that are latent in the crude index rate (which rate therefore needs
 
no adjustment for the RRi ). For only with these weights does the RRi represent
the empirical value of the factor-conditional etiogenetic proportion the proportion

b
of the cases with the antecedent that are caused by the antecedent (sect. 3.2).

It is good to learn to think about this RRi as the observed-to-expected, or Oi /Ei ,
ratio of the number of cases in the index experience of the study base:

b 
RRi D .Ci =PTi /
hX 
j
 X
PTij R0j
j
PTij
i

 X  
D .Oi =PTi / Ei PTij
j

D Oi =Ei :

Actually, this Ei is only the empirical counterpart of the actual null value
corresponding to Oi , and we hence prefer the notation in

b 
RRi D Oi =EO i :

While this standardization-based principle is impeccable in its logic, it adduces

two problems to consider. For one, as noted above, it implies the necessity
inescapable need, throughout non-experimental etiogenetic research to know what
all of the extraneous determinants of the magnitude of reference rate are. And
for another, given this knowledge merely presumptive standardization-based
isolation of the purely causal RRi is prone to be impracticable in the context of the
generally large number of potential confounders.
For the latter problem there commonly is, in principle, a reasonable solution: the
invocation of a suitable model.
4.4 The Object Imbedded in a Model 51

4.4 The Object Imbedded in a Model

As background, or the context, for etiologic/etiogenetic modeling, let us recall the

relevant essentials from section 4.3 above in reference to an antecedent of the all-
or-none type:
1. An etiogenetic study has to do with instances in a defined abstract domain of
the illness occurring in association with the antecedent whose etiogenetic role is
at issue; and in elementary terms, the natural object of study is the proportion of
these instances, in that domain, such that the antecedent actually is causal to the
outcome.
2. The empirical value for this (factor-conditional etiogenetic) proportion (P)
derives
 from
 the proportional extent to which a certain rate (R1 ) exceeds another
rate R0 :
 
PO D R1  R0 R1
  
D R1 R0  1 R1 =R0

b
D RR  1 RR
 
b
   
D O EO  1 O EO ;

where R1 is the crude index rate (with positive history for the factor at issue) and
R0 is the reference rate (with positive history for that factors defined alternative)
adjusted to the latent structure of R1 in respect to all extraneous determinants
(acausal as well as causal) of the reference rate in terms of which the latent
structures of the crude index and reference rates (R1 and R0 ) differ, these rates
characterizing an experience in the domain of the object of study; and where
O is the observed number of cases contributing to the index rate and E is the
corresponding null-expected numbers empirical counterpart.
This is the standardization approach to obtaining an empirical value for the core
object of an etiogenetic study the study result in terms of the O/E ratio serving
to derive the corresponding result for the factor-conditional etiogenetic proportion.
Involved is maintenance of the occurrence relations conditionality on extraneous
determinants of the outcomes rate of occurrence of the reference rate, that is.
This leads to consideration of the outcomes rate of occurrence as a joint
function (descriptive) of the etiogenetic determinant at issue and the presumedly
inclusive set of the reference rates extraneous determinants involved in the needed
conditionality. In the framework of such a joint function the relation of the rate to the
determinant at issue is inherently conditional on the set of extraneous determinants,
given that the form of the function is realistic.
In an extremely simple situation the etiogenetic determinant would have only
three (nominal) categories index, reference, other and the conditioning would
52 4 Etiology as the Object of Study

involve only, say, gender and age. The correspondingly designed function the
occurrence relation designed as the model for the rate, R might be as simple as

R D B0 C B1 X1 C B2 X2 C B3 X3 C B4 X4 ;

where the Bs are (presumed) parameters (constants) of Nature and the Xs are (ad-
hoc adopted) statistical variates (which the determinants they represent arent):
X1 D indicator of index history (X1 D 1 if this, 0 otherwise)
X2 D indicator of other history (so that X1 D X2 D 0 implies reference history)
X3 D indicator of male gender
X4 D age as the number of years (i.e., age/yr).
What is simple about that situation is not merely the all-or-none nature of the set
of possible etiogenetic histories and the (presumed) need to condition the relation of
R to these on gender and age only; and an added simplicity is the feature that R is a
numerical quantity not incidence density but a proportion-type rate (of incidence
or prevalence).
And for that simple situation that model is quantitatively simple in two of its
implications: that the magnitude of the effect at issue (the difference between R1 ,
involving X1 D 1 together with X2 D 0, and R0 , involving X1 D X2 D 0) is constant
(B1 ) over (i.e., unmodified by) gender and age; and that full conditioning by gender
and age is provided for by their very simple simply additive representation in
that model (without, say, X5 D X4 2 , X6 D X3 X4 , and X7 D X3 X5 ).
In terms of the fitted counterpart of this simple model we have RO 0j defined for
each of the strata by gender and age (focusing on typical age in each stratum).
P The
standardized reference rate R0 corresponding to the crude index rate R1 is j Wj RO 0j ,
where the weights derive from the index experience. Then, R1 /R* 0 D O/E implies
the empirical value for the factor-conditional etiogenetic proportion (see above).
The stratification in the derivation of the O/E ratio actually is unnecessary.

b
Associated with each index case (among the O number of these) is the corresponding
RO 1j and R O 0j implied by the fitted model; and hence the corresponding RRj
also. Thus,

EO D
X
1 RRj ;
j
b
where the summation is over the index cases (O in number), their profiles in terms
of gender and (actual) age.
A slightly but importantly modified version of that simple model is this log-linear
one:
X4
log .R/ D Bi Xi ;
0
4.4 The Object Imbedded in a Model 53

where X0 D 1. The modification is that the linear compound (of the Bs, with the
Xs the coefficients in it) now represents the magnitude of the logarithm (natural,
Napierian) of the numerical (dimensionless) rate. In terms of this model,

log .R1 /  log .R0 / D B1

log .R1 =R0 / D B1
R1 =R0 D exp .B1 / ;

where exp denotes exponential of; that is, antilog (base e) of.
Insofar as this model is correct as to its implication that the difference in log (R)
corresponding to X1 D 1 (and X2 D 0) versus X1 D 0 (and X2 D 0) is constant over
gender and age, and the bearing of gender and age on log(R) is correctly represented
in reference to the subdomain in which X1 D 0 (and X2 D 0) the implication is that
the models fitted counterpart implies

b 
RR D O=EO D exp .B1 / :

If, however, the model should additionally include

X5 D X1 X3 ;

the implication would be that

log .R1 /  log .R0 / D B1 C B5 X3 ;

implying gender modification of the RR:

RR D exp .B1 C B5 X3 /
D exp .B1 / given X3 D 0 .female gender/
D exp .B1 C B5 / given X3 D 1 .male gender/ :

This model thus implies directly the gender-specific values of the RR but the overall
RR* only through the values of R1j and R0j it thus defines (cf. above).
But, as noted above, this stratification actually is unnecessary. Log-linear models
for a rate, whether a proportion-type rate per se or the numerical element in
incidence density, are the ones of principal concern in etiogenetic research. Without
having to derive values of RO 1j and RO 0j , the values of RR
c j are obtained directly from
the fitted RR function (of its modifiers). Then, as above,

b
RR D O EO D C1
X  
cj ;
1 RR
j
54 4 Etiology as the Object of Study

b
where C1 is the number of index cases of the illness (ones occurring in association
with the antecedent at issue), the summation is over all of these cases, and RRj is
the empirical RR corresponding to the modifier profile of the jth one of these cases.
This point likely deserves to be restated, though somewhat differently. Log-linear
modeling of the rate at issue in etiogenetic research is commonly appropriate to
consider. That the model is log-linear implies no inherent restriction on the form
of the occurrence relation how the rate, or the rate ratio, is a function of age,
for example. Insofar as such a model actually is well designed (as to the form of
the occurrence relation in its referent domain), including sufficiency of the built-
in conditionality of the rates relation (descriptive) to the risk factor at issue, the
modeling provides ready definition of the critically relevant RR* (addressed above
and in sect. 4.3).
Thus far the focus here has been on models in the context of an all-or-none index
history. This needs to be supplemented by consideration of the more general case in
which

X1 D history score .numerical; ordinal/;

so that index histories involve X1 > 0 with X2 D 0 and the reference history is
represented by X1 D X2 D 0. Now the interest is in results specific to non-zero values
of X1 , and corresponding to the ith non-zero category the fitted function is evaluated
at X1 D i and X2 D 0 to derive the corresponding RRi D Oi /E i (cf. sect. 4.2).
As we noted at the end of section 4.3 above, modeling of the occurrence relation
and study of the relevant parameters imbedded in the model does away with
the problems that characterize the needed conditioning when adjusting R0 to R0 in
the framework of cross-stratification, especially when the number of strata is quite
large. For there is no feasibility problem and not much loss of efficiency in the study
of RR* in consequence of inclusion of added potential confounders into the model.
But this adduces a new problem: uncertainty about the adequacy of the model in
providing the intended conditioning of the rates (descriptive) relation to the risk
factor at issue. The solution is devoid of the desired feel for the data.
With all this said about the designed object of an etiogenetic epidemiological
study, it may be appropriate to recall one of the most important contributions of
Immanuel Kant into scientific thought. He taught that it is not a scientists role to
simply read the Book of Nature, by accruing experience in such terms as it happens
to present itself. Instead, (s)he is to define, before whatever research experience, the
terms in which (s)he needs to consult Nature. In epidemiological research, the form
of the occurrence relation at issue defining the object(s) of study is to be defined
before actually having the experience, and the role of the research experience is,
merely, to supply empirical content of that preset form.
The deployment of modeling is generally necessary in (non-experimental)
etiogenetic research but that uncertainty about the models adequacy in providing
for the intended conditionality and that lack of feel for the data can be overcome.
See section 5.4.
Chapter 5
Etiologic Studies Essentials

Abstract Given that a defined study base is a sine-qua-non for any admissible type
of etiogenetic study, and that each person-moment in it is to represent the domain
of the designed model for the cases rate of occurrence, it generally cannot be
operationally formed, nor even directly defined; it must be defined as a segment
of the population-time within a defined source population-time, a source base. The
latter, in turn, is formed by the course of a defined source population over a particular
span of time, with this population possibly defined indirectly, as the catchment
population (of cases) secondary to the directly-defined manner of case identification.
Cases of the outcome event are identified, comprehensively, in the source-base
experience, and a fair sample of the source base (of the infinite number of person-
moments constituting it) is drawn. The resulting first-stage case and base series are
reduced to instances from the actual study base (as defined), to the actual pair of
study series.
The model log-linear for the cases rate of occurrence implies its correspond-
ing logistic model to be fitted to the data (with Y D 1 and Y D 0 for the case and base
series, respectively), yielding a result for the rate ratio of interest, as a function of
the modifiers of this as they are accounted for in the model. This, in turn, provides
for calculation of the expected number of index cases in the calculation of the
essential result, that for the factor-conditional etiogenetic proportion.
These calculations under the designed model lack the intuitive appeal of those
based on cross-stratification of the data according to the confounders being con-
trolled. A solution to this problem is stratification by a unidimensional confounder
score, provided by the fitted logistic function evaluated at the reference category
of the etiogenetic determinant under study. Examination of the data in these strata
allows verification of freedom from confounding by the extraneous determinants ac-
counted for, and the calculation of the desired result is intuitive and straightforward.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 55

DOI 10.1007/978-94-007-4537-7 5, Springer ScienceCBusiness Media Dordrecht 2012
56 5 Etiologic Studies Essentials

5.1 Source Base, Study Base

Whereas the designed form of an occurrence relation for the (rate of) occurrence
of the illness at issue in a defined domain defines the object parameters for an
etiologic/etiogenetic study, and whereas an actual study of that occurrence relation
(of the object parameters in it) is to document experience of that form, the first-
order topic in the design of an actual etiogenetic study is selection of a particular
experience for that documentation.
The end result of this selection the adopted study base usually is an
aggregate of population-time from the designed occurrence relations domain, for
documentation of incidence density (of an event-type outcome), the dependence of
this rate on the determinants of it in the form of the predesigned occurrence relation.
The person-moments in this study base (infinite in number) represent the designed
occurrence relations defined domain (for the outcomes occurrence), and associated
with each of these person-moments is an/the index history or the reference history,
or, perhaps, some other history. Besides, the person-moments constituting the
study base may have been designed to satisfy various admissibility criteria of a
purely practical sort the person at the time being compos mentis and fluent in a
particular language, for example.
A study base of this type is defined in stages. Defined before the study base
itself is the source base, and the study base proper is then defined by applying
the admissibility criteria to the person-moments constituting the source base. The
source base also is defined in two stages. The first stage is the selection of and
commitment to a particular source population.
The definition of the source population may be direct that of the resident
population of a particular metropolitan area or the prescribers to a particular health-
insurance plan, for example. Alternatively, direct primary definition may be
given to the scheme of identification of cases of the illness/outcome at issue, which
makes the source population to have an indirect definition secondary to this: it gets
to be the catchment population of this scheme the entirety of those who, at any
given moment, are in the were-would state of: were the outcome event (clinical
inception of the illness) now to occur, the case would be caught by that scheme.
With the source population defined, the actual source base for an etiogenetic
study of incidence density (in causal relation to histories in respect to the risk
factor at issue) needs further definition, in terms of time. For dynamic source
populations (being open for exit, and hence having turnover of membership), such
as the residents of a particular metropolitan area, the source base is generally
defined as the source populations course over an interval in calendar time. The
same can be true of a source base formed from a previously formed, multipurpose,
cohort-type source population (which is closed for exit, even by death). But if the
source population is a cohort formed ad hoc, then the members contributions to
the population-time of the source base are prone to start from their enrollments
into the cohort, at different points in calendar time, but end simultaneously in
calendar time if not earlier, due to death or loss to follow-up.
5.2 Case Series, Base Series 57

Given the source base for an etiogenetic study of the usual type, dealing with
incidence density, the actual study base within it indeed is merely defined, not
formed by any process of admissions (of person-moments) into it. The study
population generally is dynamic even if the source population is a cohort: a persons
contribution to the population-time of the study base begins when admissibility to it
first gets to be satisfied, and it ends when those criteria no longer are satisfied; the
study population thereby has turnover of membership (which here is the meaning of
dynamic).
Exceptionally, the study base is one for a proportion-type rate, and so therefore
also is the source base in which this is imbedded. Both of these are constituted by a
series of person-moments, a finite enumerable series, instead of an aggregate
of population-time (constituted by an infinite number of person-moments).

5.2 Case Series, Base Series

As the empirical occurrence relation being documented in an etiologic/etiogenetic

study addresses the outcomes occurrence in the study base the rate of this
occurrence as a function of its determinants the study inescapably needs to involve,
for one, identification of the cases of the outcome that occur in the study base; it
needs to produce the case series that supplies the numerator inputs into the empirical
rates being documented (sect. 4.2).
Cases of the outcome are first identified in the source base, and this first-stage
case series then is to be reduced to what ultimately is needed: the series of cases
occurring in the actual study base their associated person-moments satisfying the
criteria of a person-moments admissibility into the study base (sect. 5.1 above).
For reasons of validity assurance the aim generally is to identify all of these cases,
even though an assuredly fair stochastically representative sample also would be
consistent with validity assurance. After all, an etiogenetic study ultimately is about
rate ratios, not absolute rates (sects. 4.2, 4.3, 4.4).
It bears emphasis that not only is a series of cases of the outcome at issue a sine-
qua-non element in the structure of an etiogenetic study, but the role of this case
series is to provide numerator inputs into (the calculation of empirical) rates in a
defined experience of the outcomes occurrence in the study base (sect. 4.2).
The case-control study (sect. 6.2), while still commonly deployed (ch. 2),
represents a notable failure to appreciate this role of the case series in a rational
etiogenetic study. A case series without a defined experience of case occurrence as
its referent is not a case series of a rational etiogenetic study, as this case series isnt
in the role of providing numerator inputs into rates (empirical; sect. 4.2).
More to this same effect: If one were to specify the single most basic element
the single most fundamental sine-qua-non in the structure of an etiogenetic study,
one would have to assign this status to the expressly defined study base. This point
bears emphasis because of the still-common adherence to that case-control study
concept (that trohoc fallacy; sect. 6.2).
58 5 Etiologic Studies Essentials

Given the study base and the case series providing numerator inputs into
(quantification of) the outcomes index and reference rates of occurrence in this
study base, obviously needed also is a source of the corresponding denominator
inputs; that is, needed also is a base series to accompany that case series, to provide
denominator inputs into (the documentation of) the ratio of the index and reference
rates in the study base.
When, as is usual, the study base is an aggregate of population-time (for study of
incidence density), the base series necessarily is a sample of the study base (as the
population-time of it is constituted by an infinite number of person-moments). As
a sample a stochastically representative one the base series provides numbers
(tallies) proportional (stochastically) to the population-time referents of the rates of
comparative concern; and proportionality in this sense is all that is needed, as the
etiogenetic interest is not in rates per se but only in their ratios (sects. 4.2, 4.3, 4.4).
When the study base is a series (enumerable, finite) of person-moments, census
of it for the base series always is an option in principle; but it may not be justifiable
on the ground of its inefficiency. Census may be unjustifiable if, in a study of a
proportion-type rate, there is a procedural distinction between the acquisition of the
case series and that of the base series. But there may be but a single series of the
person-moments constituting the study base, with the case series identifiable only
as a subset in this base series.
A base series is to represent the study base at large, and not merely of that
segment of the base in which a case of the outcome is not associated with the person-
moment. When the study base is one of population-time, a sample (finite) of it (of
the infinite number of person-moments in it) is not expected to include any instances
in which the outcome event is associated with a person-moment in the sample. But
in a sample and especially in a census of a study base constituted by a series
(finite) of person-moments, instances of the outcome can be involved in the base
series and to the extent they are, they belong there.

5.3 Model Fitting, Study Result

The two series from the study base the case series and the base series naturally
are documented in respect to everything that is involved in the designed occurrence
relation. And whatever may be the format of the primary documentation of those
data and the codings applied to these, in the end the data are in the form of
realizations of the statistical variates in the designed occurrence relation. In these
terms there is a data matrix with columns for Y, X1 , X2 , etc., where Y D 1 and
Y D 0 indicate the presence and the absence, respectively, of a case of the outcome
at issue (i.e., membership of the instance the person-moment in the case or the
base series).
In the usual situation in which the designed model addresses incidence density,
fitted to the data is the logistic counterpart of that model:
X
log Pr .Y D 1/ =Pr .Y D 0/ D B0 C Bk Xk :
k
5.3 Model Fitting, Study Result 59

The intercept (B0 ) in this reflects, in part, the chosen size of the base series
relative to that of the case series; but the rest of the parameters are, quantitatively,
the same as in the designed occurrence relation including, of course, the subset of
those parameters constituting the actual objects of the study (sect. 4.4).
More directly relevant for the knowledge-base of etiognosis about the rate of
morbidity is a result secondary to this direct one. The study produced an observed
number of cases of the outcome in association with the ith index history for
the etiogenetic determinant (of the outcomes rate of occurrence). This observed
number (Oi D Ci ) divided by an estimate (E i ) of the corresponding expected
number equals the crude overall index rate (Ri ) divided by the overall reference
rate adjusted to the structure of Ri :

b 
Ci =EO i D Ri =R0 D RRi

(sect. 4.3). This is the empirical counterpart (from the study at issue) of the RR that
determines the factor-conditional etiogenetic proportion (among the O D C1 cases):
.RRi  1/=RRi (sect. 4.3).
The model-fitting counterpart of this RR* is

b 
RRi D Ci =
X  
1=RRij ;
j

where the RRij set corresponds to the Ci set: it is the set of RR values from the study-
produced RR function corresponding to (the profiles of) the Ci subset of cases, and
the summation is over the members of the Ci set (sects. 4.3, 4.4).
If at issue is a proportion-type rate on the basis of an enumerable (finite) set

b
of person-moments, the designed log-linear model (sect. 4.2) is fitted to the data

(possibly a single series) as such; and the results for the RR function, RR , and
the factor-conditional etiogenetic proportion flow from this quite analogously with

b
those having to do with incidence density (above).

With that RRi set (i D 1, 2, : : : ) the principal results of the study, their
precisions need to be quantified (incl. for the purposes of its synthesis with the
corresponding results from other studies), perhaps best in terms of their associated
95% confidence intervals. That interval implies its counterpart for the factor-
conditional etiogenetic proportion of interest.
A simple way to accomplish this begins with fitting a model involving separate
indicators for the index histories with no product terms based on these and noting
the fitted value of each of the corresponding parameters .B/ O together with its SE
(standard error). Then, a 95% interval for the logarithm of RR * i can be derived as
h 
b

i
log RRi .1 2:0=w/ ;
60 5 Etiologic Studies Essentials

where w is the realization of the Wald statistic that point estimate of the
O divided by its SE. This has an obvious counterpart for the
coefficient, B,
b
overall RR .


b 
Deriving the RR and a measure of precision of this is reasonable to do even
when the study is one of hypothesis testing about the very existence of the

b
etiogenetic connection, and not yet a study on the magnitude of an established

etiogenetic connection. For, the magnitude of the empirical RR is informative
for this qualitative purpose together with the null P-value (from that Wald statistic,
perhaps). For synthesis of the result with those of other studies, relevant is the fitted
logistic function together with the SEs of the relevant coefficients in it.

5.4 Demystifying the Modeling

b
Any serious etiogenetic research is characterized by the resulting rate ratios

RR s conditionality on a reasonably complete set of extraneous determinants (of
the magnitude) of the reference rate. This generally means that the traditional cross-
stratification approach to the conditioning with its propensity to break down in
the context of multiple dimensions in the cross-stratification is now commonly
replaced by the modeling approach (sect. 4.4). But the ideal approach arguably is
one which captures the virtues of each of those two approaches while avoiding the
drawbacks of each of them.
Given a history of the all-or-none type, the notation for the frequencies in the j-th
stratum in cross-stratification (by gender, categories of age, etc.) could be taken to
be this:
D D 1 D D 0 Total
c1j c0j Cj
b1j b0j Bj

where the sizes of the subsets for this stratum of the overall case series (of size
C) and base series (of size B) are Cj and Bj , respectively; and of these, the further
subsets from the etiogenetic determinants index category (D D 1) and reference
category (D D 0) are of sizes c1j and c0j from the Cj , and b1j and b0j from the Bj .
The set of J tables like this gives the centrally relevant parameter the RR

b
determining the factor-conditional etiogenetic proportion (sect. 5.3 above) the
empirical value
b 
RR D C1 =EO 1 D C1 =
X
EO ij
j

D
X
j
c1j =
X 
c1j =RRj
j

b
X X    
D c1j = c1j = c1j =b1j = c0j =b0j
j j
X X  
D c1j = b1j c0j =b0j :
j j
5.4 Demystifying the Modeling 61

The inefficiency and its consequent instability, imprecision of this measure

arises from the susceptibility of those EO j values to considerable chance variation,
notably when the data from some of the strata are quite sparse. A remedy to this
is, in principle, reduction of the number of the strata by pooling pairs or larger
subsets of the strata, so long as this is reasonably consistent with maintaining the
intended conditionality on the stratification factors. If such pooling, consistent with
retention of the conditionality, leads to quite small a number of strata, then quite
good efficiency is achieved, after all.
The necessary condition for the poolability collapsibility of two or more
strata is (near-)constancy of the theoretical value corresponding to either the c0j /b0j
or the b1j /b0j ratio that is, of the theoretical (quasi-)rate of the outcomes
occurrence conditional on the object determinants reference category or of the ratio
of the sizes of the index and reference subsegments of the study base across strata
at issue.
This leads to consideration of replacement of cross-stratification of the data by
uni-dimensional stratification for the same purpose stratification by a scoring
function such as is involved, for the needed conditioning, in the modeling alter-
native to the cross-stratification. The modeling ordinarily addresses the outcomes
occurrence; but it could address, for the conditioning purpose at issue here, the study
determinants distribution in the study base (the relative sizes of the D D 1 and D D 0
segments of it).
This hybrid of the traditional cross-stratification approach and its modern model-
based alternative retains the merits of both of these without introducing material
problems of its own. It retains the ready intelligibility of the stratification approach
its associated real feel for the data while it also exploits the efficiency and hence
the feasibility of truly multi-dimensional conditioning characterizing the modeling
approach.
The scoring function, when used for conditioning in the meaning of equalization
of the background risks for the contrast at issue, is the fitted regression function
evaluated at the focal determinants reference category (at D D 0). With the value of
this function calculated for each of the data points (person-moments) in the database
and the strata (of risk) formed on this basis, the first concern should generally be
to verify and document for the study report the expected: that each of the
extraneous determinants (of the outcomes rate of occurrence) indeed has, within
each of the strata, a suitably balanced distribution between the index (D D 1) and
reference (D D 0) segments of the base series.

b
Upon these preparatory steps, the production of the actual main result the value

of .C1 =EO 1 / or RR proceeds as outlined above. And the precision of this result
can quantified as outlined in section 5.3.
Chapter 6
Etiologic Studies Typology

Abstract It remains commonplace to think of design options for etiogenetic studies

in terms of cohort and case-control studies, first and foremost. But it should
be understood that the essential features of etiogenetic studies are not optional.
These are features logically deduced from correct conception of what etiogenesis,
as a genre of causation, actually is (chap. 3 Abstract), and they are: study base
representing the designed (sub)domain of the cases occurrence, case series and a
sample from this study base, etc (chaps. 4, 5 Abstracts). And it should be understood
that the cohort and case-control studies, for example, do not satisfy the logical
requirements for an etiogenetic study.
We think of these purported design options as the cohort fallacy and the trohoc
fallacy, respectively, and we present the necessary corrections of each of them. We
also consider two other purported design options for etiogenetic studies, redefining
one and correcting the other. From the different points of departure we thus forge
unity: E pluribus unum (`a la seal of the U.S.), namely the etiogenetic study.
In the framework of the a-priori, singular nature of the etiogenetic study there
remain true design options in various aspects of it. The source population can be
defined either directly or indirectly (as the cases catchment population); and if
it is defined directly, it may be defined as a cohort or as a dynamic population.
The source population/base can be monolithic or fragmented/scattered. The case
identification in the source base can be directed to all cases of the illness at issue
or, for reasons of validity assurance, to cases that are severe but otherwise typical
in their manifestations. The sampling of the source base can be indiscriminate or,
for efficiency enhancement, suitably stratified. Etc. In this way, the singular nature
of the logically tenable type of etiogenetic study leaves open a multitude of design
options: E unum pluribus.

6.1 Cohort Study: Its Essence Corrected

In epidemiological usage, cohort should be understood to mean a population that

is closed, specifically closed for exits: once a member, always a member. Thus,

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 63

DOI 10.1007/978-94-007-4537-7 6, Springer ScienceCBusiness Media Dordrecht 2012
64 6 Etiologic Studies Typology

membership in the cohort deployed for all the studies in the framework of the
famous Framingham Heart Study misnomer for cardiovascular disease research
program founded on a source cohort recruited from the town of Framingham in
Massachusetts has not undergone any attrition, not even on the basis of the original
members deaths. Even though practically all of this cohorts members already are
deceased, they still constitute an undiminished source population for studies on
the basis of the in vivo data on them. The collection of all of the data was once
prospective (T > T0 ) in cohort time but it always was, and still is, retrospective
(T < T0 ) in any study time of exploiting the data.
The nature of the resident population of the town of Framingham, from which the
cohort was recruited, is different. This population is open for exits, for terminations
of membership by death, for example. As a consequence of the exits in conjunction
with new entries, this population has turnover of membership; it is a dynamic
population in this sense. The FHS could have been constructed in the framework
of this resident population of the town, as such, as its source population. Like
the cohort recruited from it, this dynamic population could have been subjected
to biennial surveys for routine data collection. The actual FHS cohort, as it got to
be documented, would now be identifiable from the database that could have been
produced from this dynamic source population.
What in etiologic/etiogenetic research is now routinely termed cohort study used
to be known as prospective study (and also as follow-up study), distinguishing it
from what was termed retrospective study (and alternatively case history study); but
the prospective-study precursor of the cohort-study term was discarded on the
ground that its antonym got to be viewed as unduly denigrating of the perceived
alternative to it, now known by the higher-sounding name case-control study.
In a cohort study (so-called, in etiogenetic research), a cohort-type study
population is recruited from the chosen source population (cohort-type or dynamic).
As of the time of the enrollments (at T D 0 on the scale of cohort time) the members
of this cohort are documented in relevant respects, most notably as to their histories
in respect to the etiogenetic determinant at issue. The members are then followed
(into prospective cohort time) to document the occurrence/non-occurrence of the
illness whose etiogenesis is being studied. And the thus-accrued data are analysed
(meaning synthesized) to express how the outcomes occurrence was related to
the divergence in the causal determinant (with certain conditionings in this; cf.
sect. 4.4) the outcome prospective and the determinant retrospective on the scale
of cohort time.
The idea in the genesis of the cohort study (on etiogenesis) was to emulate the
intervention-prognostic experiment, the intervention trial (clinical); but in such a
trial the outcomes prospective occurrence is related to prospective (sic) divergence
in the causal determinant (the interventions), both of these in cohort time. On this
basis already, this paradigm was ill-chosen without appreciation of the profound
difference between etiogenetic and interventive causations, at issue in etiognosis
and intervention-prognosis, respectively (sect. 3.1).
The genuine essentials of an etiogenetic study flow directly from the nature of
this genre of causation (sects. 3.1, 4.2; ch. 5), with no paradigmatic role for the
intervention trial in this. And as for etiogenetic causation, critically important to
6.2 Case-control Study: Its Essence Corrected 65

appreciate is its retrospective nature, from the vantage of time of the outcomes
occurrence/non-occurrence (at in association with a particular person-moment;
sect. 5.1). For this reason and in this sense, any etiogenetic study is to be thought of
as inherently being a retrospective study (while any intervention-prognostic study
inherently is a prospective study; ch. 9).
Given that in a cohort study (on etiogenesis) the causal histories are docu-
mented as of the enrollments into the study cohort so that the T0 of etiogenetic time
is made to coincide with the T0 of cohort time the outcome theoretically ought to
be ascertained at this same point in time; but this theoretically correct cohort study
is impracticable, as an infinite number of these person-moments would be required
(as the requisite size of the cohort ! 1 as duration of follow-up ! 0).
Insofar as a cohorts follow-up inherently prospective in cohort time indeed is
to be taken as definitional to cohort studies (on etiogenesis), important corrections
in the rest of the concept the cohort fallacy need to be made. The cohort must
be seen to be (not the study cohort but) only the source cohort; and its follow-up
must be seen to be (not merely the means to identify the outcome events but) the
formation of the source base, from which to identify the first-stage case series and
to draw a suitable first-stage base series, etc. as outlined in chapter 5 above.
Upon these corrections, the concept of cohort study is none other than that
of a first-principles etiogenetic study in the framework of a cohort-type source
population. This cohort feature of the study, in itself, is of no quality consequence
to the study, which actually involves a dynamic study population forming the study
base (sect. 5.1). It is consequential only if the cohorts follow-up was/is used to
document the etiogenetic histories before the person-moments in the case and base
series, and if this is materially relevant to the validity of these histories.

6.2 Case-control Study: Its Essence Corrected

A cohort study, as it has been construed, is prone to be quite onerous to carry out.
In order that the cohorts follow-up yield a reasonably large number of the outcome
events, needed generally is the enrollment and at-enrollment documentation of
quite a large cohort and its follow-up for quite a long time. The main feature of a
cohort study, says the I.E.A. Dictionary of Epidemiology (specified in sect. 3.2),
is observation of large numbers over a long period (commonly years) : : :
Much more practical is a study that, instead of the enrollment and follow-up
of a cohort, begins with a case series; and especially, if it ends with that series.
Fritz Lickint in 1929 was the first to publish statistical evidence joining lung
cancer and cigarettes [ref.]. : : : His evidence was fairly simple, constituting what
epidemiologists today call a case series showing that lung cancer patients were
particularly likely to be smokers. : : : Franz H. Muller [in his] 1939 medical
dissertation [ref.] presents the worlds first controlled epidemiological study of the
tobacco-lung cancer relationship, : : : His analysis was what we today would call
a : : : case-control study, meaning that he compared : : : the smoking behavior of
66 6 Etiologic Studies Typology

lung cancer patients with that of a healthy control group of comparable age (ref.
below; italics ours). Muller did not have to enroll and document a large cohort, and
then to follow it for years, to get his documented 86 cases and 86 controls.
Reference: Proctor RN. The Nazi War on Cancer. Princeton: Princeton University Press,
1999; pp. 183, 194-96.

Relative to the cohort study (in its uncorrected form), the case-control study
(also in its uncorrected form; above) has, apart from its practicality, the theoretical
virtue that the causal histories are specified on a scale of (retrospective) time whose
T0 is the time of outcome on the scale of the relevant, etiogenetic time.
On the other hand, though, while the cohort study has a study base of sorts
(the series of person-moments of enrollment into the cohort), the case-control
study is devoid of any inherent referent for its result in the form of study base.
The results referent is a case group together with its control group (cf. above).
There thus is no inherent specification of a population-time, or of a series (finite,
enumerable) of person-moments, for which rates or rate ratios are documented.
And as there is no inherent conception of study base as the referent of the study
result, there is no base series to supplement the case series, to provide information
about the denominator inputs into the compared rates. The case-control study
is, thus, grossly at variance with the essentials of etiogenetic studies set forth in
sections 5.1 and 5.2. Seen to represent the reverse of the cohort study, it has
been alternatively termed the trohoc study (trohoc being the heteropalindrome
of cohort); and the profound fallacy in it may thus be termed the trohoc fallacy.
As the comparison of the index and reference rates of the outcomes occurrence
in a defined study base is replaced by comparison of cases with controls, which
comparison does not correspond to the causal contrast (cause present vs. its defined
alternative present, in the study base), strange things happen things that should
have been interpreted as indicating that the comparison is anomalous. One of these is
the phenomenon of overmatching, which is prone to occur in case-control studies
but not in cohort studies. Had Muller in his case-control study (above) matched
his controls to his cases according to some close correlate of smoking (such
as habitual match-carrying), the distribution of his controls by smoking history
would have been very similar to that of his cases, while no matching of non-
smokers to smokers, by whatever correlate of smoking, in a cohort study would
have the corresponding consequence in respect to their relative rates of subsequent
occurrence of lung cancer.
Closely related to this anomalous consequence of matching in case-control
studies is another one. Matching of the controls to the cases in these studies
has commonly been thought of as being a means to prevent confounding. But
prevention of confounding (distinct from control of it) can only be a matter of
seeing to it that the index and reference segments of the study base have suitably
balanced similar distributions according to the potential confounder an
extraneous determinant of the outcomes rate of occurrence in the reference segment
of the study base (sect. 4.3). And given the absence of study base in the concept, a
6.3 Cross-sectional Study: Its Essence Redefined 67

case-control study does not provide for detection of actual confounding nor even
for meaningful thinking about it (about this feature of the study base).
The case-control study thus requires a fundamental correction in terms of
introduction of the ever-necessary concept of study base (sect. 5.2). And secondary
to this, it requires the understanding that cases are not a group of people but a series
of person-moments with this study base as its referent by representing the entirety
of the cases occurring in it and that for the controls for it to be meaningful at
all, they actually need to constitute a fair sample of this study base, to form a base
series in this meaning (cf. sect. 5.2).
The way the cases of the outcome (illness) are identified in a case-control study
implies its corresponding catchment population as the studys source population.
This is the population (dynamic) whose membership is defined by the were-would
state of: were the outcome event now to occur, it would be caught by the studys
case ascertainment scheme (cf. sect. 5.1). This can be the way indirect of defining
the source population for the etiogenetic study that comes about as a result of the
necessary corrections introduction into the case-control study. For the source base
defined by the case ascertainment over a span of time the case ascertainment is
complete by definition, and the challenge is fair sampling of that source base for the
first-stage base series.
Alternatively, upon the correction, the source population is defined directly (as
the adult resident population of a particular metropolitan area, say; sect. 5.1), and
complete case ascertainment for the source base is now the challenging part (while
fair sampling of it is not).

6.3 Cross-sectional Study: Its Essence Redefined

Under Cross-sectional study, the I.E.A. dictionary (specified in sect. 3.2) gives
this:
A study that examines the relation between diseases (or other health-related characteristics)
and other variables of interest as they exist in a defined population at one particular time.
The presence or absence of disease and the presence or absence of the other variables (or,
if they are quantitative, their level) are documented in each member of the study population
or in a representative sample at one particular time. The relationship between a variable and
the disease can be examined (1) in terms of the prevalence of disease in different population
subgroups defined according to the presence or absence (or level) of the variables and (2)
in terms of the presence or absence (or level) of the variables in the diseases versus the
nondiseased. Note that disease prevalence rather than incidence is normally recorded in a
cross-sectional study [ref.]. The temporal sequence of cause and effect cannot necessarily
be determined in a cross-sectional study.

We suggest that, actually, the essence of the cross-sectional etiogenetic study is,
simply, this: The study base is formed as a single series of person-moments, as in a
cohort study, but outcomes associated with these person-moments are ascertained
with no follow-up the outcome being presence/absence of prevalent case of the
illness at those person-moments.
68 6 Etiologic Studies Typology

Cross-sectional study is rather commonly presented as one element in the triad

in which the other elements are cohort study and case-control study, but as the
least eminent element in this triad. As we define it (above), it is distinguished from
cohort study by the absence of follow-up (for outcome determinations); and the
absence of (distinctly) separate formations of the case group and the control
group distinguishes it from case-control study.
Absence of follow-up means that cross-sectional study is not longitudinal in
time (diachronic) in the meaning that cohort study inherently is (by the follow-up
in it); but it is just as amenable to consideration of longitudinal histories as cohort
study (and case-control study also) is. Thus the meaning of cross-sectional
here is not that the outcome-determinant relation is non-longitudinal (synchronic)
by definition, nor do the person-moments constituting the study base inherently
represent a cross-section of the source population at a particular point in calendar
time or on any other scale of time. In this meaning (cf. quote above), the term is a
plain misnomer.
Because of the absence of follow-up, the person-moments constituting the
study base are classified as of those very moments as for the occurrence then
presence/absence of the outcome at issue; and at issue thus inherently is, as we put
forward in the definition above, an outcome state (rather than event). Its occurrence
therefore is documented in terms of its rate of prevalence this, naturally again, in
relation to the etiogenetic factor, conditionally on extraneous determinants of the
outcomes prevalence (in the defined domain of study, represented by each of the
person-moments constituting the study base).
When the designed occurrence relation (incorporating the object parameter[s] for
the study) addresses prevalence (of a state-type outcome), a single series of person-
moments obviously is an admissible option for the study base. And in the framework
of this option, a subset of this series constitutes the case series for the study, while
the series in its totality can serve as the base series with sampling of that study
base an available option for the latter (sect. 5.2).
With this conception of the essence of the cross-sectional etiogenetic study
and this understanding of the implications of this essence, this least eminent
member of the purported triad of fundamental types of etiogenetic study is one
of major distinction: Different from the cohort study and the case-control
study, the definitional essence of the cross-sectional study embodies no need for
corrections upon the necessary correction of the definition of its essence.

6.4 Semi-cohort Study: Its Essence Corrected

As we noted in section 1.1, epidemiological practice has been particularly eminent

in occupational medicine, and so consequently also has been epidemiological
research for this discipline of medicine. The generic aim of this research has been
the identification of health hazards relatively common in, if not specific to, work
environments (as the basis for illness-preventive actions in occupational settings).
6.5 E Pluribus Unum 69

A common type of etiologic/etiogenetic study on incidence in this framework

has been one that is neither cohort nor case-control study but something that
might be termed semi-cohort study, as it involves an index cohort but no reference
cohort: A cohort of workers representing, at its enrollment, the index category of
the hypothesized etiogenetic determinant for a particular illness typical has been
exposure to a potentially carcinogenetic airborne substance in the etiogenesis of
lung cancer is recruited and suitably documented at enrollment. It is followed to
identify and document occurrences of the outcome say deaths from lung cancer
as proxies for clinical inceptions of the disease. A given number of these events is
observed in the course of the cohorts follow-up, and the corresponding expected
number in derived by applying the local or national general population rates of the
events occurrence, specific for gender and age, to the various gender-age strata
of the cohorts follow-up time. The thus-derived O=EO ratio is the standardized
mortality (or morbidity) ratio, SMR, result of the study (cf. sect. 4.3).
For this quite primitive study well recognized to be marred by the healthy
worker effect an improved counterpart free of that anomalous effect would
have as its source population the union of two cohorts: such an index cohort
together with a reference cohort of workers who at cohort T0 represent not only
the determinants reference category but also similar occupational exposure, pre-
and post-T0 , to extraneous causes of the outcome and, just as importantly, similar
distributions by other unquantifiable extraneous determinants of the outcomes rate
of occurrence (e.g., socio-economic status, if relevant). With the source base thus
formed as the union of the two cohorts population-time of follow-up the rest
of the elements of a first-principles etiogenetic study flow from it: the source base;
the first-stage case and base series; the reduced, second-stage counterparts of these;
etc. (ch. 5). The etiogenetic histories in this corrected counterpart of the semi-cohort
study are not defined as of cohort T0 any more than in the corrected counterpart
of the cohort full-cohort study.

6.5 E Pluribus Unum

As it is at present, design of an etiologic/etiogenetic study is thought fundamentally

to involve the choice among cohort study, case-control study, and cross-
sectional study in the main, with the understanding that there also are some other
fundamental options to consider (such as the semi-cohort study; sect. 6.4 above).
But implicit in the foregoing in this chapter is the idea that the principal
ones among these purported fundamental options the cohort study and the
case-control study are logically inadmissible (sects. 6.1, 6.2) and that the
cross-sectional study, while logically admissible, has a raison detre only when
the outcome phenomenon is a state (sect. 6.3), while it ordinarily is an event
(sects. 6.1, 6.2, 6.3).
In the spirit of the Latin words in seal of the U.S., implicit in the foregoing
is the idea that a wholesome unum/unity can be forged from the now-perceived
70 6 Etiologic Studies Typology

pluribus/variety of principal types of etiogenetic study; that a certain fundamen-

tal unity is inherent in the logically admissible types of population-level non-
experimental study on the etiogenesis of an illness; that this unity is to be understood
to be in the nature of those studies a priori and not as a consequence of judgements
and their consequent decisions in the designs of the studies; that one is to know
what features an etiogenetic study inherently has, before setting out to design its
particulars in this a-priori framework of inescapable essentials of it.
And to say it again, a first-principles non-experimental etiogenetic study inher-
ently has these core features:
1. There is an expressly defined study base, each of the person-moments in
it representing the designed occurrence relations domain (abstract) and also
consistency with whatever pragmatic admissibility criteria besides.
2. There is a case series identified from the study base, intended to be representative
of all of the cases of the outcome phenomenon occurring in the study base; that
is, having the study base as its referent in this meaning.
3. There is a base series as a fair sample of the study base; that is, a series having,
in this sense, the study base as its referent (the study base being, a priori, the
referent of the study result).
4. The elements (person-moments) in the case series and base series are docu-
mented in all relevant respects, that is, not only as to the series identification
(case series vs. base series) but also as to histories in respect to the determinant
at issue (generally on a scale with T0 the time of the person-moment in the series),
the confounders in the designed occurrence relation, and such modifiers of the
rate ratio as were designed into the occurrence relation defining the objects of the
study.
5. From these data is deduced (synthesized) the empirical rate-ratio for the out-
comes occurrence (in the study base), index category versus reference category,
conditional on the potential confounders, specifically the confounder-conditional
rate ratio in the meaning of the observed-to-expected ratio that determines the
empirical value for the factor-conditional etiogenetic proportion.

6.6 E Unum Pluribus

Even though the essence of etiologic/etiogenetic type of causation (sect. 3.1)

logically implies a fundamental unitarity/singularity in the nature of all first-
principles etiogenetic studies (sect. 6.5 above) a set of features that an etiogenetic
study should be understood to have a priori and not as results of ad-hoc decisions in
study design there nevertheless remains a multitude of genuine design options in
this a-priori framework of unitarity.
These options do not imply a typology of the studies as studies per se (`a la
cohort study, etc.) but options in the design in respect to only particular aspects
of the fundamentally singular type of etiogenetic study. And the genuine options
6.6 E Unum Pluribus 71

in any given aspect of an etiogenetic represent logical alternatives to each other

as would be cohort study versus dynamic-population study, cross-sectional study
versus longitudinal study, and case-control study vs. cases-alone study, were one
to continue classification of studies themselves according to select aspects of them,
considered in isolation. (A car with automatic transmission is not classified as and
said to be an automatic car; and a car with four-wheel drive is not seen to be an
alternative to it. Its logical alternative is understood to be a car with a non-automatic
transmission; but even more is understood, namely that at issue really is not types
of car but of transmission in cars.)
Examples of the design options in etiogenetic studies follow, some of them
harkening back to the studys objects design, others pertaining to its methods
design:
Outcome: event versus state (e.g., clinical inception vs. subclinical presence)
Outcome: any case versus select subtype of case
Source population: cohort-type versus dynamic
Source population: direct definition versus indirect definition (as catchment
population)
Source base: prospective versus retrospective in study time (in which T0 is the
time of the studys inception)
Base sampling: statistical (using sampling frame) versus using extraneous out-
comes
Base sampling: indiscriminate (unmatched) versus discriminate (matched or
otherwise stratified)
Etiogenetic histories: recorded before outcome versus ascertained after outcome
(in case and base series)
Confounding: prevention versus control (upon documentation)
Control of confounding: standardization versus modeling versus hybrid of these
(sect. 5.4)
Several of those items are, implicitly, specific to etiogenetic studies of the usual,
non-experimental type. As for experimental etiogenetic studies, which are of the
form of intervention studies, suffice it to note here that they inherently involve a
cohort-type study population, study base that is prospective in cohort and study
time, and prevention of confounding at cohort T0 at least; and that they obviously
lend themselves to identification of the case series directly from the study base and
to statistical sampling of this base for the base series as well as to pre-outcome
documentation of these two series in respect to the contrasted etiogenetic histories,
inter alia. See section 9.3.
Chapter 7
Etiologic Studies Objects Design

Abstract One of Immanuel Kants most seminal teachings was the idea that a
scientist does not learn by simply observing, reading Nature; that (s)he learns by
reading into Nature the terms in which (s)he thinks about Nature. It is in these terms
that (s)he, in a study, interrogates Nature seeks tentative answers to questions
(s)he poses.
In respect to population-level etiogenetic research this means that, preparatory
to a study, the researchers think, deeply, about the rate of occurrence of a health
phenomenon the event of a cancer becoming clinically manifest, say in a
particular domain initially defined by a range of age alone, perhaps. And they
think about the phenomenons rate of occurrence in causal relation to something in
the peoples past their diets antioxidant content, perhaps.
With such a sketchy point of departure, the investigators need to proceed to refine
their idea, in that example the particular meanings they elect to associate with the
generic terms in: the (rate of) occurrence of cancer in causal relation to histories
in respect to their diets antioxidant content, this in a particular domain of the
occurrence. They need to decide whether their thinking in this is specific to a
particular type of cancer and a particular type of antioxidant, or whether it is generic
in one or both of these respects. They need to specify their thinking in respect to the
range of time for the etiogenetic role of dietary antioxidants, retrospectively as of the
outcome event (its occurrence/non-occurrence). And they need to decide whether
the domain of the study is to be age-restricted and one of no previous clinically
manifest cancer, for example.
Upon decisions like these, the investigators go on to design a statistical model
(log-linear) for the outcomes occurrence in the defined domain, with determinants
in the form of statistical variates (Xs). The parameters of Nature constituting the
objects of study are imbedded in this model, the design of this model amounting to
the studys objects design.
A future with examples like this is envisioned in this chapter. As of now, a studys
objects design is not even in the common vocabulary of epidemiological research.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 73

DOI 10.1007/978-94-007-4537-7 7, Springer ScienceCBusiness Media Dordrecht 2012
74 7 Etiologic Studies Objects Design

7.1 Some Considerations in the Design

In epidemiology in the meaning of a discipline of research (sect. 1.3), and in the

thinking of epidemiological researchers likewise, there naturally is much concern
for the methodology of the studies, of etiologic/etiogenetic studies in particular
for the theory of methods design of these studies.
By contrast, and very unjustifiably, objects design has not yet become a topic,
even, in the theory of etiogenetic research, much less one perceived to also require
a theoretical framework. In line with this, while editors of epidemiological journals,
like those of other medical journals, expect any study report to include a section
on Methods, they generally expect there to be only an Introduction rather than a
section on Objects immediately in front of this.
As it thus is, the objects of any published etiogenetic study need to be inferred
from the form of the result (distinct from the empirical content of that form in
the result). For to the extent that the design of an etiogenetic study actually does
involve or is preceded by design of the objects of study, this determines, to the
corresponding extent, the form of the result of the study.
Akin to the a-priori singularity in the methodology of etiogenetic studies
(sect. 6.5), the object of an etiogenetic study also has a set of generic elements as
an a-priori given. At issue inherently is the occurrence of a defined phenomenon
of health in a defined domain (human), that phenomenons/outcomes rate of
occurrence (momentary, generally incidence density) in that domain in relation to
a particular determinant (retrospective) of that rate, its index category or categories
relative to its reference category, this relation in terms of rate ratio(s), conditionally
on certain extraneous determinants of the outcomes (rate of) occurrence for causal
interpretability of it.
So, the first component topic in an etiogenetic studys objects design naturally
is design of the actual outcome phenomenon per se, conceptually, usually in
the context of its conceptual approximation specified a priori. For example, if the
prompting for the study is that evidence from previous research has pointed to the
possibility that diets relatively poor in a given antioxidant are etiologic/etiogenetic
to a particular type of cancer, while perhaps not to other types of cancer, the first-
order object-design question is, whether the outcome phenomenon should be taken
to be that cancer, specifically, or whether it should be cancer unspecified (cf. sect.
2.2).
Analogously for the (potentially) etiogenetic factor. So if previous research
suggests that paucity of a particular antioxidant in the diet is oncogenetic, while
some others appear not to be, the corresponding object-design question is, whether
the etiogenetic histories should be specific to this particular antioxidant or whether
they should, instead, address a suitable overall measure of the diets antioxidant
content (incl. that from dietary supplements; cf. sect. 2.2).
With the outcome and etiogenetic factor designed, the occurrence relations
domain that of the outcomes occurrence, as it will be addressed in the study
becomes a topic in the studys object design. So if the study is about diet deficient
7.2 Some Principles of the Design 75

in a particular antioxidant, or antioxidants overall, in the etiogenesis of a particular

cancer, or cancers overall, the domain design involves consideration of whether
childhood should be involved in the domain; and if not, whether the domain
should involve restrictions in the adult range of age and in some other respects,
such as freedom from any previous overt case of the cancer, or from any cancer,
or, even, from other extraneous but potentially oxidation-caused illnesses (notably
atheromatotic ones).
With not only the outcome and (potentially) etiogenetic factor but also the study
objects domain specified, the temporal aspect of the occurrence relation becomes
the topic in the studys object design. The question is, where in time retrospective
from the time of the outcome, the occurrence/non-occurrence of the phenomenon
is the pathogenetic process whose etiogenesis is at issue in the study. Does it take
place in, and are the etiogenetically relevant histories correspondingly about, diets
prevailing in, say, the last 5 years before the outcome? or are they to be about, say,
the first three or four decades of life (with the study domain perhaps about the sixth
or seventh decade of life)?
The generally most challenging question remains: what are all the extraneous
determinants of the outcomes (rate of) occurrence, the distributions of which could
confound the study base and, in this case, would call for focus on the occurrence
relation conditional on these?
Decisions concerning the study objects design in the particular example of
studying dietary antioxidants in the etiogenesis of cancer obviously have great
bearing on the studys methods design and, ultimately, on the meaning of the studys
result, the conceptual form of which will be that of the designed object of study (cf.
above).

7.2 Some Principles of the Design

The designed form of the object of an etiologic/etiogenetic study represents

particulars in the framework of the generic form that it, and hence the study result,
is to have a priori (sect. 7.1 above). The result of the study will be about the outcome
phenomenons rate of occurrence in the study base, its index segment(s) versus its
reference segment, each person-moment in the study base satisfying the criteria
for the designed domain of the objects of the study (parameters in the designed
occurrence relation) and the empirical rate-ratio being conditional on all of the
designed potential confounders.
The result will inherently be quantitative and the magnitude of the empirical
rate-ratio matters even when the study is one of hypothesis-testing rather than
quantification. And the result will be, inherently, descriptive of the particularistic
counterpart, in the study base, of the occurrence relation that is designed for its
abstract domain.
The intent in the study objects design most notably in the way the outcome-
antecedent relation is made conditional on extraneous determinants of the out-
76 7 Etiologic Studies Objects Design

comes (rate of) occurrence is that the study result the empirical occurrence
relation will be informative about the causal connection at issue. In reality,
however, faulty objects design (just as faulty methods design) can make the study
result misinformative about the causal connection, about its strength in meaningful
terms in a given direction, even apart from the potential incompleteness of its
conditionality on all of the potential confounders.
In what follows, we address the principles of an etiogenetic studys objects
design in the contexts of two generic examples, very different in kind.

The first example here is about something that we already touched upon in
section 2.4 explicitly and in section 7.1 implicitly, namely the etiogenesis of prostate
cancer. In section 2.4 we quoted from a review of studies there had been
several cohort and case-control studies but no experimental ones on dietary
fat in the etiogenesis of this cancer, and here we consider one of each of these
two purportedly principal types of etiogenetic study (ch. 6). To these we add a
subsequent experimental study on the same factor but in relation to a different
cancer. Besides, as for diet deficient in the antioxidant selenium in the etiogenesis of
prostate cancer, we examine another case-control study and also the experimental
study for which it was a partial prompting (there having been no other case-control
studies nor any cohort studies before the trial). These five studies are:
1. Shuurman AG et alii. Association of energy and fat intake with prostate carci-
noma risk. Results from the Netherlands cohort study. Cancer 1999; 86: 1019-27.
2. Hayes RB et alii. Dietary factors and risks for prostate cancer among blacks
and whites in the United States. Cancer Epidemiology, Biomarkers & Prevention
1999; 8: 25-34.
3. Beresford SAA et alii. Low-fat dietary pattern and risk of colorectal cancer. The
Womens Health Initiative Dietary Modification Trial. JAMA 2006; 295: 643-54.
4. Yoshizawa K et alii. Study of prediagnostic selenium level in toenails and the
risk of advanced prostate cancer. Journal of the National Cancer Institute 1998;
90: 1219-24.
5. Lippman SM et alii. Effects of selenium and vitamin E on risk of prostate
cancer and other cancers. The Selenium and Vitamin E Cancer Prevention Trial
(SELECT). JAMA 2009; 301: 39-51.
For broadest orientation here, it needs to be appreciated that, in practically
relevant terms, prostate cancer (or any cancer, for that matter) begins with the
inception of sickness from it (once the cancer becomes overt clinical in its
progression), and only this event is subject to the formation of the case series in
an etiogenetic study. Any study of the etiogenesis of prostate cancer thus must
be understood to actually be about the etiogenesis of (the event of) the onset of
symptoms and/or signs of it. All five of those studies addressed cancer in the
meaning of the event of its clinical inception.
Related to this, and of equally fundamental relevance to the objects design of
a study on the etiogenesis of cancer is a duality of phases in the development
of symptomatic cancer: the first, pathogenetic phase of this (of pathologic changes
7.2 Some Principles of the Design 77

culminating in the inception of the cancer) is to be distinguished from the second,

progression phase of it (during the cancers latent presence). An etiogenetic
influence on the pathogenetic phase is termed initiation of the cancer, its counterpart
in the progression phase being termed promotion of the cancer of its progression.
A designed occurrence relation addressing one of these does not thereby also
address the other; that is, evidence about one of these (from a given study) is not
evidence for or against the other when studying fat-rich or antioxidant-poor diet in
the etiogenesis of clinically manifest prostate (or any other) cancer.
The objects design thus needs to be directed, expressly, to the study of initiation
or promotion or both, notably because the healthcare implications are very different
according to which one of the two phases of the overt cancers etiogenesis is at
issue. Initiation occurs, and initiators thus are to be avoided, in the earliest decades
(three or four, say) of life already, while promotion occurs and promoters are to be
avoided in middle age and later (as of the fifth decade, say).
This means the need to design the study objects domain, and also the outcome-
determinant temporal relation in the object, in accordance with whether at issue is
initiation or promotion or both in the etiogenesis of the overt form of the cancer
(cf. sect. 7.1 above).
All five of those example studies were ones of etiogenetic hypothesis-testing, as
is quite characteristic of epidemiological research in general (as the research tends
to be slow to produce even qualitative knowledge on what it addresses; ch. 2). So,
of most fundamental significance to their objects designs was the broadest nature
of the hypothesis. Three questions about the hypothesis concerning the etiogenesis
of overt cancer are of particular relevance: Is it, as discussed above, about initiation
or about promotion of the development of the cancers, or both? Is it, as discussed
in section 7.1 above, specific to a particular type of cancer or about cancers quite
generally? and, Is it, as also discussed in section 7.1, specific to a particular factor
in a given class of them (and its defined alternative) or about a whole class of agents
without distinctions among them?
Even though all five of those studies were, as noted above, about the etiogenesis
of the inception of the symptomatic stage in the progression of the cancer at issue,
none of them was specific, in the study report, on whether the study was about
initiation or about promotion of the malignant process, or about both of these, per
the nature of the hypothesis.
The determinant histories in those studies, as of the time of outcome, were only
about the last few years (rather than decades), as though tested was a hypothesis
specifically about promotion of the cancer. Yet, the antioxidant hypothesis at least is
expressly about the initiation of cancer (about diet deficient in antioxidants allowing
unopposed accumulation of the oxidative damage to the genome that leads to the
development of activated oncogene together with inactivated suppressor gene and,
thereby, to initiation of the cancer).
Etiogenetic studies that are insufficiently longitudinal in (the retrospective
depth of) their objects are, of course, relatively practicable to carry out; but this
flaw in their objects design can be serious to the point of making the study result
misleading. This problem presumably attends the statement in the PDQ Cancer
78 7 Etiologic Studies Objects Design

Prevention Overview web page of the U.S. National Cancer Institute, that The
results of the [SELECT; ref. 5 above] indicated that taking daily selenium or vitamin
E or both did not reduce the incidence of prostate cancer compared with placebo
[ref.]. For recruited into that trial were men in their 50s of age or older (rather than
young children) and their median follow-up was 5.46 years (rather than decades).
Thus, by its nature that trial actually was solely about the cancers promotion, while
the hypothesis presumably was about its initiation.
The principle that thus was violated in all five of those example studies is this:
A meaningfully designed object for an etiogenetic hypothesis-testing study may
be infeasible to study (validly); and when it is, it should not be studied through
ersatz objects that actually are meaningless in respect to the biological essence of
hypothesis. The problem is much worse than futility: it is risking misunderstanding
of the study result, not understanding that it is meaningless in respect to the actual
hypothesis. It should be understood that some objects of desirable and meaningful
epidemiological knowledge are not subject to practicable population-level study;
that in respect to them it is necessary to settle for such insights and population-
level surmises as can be derived from more basic epidemiological research (cf.
sect. 1.2).
Apart from their temporal aspects, the causal contrasts in those five studies, from
# 1 to # 5 respectively, were these: a given level of fat intake (kcal/day) vs. another,
also conditionally on total energy intake; a given level of foods high in animal fat
(grams/day) vs. another, and the same in terms of such foods as a proportion of
total energy intake; reductions in fat together with increase in vegetables, fruits, and
grains vs. continuation of usual diet; a given level of toenail selenium content vs.
another, also conditionally on calcium intake and lycopene intake; and a given level
of supplementary antioxidant intake vs. placebo intake.
As in studies # 1 and # 2, meaningful quantification of fat intake indeed is
conditional on total energy intake, perhaps expressed as the proportion of the
latter. But such quantification remains meaningless except when at issue actually is
conditionality on or proportion of total energy intake from fats and, isocalorically,
specified alternative nutrients (carbohydrates, most notably), with total energy
intake a possible added element in the designed occurrence relation. The contrast in
# 3 is a bit of an approximation to what thus is needed, while that in # 4 is sensible,
though only isocalorically. Concerning diet supplementation with antioxidants, # 5
is impeccable as for the meaning of the factor contrast (when not considering its
temporal aspects).
The principle that in this aspect of those example studies object designs was
incompletely heeded is this: While the causal (index) histories in an etiogenetic
study are to be explicitly defined and meaningful as such, meaningful explicit
definitions also are to be given to the alternative(s) to these. The proper concept
of the alternative antecedent the reference history in an etiogenetic contrast is
not simply absence of the index history at issue.
Concerning the determinant contrast, the example studies considered here raise
this question: Insofar as it is reasonable and indeed preferable to study the effect
(etiogenetic) of total fat content of diet, without distinctions between, say, fats from
7.2 Some Principles of the Design 79

meats and dairy products as components of this, is it not better, by analogy, to study
total antioxidant content of diet instead of studying particular components of this?
(Cf. sects. 2.2, 7.1.)
And in the same vein, as the fat and antioxidant contents of diet are hypothesized
to be etiogenetic to various types of cancer, the question that arises is: Shouldnt the
object of study focus on the cancers unspecified instead of focusing on a given one
of them in any given study? (Cf. sects. 2.2, 7.1.)
The principle bearing on answering these two questions is this: Uncalled-
for distinction-making among potentially etiogenetic factors and among health
outcomes conduces to uncalled-for complexity and its consequent confusion and
retardation of progress, and it is, therefore, to be avoided. In those five example
studies there generally wasnt even a practical, study-efficiency reason to focus on
particular antioxidants or particular cancers instead of composites of these. And,
by the way, in studies of smoking in the etiology of cancer, progress would have
been even slower than it actually was (sect. 2.7) had distinctions been made among
particular brands of cigarette, while the initial focus on cancers of the airways was
obviously well-justified (while concern for any cancer, without distinctions, would
have further retarded the progress).
A technical problem that attends the focus on a component in a family of
conceptually related factors of etiogenetic potential is the need to control all of
the others as potential confounders in any non-experimental study on the singled-
out component. In the case-control study on selenium level (as measured from
toenails) in the etiology of prostate cancer (# 4 above), lycopene intake but intake
of no other antioxidant was controlled, and the operational meaning of this one type
of extraneous antioxidant intake (and calcium intake too) was left unspecified.
The principle that thus comes to focus is this: In non-experimental testing of an
etiogenetic hypothesis, histories of potential confounders need to be conceptualized
in truly meaningful terms and accurately documented in these terms (for full control
of them). When this is not possible, the object of study needs to be modified to
eliminate such a problem or, the project is to be abandoned altogether. Study of the
aggregate intake of antioxidants, without focus on any given one of the components
in this, is free of confounding by intakes of extraneous antioxidants. (Different from
potential confounders, etiogenetic hypothesis-testing does allow some fuzziness in
the histories involved in the causal contrast[s].)
While an etiogenetic study is about the rate of the outcomes occurrence in causal
relation to the etiogenetic determinant of this, ultimately in terms of the rate ratio for
the index category versus reference category contrast(s), those five example studies
were quite variable in what their respective results were said to be about. Reported
from the cohort study (# 1) was rate ratio. From the two case-control studies
(# 2 and # 4) the results were said to be odds ratios, and from the two experimental
studies (# 3 and # 5) reported were hazard ratios.
The principle pertaining to that variety is this: When the health phenomenon at
issue in an etiogenetic study is an event (rather than a state) and the etiogenesis at
issue is not very acute as was the case in all five of those studies the proper
comparative measure dictating the objects of study is rate ratio in the singular
80 7 Etiologic Studies Objects Design

meaning of incidence density ratio (in a defined abstract domain); and the empirical
counterpart of this from any study thus also properly is an IDR, this in reference to
the population-time constituting the study base and the contrast of its index segment
against its reference segment defined as of the person-moments constituting the
study base (and not as of the persons entries into the study base).

While the examples above were about diet excessive in fats or deficient in
antioxidants in the etiogenesis of overt, symptomatic cancer, our added example
is a related one and equally important: lack of screening really lack of screening-
associated presymptomatic treatments in the etiogenesis of fatality from a cancer.
Illustrative of the state of objects design in this research is a review of the results
of this research in respect to breast cancer:
Demissie K et alii. Empirical comparison of the results of randomized trials and case-control
studies in evaluating the effectiveness of screening mammography. Journal of Clinical
Epidemiology 1998; 51: 81-91.

The idea was to derive summary risk estimates separately from each of the two
types of study, for comparison in the spirit that, in principle, the same parameter was
being quantified in each of the studies, even if the results of the two types of study
may differ on account of incomplete adherence to the intervention categories in
the randomized trials.
The authors repeated the prevailing idea that The gold standard for evaluating
screening programs is the randomized controlled trial (RCT) (sect. 2.6), adding
that Case-control studies are easier to perform but their role in this area is
controversial. They gave several references to articles (and to one book) in which
the principles of the case-control studies on screening have been delineated.
Before getting to the particulars of that review, we give our introduction into
the principles concerning the respective generic types of object that actually are
addressed in those two types of study on the intended consequence of screening for
a cancer.
That gold standard RCT of epidemiological researchers and policy advocates
regarding screening for a cancer is one in which seemingly healthy persons are
randomly assigned to one of the two arms of the trial. One of these commonly is
simply that of usual care (undefined, variable) in respect to the cancer, while in
the other arm the experimental one of the study the care is defined to the
extent that the participants are invited, or actually scheduled, to undergo periodic
screenings for the cancer in terms of an initial test, with positive result of this
test prompting referral to clinical care (for further diagnostics, undefined, and
possible early treatment, undefined, upon rule-in diagnosis, undefined). The trial
design involves specification of the time interval between successive rounds of the
testing and also the number of these rounds, together with the duration of follow-up
(as of randomization). The concern is to document the mortality from the cancer
over the duration of the follow-up in each of the two arms of the trial with a
view to documentation of the proportional reduction in that mortality (due to the
screening/testing).
7.2 Some Principles of the Design 81

The most eminent example of these gold standard trials now is the most recent
one, the National Lung Screening Trial in the U.S. Pertaining to it, the National
Cancer Institute of the U.S. released, on 28 October 2010, a statement saying,
among other things, that The primary goal of the NLST was to determine whether
[sic] three annual screenings with low-dose helical computerized tomography
(LDCT) reduces mortality from lung cancer relative to screening with chest x-ray
(CXR). The trial had been designed to have 90% statistical power for detecting 20%
reduction in such mortality. The essential result prompting the trials termination
was that in the LDCT and CXR arms of the trial the respective rates of death
from lung cancer over the 58 years of follow-up since entry into the trial were
245.7 and 308.3 per 100,000 person-years, and this was taken to imply
a (308.3  245.7)/308.3 D 20.3% reduction in lung-cancer mortality by LDCT
relative to CXR (Table 3 in the statement). The relative deficit in mortality from
lung cancer in the LDCT arm exceeded that expected by chance, even allowing for
the multiple analyses conducted during the course of the trial, presumably taken to
imply that the reduction was not due to chance.
Some editing is called for here. First, the mortality from lung cancer in the LDCT
arm having been 20.3% lower than in the CXR arm does not mean that LDCT (really
its associated treatments) reduced the mortality by 20.3% relative to what it would
have been in the people in this arm of the trial had CXR been used instead of LCDT.
And second, even though the trial was designed for detecting 20% reduction in such
mortality (with 90% probability) and was stopped with the idea that the mission
had been accomplished, in truth there was no such detection, nor could there have
been. Reduction is an effect, and effect is not a phenomenon, subject to observation
and, hence, detection and quantification (sect. 4.2). The reduction might have been
10%, or perhaps 40%; its specific magnitude is unknowable, that 20.3% being a
composite of the reduction together with the workings of bias and chance. (The
95% confidence interval, not given in the NCI statement, is 631%.)
The statistical power calculation alluded to in that statement implies that
associated with three annual screenings with [LDCT], instead of CXR, there is, if
any, a particular degree of proportional reduction in mortality from lung cancer a
constant proportion over time as of the initiation of the screening, fully operative in
whatever duration of follow-up.
Related to this is our first principle concerning the mortality reduction in a
trial such as the NLST: The proportional reduction in mortality from the cancer
is nothing like a constant over time from the beginning of the screening (for the
generally short duration of it) to the end of the follow-up (for an arbitrary duration
of it). It thus is logically inadmissible to quantify the reduction by pooling the
experience across the entire duration of the follow-up. The proper concern in a trial
like this is to address the incidence density of death from the cancer as a function of
time since the initiation of the screening. And that function is, of course, different
for different durations of the screening.
This leads to our related, second principle: Reduction in mortality from a
cancer subsequent to screening can occur only if the cancers treatments under
the screening those early treatments are more commonly curative than those
82 7 Etiologic Studies Objects Design

in the absence of screening. In fact, attainment of enhanced curability by earlier

treatments is the very idea in screening for a cancer. Thus the parameter of Nature
that should be viewed as the proper object of any study on the intended consequence
of screening for a cancer is one that meaningfully quantifies the gain in the cancers
curability rate when screening-associated early treatments replace the treatments
on already symptomatic cases in the absence of screening. This is a proportion of
the cases of the cancer that are fatal in the absence of screening, the proportion of
these otherwise fatal cases that are curable by screening-associated early treatments.
Whatever may be the adopted design parameters for a trial such as the NLST in
respect to the number of rounds of screening and the duration of follow-up, the
proportional reduction in mortality as it is addressed in them (above) is prone to be
much smaller than the curability gain just defined. But with sufficiently long-term
screening, and focus on the appropriate segment of the follow-up, the magnitude of
the curability gain can be studied in terms of a trial contrasting screening with no
screening (ch. 11) insofar as the necessary long-term adherence to the contrasted
regimens can be effected.
This, in turn, leads to our third principle: A quantitatively meaningful etiogenetic
study on death from a cancer, with lack of screening for it the etiogenetic factor,
can be based on a case and base series from the relevant segment of follow-up
in a screening trial with sufficiently long-term screening (and close adherence to
the schedule). In the case series, the relevant history is about whether the person
was under screening at the time of the cancers detection (by virtue of being in the
screening arm of the trial, irrespective of whether the diagnosis was derived on the
prompting of a positive result of the initial test at issue or due to symptoms emerging
between the scheduled tests). The corresponding histories in the base series involve
some subtlety in respect to the corresponding times to which they refer, as set forth
in section 11.5. The rate ratio (incidence-density ratio) from this study translates to
(an empirical value for) the curability gain from early treatment.

With these principles as the background we now return to that review meta-
analysis by Demissie et alii (ref. above).
That reports Table 1 was entitled Design, population and characteristics of
screening approach of randomized control [sic] trials. As we noted above, critically
important determinants of the magnitude of such mortality reduction as is addressed
in these screening trials are the duration of the screening and the duration of the
follow-up (as of randomization) these, naturally, in addition to the nature of the
entire regimen to pursue early diagnosis about the cancer, including the time interval
between successive rounds of this regimens application. That table reveals no
specifics about the diagnostic regimens (not even their initial tests and the definitions
of their positive results). The number of rounds of the screening is specified for
each study (the range is from 1 to 7 among the trials), and specified also is the
time interval between the rounds of screening (it was 1233 months). It thus can
be deduced that the durations of the screening ranged from 2 to 12 years. The
duration of follow-up is specified, ranging from 7 to 16 years. Given the trials
great variability in such result-relevant aspects of their respective methodologies, no
7.3 The Result of the Design 83

shared measure of mortality reduction much less the practice-relevant parameter

of Nature, the curability gain (regimen-specific) was addressed as the object of
study.
There thus was no justification for deriving a single summary risk estimate
from these gold standard studies, trials that actually quantified a variety of highly
arbitrary measures of mortality reduction and with adherence to the screenings as
low as 61% at one of the baseline rounds and 4% in one of the scheduled repeat
rounds of screening. (Cf. sect. 2.6 re screening.)
The case-control studies (on lack of screening in the etiogenesis of death from
breast cancer) had been conducted in settings into which routine availability of
screening mammography had been introduced. Of particular note is the statement
under Methods that Exposure was defined as participation in one or more screening
examinations from the beginning of the screening program up to the time of
diagnosis of breast cancer in the case and the comparable time in the control. (Cf.
third principle above).
Both trials (RCTs) and etiogenetic studies (non-experimental) on screening for
a cancer need to address the same parameter of Nature characterizing the extent
to which screening for a cancer has its intended consequence; they both need to
quantify the gain in the cancers curability afforded by screening-provided earlier
detections of (earlier rule-in diagnoses about) the cancer and the respective
regimens of screening (pursuits of early diagnosis, and not merely the initial tests
in it) as well as the treatments need to be both defined and comparable (i.e.,
practically the same).
But: no meaningful quantity was addressed in any of the studies reviewed by
Demissie et alii and, hence, in this review of those studies.
Principles of the study of the meaningfully construed intended consequence of
screening for a cancer the gain in the cancers rate of curability we address
further in chapter 11.

7.3 The Result of the Design

Design of the objects of an etiologic/etiogenetic study the parameters of Nature

to be addressed in the study is the result of designing an occurrence relation
for the outcome phenomenon together with the domain for this (sect. 4.4). For, the
studys object parameters are imbedded in this adopted model, generally along with
some nuisance parameters having to do with making the outcomes occurrence in
relation to the etiogenetic determinant suitably conditional to imply (as best can be
judged) that relations causal nature (sect. 4.4).
It deserves emphasis that this model, while medical in its substance, is statistical
in form. Medically one thinks about the outcomes rate of occurrence in the
domain, and specifically about the way in which this rate depends, causally, on
the history in respect to the (potentially) etiogenetic factor at issue; and the rate-
ratios modifiers along with potential confounders of the causal relation also enter
84 7 Etiologic Studies Objects Design

into this medical thinking. In the designed occurrence relation, the outcomes rate of
occurrence is represented by a chosen occurrence measure, ordinarily the logarithm
of the numerical element in the outcome events incidence density (sect. 4.4). The
magnitude of this measure is formulated as a function (linear) of a set of statistical
variates (numerical) adopted ad hoc as representations of the person characterizers
on which this magnitude depends according to the design of the occurrence
relation that defines the objects the object parameters of the study.
We emphasize this because it represents a major departure from the prevailing
thinking and practice in etiogenetic research. As it is, quite sharp a distinction is
being made between study design and the studys data analysis, so that the study
design does not define the parameters the objects of study the empirical values of
which are to be derived by suitable synthesis (sic) of the data by suitable fitting of
(the operational counterpart of) the designed occurrence relation to the data. We thus
emphasize the need to integrate the medical and statistical aspects of the research
just as the material and mathematical aspects are integrated in physics (in, e.g.,
Maxwells equations on electromagnetic radiation).
We thus emphasize the need to think of etiogenetic research as inquiry into the
magnitudes of defined statistical parameters of Nature, whether qualitatively (as to
existence of deviations from their null values) or quantitatively (as to the magnitudes
of those deviations).

7.4 Implications of the Design

The here strongly advocated explicit, up-front design of the objects of study, and
this, specifically, as a matter of design all the way to the statistical form of the
occurrence relation implying the statistical objects of (the statistical parameters
for) study, has as its most straightforward implication the conceptual form of the
result that the result is empirical rate-ratio as a function of certain modifiers of its
magnitude, as we already have noted.
A related, minor implication is that an etiogenetic studys objects design implic-
itly defines the objective of the study: to produce evidence about the magnitudes of
the parameters that were designed to constitute the objects of the study.
Very important is the objects designs bearing on the studys methods design.
Suffice it here to refer back to sections 2.2 and 7.2 and, for example, the
discussions there of how seriously misleading has been all of the research
purportedly critically relevant randomized trials included on diets deficient in
their antioxidant content in the etiogenesis of cancers. Without due regard for
deliberate and thoughtful and also knowledge-driven objects design, these trials
methodologic routines have implied results in reference to very recent diets (or diet
supplementations) on the scale of etiogenetic time (retrospective as of the time of
the occurrence/non-occurrence of the cancers entering the clinical phase of its
development). Proper attention to objects design, we argued, would call for the
causal contrast to refer to the earliest decades of life, with the domain for the
7.4 Implications of the Design 85

outcomes occurrence representing rather late decades of life. And we noted, too,
that insofar as the nature of a meaningfully designed set of objects makes the study
methodologically impracticable, refraining from study is preferable to a study that
is inherently misleading on account of the form of its result.
Finally, what has here been said about objects design for etiogenetic studies
implies the need to stop classifying etiogenetic researchers as being either epi-
demiologists or biostatisticians, as this fosters disintegration of the medical and
statistical aspects of the research, while integration of these two really is called for
(cf. above). And once the necessary integration takes manifestation in the objects
design in terms of an occurrence relation with statistical variates and statistical
parameters, Statistical Methods vanishes from the study report while Objects Design
enters.
Chapter 8
Etiologic Studies Methods Design

Abstract Once the studys objects design has defined what the investigators are
concerned to shed light on the values of what parameters of Nature in what do-
main, conditionally on what extraneous determinants on the outcome phenomenons
rate of occurrence they can turn their attention to the studys methods design. In
it they define how they aim to measure those parameters values ultimately by
means of fitting the logistic counterpart of the objects-defining log-linear model for
the outcomes occurrence to the data (the realizations of Y, X1 , X2 , : : : ) on the
studys case and base series. The resulting empirical rate-ratio function translates
into the result for the etiogenetic proportion of interest (chap. 4 Abstract).
The methods design of an etiogenetic study proceeds in two distinct stages.
In the first stage, the objects-defining occurrence relation that was designed in
reference to Nature in the abstract is adapted to the actual study, translated into the
operationalized model. The independent variates in the logistic model to be fitted
are defined in reference to this counterpart of the theoretical model. The second
stage of the methods design defines the means by which empirical content of that
form is judged to be best obtained. While the essence of the study has an a-priori
definition, many particulars of it remain to be specified/designed (chap. 6 Abstract).
This second stage of the studys methods design is governed by the imperative
of validity assurance, descriptively, together with the desideratum of efficiency
optimization. Both of these, in turn, are governed by principles of the studies
methods design.
Apart from efficiency, a determinant of the studys informativeness is its size;
but this is not subject to principles-guided optimization, with one exception: when
a violation of principle is detected in the studys objects design or in its methods
design in respects other than study size (number of datapoints in the two series), the
optimal indeed, correct size of the study is zero.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 87

DOI 10.1007/978-94-007-4537-7 8, Springer ScienceCBusiness Media Dordrecht 2012
88 8 Etiologic Studies Methods Design

8.1 Some Considerations in the Design

The methods design of an etiologic/etiogenetic study should be understood to be

constrained by the methodologic implications of the objects design for the study.
After all, the studys methodology is the means to the studys preset end the
objective of attaining (new) evidence about the magnitudes of the studys object
parameters. And another overarching constraint, just as critical, is the need to
appreciate the singular essence of the study (sect. 6.5), even though this leaves
for consideration the menu of design topics and their corresponding design options
(sect. 6.6).
For the actual study, the elements in the designed occurrence relation need to
be given their respective operational definitions. For example, while the study
domain (abstract) generally is one of no previous occurrence of the outcome at
issue, the operational criteria for this need to be defined. And, as another example,
when the outcome at issue (as is typical) is the event of the illness making, in its
progression, the transition from a latent, preclinical stage to the overt, clinical stage,
the operational counterpart of this needs to be defined.
The statistical variates (Xs) in the designed occurrence relation are defined in
reference to various person characterizers in the abstract; but those variates need to
be replaced by their operational counterparts, ones in reference to the operational
counterparts of the conceptual determinants of the outcomes (rate of) occurrence
(in the abstract domain). For example and most notably, definition is needed for the
way the histories on the etiogenetic determinant are to be ascertained, preparatory
to these ascertainments and their translations into realizations of the corresponding
variates in the definition of the objects of study.
All of these considerations, and other similar ones also, are directly related to the
designed occurrence relation, including the potential confounders in it. But they
indeed are ones of the studys methodology and not ones of reformulation of the
objects of the study. One does not, in this research, study the operational, but
one necessarily is operational in studying the non-operational, the abstract. The
beginning of this, in etiogenetic research, is that something noumenal (causation)
is being studied in terms of matters phenomenal (sect. 4.2) empirical.
In the rest of the methodology, the first set of considerations generally relates to
the source population; and in this, the very first component consideration generally
should be whether it can be simply selected/defined, or whether, instead, it needs to
be operationally formed. For, a usable source population is one in which sufficiently
well-ascertainable histories in relevant respects most notably as to the etiogenetic
determinant at issue are associated with each person-moment in it; and it can be,
as in the Framingham Heart Study and the Nurses Health Study, that facts on the
source population (a cohort in each of these examples) need to be prospectively
observed and documented to make the histories for the case and base series of
subsequent etiogenetic studies ascertainable and, hence, documentable.
If this preparatory facts-recording (burdensome, time-consuming) is deemed
not to be needed, then the highest-priority considerations generally have to do
8.2 Some Principles of the Design 89

with the way in which the source population will be selected/defined. The chosen
definition may be direct; or it may be indirect, as the catchment population
secondary to the way in which cases of the outcome are identified (to which direct
definition is given).
Either way, an important methodological consideration is the operational defini-
tion of the illness at issue, notably as to whether it should be restricted to cases are in
their manifestations typical and also severe for reasons of helping to assure validity
of the study. This restriction in the operational definition of a case of the illness
facilitates complete case identification from a directly defined source population;
and it makes more concrete the source population with indirect definition of it,
thereby facilitating its fair sampling of the source base (for the first-stage base series;
sect. 5.2).
Among the orientational considerations also is this: Should the conditionality
of the outcomes rate of occurrence on potential confounders, called for by the
studys objects design, indeed be pursued in terms of documentation and control,
or should this be replaced by prevention of the confounding by redesign of
the occurrence relation or suitable design of the study base as representation of
the domain? For example, as smoking represents a generally poorly controllable
potential confounder, the solution may be restriction of study domain to life-long
non-smokers. As another example, confounding by the level of fever in the study
of aspirin use in the etiogenesis of Reyes syndrome in the domain of febrile
illness in childhood may be prevented by contrasting aspirin use with the use
of another antipyretic (acetaminophen, say). On the other hand, prevention of
confounding by socio-economic status in the study of etiogenesis by occupational
exposure to a toxin may be accomplished by forming the source population of
workers in such a way that the exposure at issue can be presumed to be, in this
population, uncorrelated with SES (cf. sect. 6.4). In experimental causal research,
the preeminent means of prevention of confounding, specific to T0 of cohort and
scientific time, is randomization, while prospective confounding may be prevented
by blinding (which may require placebo comparison).
These examples might suffice to convey the point that even though cohort,
case-control, and cross-sectional study do not represent defensible design
alternatives to each other in etiogenetic research (sect. 6.5), there really are
options very notable ones in the studys methods design beyond the operational
reformulation of the objects-defining occurrence relation together with its domain
(sect. 6.6).

8.2 Some Principles of the Design

The overarching principle in an etiologic/etiogenetic studys methods design

in the operationalization of the designed occurrence relation and in the design
of the study proper is the imperative to assure a reasonable degree of validity
90 8 Etiologic Studies Methods Design

for the study, complete validity generally being unattainable. Another, secondary
concern desideratum rather than an imperative is maximization of the studys
efficiency, within the limits of the attainable.
An etiogenetic studys validity can be, and needs to be, thought of in two stages.
The study is descriptively valid to the extent that the object parameters empirical
values in its result would converge to those of the corresponding parameters
of Nature (defined by the designed occurrence relation) were the study size to
approach infinity (so that chance imprecision of the empirical values would be fully
eliminated). Insofar as those parameters have locally (in the studys place and at
its time) specific values, then this amounts to definition of the studys local validity,
in descriptive terms. And the study is causally (etiogenetically) valid to the extent
that its result is not only descriptively valid but also free of confounding.
Lack of validity in an etiogenetic study is, most broadly, of one or more of three
types. It is constituted by selection bias and/or documentation bias taking away from
descriptive validity and, of course, by confounding bias as a potential added element
in the lack of causal validity.

Selection bias results from violation of one or both of two principles. One of these
principles is that commitment to the source base must be made independently of
even a hunch about what the result from the study base in it, specifically, would
be; and by the same token, the commitment must be adhered to independently of
the findings from what was designed to be the study base. The associated other
principle is that a study base (within the source base) must be the result of the
study subjects entries into and exits from it independently of the outcome itself.
This generally means entries and exits independently of precursors (prodromata) of
the outcome. However, for validity of etiogenetic study, actually required is only
that there not be differential dependence on precursors of the outcome between the
contrasted experiences.
Selection bias of that first kind is akin to result-dependent submission of the
study report for publication, and to the study reports result-dependent acceptance
for publication. Both of these violations of principle very common and knowingly
committed! conduce to publication bias in the aggregate of published studies
on any given (set of) object(s) of study, and this in turn makes for unavoidable
selection bias in the study base of any derivative study on the magnitudes of
the object parameter(s) (sect. 13.1). Publication bias could be eliminated by a simple
innovation in medical journalism (sect. 13.6).
Selection bias of that second kind is tantamount to the study base being distorted
by failures in the implementation of the protocol. By the same token, its prevention
is a matter of stringent adherence to the protocol in its implementation, so long as
the protocol itself is sound.

Documentation bias, in sharp contrast to selection bias, is not at all about the
study base per se. Whatever the study base is, the occurrence relation in it the
operationalized counterpart of what was designed in reference to the abstract
should be documented correctly, validly in this sense. In this documentation, bias
8.2 Some Principles of the Design 91

is a consequence of contravention of one or more of the three principles of it:

the production the two series (case and base series) must be valid for their shared
referent, the study base; these two series must be correctly documented in respect
to the person-characteristics represented by the realizations of the statistical variates
in the operationalized version of the designed occurrence relation; and the study
objects operationalized version must accord with the theoretical one.
With secondary definition of the source population (as the cases catchment
population), the case series inherently is complete for the source base and, hence,
for the study base, but valid sampling of the indirectly defined source population
(at whatever time) is a major challenge; and when the source population is defined
directly, there generally is a sampling frame for its valid sampling, but complete
(and valid incomplete) identification the cases occurring in the source base poses a
major challenge.
When direct definition is given to how the cases are identified, a first step in
making the catchment population more concrete (for more valid sampling of it) can
be basing it on canvassing all of the care facilities in which the cases are diagnosed
within a contiguous, large region (a metropolitan area, say) and restricting the cases
admissibility (and thereby the study base) to residents of that region.
More to the same effect is changing the operational definition of case to one that
is severe and clinically typical. For if the case is severe, the person is more likely to
seek medical attention; and if it is clinically in its symptoms and signs typical,
its recognition for what it is, is more likely.
With these arrangements for the development of the case series for an etiogenetic
study, the source base becomes quite concrete in its definition; in fact, it gets to be
indistinguishable from one in the context of the source population having been given
of the primary, direct definition. And if the source population is defined directly,
its associated challenge of valid case identification would be more manageable
by pursuing it in the manner just outlined for it in the context of the source
populations secondary, indirect definition. With these arrangements the otherwise
major duality constituted by the two ways of defining the source population becomes
moot.
Even though a concrete definition of the source population generally provides for
statistical sampling of the source population/base, it may be preferable, for validity
assurance, to sample it by the use of extraneous outcome events, to help assure
that the reported histories in the base series are of the same accuracy as in the
case and base series. Valid outcomes for this purpose are ones whose occurrence
is uninfluenced by the etiogenetic determinant at issue.
When sampling the base by suitable extraneous outcome events, two important
questions relate to this: How many types of such an outcome? and, How identified?
We hold that a good number generally is three, as a number larger than one allows
checking their postulated interchangeability under the premise that each of them
is suitable, while an unduly large number has two drawbacks: it takes away from
the informativeness in these interchangeability assessments, and it imposes undue
burdens to colleagues in their judgements about the tenability of the premise that
each of the events is suitable for the purpose.
92 8 Etiologic Studies Methods Design

The identification of the extraneous outcome events for the base series
generally all of them, as defined, occurring in the study base is to be, in
relevant respects, completely analogous to that for the case series: definition of the
admissible cases, canvassing all of the care facilities for them (sic), etc. (see above).
This makes (the common practice of) matching by hospital inadmissible whenever
the care facilities for the extraneous cases differ from those for the cases of interest.
The distribution of the case series (by a correlate of the etiogenetic history) can
be used as a (partial) guide for stratification of the sampling for the base series, the
other guides being the histories varied distributions and the samplings varied unit
costs across the strata. (See efficiency below.)
If the base sampling is stratified by person-characterizers other than what
already is involved in the designed (and operationalized) occurrence relation, the
stratification factors need to be added to those in the pre-designed occurrence
relation (i.e., the result is to be made conditional on these, too).

Confounding bias in the result of an etiogenetic study has two possible sources:
the designed theoretical occurrence relation may (and commonly does) fail to
make the objects of study sufficiently conditional on extraneous determinants of
the outcomes (rate of) occurrence; and the operationalized counterpart of this
conditioning may be deficient as a representation of that conditioning.
Insufficient conditionality on extraneous determinants in the theoretical occur-
rence relation is not inherently a matter of the set of accounted-for potential
confounders being incomplete. Another source of the problem is inadequate
conceptualization and hence inadequate representation of a determinant that
is accounted for. For example, history of smoking likely is, generally, quite
inadequately accounted for by inclusion, in the log-linear model, of a single term
for pack-years of cigarette smoking. The same is true of a single term for years of
education in conditioning for socio-economic status. And when use of aspirin has
been studied in the etiology of Reyes syndrome in the domain of febrile illness
in childhood, a major concern has been potential confounding by the level of the
fever, the meaning of which is obfuscated by the antipyretic effect of the aspirin
use.
When problems in the conceptualization and/or documentation of a confounder
(or in the modeling of its role as a determinant) are unsurmountable, making control
of confounding (characterizing the study base) impracticable, the need is to consider
possible attainability of prevention of confounding (from characterizing the study
base); see section 8.1 above.

Efficiency optimization, like validity assurance, for an etiogenetic study begins in

the studys objects design, as for the domain in the occurrence relation. For a start,
the domain must satisfy the most elementary requirements for the study objects
(parameters) studyability: the domain must be one in which the outcome of interest
occurs; and it must also be one in which the determinant of interest varies. Thus,
when studying the occurrence of deep-vein thrombosis (or its consequent pulmonary
embolism) in causal relation to recent use of an oral contraceptive, both of these
8.2 Some Principles of the Design 93

elementary requirements are satisfied only by women of childbearing age. Study

in a domain that does not satisfy these simply qualitative requirements would be
extremely inefficient. Its information yield in relation to its cost (monetary and
other) would be extremely low, as its informativeness would be nil.
The efficiency-optimal rate of occurrence of the outcome (in the domain of the
study) is a topic of some subtlety. If the cost of assembling (and documenting) a
given number of cases of the outcome is the same in a lower-rate domain as in a
higher-rate one, more efficient for hypothesis-testing is prone to be a study in the
lower-rate domain, given also that the distribution of the contrasted histories on the
determinant of concern is the same in these two domains. For where other causes
are less common, the rate-ratio for the cause at issue tends to be more pronounced
in its deviation from unity (as this cause is a more salient one in the relative absence
of others).
To wit, the role of in-utero exposure to mothers use of DES (diethyl stilbestrol,
to sustain pregnancy) in the etiogenesis of clear-cell carcinoma of the vagina was
evident from a mere one dozen cases of this disease because this exposure is a quite
unique cause of this very rare disease. If the disease were much more common, it
would have a high etiogenetic proportion for causes other than in-utero exposure to
DES; the proportion of the cases with this DES antecedent would be much lower;
the factor-conditional etiogenetic proportion for this DES exposure would be much
lower; and the rate-ratio pertaining to it would be much closer to unity. A dozen
cases of the disease would, thus, be much less informative about the existence of the
DES etiogenesis (when considered in conjunction with a suitable base series).
As for the contrasted histories variability in the study domain, the efficiency-
optimal distribution naturally is that of maximal variability near-equal frequencies
for the index and reference histories (with no other histories). Thus, a particular
choice of the domain is efficiency-enhancing if it means greater variability of the
etiogenetic histories, ceteris paribus.
The variability of the contrasted etiogenetic histories is subject to methodologic
enhancement as well. In particular, when a cohort is formed to serve as the source
population, the histories variability in the prospective study base should not be
determined solely by this cohorts source population (generally dynamic). Instead,
it should be enhanced by means of selective enrollments into the cohort with a view
to efficient variability of the histories at cohort T0 , aiming at either the two-point
design (accenting, equally, the histories extremes) or the three-point design (with
a third group in the middle).
In contravention of this principle, very notably, the cohort for the ever-so-eminent
Framingham Heart Study was selected as a representative sample of the targeted
segment of the towns population, the consequence being the very opposite of what
would have been efficient: accented got to be the middle instead of the extremes.
(Nature has no propensity to optimize the efficiency of epidemiological studies: it is
left for the investigators to pursue.)
When the decision has been taken to conduct an etiogenetic study in a given
setting, there commonly is an opportunity to enhance the studys informativeness
(its results precision) by increasing the size of the case series (by increasing the
94 8 Etiologic Studies Methods Design

size of the study base) and/or by increasing the size of the base series. This raises
the question about the efficiency-optimal relative sizes of these two series when
both of them can be expanded; and the answer is that optimal is the size ratio that
is the inverse of the square root of the unit-cost ratio between the two series. But if
the size of the case series is a given, the studys informativeness can be increased
only by increasing the size of the base series which does not provide appreciable
further increase once the base series already is, say, fivefold in size relative to the
case series. A large base series is prone to be cost-ineffective, inefficient.
Among the distributional determinants of an etiogenetic studys efficiency also is
the designed distribution of the base series across strata of the study base. It remains
commonplace to match the base series to the case series, meaning that it is chosen so
as to have identical distributions for the two series, by gender and age, for example;
but this is not efficiency-optimal (while irrelevant for validity). Efficiency-optimal
is sampling that is proportional to informativeness and inversely proportional to
the square root of unit-cost of the sampling; and a stratums informativeness is
proportional not only to the number of cases from it but also to the product of
the proportional sizes of the index and reference segments of the stratum-specific
segments of the study base.

8.3 Implications of the Design

Once the objects design for an etiologic/etiogenetic study has been followed by its
corresponding methods design, the investigators are ready to produce the study
protocol, which defines, in all relevant detail, the study methodology all the way
to the synthesis of the collected data into the study result together with its associated
measures of the imprecisions of the empirical values of the object parameters.
The study protocol has direct bearing on the study report, the core of which is
evidence about the object parameters in dual terms: the study result together with its
associated measures of imprecision for one, and the genesis of these numerics for
another with description of the salient features of the protocol the core input to the
latter, supplemented by information about deviations from it in the implementation
of the designed methodology.
The study protocol commonly is to be supplemented by the studys manual of
operations or, as a modern counterpart of this, a web-based management system,
governing and guiding the study procedures with a view to quality assurance in
respect to adherence to the protocol.
Insofar as the operationalized version of the occurrence relation is true to the
theoretical one it represents and the rest of the methodology indeed is, as intended,
practicable and successfully implemented, the ultimate implication of this is that the
study result will be, in essence, descriptively unbiased (sect. 8.2 above). The extent
to which a descriptively unbiased result also is causally unbiased is a question of
the studys methodology; but it also is a question of its objects design (sect. 8.2
above).
8.3 Implications of the Design 95

The study results degree of causal unbiasedness together with its precision
determine the burden of evidence from the study. The precision is determined by
the studys efficiency together with its size.
It bears emphasis that there are not, nor can there be, tenable principles of
optimizing the size of an etiogenetic study the number of datapoints (person-
moments) in its study (case and base) series save for one: for a study with
an identifiable violation of tenable principles in its objects design or methods
design, the optimal size is zero. Any scheme of sample-size determination is but a
mapping of an arbitrarily selected value for a measure of the results precision to its
corresponding size of the study; and the result derived from it is commonly changed,
drastically, by reviewers of the grant application to zero (with corresponding
adjustment of the budget).
Chapter 9
Etiologic Studies Intervention Counterparts

Abstract In regard to population-level epidemiological research, the concern in

this text has thus far been with the overwhelmingly most common genre of this,
namely, causal research of the etiogenetic rather than interventive genre (chap. 3
Abstract); but the causal research can also be of the latter type, notably in studying
the health effects of preventive interventions vaccinations, for example.
While etiogenetic research generally is about unintended effects, current, of
matters in peoples past (constitutional, environmental, behavioral), interventive
research is mainly about intended effects, prospective, of actions (interventive) taken
prospectively, with the aim of changing the future course of health for the better. It
thus is, mainly, about effectiveness of interventions.
Intervention being a purposive action, with a defined alternative (possibly
inaction), it is subject to experimental study, in which the choice of intervention
is governed by the concern to learn about alternative interventions relative effects
rather than to achieve an interventions known (or presumed) effect.
In an intervention study, even if only quasi-experimental, the objects of study
have to do with a domain of indications (and absence of contraindications) for
the contrasted interventions. They are imbedded in a designed occurrence relation
for prospective intervention-time, with the determinant of interest characterized by
prospective divergence between the contrasted interventions, defined as algorithms
of intervention throughout the designed time horizon for the outcomes occurrence.
An intervention study, like an etiogenetic study, should lead to a case series and
a base series from the study base, which here is the study cohorts population-time
of follow-up. And in particular, the base series should be a simple, representative
(rather than stratified) sample of the study base. For this allows the resulting rate-
ratio function to be translated into its corresponding incidence-density function,
which in turn can be translated into the corresponding prognostic probability
function for the outcome of interventive concern.
Experimental intervention research raises particular ethical issues, which remain
incompletely understood and poorly dealt with.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 97

DOI 10.1007/978-94-007-4537-7 9, Springer ScienceCBusiness Media Dordrecht 2012
98 9 Etiologic Studies Intervention Counterparts

9.1 The Eminence of Experiments

Etiologic/etiogenetic research generally is non-experimental (ch. 6), meaning that

the causal determinants histories in the study base are not artificially experimen-
tally arranged for the purposes of the study; the history associated with any given
person- moment in the study base got to be what it is quite independently of the
study. The particular reasons for the non-experimental nature of etiogenetic studies,
not shared by their intervention counterparts, are scientific-practical for one and
ethical for another.
An experimental study base in etiogenetic research would commonly need to be
deeply prospective in study time. For, from the vantage of the person-moments in the
study base, the experimentally arranged etiogenetic histories would commonly need
to be quite longitudinal, and this requires at least equally longitudinal prospective
work in an experimental etiogenetic study. Apart from the tedium, a major problem
is that long-term adherence to the experimental regimen, whatever it is, would
commonly be unattainable.
The ethical obstacle to experimental research on etiogenesis generally is the
qualitative nature of the causal contrast: even in hypothesis-testing there generally
cannot be experts equipoise between the merits of a person being assigned to one
of the determinants index categories versus the reference category of it the former
typically representing a (prospective) health hazards without countervailing benefit,
potentially at least, while the hazards would assuredly be absent in the determinants
reference category. Even if there were properly informed volunteers for participation
in such a study, solicitation of their participation is generally regarded as ethically
inadmissible (sect. 9.4).
These prohibitive ethical considerations do not apply to hypothesis-testing
about (potential) etiogenesis by quantitatively suboptimal but unknown levels of
something that generally is (or at least is presumed) to be needed for maintenance of
good health things such as fluorine in ones drinking-water environment (sect. 4.1)
and antioxidant intake as a matter of ones behavior in respect to diet and dietary
supplements (sect. 2.2).
In contrast to etiogenetic research, epidemiological intervention research gen-
erally is experimental even on the population level (and not only on the bench
level). One practically enabling factor is that experimental intervention is no more
of an artifact than is intervention in actual practice of healthcare (preventive). Even
the general ethical obstacle to experimental etiogenetic research on humans (above)
is absent from truly pragmatic intervention research as experts equipoise is inherent
in the contrast, which is between/among candidates for the intervention of choice.
And whereas the outcome phenomena generally are prospective in study time, so
inherently also are the interventions (different from causal histories in etiogenetic
studies).
Finally, experimentation in the particular meaning of a randomized contrast may
actually be needed. It may be needed for prevention of confounding by indication,
which is a topic peculiar to the intentional causation by intervention, distinct from
9.2 The Objects of the Studies 99

the generally unintentional causation in the etiogenesis of illnesses as they naturally

occur, and which tends not to lend itself to documentation and control based on this.
Another, obvious, need for experimentation and it can just as well randomized
attends to the situation in which at least one of the contending interventions of
choice is not (yet) in routine use.

9.2 The Objects of the Studies

In etiologic/etiogenetic studies, quite generally, the histories pertaining to a given

contrast can readily be supplemented by other, unrelated contrasts. Thus, a contrast
which in itself justifies an etiogenetic study generally justifies addressing some
ancillary contrasts, their study by means of the same case and base series: several
substantively distinct studies can be efficiently conflated into what procedurally is,
in essence, a single study.
In intervention studies of the usual, parallel type, the counterpart of etiogenetic
studies necessary case series together with its corresponding base series is their
necessary study cohort together with the intervention contrast in it, prospective,
as of cohort T0 . And while intervention studies generally are justified by inquiry
into the compared interventions relative effectiveness in producing the intended
effect, this is readily supplemented by consideration of other, ancillary outcomes,
generally ones that have to do with the interventions (potential) unintended effects.
Again, several substantively distinct studies can be conflated into what procedurally
is, in essence, a single study. We discuss the alternative to these parallel trials the
before-after trials, that is at the end of section 9.3 below.
The objects of any population-level epidemiological study are defined, as we
have set forth, by a designed occurrence relation a health phenomenons rate
of occurrence as a function of determinants of this rate, in a particular domain of
this occurrence. We have not, however, emphasized that both the domain and the
occurrence relation in it are defined on the scale of scientific time, which in turn
is defined by its inherent zero point, T0 . This scale of time is distinct from such
particularistic scales of time as study time (T0 : the time of the inception of data
collection) and cohort time (T0 : the time of entry into the cohort).
While in the study base of any etiogenetic study the scientific time at each of its
person-moments is the time of the outcome and, thus, T D T0 in etiogenetic time
(retrospective, T < T0 D 0), in any interventive study the scientific time is some
T > T0 in intervention-prognostic time at each of the person-moments in its study
base, T0 being the time as of which the intervention status diverges (this being the
time, also, of cohort T0 for the study population). In each of these two types of study,
a person-moment in the domain of the object of study is one of occurrence/non-
occurrence of the outcome phenomenon in question.
While the object of any given etiogenetic study is the outcome phenomenons
rate of occurrence at T0 of etiogenetic time, this in causal relation to retrospective,
pre-T0 divergence in the causal determinant, that of any given interventive study is
100 9 Etiologic Studies Intervention Counterparts

the outcome phenomenons prospective, post-T0 rate of occurrence on the scale of

intervention time, this in causal relation to prospective, post-T0 divergence in the
causal determinant.
In etiogenetic studies the designed temporal relation between the outcomes
occurrence and its causal determinant is, in principle at least, dictated by Nature
the histories need to cover, in principle, the entire range of time (retrospective)
in which the factor can have exerted its etiogenetic effect. If the object of study
is designed to focus on a limited segment of the full range of etiogenetic time,
then histories in respect to the other segments represent potential confounders.
But in interventive studies, by contrast, the corresponding temporal relation is
dictated, in its entirety, by the investigators. The investigators commonly opt for
studying the outcomes (rate of) occurrence in a limited, early segment (prospective)
of the scientific time, occurrence in this period in causal relation to the causal
(interventive) determinant of this as of the scientific T0 with the outcomes timing
quite possibly an element in the designed occurrence relation and with no role for
the later ranges of prospective time; the object function is understood to pertain only
to a limited range of prospective time.
Another difference relates to this. In an etiogenetic study, concerning a given
(potentially) etiogenetic factor, there may be concern to distinguish among various
periods of etiogenetic time recent, intermediate, and distant past, say, each
expressly defined leading to their corresponding, separate etiogenetic determinants
for joint inclusion in the designed occurrence relation. In an interventive study there
generally is a counterpart of this but there is no imperative to address the entire range
of the relevant prognostic time: in particular, no confounding arises from whatever
may be happening in the segment of time that is not addressed.

9.3 The Methods in the Experiments

The source population in an intervention experiment, very different from its

counterpart in non-experimental etiologic/etiogenetic studies, is, in a segment of
its prospective course in study time, a source for identification of candidates for
enrollment into the study population, inherently the study cohort. These are persons
who, at the time of the identification, are seen to apparently represent presence of
the (designed) indication for the intervention(s) and freedom from the (designed)
contra-indications for it/them, while also apparently satisfying whatever other (de-
signed) criteria of admissibility into the study cohort. (Those admissibility criteria
are defined by the operationalized version of the initially designed occurrence
relation; cf. sect. 8.1.) The source population can be, for example, a segment of
the listeners of a set of radio stations.
Given an identified case (person-moment) of apparent eligibility for enroll-
ment into the study cohort, the investigators may but need not proceed to
solicitation of the persons potential volunteering to the enrollment, following
whatever non-scientific preliminaries to this (sect. 9.4 below). If the consent to
9.3 The Methods in the Experiments 101

(potential) enrollment is sought and obtained, this may initially be followed only
by tentative enrollment. For, the admissibility criteria may need to be assessed more
comprehensively and/or more closely; and there may be a need to furtively test the
persons adherence to an agreed-upon plan for some simulated intervention.
The process leading to a persons definite enrollment into the study cohort is
completed if, and when, the assignment experimental, generally randomization-
based to a particular intervention occurs. This is the study cohorts membership-
clinching event, after which there will be no exit from the membership (sect. 6.1).
This study population is formed from, rather than embedded in, the source popula-
tion (cf. sect. 5.1), and it thus has its (prospective) course independently from that
of its source population (which usually is open dynamic rather than closed).
Two tasks are executed at the time of a persons enrollment into the study
cohort, at this T0 of cohort time as well as of actual scientific time. One of these
tasks naturally is documentation of the relevant facts which, in the operationalized
(methodological) version of the initially designed occurrence relation, have this T0
as their temporal referent. And the other one, just as naturally, is the assignment
of the intervention, possibly with blinding of the assignment (for prevention of
prospective confounding and/or assurance of validity of outcome assessment). If one
of the interventions actually is no intervention at all, it may need to be represented
by placebo intervention in the context of blinding.
In the post-T0 period, too, there are two tasks to be executed. One of these again is
a matter of observation and it related documentation, while the other one is in the na-
ture of execution and enforcement, having to do with the commitments of the study
subjects and, notably, those of the research personnel too. The documentation has to
do with protocol adherence for one, and with the outcomes both effectiveness- and
safety-related for another. And the execution and enforcement of the intervention
commitment is a post-T0 task even in the context of an acute intervention, given that
the protocol specifies this as the only (sic) intervention (so long as the documented
outcomes may be affected by extraneous interventions).
Given that all that is to be performed in the post-T0 (post-randomization) period
in an intervention experiment, follow-up is too passive-sounding as a term for
it. To wit, studied are effects of actual interventions, not those of intentions to
be subjected to an intervention (or to its defined alternative; sect. 3.1). Effects
in medicine, and in its simulated practice in intervention experiments likewise,
characterize actions, not mere intentions to act; and they can be unintended as well
as intended effects of actions, never of mere intentions to act.
Once the data have been collected and entered into the trials database, the
outcome events of a given type (those relevant of effectiveness, say) that occurred in
the study base can be identified from the database, and as a supplement to this case
series can be selected a suitable base series as in a logically construed etiogenetic
study (sect. 6.5). For each of these two series, the database allows the identification
of the type of the intervention and the duration of this (since T0 ); and the same is
true of all other relevant information (incl. that documented at T0 ).
Fitting the logistic counterpart of the designed log-linear model for the outcome
events incidence density to the data (the designed statistical variates realizations)
102 9 Etiologic Studies Intervention Counterparts

in the two series, results in a documentation of the experience in terms of the events
incidence-density ratio as a function of intervention time (T), type of intervention,
and the persons prognostic indicator profile at T0 .
But as at issue actually is not an etiogenetic study but an interventive one, the
interest is not in the (causal) rate ratio but the rate proper, the events incidence
density as a joint function of all of those determinants of its level. To derive this
function for the incidence density of the outcome event at issue, a representative
(simple random) sample of the study base is to be drawn to constitute the base
series. Then, the desired function for incidence density can be derived as

ID D .B=PT/ exp .L/ ;

where B is the size of the base series, PT is the amount of population time
constituting the study base, and L is the linear compound arising from the logistic
models fitting to the data on the case and base series. The corresponding empirical
function for cumulative incidence and risk from time T D 0 to T D t is
Z t

CI0;t D 1  exp  IDt dt :

0

When the outcome at issue is a state of health and the concern thus is to
derive the corresponding prevalence function, the base series need not be derived
by representative sampling; the only requirement is that it be independent of the
presence/absence of the outcome state at the sampled person-moment. For, the
designed prevalence function is fitted to this single series, with no involvement of
quasi-rates and rate ratio. The logistic model for the prevalence is fitted to the data.
The logit of the prevalence at T D t, in the study base, is taken to be the resulting
functions realization at this point in time since the interventive T0 .
In an intervention trial of that usual, parallel type, any intervention contrast is an
asymmetrical one, just as is any etiogenetic contrast. The concern is not to compare
one intervention with another one. Instead, the concern is with only one intervention
at a time, with its effect on the (rate of) occurrence of the outcome at issue when this
intervention the index intervention in the contrast is present in lieu of its defined
alternative the reference intervention in the contrast. The role of the reference
arm in the trial is to provide for surmising what the outcomes rate of occurrence in
the index arm would have been had this subcohort been subjected to the reference
intervention instead of the index intervention (cf. sect. 4.3).
This parallel arrangement is needed when, as is usual, one of these two
conditions obtains: the alternative indeed is an actual intervention, another possible
intervention of choice; or it is absence of any actual intervention but the outcomes
rate of occurrence in the absence of any intervention cannot be presumed to be
(practically) constant over the follow-up time in the trial.
By the same token, there is no need for a reference arm when an intervention
is contrasted with no intervention and the outcomes rate of occurrence can be
presumed to be (practically) constant over a period of time before and after the
9.4 The Ethics in the Experiments 103

initiation of the intervention, given no effect of the intervention. In this situation,

a before-after trial will do: the change in the outcomes (rate of) occurrence from
before the intervention to what it is after (the initiation of) the intervention is a
measure of the interventions effect as such (being that practically no change can be
presumed to occur in the absence of the intervention).

9.4 The Ethics in the Experiments

In population-level epidemiological research, the issues in the studies objects

design and methods design correspond to ontology and epistemology in philosophy,
the mother of all of science. For, in ontology philosophers contemplate what is real
and hence possibly worth knowing about, while in epistemology they contemplate
how knowledge about a real thing (entity, quality/quantity, or relation) could be
gained.
Ontology and epistemology are quite unrelated to a third eminent branch of
philosophy, to moral philosophy or ethics. In this the concern is to understand what
is morally good or bad, right or wrong, in human action, as it has to do with other
creatures, humans in particular. Population-level epidemiological research is human
action, of course; and it has to do with other humans, in two fundamental ways: it
seeks knowledge (scientific) for the benefit of humans at large, and to this end it can
involve imposition of something undesirable on some humans. Philosophy of ethics
therefore has direct bearing on research on humans, on its experimental genre in
particular.
One of the two principal branches of ethics, in philosophy, is characterized as
being teleological (Gr. telos, end) in its fundamental principle: a chosen action
is ethical if it is intended to maximize the aggregate of happiness among those
who are affected by it. In teleological terms, thus, for a planned epidemiological
experiment on the effect of some intervention on the population level to be ethical,
required is, first, that the studys design as for both objects and methods is, as
best the investigators can tell (upon all the necessary inputs and reflections), free of
recognized errors; second, that its execution will be made as flawless as possible;
and third, that the resulting advancement in the knowledge-base of epidemiological
practice is expected to enhance human happiness more than the study subjects
participation in the trial, together with whatever other costs of the trial, will take
away from it. The operative criterion of teleological ethics indeed is this sincere
effort to choose the course of action that maximizes the aggregate happiness of those
who will be affected by the study, not success in this effort. An epidemiological
experiment on humans thus is not teleologically unethical just because it produced
only unhappiness (due to, e.g., some common misunderstanding in the studys
objects design; sects. 7.2, 7.3).
The other principal branch of ethics, in philosophy, is characterized as being
deontological (Gr. deon, binding). It seeks to define ethical obligations, duties. As
for human experimentation in epidemiological (and other) research, the overarching
104 9 Etiologic Studies Intervention Counterparts

ethical obligation, now strictly binding to the researchers, has to do with the study
subjects informed consent to the participation: without such consent from each
of the study subjects, the study is deontologically unethical. This is commonly
accompanied by deontological guidelines for the maintenance of confidentiality of
the information collected on the study subjects.
Informed consent in intervention-prognostic research on humans is commonly
thought of as pertaining, specifically, to each study subjects enrollment into a trial,
with experts equipoise about the interventions relative merits seen to be an added
prerequisite for ethical solicitation of the enrollment; but both of these ideas are
untenable, arguably at least.
As the decision to enter into a trial is to be an informed one, so logically should
also be continued participation in it, notably when new information accrues from
the study itself and/or from concomitant other studies: the information for the
consent should be updated, as indicated. But as it now is, the trials Data Safety
and Monitoring Board or its equivalent, in the light of merely the accruing data
in the trial itself and in the context of its own valuations, periodically decides to
continue the trial until, perhaps, abruptly calling it to a halt (notably as, suddenly,
there is a statistically significant difference, as P < ).
And equipoise, insofar as it should be seen to be relevant at all, should be that
of each potential study subject, whose subjective valuations matter in addition to
whatever objective information is supplied by experts. But actually, a potential study
subject should be free to volunteer for the advancement of the knowledge-base of
the practice of epidemiology (sect. 1.1) even in the face of whatever disutility this
action may subjectively seem to entail to the volunteer. By the same token, there
should not be any deontological prohibition of solicitation of participation in an
experiment, given provision of all of the relevant information about it. After all,
the investigators experimentation on him- or herself is not being considered to be
ethically inadmissible (see sect. 12.3).
Overall, the ethics of human experimentation in epidemiological research re-
mains incompletely developed even in respect to its broadest principles, and much
could be said about disingenuous heeding of such ethical deontological norms
as generally are being formally upheld.

9.5 Quasi-experimental Intervention Studies

The implementation of an intervention experiment in population-level epidemiolog-

ical research is, commonly, quite challenging, from the recruitment of volunteers
for enrollment into the study cohort through all the necessary work on it as for
intervention execution and data collection commonly over a long period of time,
inescapably prospective from the vantage of the trials initiation.
This raises the question of whether intervention effects really need to be studied
in the framework of experimental simulation of practice, instead of setting out to
9.5 Quasi-experimental Intervention Studies 105

learn from experiences in actual practice, documented for the purposes of practice
(rather than science) given that the compared interventions actually are being used
in actual practice.
In contemplating this alternative to experimentation, it is to be borne in mind
that the objects of epidemiological intervention studies have a particular structure
(sect. 9.2); and that, consequently, the intervention studies themselves have a
structure characteristic of them (sect. 9.3.).
The question about possible non-experimental alternatives to intervention exper-
iments in population-level epidemiological research thus takes this form: Can quasi-
experimental alternatives to those experiments, when distinctly more feasible to
construct, also be without appreciable compromise of validity? Such studies would
be, by definition, like their experimental counterparts as for the structure of them,
but they would not actually be experiments in terms of that which is definitional to
intervention experiments as distinct from their quasi-experimental counterparts: the
genesis of this structure, specifically the adoptions of the interventions for the (sole)
purpose of learning about their effects (sect. 9.1).
In quasi-experimental intervention studies the interventions are ones adopted so
as to elicit their already known, or at least presumed, intended effects. In these,
the intervention-study structure is achieved not by constructing it but, instead, by
suitable selection of the study subjects so as to achieve the needed structure as
though an experimental protocol had been followed.
The central reservation about quasi-experimental intervention studies relates,
naturally, to the very feature that makes them only quasi-experimental in the study
of intended effects (of interventions). Non-experimental, non-randomized studies
of the intended effects of alternative interventions are prone to be subject to incom-
pletely documentable and, hence, less than adequately controllable confounding by
indication in studies of the interventions intended effects, notably when used are
data recorded for the purposes of actual practice (rather than for research).
Ill-documented particulars of the indications severity bear on the risk of the
outcome at issue, and they may also have differential bearing on the choice of
intervention. This problem of potentially uncontrollable confounding by indication
generally is, however, much lesser in epidemiological research on preventive
interventions than in clinical research on therapeutic interventions, on account of
the generally simpler indications in the former.
Quasi-experimental intervention studies do have their place in the development
of the scientific knowledge-base of epidemiological practice, and the place is
characterized by simple indication for acute interventions and interest in long-term
effects of these, so long as the alternative interventions have been in use for long-
enough period of time.
Chapter 10
Causal Studies Acausal Counterparts

Abstract In community-level preventive medicine the core mission of reducing

morbidity from various illnesses involves two principal components: reduction of
the occurrence of causes of illnesses and reduction of peoples susceptibilities to
such causes as remain.
When decisions about these actions quitting smoking or undergoing vaccina-
tion, for example are taken by individual members of the cared-for population
in response to community-level health education peoples rational decisions
about them require assessments of their personal risks for the illnesses that are
at issue. Insofar as the knowledge-base for this is available from population-
level epidemiological research, acausal/descriptive individuals in the cared-for
population can perform their personal risk assessments on the basis of a web-based
system arranged by the health educators.
The knowledge-base of these risk assessments is constituted by risk functions.
For a given one of these, the domain is one in which the risk assessment is done,
and subdomains within this are defined by risk/prognostic indicators for the health
outcome at issue. The risk function, thus, expresses the cumulative probability of
the health events occurrence over prospective/prognostic time as a function of that
time (T) jointly with the prognostic indicators at the time of the risk assessment
(at prognostic T0 ). The risk function is predicated on no change in the risk factors
(causal) over the time horizon addressed by the function.
In an acausal prognostic study, as in an intervention-prognostic study, a cohort is
enrolled from the risk functions domain of application, and the study base is formed
by this cohorts prospective course (away from the domain of application). A case
series from this study base is coupled with an unstratified sample of this population-
time, to be able to assess incidence-density function subject to translation into
the corresponding function for cumulative incidence and hence risk from
T0 to T.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 107

DOI 10.1007/978-94-007-4537-7 10, Springer ScienceCBusiness Media Dordrecht 2012
108 10 Causal Studies Acausal Counterparts

10.1 The Focus on Prevention-oriented Research

Epidemiological practice has traditionally been public-health medicine in the

principal meaning of community-level preventive medicine, the broadest modalities
in this being education, regulation, and service. This community segment of
preventive practice has been supplemented by clinical preventive medicine, while
preventive medicine at large has been supplemented backed up by therapeutic
and rehabilitative clinical medicine.
Accordingly, a secondary concern in community health-education has been
optimization of peoples deployment of clinical care prevention of its underuse
and, also, overuse. In his Tracking Medicine: A Researchers Quest to Understand
Health Care (2010), John Wennberg makes an evidence-based case for overuse of
available healthcare being counterproductive to population health: Our research
finds support for [the hypothesis that] greater care intensity is associated with higher
mortality rates and poorer objective measures of process quality (p. 160).
In the modern context of national health insurance, clinical medicine, too, has
become a component in public-health medicine, in fact the principal component
by far; and this has substantially enlarged the scope of epidemiological practice of
public-health medicine. In the main this new segment of epidemiological practice is
a matter of something other than preventive medicine; it is documentation of rates
of occurrence of phenomena of care providers actions within the jurisdiction of the
epidemiologists concern, for clinicians evaluation of the quality of the care and
possible corrective actions by administrative and/or policy-making officials. This
fact-finding, addressed in section 14.5, is commonly misconstrued as research and
termed health services research.
As epidemiological practice thus interfaces with clinical practice, its knowledge-
base has a partial overlap with that of clinical medicine. From this it does not follow,
however, that epidemiological research (scientific) should extend beyond what is
needed for the scientific knowledge-base of community-level preventive medicine.
For, whatever the community doctor needs to know about clinical medicine about
screening for a cancer (ch. 11), for example (s)he can and should obtain from the
relevant clinicians.
Community medicine in the form of service can be a matter of preventive inter-
vention on individuals dealt with one at a time, as in clinical medicine. Community
projects/programs of vaccination are prime examples of this. But there are no
corresponding community-level projects or programs of therapeutic or rehabilitative
services, as these like vaccinations need to be provided individually but never
indiscriminately different from vaccinations.
The outlook in play here in terms of which individuals in the cared-for
population take personal decisions in the light of what they have come to understand
are their individualized risks and the susceptibilities of these for reduction, and their
personal valuations of these is no different from that in clinical medicine, whether
10.2 The Acausal Knowledge-base of Prevention 109

preventive or therapeutic. But we must point out that in a very eminent little book,
The Strategy of Preventive Medicine (1992), Jeffrey Rose this highly esteemed
professor of epidemiology argued otherwise. He accented the importance of the
cumulative effect on the population level of even minor gains by the individuals
involved. In this books Preface he wrote, The essential determinants of health of
society are thus to be found in mass characteristics: : : : effective prevention requires
changes which involve the population as a whole.

10.2 The Acausal Knowledge-base of Prevention

Women in the population that an epidemiologist serves do not look for guidance
from the epidemiologists health-education in respect to reduction of their risks for
prostate cancer; but they do look for guidance in respect to prevention of coronary
heart disease and stroke, for example. Interest in, and relevance of, prevention-
oriented measures of maintenance of low-risk lifestyle naturally depends on the
perceived level of risk for the illness at issue. An epidemiologists health-education
thus needs to provide for individual risk assessments in the community (s)he serves.
While rate inherently is a population-level concept in epidemiology, risk inher-
ently has to do with individuals (as members of a population perhaps). Risk is, for
an individual, the probability in the objective, relative frequency meaning of this
that the outcome adverse at issue will come about. It inherently is conditional
on whatever is the known risk profile prognostic profile of the particular person
at the time of the risk assessment, the prognostication in this sense. The (profile-
conditional) probability/risk of the outcomes occurrence is defined in reference to
a particular risk period for this, and it is the proportion of instances of the risk profile
in general in the abstract such that the outcome will occur within this period.
Risk per se is acausal, having to do with this proportion conditionally on a particular
intervention (such as none) with no causal contrast involved. It also is, for scholarly
purposes at least, conditional on surviving extraneous causes of death over the risk
period at issue.
In preventive medicine, risk assessment is relevant for decisions about
prevention-oriented actions in two stages. The first stage is assessment of
background risk, meaning risk conditional on no (new) prevention-oriented action
no change in, say, dietary habits. And if this risk is assessed and deemed high enough
to entertain such a change, change-conditional risks also need to be assessed for
assessment of the changes in risk over various periods that would result from the
action being considered. These two types of risk, considered jointly in respect to
a given potential action on the part of a particular individual at a particular time,
imply the causal risk difference relevant to the decision about the action.
110 10 Causal Studies Acausal Counterparts

10.3 The Objects of the Acausal Studies

The generic object of study for risk assessments with a view to potential prevention-
oriented action, for individuals decisions about the action, is a risk function
designed for a defined domain of risk assessment. This function may address the risk
conditionally on no action (preventive), or the risk may be formulated as a function
also of a potential action/ inaction with a view to the individualized risk difference
by the action, reflecting the effect (preventive) of this action when adopted in lieu
of no action.
Regardless of whether potential prevention-oriented action is involved, the risk
is modeled as a function of prognostic indicators relevant to consider as definers of
subdomains of the functions domain; and the function generally needs to define the
risk not for a single risk period but as a function of prognostic time as to the form
of those functional dependencies of the risk.

10.4 The Acausal Studies Proper

An experiment (trial) designed to address the effectiveness of a prevention-

oriented intervention inherently provides for addressing intervention-conditional
risk functions (sect. 9.3); and by the same token, the corresponding quasi-
experiments (sect. 9.5) also do.
Ordinarily, a risk-assessment study involves enrollment a cohort of study subjects
as representatives of the domain of risk assessment, from among candidates for the
enrollment identified in a chosen source population. For efficiency of study, the
enrollment is selective with a view to increasing the variability of the prognostic
indicators realizations at entry into the study cohort. At entries into the cohort, at
cohort and prognostic T0 , the study subjects prognostic profiles are documented.
And following the entries, the cohort members are followed principally with a view
to identifying and documenting the end point of the follow-up, the illness event at
issue. The empirical counterpart of the designed risk function is derived from the
resulting data as discussed in section 9.3.
That is the essence of the study so long as studied is the risk in the absence of
any knowledge about measures to prevent the illness, ones that might be applied
by the study subjects in the course of the follow-up. And it also is the studys
essence when an added feature is (random) assignments to particular ones of
prevention-oriented interventions, together with implementations/enforcements and
documentations of these (sect. 9.3). Otherwise the study is tantamount to a quasi-
experimental intervention study, with its challenges in validity assurance (sect. 9.5).
An intervention-conditional, acausal risk function should be not only
valid/unbiased but also maximally discriminating among different levels of risk
within the study objects domain. As a measure of the empirical functions
discriminatingness/informativeness, some use the area under the ROC (receiver
operator characteristic) curve.
10.4 The Acausal Studies Proper 111

More meaningful as a measure of a risk functions informativeness

(validity-conditional) might be taken to the standard deviation of the function-based
risk values divided by the maximum that this SD could possibly be. Empirically,
this informativeness index is
   1=2
OI D V
O N O 
R 1  RO ;
N 1

where, based on a representative cohort from the functions domain, V O is the

unbiased estimate of the unit variance of the N values of risk from the function and
RO is the empirical value for the unconditional risk (proportion with Y D 1) among
the N instances.
Chapter 11
Studies on Screening for a Cancer

Abstract An epidemiologists concern for the health of the cared-for commu-

nity/population actually is not limited to community-level preventive medicine,
without regard for whatever may be going on in the clinical segment of healthcare
at large. (S)he understands that the peoples health public health in this meaning
of the term is much influenced by the aggregate of clinical healthcare (preventive,
therapeutic, rehabilitative) in the community mostly for the better but, also, for
the worse. And as a consequence, (s)he is concerned to optimize peoples use of
clinical care by means of community-level health education aimed at prevention of
both under- and overuse of clinical care.
One line of this education has to do with screening for risk factors for illnesses
(e.g., the BRCA 1 & 2 mutations with a view to detection of high risk for breast
cancer), preparatory to decision about preventive interventions (e.g., by tamoxifen
use) and/or screening for the illness at issue (breast cancer in this example) that
is, pursuit of early, latent-stage diagnosis (rule-in) about the illness with a view to
enhanced curability by early treatment.
Screening is directed to individuals one at a time, and it thus is clinical healthcare,
by definition. Screening for a risk factor is preventive care, while screening for an
illness is therapeutic therapy-oriented care (when opportunity for successful
prevention already is passe). The yield of screening is information, and this in itself
has no effect on the persons course of health. It thus is not an intervention.
This chapter addresses, in the main, the epidemiologic researchers misunder-
standings in regard to these orientational concepts and the consequent serious
misunderstandings about research on the intended consequences of screening, for a
cancer in particular. Epidemiologists should come to understand that the knowledge-
base, just as practice, of screening is a clinical matter.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 113

DOI 10.1007/978-94-007-4537-7 11, Springer ScienceCBusiness Media Dordrecht 2012
114 11 Studies on Screening for a Cancer

11.1 Introductory Example

The risk for breast cancer overt, life-threatening cancer of a breast is very high
for women with certain exceptional mutant BRCA 1 and/or BRCA 2 genes. This
topic is of considerable healthcare concern, as screening for these very major risk
factors is practicable, as effective preventive interventions (by tamoxifen use or
even by bilateral mastectomy) are available options, and as screening for the cancer
(with a view to enhanced curability) is practicable all of this, naturally, in clinical
medicine.
Even though this care is, by its very nature, a matter of clinical, rather than
community, medicine, it nevertheless is a concern from the vantage of public
health and its attendant public policies, now that clinical healthcare has been
brought to the public domain via the advent of national health insurance. On the
other hand, though, the promulgators of the public policies are constrained by the
social contract that leaves the medical professions autonomous free of societal
domination in specifically medical aspects of the care.
This topic differs quite profoundly from the ones addressed in section 4.1 in that
the public policies about it are not legalistic-regulatory and directed to the publics
behaviors or environments outside the realm of (professional) healthcare. They are,
instead, directed to the relevant healthcare (clinical), by physicians. But consistent
with autonomy of the medical professions, the public policies are not normative
stipulations concerning the care; they are about societal reimbursements of the costs
of the care.
So, what do the promulgators of healthcare policies (on reimbursements) princi-
pally need to know about clinical care related to this genetic aspect of breast cancer?
and how will they get to know it? It is relevant to note first, in the particular context
here for a start, that they do not need to know about the etiogenetic proportions
for those genotypes in cases of breast cancer associated with those genotypes, from
causal research of the etiogenetic genus. The frequency prevalence of each of
those genotypes is a matter not of human behavior (unpredictable, capricious, as in
the first three of the four examples in sect. 4.1) nor of peoples environments (also
quite variable) but of human biology, and these rates are knowable from genetical-
epidemiological research. But again, policy promulgators need not know about these
proportions either.
Instead, the policy-makers need is to know about the risk of coming down
with overt breast cancer say, life-time cumulative risk of this among women
with those genetic traits, this naturally from descriptive, acausal research for risk
assessment (ch. 10). And in particular, the policy-makers need to know much more
from specifically clinical research so as to have the scientific basis, supplemented
by economic facts, for assessing the cost-effectiveness of the care, this being central
to the perceived justifiability/unjustifiability of societal reimbursement of its cost.
11.2 Epidemiologists Outlook 115

11.2 Epidemiologists Outlook

Concerning a particular generic type of cancer breast or lung cancer, say

epidemiologists concern generally does not focus on the rate of incidence of
diagnosis (first rule-in dgn.) about it, nor on the rate of the cancers prevalence,
but, instead, on the rate of incidence of death from it on mortality from the cancer,
that is.
The concern being reduction of that mortality (by community-oriented means),
one avenue to this end is promotion of early detection of the cancer, this on the
premise that provision for early treatment provides for more common cures of the
cancer and, thereby, reduction in mortality from it. This is a matter of promoting
screening for the cancer pursuit of diagnosis (rule-in dgn.) about the cancer in
its preclinical, latent stage of development, baseline screening followed by periodic
repeat screenings. This promotion is adopted into epidemiological practice in regard
to a particular segment of the cared-for population as a matter of public policy if the
screening in this subpopulation is deemed to produce enough life-saving mortality-
reducing benefit to justify its attendant harm and monetary cost.
In this promotion there are two fundamental options. One option is community
education focusing on the screening in the framework of clinical medicine, while an
alternative to this is internal to community medicine: community education about,
and community-level provision of, the initial testing, with referral of the test-positive
persons for further diagnostic work-up in the clinical framework (together with early
treatment in the event of the cancers rule-in dgn.).
Epidemiologists preference, at present, generally is for the latter; and associated
with this is viewing community-level provision of the first diagnostic test as a
community-level intervention, and a preventive (sic) intervention at that (Apps. 1, 2).
From this major fallacy is deduced the idea that study of the mortality-reducing
effectiveness of screening is, ideally at least, a matter of randomized trials and,
notably, clinical (sic) trials at that.
Three features generally characterize epidemiologists design of these trials,
quite regularly touted as representing the gold standard in this research: arbitrary
(and commonly very small) number of repeat screenings; attention to deaths from
the cancer throughout an arbitrary duration of follow-up (post randomization); and
disregard for the timings of these deaths by focus on single-valued proportional
reduction (apparent) in mortality from the cancer over the duration (arbitrary) of
the follow-up (cf. sect. 7.2). As the community-level intervention on any given
individual never extends beyond the initial test, yet another feature of these trials
generally is absence of any protocol for the rest of the diagnostic work-up and for
the treatment upon diagnosis (rule-in) about the cancer.
116 11 Studies on Screening for a Cancer

11.3 The Gold Standard Trial

A hypothetical example of a screening intervention experiment, for simplicity

focusing on the mortality implications of the baseline round of screening for a
cancer, is addressed in Table 11.1 here. More specifically, of course, the focus in this
special case of the gold standard study on screening for a cancer is on mortality
from cases of the cancer latent cases that are detectable by the screening (by the
baseline version of its regimen/protocol or by the aggregate of the corresponding
unregimented practices; cf. above). At issue is early detection (under screening)
versus no early detection (under no screening), mortality in relation to this.
According to that table, its Part A, the fatalities from those detectable cases
in the absence of screening-based early detection occur 2-12 years later; 10%
2-4 years later, 30% 4-6 years later, and 60% 6-12 years later. And that Part A of
the table also shows that of the potential fatalities in those three successive periods
of prospective time, 10, 40, and 90%, respectively, are preventable by screening-
associated early treatments. (The later is the potential death from the cancer, the
earlier is the cancers stage of development at the time of the potential screening;
and as a consequence, the later is the potential fatality, the greater is the cancers
curability and the deaths consequent preventability by early treatment.)
Part B of that table shows the implications of the time-specific preventability
rates by early treatment in otherwise fatal (but detectable) cases, implied by Part
A. The focus is on the reduction in cumulative mortality, separately for the cases
that are detectable by screening at baseline and for these cases in combination with
those that are not yet detectable at the time of the baseline screening. While the
former proportional reductions are shown to be, at 6, 12, and 20 years, 13, 69, and
69, respectively, the latter ones are less than 13%, much less than 69%, and very
much less than 69%, respectively.

Table 11.1 Hypothetical example concerning baseline screening

Part A. Fatality implications of the screening (of its associated early trmt) in
cases detectable at baseline
Detectable cases fatal under no screening: Reduction in detectable cases fatal
time to death outcomes under screeninga
<2 yr: 0% <2 yr:
2-4 yr: 10% 2-4 yr: 10%
4-6 yr: 30% 4-6 yr: 40%
>6 yr: 60% >6 yr: 90%
Part B. Reduction in cumulative mortality
1. Baseline-detectable cases 2. Baseline- and later-detectable cases
6 yr: 0.1  0.1 C 0.3  0.4 D 13% 6 yr: <13%
12 yr: 0.13 C 0.6  0.9 D 69% 12 yr: 69%
20 yr: 69% 20 yr: n69%
a
This is the rate of curability, by early treatment, of otherwise fatal detectable cases
11.4 Meaningful Types of Trial 117

Now, insofar as one in this example wishes to think about screening-associated

reduction in cumulative mortality from the cancer (consequent to early treatments),
one must be thinking about the screen-detectable subset of the cases (as among the
rest of the cases there is, at baseline, no screening-based reduction in the prospective
fatal outcomes). In the example addressed in Table 11.1 here, this means that ones
interest is focused on that 69% the proportional reduction in otherwise fatal
outcomes of the detectable cases.
That 69% reduction in fatalities from those baseline-detectable cases translates
into the same reduction in the cumulative mortality from the cancer for that subset
of cases at 12 years of follow-up and later (but not earlier) but this reduction is
not manifest at any time in the follow-up of screened and unscreened subcohorts
in a randomized trial on the screening (baseline scr. vs. no scr.). Manifest in
the trials screening arm are only cumulative rates of mortality from the cancer
for the baseline-detectable cases in combination with ones that only later become
detectable (by repeat screening); the manifest reductions in the cumulative rate of
mortality from the cancer are smaller than the curability rate of otherwise fatal cases,
that 69%, at whatever time after the baseline screenings, some of them much smaller.
(See Part B of Table 11.1.)
Upshot: Nothing quantitatively meaningful is addressed in this gold-standard
trial on screening for a cancer (the way they are designed; sect. 11.2 above).

11.4 Meaningful Types of Trial

As we repeatedly let on in the two sections above, screening for a cancer, in itself,
does not reduce mortality from the cancer as, contrary to common claim, it is not
an intervention. Only the screening-afforded early treatments (in lieu of later ones)
have the potential for that consequence (in causal terms), on the ground of enhanced
curability of otherwise fatal cases of the cancer.
And we also showed, in section 11.3 above, that this curability gain is not
manifest in the proportional reduction in cumulative mortality from the cancer in
such screening trials as ordinarily are designed and carried out; that such a mortality
reduction generally grossly underrepresents the proportion of cases that are curable
by early treatment among otherwise incurable, fatal cases of the cancer; that, in
fact, nothing quantitatively meaningful is studyable by means of such trials. And
qualitatively, the mere existence of the reduction generally should be understood not
to be a hypothesis but a corollary of the attainability of earlier, latent-stage diagnosis
and the enhanced curability consequent to this.
Harkening back, again, to the example in section 10.3 above, that curability rate
(69%) for early-detected, otherwise fatal cases obviously is, in principle, studyable
by means of an actual intervention trial. In it, some of the baseline screen-detected
cases (a random subset) of the cancer would be treated without delay, others only if
and when they progress to overt, symptomatic disease.
118 11 Studies on Screening for a Cancer

In the absence of any overdiagnoses taking practically benign lesions to be

life-threatening the relative magnitudes of the cumulative incidence of death from
the cancer in the trials two arms, once leveling off (after 12 years of follow-up),
would obviously give an empirical measure of the proportion of the cases such that
only early treatment is life-saving. And as a matter of fact, even if there would be
overdiagnosis, these asymptotic levels of the cumulative incidence would not need
to be corrected for the proportion of overdiagnosed cases (the frequency of which
could be studied on the basis of the delayed-treatment arm of the trial) in order
that their relative magnitudes fairly represent the curability gain from the screening-
associated early diagnoses.
That type of trial would be quite straightforward, as the focus would be on
the true causal contrast not screening versus no screening but early, latent-
stage treatment versus treatment delayed until the disease is overt, symptomatic
and addressed would be the contrasted treatments relative effects preventive
on all of the cancers naturally fatal outcomes, across all of their timings, and
without admixture of deaths from cases of the cancer outside the causal contrast
(cf. sect. 11.3).
In these terms, study of reduction in mortality from the cancer in the meaning
of case-fatality rate would be meaningful, completely so, separately for baseline
and repeat screenings. But of course, recruitment of well-informed volunteers would
tend to present a practically insurmountable problem of feasibility.

A randomized screening trial is profoundly different from that screening-related

intervention trial. It contrasts screening with no screening; and while it generally
has been quantitatively meaningless (if not downright misleading and grossly so;
sect. 11.3), with proper design and its corresponding execution it too would be
properly informative about the curability gain and its consequent mortality reduction
attendant to the cancers early detection. But the requirements are quite onerous,
and it is important to appreciate that even an ideal trial of this type just like the
intervention trial above can never meaningfully address the mortality consequence
of the screenings introduction into a community. Both of these topics we address
below.
For an RST to be quantitatively meaningful, there needs to be a segment in the
cohort time of follow-up such that all members of the screened subcohort have, in
this period, histories of full survival benefit of the screenings (periodic) really,
to say it again, of screening-associated early treatment of the cancer. It is in that
segment of the trials cohort time that the screened subcohorts mortality from the
cancer is reduced to the extent in the proportion that the early interventions cure
otherwise fatal cases of the cancer.
Harkening, again, back to the example (hypothetical) in section 11.3 above, the
mortality-reducing consequence of the baseline screening (with no screening as the
alternative) occurs, completely, in the cohort time interval of 212 years (after the
cohort T0 , the time of the randomization to the baseline screening, only , or to
no screening); only in this period of time could the baseline screening bear on
11.4 Meaningful Types of Trial 119

the cancers outcome. But in this segment of the cohort time the attained rate
of cures at baseline, in otherwise fatal cases, is not manifest in (the complement
of) the ratio of the cumulative rates of mortality from the cancer (in one minus
this ratio), as included in the screened-arm experience are deaths from cases that
only subsequently were detectable but were not detected by the screening. The
mortality ratio in this segment of follow-up is, thus, substantially diluted toward
unity (from the 1.00  0.69 D 0.31; cf. sect. 11.3 above).
Now, consider an RST that involves also regular repeat screenings annual ones,
say and suppose the latent cases detectable by these screenings also would have
their fatal outcomes 2-12 years following the detections in the absence of early
treatment. In this RST there can be a period of follow-up time in which deaths
from the cancer in the trials screening arm are reduced by a proportion identical
to the screening-associated rate of cures of otherwise fatal cases. This period, if
there, begins at 12 years of follow-up on the condition that the screenings duration
exceeds 12  2 D 10 years; and the length of this period is the same as the duration
of that exceedance.
In this period of maximal reduction in the mortality, if it exists (by the duration of
the screening), all deaths from the cancer in the trials screening arm are associated
with a history of diagnosis under the screenings repeat screenings and their
associated early treatment. The diagnoses about these cases were either screening-
based or prompted by symptoms appearing between scheduled screenings. In these
cases the early diagnosis its associated treatment failed to halt the cancers
progression to its fatal outcome.

Addressed in the foregoing has been study of the extent of a cancers enhanced
curability by means of experimental experience with the way case-fatality rate
from the cancer is affected by screening (by its associated early treatments).
With this clinical matter settled (more or less), the genuinely epidemiological
question remains, as it is about the mortality consequence of the introduction of
a public-health program to reduce the deaths incidence density (sect. 11.2). The
program could be one of mere reimbursement of the cost of the initial test in the
diagnostics constituting the screening, coupled with education of the public about
the screening (in the clinical domain); or it could be education about, and provision
of, community-level service as a matter of providing the screenings initial testing
(sect. 11.2).
So, the research clinical research outlined above actually is about curabil-
ity implications of mortality experience with experimental screening, while the
question relevant to epidemiological practice is about the mortality implications of
that curability, this in reference to the contemplated public-health program(s). This
question is not addressed by this research nor is it subject to being answered by any
epidemiological research. But to the extent depending on indications and quality
assurance the screening is clinically justified, it also is public-health-justified,
now that clinical medicine, too, is public-health medicine (given national health
insurance). For more on these challenging matters, see Appendix 2.
120 11 Studies on Screening for a Cancer

11.5 Meaningful Etiologic-type Study

Lack of screening for a cancer (of its associated early treatment) is etio-
logic/etiogenetic to death from the cancer. The extent to which this is so naturally
is studyable only where the screening has been available (but not routinely used).
And it is studyable to learn about the here-essential parameter of Nature the
curability gain from screening-based early diagnoses only in settings in which the
screening has been available long enough.
These opportune settings ones in which the screening-afforded gain in curabil-
ity could be quantified by means of etiogenetic studies already are quite common;
and they really should be made use of for this purpose, for the serious problems
inherent in the gold standard trials in this quantification (sects. 7.2, 11.3, 11.4) are
not shared by their etiogenetic counterparts.
The study base needs to be defined (as to age, i.a.) in such a way that associated
with each person-moment in it there was an opportunity for the screening in the
relevant past, 212 years ago in the example in section 11.3 above. Thus, if the
cancer at issue in this example is that of breast and it was introduced over 12 years
ago to women 5070 years of age and has continued to be available only to women
in this range of age, the study base needs to be restricted to women 6272 years of
age; for, women younger than 62 years of age did not have access to screening as
long as 12 years ago, and women more than 72 years of age did not have that access
as recently as two years ago. If the screening for that range of age was introduced
less than 12 years ago, no-one in the population had access to the screening for the
entire period of 212 years ago, so that a valid study base cannot yet be defined in
this setting.
For the case series (of death from the cancer, in a suitably defined study base
as to age, i.a.), the index history need not be that of regular screening throughout
the period of 212 years prior to the death (to continue the example). A sufficient
substitute for this has to do with the time of the cancers diagnosis (rule-in dgn.): at
the time of the diagnosis the person was under screening, so that the diagnosis either
resulted from screening or was a symptoms-prompted interim diagnosis within a
regular-type cycle of the screening. (Additional screenings outside this particular
cycle could not have had any potential role in prevention of the death, so that
history about those screenings is irrelevant in the definition of the index history.)
The corresponding reference history in the case series is: diagnosis while not under
screening; that is, diagnosis not preceded by on-schedule initial test, with negative
result of the screening.
Definition of the index history for the base series presents the challenge that there
is no counterpart for the routine positive history of diagnosis (rule-in) about the
cancer in the case series, no inherently corresponding time on which to focus it and,
consequently, the reference history also. The solution of choice to this dilemma is
to use the case series as the guide to this timing: as a case (of death from the cancer)
has occurred at a particular time in age and calendar time, this prompts a number of
probes into the study base, into the same stratum of time (age and calendar time);
11.5 Meaningful Etiologic-type Study 121

and for this set of person-moments in the study base, the screening history is defined
in reference to the same time at which the first manifestation of the cancer (positive
test result, or symptom) occurred in the member of the case series.
With the case series from the source base restricted to the limited range of age
that is to characterize the study base (see above) and with the base series time-
matched to this, another restriction also is of note here. Study of the curability gain
requires that only one detected case of the cancer characterizes the histories in the
case series, with none the counterpart of this in the base series.
If this is all that is in the scientific essence of the study design, then the data from
the study lead to a set of 2  2 tables, each specific to a confounder stratum at the
time of the outcome:
I R Total
c1j c0j Cj
b1j b0j Bj

where I and R denote the index and reference history (of screening), respectively,
and Cj and Bj are the stratum-specific sizes of the case and base series, respectively
(cf. sect. 5.4). And the confounder-adjusted empirical measure of the rate ratio (for
death from the cancer, contrasting I with R) is

b
RR D
X
j
c1j =
X 
j
b1j c0j =b0j


(cf. sect. 5.4); and the corresponding empirical measure of the curability gain from a

b
round of screening (from screen diagnoses when added to interim diagnoses within
the round) is the complement of this, 1  RR .

b
If the data are examined separately by categories of the etiogenetic time
(retrospective as of the outcomes in the two series), presumably the RRs for more
distant times are seen to be lower than those for more recent times (see Table 11.1
in sect. 11.3). This, while true, is irrelevant for the quantification of the curability
gain so long as the rate of the screenings use has been essentially constant over that
range of retrospective time as evinced by the I/R ratio in the reference series over

b
that span of time. Otherwise averaging of the time-specific RRs their logarithms
is called for, with the RR the antilog of this average (unweighted).
The foregoing implicitly involves the premise of no confounding in the study
base according to extraneous determinants other than calendar time and age of the
incidence density of death from the cancer. But insofar as there is, in the study base,
potential confounding by other factors, they need to be accounted for in the model
for the incidence density and, accordingly, in the logistic model fitted to the data
perhaps leading to stratification according to a confounder score (sect. 5.4).
As this is a non-experimental study on causation (of the etiogenetic sort),
the question is, as always in this genre of studies, whether all of the potential
confounders were identified and adequately documented, for control in the synthesis
of the data to the result of the study. Now, undergoing a round of screening or
122 11 Studies on Screening for a Cancer

refraining from this is a purposive course of action, and the indications of risk
bearing on the decision about this action are known to the decision-makers and
hence knowable to the investigators. The question is, though, whether the facts about
these can validly be ascertained for both of the study series from the next of kin in
both of them.

In closing here, the comparative big picture. If a randomized screening trial is

designed for long-enough continuation of the screening and the mortality contrast is
focused on the right segment of follow-up time, it is still prone to produce a highly
downward-biased empirical value for the curability gain due to poor adherence to
the assigned long-term screening and non-screening. In an etiogenetic study, such as
was outlined above, the principal challenge in validity assurance is confounding by
indication (for the screening), due to absence of randomization. But such residual
confounding as on this basis will characterize the result is negative confounding,
again biasing the empirical value of the curability gain downward but much less
so than the bias from the best possible screening trial.
This leads to the need to appreciate fusion of these two types of study: when the
experimental practice in an RST has continued long enough, this trials population-
time of follow-up constitutes a suitable source base for an etiogenetic study such as
was outlined here yielding a result less biased than the one from the right segment
of the follow-up in the intention-to-screen spirit of the trial, as the bias from the
(quite considerable) non-adherence to the protocol is avoided.
But as it is at present, reviews tend to focus on the RSTs alone; and in this,
they tend to produce derivative results on mortality reduction as though the same
parameter value were assessed with whatever duration of the screening and follow-
up, and with no regard for the non-adherence to the contrasted regimens of screening
versus non screening. See Appendix 2.
Chapter 12
Some Paradigmatic Studies

Abstract Thomas Kuhn, the eminent historiographer of science, is particularly

well known for his description of the role of paradigms and paradigm shifts in
science, especially in sciences in which eminent controversies are common social
science and psychology, in particular. Paradigms, he says, are model problems
and model solutions that centrally influence research in the framework of normal
science. Along this path there is prone to develop a crisis, leading to revolutional
science and its associated new paradigm(s).
In chapter 2 it became evident that modern chronic-disease epidemiological
research is, in respect to such central problems as nutritional etiogenesis of
cardiovascular disease and cancer, and the intended consequences of screening
for a cancer, highly controversial about the implications of the accrued, very
extensive evidence. Arguably at least, however, epidemiological research is, now,
in a Kuhnian crisis even though this is not commonly being recognized by the
community of researchers concerned with problems of community medicine. As
for model problems, screening for a cancer continues to be viewed as a topic
quite eminent in purportedly epidemiological (rather than clinical) research. And
the controversies of social science may come to add to those endogenous to the
biomedical research on the etiogenesis of illness, now that social epidemiology is
on the ascendency.
Population-level epidemiological research does lend itself to normal-science
paradigms in respect to theory, but the prevailing methodologic ones are in need
of revisions while paradigms of the studies objects design are not only underde-
veloped but in all essence non-existent. The randomized trial adopted from clinical
research into epidemiological intervention studies is, now, a very false paradigm
for research on screening for a cancer, having earlier been a false paradigm of the
cohort study on etiogenesis.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 123

DOI 10.1007/978-94-007-4537-7 12, Springer ScienceCBusiness Media Dordrecht 2012
124 12 Some Paradigmatic Studies

While paradigms for research are important and tenable paradigms are
particularly desirable, they should not be seen to define the boundaries of productive
research. Some problems require extraparadigmatic solutions, even ones worthy of
Nobel Prize for Physiology and Medicine.

12.1 The Role of Paradigmatic Studies

A student preparing for a career in epidemiological research will do well studying

some of the history and philosophy of science in general, orientationally through
Thomas Kuhns The Structure of Scientific Revolutions (1970).
Based on his review of the history of social and psychological as well as natural
science, Kuhn adduces the concept of normal science, defining it as research
firmly based upon one or more past scientific achievements, achievements that some
particular scientific community acknowledges for a time as supplying the foundation
for its further [studies] (p. 10). He explains that: Today such achievements are
recounted, though seldom in their original form, by science textbooks, elementary
and advanced. : : : [These writings serve] for a time implicitly to define the
legitimate problems and methods of a research field for succeeding generations of
[researchers in the field] (p. 10).
Such achievements, ones that provide models from which spring particular
coherent traditions of scientific research Kuhn terms paradigms (p. 10). Men
whose research is based on shared paradigms are committed to the same rules and
standards for scientific [research]. That commitment and the apparent consensus it
produces are prerequisites for normal science, i.e., for the genesis and continuation
of a particular tradition (p. 11).
History suggests that the road to a firm research consensus is extraordinarily
arduous (p. 11). Thus, fundamental disagreements characterized, for example, the
study of motion before Aristotle and of statics before Archimedes, the study of heat
before Black, of chemistry before Boyle and Boerhaave, and of historical geology
before Hutton. In parts of biology the study of heredity, for example the first
universally received paradigms are still more recent; and it remains an open question
what parts of social science have yet achieved such paradigms at all (p. 11).
Once a paradigm an accepted model or pattern (p. 23) has come into being,
it is an object for further articulation and specification under new or more stringent
conditions (p. 23).
Researchers in a given field of science agree in their identification of a paradigm
without agreeing on, or even attempting, to produce, a full interpretation or
rationalization of it. Lack of a standard interpretation or of an agreed reduction
to rules will not prevent a paradigm from guiding research (p. 44; italics in the
original). Scientists work from models acquired through education and through
subsequent exposure to the literature often without quite knowing or needing
to know what characteristics have given these models the status of community
paradigms (p. 46).
12.2 Paradigmatic Causal Studies 125

In the development of any science, the first received paradigm is usually

felt to account quite successfully for most of the observations and experiments
easily accessible to that sciences [researchers]. Further development, therefore,
ordinarily calls for the construction of elaborate equipment, the development of an
esoteric vocabulary and skills, and a refinement of concepts that increasingly lessens
their resemblance to their usual common-sense prototypes. That professionalization
leads : : : to an immense restriction of the scientists vision and to a considerable
resistance to paradigm change (p. 64; italics ours).

12.2 Paradigmatic Causal Studies

In epidemiological research at present, normal-science paradigms dominate eti-

ologic/etiogenetic research in terms of cohort study and case-control study
in particular (sects. 2.6, 6.1, 6.2); intervention research in terms of randomized
controlled trial (ch. 9); and screening research on a cancer also in terms of RCT
(sect. 11.3).
The cohort study and case-control study are different not because of a corre-
sponding difference in what is being studied; the distinction is merely methodologic.
Yet the respective results appear to be different. Routinely reported from a cohort
study now is a result in terms of relative risk or, alternatively, risk ratio or rate
ratio, while reported from a case-control study most commonly now is something
quite peculiar to this type of study: odds ratio.
From an RCT on an actual intervention the reported result now quite routinely
is a hazard ratio, while from an RCT on screening for a cancer the reported result
is (proportional) reduction in mortality (from the cancer, due to the screening;
sects. 7.3, 11.3).
This illustrates what we quoted above as a general observation of Thomas
Kuhn, when applied specifically to epidemiological researchers; they agree in their
identification of a paradigm without agreeing on, or even attempting to produce,
a full interpretation or rationalization of it. As we have let on, we do not regard
those paradigms, now generally treated as received truths, as having a tenable
rationale for reasons that we have put forward (sects. 6.1, 6.2, 6.3, 6.4, 6.5, 9.3,
11.3) and leave for the reader to weigh and consider (Bacon; sect. 1.4).

Below, we sketch alternatives to those prevailing paradigms, all of them concerned

with causal research. These sketches do not bring forth anything essential we
havent yet said or implied. Rather, they represent an attempt at crystallization of
what weve already said, at concentration of the readers mind on some essentials in
introductory but nevertheless critical study of the theory of epidemiological research
(cf. Preface).
We are, as has become apparent, of the mind that in etiogenetic research it is
time for a paradigm shift (Kuhn) away from the cohort and case-control studies,
126 12 Some Paradigmatic Studies

which we see as representing the cohort and trohoc fallacies, respectively (sects. 6.1,
6.2). We suggest replacement of these by a singular conception of the nature of
(population-level) etiogenetic studies: the etiogenetic study (sect. 6.5).
As for intervention research, we can be seen to adhere to the prevailing,
randomized-trial paradigm, except for a proposed modification of it. We propose
that the object of the trial should not be a hazard ratio but an intervention-
prognostic probability function, and we delineate the way in which this can be
studied in the framework of such data as already are routinely collected in these
trials (sects. 9.2, 9.3).
Mass application of a screening test is taken to be an important type of
intervention preventive intervention by epidemiologists and by promulgators
of policies for healthcare (sect. 2.6, i.a.). For this reason we address screening
research specifically, research on screening for a cancer in this book (ch. 11,
i.a.). But we hold those ideas of epidemiologists and promulgators of healthcare
policy to be mistaken, and screening for a cancer to be a topic first and foremost a
diagnostic topic in clinical medicine (sects. 11.1, 11.2, i.a.). It is only because of
that common misunderstanding that we address screening (for a cancer) in this book
on epidemiological, rather than clinical, research. Further attention to it we give in
Appendix 2.

Nutritional epidemiology was the principal focus in most of the chapter on

epidemiological knowledge at present (ch. 2), due to the centrality of thinking
about nutrition (incl. nutritional supplements) in modern preventive medicine in
industrialized countries in particular.
The big picture that emerged had these features of particular note here: nutritional
etiology/etiogenesis of a phenomenon of health can be very chronic (as in, e.g., overt
cases of various cancers) or quite acute (as in, e. g., weight gain); and research on
nutritional etiogenesis of illness has had notable deficiencies in its objects designs
in respect to the temporal aspect of the etiogenesis and, also, in the formulation of
the causal contrasts in the objects of study (chs. 2, 7).
A paradigmatic, exceptionally successful nutritional-etiogenetic study would be
an experimental one. For, the causal contrasts in this research are too intricate
(sect. 7.1) for proper documentation from non-experimental diets. Of course, only
relatively short-term experimentation would be practicable, and the here-envisioned
paradigmatic experiment thus is instructive about what hypotheses are, and which
ones are not, subject to truly meaningful human-level testing. (Science does have its
limits, and human experiments on, e.g., antioxidant deficiency in the etiogenesis of
various cancers have been unrealistic and, hence, misleading; sects. 2.2, 7.1).
As Gary Taubes explains in his Good Calories, Bad Calories (2007), a practi-
cable experiment, suitably designed and executed, could settle the now-so-topical,
burning question about the dietary etiogenesis of weight gain and, hence, obesity
whether its excess calories that make us fat (the conventional wisdom) or
purely the effect of carbohydrates on insulin and insulin on fat accumulation (the
carbohydrate hypothesis) (p. 466).
12.3 Successful Extraparadigmatic Studies 127

At issue is a crucial experiment with implications much beyond obesity: if the

carbohydrate hypothesis is correct in respect to obesity, it thereby presumably
is correct in respect to the genesis of the metabolic syndrome, seen to be at
the root of obesity (sect. 2.4) and, independently of this, of many other illnesses
besides: atheromatosis manifesting in cardiovascular disease (sect. 2.1); cancers
(sect. 2.2); ageing (acceleration of its progression; sect. 2.3); dementias of old
age (incl. Alzheimers); hypertension (sect. 2.5); and yet other illnesses too as
Taubes explains in that remarkable book (with the subtitle, Fats, Carbs, and the
Controversial Science of Diet and Health).
The paradigmatic study suggested here the crucial experiment would be about
changes in weight and in level of insulin (and perhaps changes also in levels of
glucose, triglycerides, and fractions of cholesterol) in blood (plasma) over a span
of time (months), in (sub)cohorts on different diets (randomly assigned) for that span
of time. The study subjects admissibility criteria would not be of critical importance
to this study, but the definition of the set of diets would be.
The diets in this experiment would represent a contrast, for one, in the daily
quantity of dietary energy intake conditional on the type-composition of the sources
of energy (incl. beverages); and, conversely, there would be a type-composition
contrast conditional on the quantity of energy. The paradigmatic experiment would,
thus, represent a factorial design in this sense. The two levels of energy intake would
represent an increased and the prevailing rate of energy intake, respectively; and the
two type-compositions would represent an increased and the prevailing proportion
of energy from carbohydrates.
The carbohydrate hypothesis about the etiogenesis of obesity that would be
addressed by this study was widely, if not universally, held to be true until the
1970s (Taubes, p. 461), when the officially engineered shift away from fats in diets
and correspondingly increased consumption of carbohydrates started in the U.S.,
followed by the major epidemic of obesity. The paradigmatic study just outlined
and in essence envisioned by Taubes (p. 466) could re-establish the carbohydrate
conception of the etiogenesis of obesity and many other diseases of civilization.

12.3 Successful Extraparadigmatic Studies

A student being introduced into epidemiological research should not be left with the
idea that hypotheses about etiology/etiogenesis derive solely from epidemiological
studies on humans (with populations as units of observation perhaps; cf. sects. 2.1,
2.2, 2.6), or that testing of such hypotheses is best done by means of non-
experimental etiogenetic study (sect. 6.5). The route to success may well be
extraparadigmatic. In what follows, we illustrate this by the genesis of the idea that
the etiogenesis of peptic ulcer involves a role for a bacterium.
The research, in Australia, that led to the discovery of the role of Helicobacterium
pylori as the principal agent in the etiogenesis of peptic ulcer, gastric and duodenal,
128 12 Some Paradigmatic Studies

led the investigators to Stockholm, to receive the Nobel Prize. In outlining this
research, we draw from and direct the student to two articles:
1. Van der Weyden M et alii. The 2005 Nobel Prize in physiology or medicine.
Med J Aust 2005; 183: 612-4.
2. Atwood KC. Bacteria, ulcers, and ostracism? H. Pylori and the making of a
myth. Skeptical Inquirer Magazine 2004; vol. 28.6. Available at: http://www.
csicop.org/si/show/bacteria ulcers and ostracism h. pylori and the making of
a myth/. Accessed November 9, 2011.
The essentials of this research, according to reference 1 above, illustrate some
of the human hallmarks of revolutionary research. These include:
being at the right place at the right time, and seeing what other people had seen
but thinking of what nobody else thought [ref.];
the role of serendipity;
a passion for research that abandons personal safety with self-experimentation;
and
the inevitable resistance of the medical establishment as research undermines
current dogma.
That last point is, however, disputed, seen to be a misrepresentation of what
merely was the skepticism that is inherent in the scientific outlook (ref. 2).
Notable about the genesis of the hypothesis is that there were no insights
from basic science suggesting it, nor did the hypothesis arise from data on the
occurrence of peptic ulcer in any human population. It arose from Barry J. Marhalls
review of J. Robin Warrens records on his 25 cases of endoscopic biopsy, many of
them showing mysterious, spiral micro-organisms. The diagnoses in these cases
were duodenal ulcer in two, gastric ulcer in seven, gastritis in 12, and erosions and
scars in four (ref. 1).
The two future Nobel laureates next conducted a prospective study on 100 cases
of endoscopy, in which they, as the first researchers ever, came serendipitously
upon successful cultivation of the mysterious organism, which they named Heli-
cobacterium pylori. In the resulting data there was a strong association between
gastritis and the presence of this bacterium. It was found in all patients with
duodenal ulcer and 80% of patients with gastric ulcer. In contrast, the presence was
rare in patients with non-steroidal drug-related ulcers (ref. 1).
Then, Marshall and his colleagues showed that bismuth salts (which had been
used to treat gastritis and peptic ulcer disease for many years) killed H. pylori in
vitro; and, in clinical studies, that bismuth cleared H. pylori but the infection would
recur unless metronidazole was added to the regimen (ref. 1).
While the first prospective studys acceptance for publication by The Lancet was
an exceptionally protracted process following rejection by the Gastroenterologic
Society of Australia Marshall carried out another type of study as well, very small
and very dramatic. He asked [a colleague] to perform a gastric biopsy on him and
then ingested a pure culture of H. pylori (109 organisms). All was well for 5 days, but
then he developed halitosis [bad breath], morning nausea, and recurrent vomiting of
12.3 Successful Extraparadigmatic Studies 129

acid-free gastric juice. A gastric biopsy on Day 10 showed severe acute gastritis and
many H. pylori. The symptoms spontaneously resolved after 14 days, : : : (ref. 1).
This constituted highly suggestive evidence that the organism caused gastritis.
But it was far from conclusive, because it involved a single subject and was reported
by the very author most wedded to the hypothesis. Thus, replication by others would
have been required. Perhaps more important was that the subject, who was none
other than Marshall himself, failed to develop an ulcer. Note also that the disease
resolved without treatment (ref. 2).
What finally convinced doubters of both cause and treatment was something
that by its very nature took several years to establish. : : : The first trial that was
both large enough and rigorous enough to be noticed was conceived by Marshall
and Warren in 1984 : : : [They] reported that the recurrence rate of duodenal ulcer
was much lower in patients whose H. pylori were eradicated than in those whose
bacteria were not, : : : Since the authors were the original proponents of the bacterial
hypothesis, moreover, any firm conclusions would first require confirmation by
others. This was not ostracism; it was appropriate scientific skepticism (ref. 2).
While Marshall and Warren have irrevocably changed clinical practice and have
alleviated much human suffering (ref. 1), they also have made a major contribution
to the discipline of etiogenetic epidemiological research: they have, unwittingly,
reminded both teachers and students that this research, when successful in its
hypothesis generation, and testing of the hypothesis too, is not inherently a matter
of competent adherence to established paradigms, or to what used to be termed
the epidemiologic method. Their work is paradigmatic of the potential prowess of
extraparadigmatic etiogenetic research.
Chapter 13
From Studies to Knowledge

Abstract Whereas the aim of basic epidemiological research is to provide for

the development of products for use in the practice of community medicine, in the
population-level counterpart of this the aim is to advance the knowledge-base of the
practice.
The population-level research produces evidence about (the magnitudes of) the
parameters that constitute the objects of any given study, original or derivative; it
does not produce knowledge about the objects of study. The evidence, is, however,
of little value if it does not serve to advance knowledge about the objects of study.
The knowledge, to the extent that it is achieved, serves as an input to decisions in
the practice; but it does not in itself translate into rational decisions in the practice
and, thus, not into guidelines or even recommendations for practice.
Major lines of population-level epidemiological research have, thus far, led
to more controversy than knowledge; and where, exceptionally, the need for
knowledge has been particularly keenly felt, very elaborate projects of translating
evidence into knowledge have been mounted.
For the fact that the accruing evidence does not, by the very nature of science,
inherently lead to knowledge relevant to the practice of community medicine there
is, we suggest, an identifiable and removable cause. We believe it to be
the prevailing nature of medical journalism. Journals of epidemiology now are
not focusing on the concerns of practitioners of community medicine, including
updatings of the knowledge-base of this. They play into the common fallacy or
conceit that practitioners of medicine are scientists interested in the continual
accrual of evidence and on this basis updating their personal opinions about the
objects of study.
In the last section of this chapter we outline how epidemiological journal-
ism should be innovated so that there would be a natural progression from
epidemiological research to epidemiological knowledge; and in Appendix 3 we
outline how this natural process could be supplemented by work in the style of
WHOs International Agency for Research on Cancer.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 131

DOI 10.1007/978-94-007-4537-7 13, Springer ScienceCBusiness Media Dordrecht 2012
132 13 From Studies to Knowledge

13.1 Original Study versus Derivative Study

In epidemiological research, studies pieces of research, research projects (distinct

from research programs) on a given set of objects of study are of two fundamental
kinds. An original study is either the first study on the objects at issue or it is one
of this studys replications; and given a set of completed studies on the same set of
objects, a derivative study is one based on these.
In the ideal world of epidemiological research, all of this would be very
straightforward. The first one of the original studies would always address a well-
designed occurrence relation and have a well-designed methodology to study it;
the execution of the methods design would be flawless; and all of this would be
replicated with flawless objectivity. The first original studys replications and the
reports of these would differ from the first study only in terms of their spatio-
temporal particulars and amounts of information (efficiency, size), never in their
scientific substance. Replications of the first study would, thus, merely contribute to
the available amount of information about (the magnitudes of) the object parameters
under study. And at any given stage of the accumulation of the information, a single
derivative study would suffice, simply synthesizing the information from all of the
completed original studies (as a matter of information inverse-variance weighted
averaging of the parameters empirical values from the original studies).
Actual epidemiological research is, still, far from this ideal, in all of the elements
in the ideal. Emblematic of this are the prevailing practices in the ultimately relevant,
derivative research. In the absence of the availability of a study base for simple
synthesis of the results of the entirety of original studies, all well-designed and well-
executed, on an expressly defined set of objects of study (for an expressly defined
domain of this), systematic reviews draw from the original studies on a given,
loosely-defined topic substituting for a given, expressly-defined object function
(for a given domain) involve the investigators judgements about the admissibility
of particular original studies as inputs into the derivative study; and then, the results
are synthesized across varying objects of study (forms of the study result) and
methodologies (ultimately as a matter of the designs implementation) with the
synthesis possibly degenerating into mere tallies of positive and negative studies,
in disregard of, even, the respective amounts of information (cf. sect. 2.6). Things
are even worse: the study base for these reviews generally is seriously selection-
/publication-biased (cf. sect. 8.2).
In the needed improvements the beginning naturally would be in the realm of
original studies and, in these, in their objects designs, for we are, alas, unaware of
even a single epidemiological systematic review that could focus on expressly
defined, single occurrence relation for an expressly defined domain the set
of studies on this. Equally eminent as a needed improvement is elimination of
publication-bias as an engrained feature of epidemiologic-scientific journalism (cf.
sect. 13.6).
13.3 Interpretation of the Evidence 133

13.2 The Evidence from a Study

An epidemiological study (on the human level) is a project to produce evidence on

(the parameters involved in) an epidemiologically meaningful occurrence relation.
The evidence from such a study is imbedded in the report from it (along with content
of no evidentiary burden, starting from specification of the authors of the report and
commonly ending in a conclusion). The reported evidence is either original or
derivative in nature (sect. 13.1 above).
An epidemiological study of the original variety can be thought of as a single,
unreplicated measurement of the magnitudes of the parameters in the object function
of the study; and a derivative study based on a set of original studies can be thought
of as making use of replicated measurements on the parameters values in deriving
the best possible single reading on each of them. This is true of hypothesis-testing
as well, in which the measurement (multidimensional) is used for discrimination
between the null and non-null values of the parameters in the object function.
Evidence about the magnitude of the object of a measurement is first and
foremost constituted by the result of the measurement, in epidemiological research
(on the human level) by the set of empirical values of the object parameters from
the study (original or derivative). But a critically important added element in the
evidence is specification of the methodology in the production of the result, as this
bears on the results evidentiary burden through its bearing on the results (precision
and) validity.
In reports on epidemiological research it is always possible and, hence,
necessary to supplement the object parameters empirical values with measures
of imprecision of these (standard errors or confidence intervals). These are a
reflection of the studys degree of informativeness (about the parameters values),
determined by the studys efficiency together with its size. They thus represent not
the studys results per se but one of the two important qualities of these.
For a study results degree of bias lack of validity no measure can be derived
from the study; if it could, the corresponding requisite correction would be made in
the result. The degree of bias in the result remains a matter of judgement, by whoever
is concerned with the result from the study the systematic review investigators
in the context of any original study, and the relevant scientific community as for a
derivative study.

13.3 Interpretation of the Evidence

Evidence from an epidemiological study is, as just set forth, imbedded in the report
from the study; and specifically it is, so we implied, constituted by words and
numbers describing the studys results on the magnitudes of the parameters in
134 13 From Studies to Knowledge

the designed object of the study, these together with description of the genesis of
these results, verbal and numerical description of this too, augmented by measures
of the consequent imprecisions of the parameters empirical values.
Interpretation of this evidence we take to be formation of a view of what this
reporting of the evidence actually means, what the thus-communicated evidence
actually is. We thus sharply distinguish between interpretation of evidence, in this
meaning of it, and inference from the evidence (sect. 13.4 below).
In the context of the present level of development of the shared concepts and
terms among epidemiological researchers, the interpretation of the report of, and
hence the evidence from, epidemiological studies can be challenging, insofar as
avoidance of misunderstanding is the concern (sect. 2.6). To the extent that this
is the case, it is a serious problem. For, Where concepts are firm, clear and
generally accepted, and the methods of reasoning and arriving at conclusion are
agreed between men (at least the majority of those who have anything to do with
these matters), there and only there is it possible to construct a science, formal or
empirical. So wrote Isaiah Berlin in his venerable The Proper Study of Mankind:
An Anthology of Essays (Hardy H, Hansheer R, Editors; 1997; p. 61).
Emblematic about the status quo are some central routines in epidemiological
study reports at present. There still eminently is cohort study and case-control
study, rather than merely the etiologic/etiogenetic study (sect. 6.5). And even
though these two are thought of as alternatives to each other, reported from the
former usually is a result in terms of relative risk, with odds ratio the common
counterpart of this from the latter (sect. 2.6) both of these terms properly
referring to a parameter rather than the result of a study (which in the context
of an etiologic/etiogenetic study generally is the ratio of two empirical rates,
two incidence densities). Significantly increased risk in the determinants index
category is a common expression for the strictly particularistic fact that the rate
in the study experience was higher in this category, and that the difference was
statistically (sic) significant the actual risk implications of this, if any, being
a matter not of experience per se but inference from it (sect. 13.4 below). Just
as unobservable/undocumentable as causal increase in risk or rate is, of course,
reduction (inherently causal) in it, also commonly reported as a purported result
of an epidemiological study (sect. 2.6). Neither one of them can be found in, or
shown by, an epidemiological study.
Particularly great challenges of interpretation are exemplified in Appendix 2.

13.4 Inference from the Evidence

An epidemiological researcher, and any consumer of the research likewise, faces a

profound yet unavoidable limitation of the studies, derivative ones included. How-
ever well-designed be the object function of the study and however good the design
of the studys methodology as well as execution of this, the inescapable problem is,
for a start, that the object of study has an abstract placeless and timeless domain
13.5 Knowledge from the Evidence 135

as its referent, while the study result inherently has a particularistic referent (i.e.,
the study base, spatio-temporally specific). This constitutes a notable challenge for
learning about descriptive acausal, purely phenomenal objects of study already.
The challenge is greater yet when the object of study is causal noumenal in this
sense and the study proper is, as is commonplace, non-experimental and subject
to biases even in descriptive terms.
Thus, however correct may be someones interpretation of the evidence from a
study (as to what the report from the study means in this respect; sect. 13.3 above),
his/her inference about the object of study in the light of the evidence remains a
matter of subjective judgements, leading only to his/her personal opinion about the
abstract truth in question.

13.5 Knowledge from the Evidence

Scientific research, epidemiological like any other, is a quest for scientific knowl-
edge for presumably truthful insights into the abstract.
Whereas evidence from epidemiological research on any given object of inquiry
provides for only personal opinions about the truth at issue (sect. 13.4 above),
the challenge that immediately results from this is reasonable conceptualization
of the nature, the essence, of epidemiological knowledge. And with this resolved,
it remains to understand how the evidence on any given object of desired, epi-
demiological knowledge can be translated into that knowledge. These challenges
have remained too little addressed and incompletely resolved, with the consequence
that epidemiological knowledge, even on obviously important topics, still tends to
remain ill-defined or controversial (ch. 2).
Pre-scientifically it used to be known firmly, for a very long time that disease
commonly is a matter of aberrant distribution among the four humors in the body.
It also was known that miasma vapors emanating from the soil, swamps, and
decaying organic matter are causal to disease, and that disease can be alleviated
(or perhaps even cured) by blood-letting.
In this modern era of health science it has just recently become known, from
epidemiological research, that dietary fats are causal to cardiovascular disease, while
research previously had persuaded experts that it was dietary carbohydrates that had
this effect; and the latter may get to be the knowledge in the future as well (sect. 2.1).
Epidemiological knowledge, even in this era of epidemiological science (in the
research meaning of this), is not inherently true; it is, even, subject to being a denial
of previous scientific knowledge science being, by its nature, self-corrective rather
than simply (linearly) progressive. It is, first-off, knowledge in the sense of experts
(in the relevant scientific community) having come to a more-or-less common,
shared belief about the truth at issue. These experts can defend their shared belief;
their knowledge is active in this sense.
Secondarily, there is others knowledge about the experts shared belief, passive
knowledge in this sense, knowledge that cannot be justified by those who possess
136 13 From Studies to Knowledge

it. Most people know that dietary fats are causal to cardiovascular disease, but only
experts on the subject can justify this belief (with the justification subject to being
mistaken).
The public-health authorities of Denmark, in 2011, adopted a public policy
consequent to their second-hand knowledge secondary to first-hand, expert
knowledge about the health consequence of fats in the diet. They introduced a
tax on purchases of foods high in saturated fats. This policy may well be ill-founded
and even counterproductive: the result may well be, for one, an epidemic of obesity
in Denmark, just as there already is in the fats-averse U.S., for example.
Express formulation of expert knowledge on matters epidemiological has not
been prone to emerge spontaneously in the relevant scientific community; it
has required deliberations and formulations and by topic-specific expert com-
mittees. But this, too, has been problematic, as that which emerges is prone to
be dependent on which set of experts constitutes the committee, among other
factors.
Spontaneous movement from evidence to knowledge within the relevant sci-
entific communities could be, so we believe, substantially enhanced by suitable
innovations in medical journalism (sect. 13.6 below); and related to this, the culture
of task forces on preventive healthcare could and should be substantially
improved (Apps. 1, 2, 3).

Two examples of express efforts to derive knowledge from the evidence produced
by epidemiological research (incl. bench research; sect. 1.2) are particularly
instructive: the work on the Smoking and Health report (1964) of an ad-hoc
advisory committee to the Surgeon General of the Public Health Service of the U.S.
Department of Health, Education, and Welfare; and the work on the continually
accruing IARC Monographs on the Evaluation of Carcinogenic Risks in Humans,
IARC being the International Agency for Research on Cancer within the World
Health Organization.
The smoking-and-health committee received from the National Library of
Medicine a total of 1,100 articles (p. 14). In addition to the special reports
prepared under contracts, many conferences, seminar-like meetings, consultations,
visits and correspondence made available to the Committee a large amount of
material and a considerable amount of well-informed and well-reasoned opinion
and advice (p. 15).
There were an uncounted number of meetings of subcommittees and other lesser
gatherings. Between November 1962 and December 1963, the full Committee held
nine sessions each lasting from two to four days (p. 15).
All members of the Committee were schooled in the high standards and criteria
implicit in making scientific assessments; if any member lacked even a small part
of such schooling he received it in good measure from the strenuous debates that
took place at consultations and at meetings at the subcommittees and the whole
Committee (p. 19).
It is advisable, however, to discuss briefly certain criteria which, although
applicable to all judgements involved in this Report, are especially significant for
13.5 Knowledge from the Evidence 137

judgements based upon the epidemiologic method : : : When coupled with the other
data, results from epidemiologic studies can provide the basis on which judgements
of causality can be made (pp. 1920; italics ours).
In carrying out studies through the use of this epidemiologic method, many
factors, variables, and results of investigations must be considered to determine first
whether an association actually exists between an attribute or agent and a disease.
: : : If it be shown that an association exists, then the question is asked: Does the
association have a causal significance? (p. 20).
To judge or evaluate the causal significance of the association : : : a number of
criteria must be utilized, no one of which is an all-sufficient basis for judgement.
These criteria include:
(a) The consistency of the association
(b) The strength of the association
(c) The specificity of the association
(d) The temporal relationship of the association
(e) The coherence of the association
These criteria were used in various sections of this Report. the most extensive and
illuminating account of their utilization is found in chapter 9 in the section entitled
Evaluation of the Association Between Smoking and Lung Cancer (p. 20).
The application of these criteria led to these Conclusions in respect to lung cancer
(p. 196):

1. Cigarette smoking is causally related to lung cancer in men. : : : The data for women,
though less extensive, point to the same direction.
2. The risk of developing lung cancer increases with duration of smoking and the number
of cigarettes smoked per day, and it is diminished by discontinuing smoking.
3. The risk of developing cancer of the lung for the combined group of pipe smokers, cigar
smokers, and pipe and cigar smokers is greater than in non-smokers, but much less than
for cigarette smokers. The data are insufficient to warrant a conclusion for each group
individually.

Three remarks may be called for here. First, at the time of this Report, epidemi-
ology was commonly defined as research by means of the epidemiologic method
(cf. above), and its principal types were termed, as in this Report, prospective
study and retrospective study (which later got to be termed cohort study and
case-control study, respectively).
Second, the criteria for causal inference deployed by the Committee were the
considerations famously elaborated by the statistician A.B. Hill the year after
the Reports publication. (He was not a member of the 10-member Committee;
its statistician was W.G. Cochran.) But those criteria/considerations are largely
untenable. To wit, the strength of a non-causal association can be very strong, as
in the case between match-consumption and lung cancer, for example; and this
association exhibits a strong dose-response pattern a topic under Coherence of
association (p. 123).
138 13 From Studies to Knowledge

Third, curiously missing from the criteria is the biological plausibility of the
associations causality, which Hill included, and the plausibility of confounding
as an explanation of the association, which Hill did not include. The section
immediately in front of Conclusions is entitled Other Etiologic Factors and
Confounding Variables (p. 183 ff), but it simply expands on the point that
a causal hypothesis for the cigarette-smoking-lung cancer relationship does not
exclude other factors (p. 193). The ones addressed are Occupational Hazards,
Urbanization, Industrialization, and Air Pollution, Previous Respiratory Infec-
tion, and Other Factors. There is no word about any of these or any other
confounders as possible explanations of the association between smoking and lung
cancer.

As for the IARC Monographs, their Preamble http://monographs.iarc.fr/ENG/

Preamble/CurrentPreamble.pdf is instructive reading. The Preamble is primarily
a statement of scientific principles, rather than specification of working procedures
(p. 1). It deserves to be quite extensively quoted here.
In these monographs, the potentially carcinogenic agents are viewed in very
inclusive terms: specific chemicals, groups of related chemicals, complex mixtures,
occupational and environmental exposures, cultural and behavioural practices,
biological organisms and physical agents. This list of categories may expand as cau-
sation of, and susceptibility to, malignant disease become more fully understood
(p. 2).
Although the Monographs have emphasized hazard identification [a hazard
being an agent that is capable of causing cancer under some circumstances],
important issues may also involve dose-response assessment. In many cases, the
same epidemiological and experimental data used to evaluate a cancer hazard can
also be used to estimate a dose-response relationship (pp. 23).
The Monographs are used by national and international authorities to make risk
assessments, formulate decisions concerning preventive measures, provide effective
cancer control programs and decide among alternative options for public health
decisions. : : : These evaluations represent only one part of the body of information
on which public health decisions may be based. : : : Therefore, no recommendation
is given with regard to regulation or legislation, which are the responsibility of
individual governments or other international organizations (p. 3).
Working Group Members generally have published significant research related
to the carcinogenicity of the agents being reviewed, and IARC uses literature
searches to identify most experts. Working Group Members are selected on the basis
of (a) knowledge and experience and (b) absence of real or apparent conflicts of
interests. Consideration may also be given to demographic diversity and balance of
scientific findings and views (p. 4).
Approximately one year in advance of the meeting of a Working Group, the
agents to be reviewed are announced on the Monographs programme website (http://
monographs.iarc.fr) and participants are selected by IARC staff in consultation
with other experts. Subsequently, relevant biological and epidemiological data are
13.5 Knowledge from the Evidence 139

collected by IARC from recognized sources of information on carcinogenesis, : : :

Meeting participants who are asked to prepare preliminary working papers for
specific sections are expected to supplement the IARC literature searches with their
own searches. : : : The working papers are compiled by IARC staff and sent, prior
to the meeting, to Working Group Members and Invited Specialists for review. The
Working Group meets at IARC for seven to eight days to discuss and finalize the
texts and to formulate the evaluations. : : : Care is taken to ensure that each study
summary is written or reviewed by someone not associated with the study being
considered. : : : IARC Working Groups strive to achieve a consensus evaluation
(pp. 56).
Several types of epidemiological study contribute to the assessment of carcino-
genicity in humans cohort studies, case-control studies, correlation (ecological)
studies, and intervention studies. Rarely, results from randomized trials may be
available. Case reports and case series of cancer in humans may also be re-
viewed. : : : In correlation studies, individual exposure is not documented, which
renders this type of study more prone to confounding. In some circumstances,
however, correlation studies may be more informative than analytical study designs
(pp. 89).
It is necessary to take into account the possible roles of bias, confounding and
chance in the interpretation of epidemiological studies. : : : Potential confounding
: : : should have been dealt with either in the design of the study, such as by
matching, or via the analysis, by statistical adjustment. In cohort studies [our semi-
cohort studies; sect. 6.4], comparison with local rates of disease may or may not
be more appropriate than those with national rates. : : : Detailed analyses of both
relative and absolute risks in relation to temporal variables, such as age at first
exposure, time since first exposure, duration of exposure, cumulative exposure, peak
exposure (when appropriate) and time since cessation of exposure, are reviewed
and summarized when available. Analyses of temporal relationships may be useful
in making causal inferences. : : : After the quality of individual epidemiological
studies of cancer has been summarized and assessed, a judgement is made of the
strength of evidence that the agent in question is carcinogenic to humans. In making
its judgements, the Working Group considers several criteria for causality (Hill
1965). A strong association (e.g., a large relative risk) is more likely to indicate
causality than a weak association, : : : If the risk increases with exposure, this is
considered a strong indication of causality, : : : (pp. 911).
Upon the review of evidence also from experimental animals as well as mech-
anistic and other relevant data (p. 15), the body of evidence is considered as a
whole, in order to reach an overall evaluation of the carcinogenicity of the agent in
humans. : : : The categorization of the agent is a matter of scientific judgement
that reflects the strength of the evidence derived from studies in humans and
in experimental animals and from mechanistic and other relevant data (p. 22).
The categories are: Group 1: The agent is carcinogenic to humans : : : Group
4: The agent is probably not carcinogenic to humans (pp. 22-3; italics in the
original).
140 13 From Studies to Knowledge

Looking back at these excerpts from the Preamble to the IARC Monographs, the
points of instructive note include these:
1. Whereas we think of carcinogenic factors in the categories of constitutional,
environmental, and behavioral, IARCs classification of carcinogenic agents
is quite different from this. Those specific chemicals, groups of related
chemicals, complex mixtures are, to us, potential carcinogenic factors only
when they are constituents (acquired) of a persons constitution, or subject to
becoming constitutional on account of a persons behavior or environment;
occupational factors are, to us, either behavioral or (micro-)environmental;
exposures are, to us, all environmental (even if as a result of behavior);
behavioural practices are not, to us, alternatives to cultural ones but,
potentially, cultural in their origin; all organisms are, to us, biomedical; all
carcinogenetic agents are, to us, chemical, physical, or biological, involved
in the constitutional, behavioral, or environmental carcinogenic factors (such as
inflammation, smoking, and air pollution); and causation is, to us, that which
a cause may bring about, depending on susceptibility to the cause, without the
latter being a carcinogen (while sunlight is a carcinogen, albinism is a related
susceptibility factor but not a carcinogen).
2. IARC distinguishes between epidemiological and experimental data, imply-
ing that epidemiological data inherently are non-experimental. To us, research
data are epidemiological if the intent in their production was to advance the
practice of community medicine, with the research quite possibly experimental,
including in the laboratory (sect. 1.2).
3. IARC is concerned not only with identification of carcinogenic hazards but also
with quantification of carcinogenic effects dose-response assessment. The
Monographs have emphasized the former, understandably, as epidemiologi-
cal research has tended not to get past hypothesis testing (ch. 2).
4. IARC sees itself in the service of public-health authorities in their decisions
about preventive-oncologic measures, but: no recommendation is given. This
is in sharp contrast with what is done by the Canadian Task Force on Preventive
Health Care and by the Preventive Medicine Task Force in the U.S.; and we
think that the IARC is right on this important matter. See Appendices 1 and 2.
5. IARC has a very elaborate scheme of assuring the greatest possible validity
of its Conclusions, which again is instructive of how major is the transition
from evidence to knowledge, seen to result only from experts consensus
evaluation of the evidence. As the Working Groups have to strive to achieve
consensus, the question is, How concordant would tend to be the conclusions
of two or more similarly constituted and similarly working expert groups,
addressing the same question.
6. IARC reflects our contemporary normal-science conception of types of epi-
demiological study on humans. Exceptionally, though, it prefers to use
correlation study as the term for what now is commonly known as ecological
study. The preference is curious, as the adjective is no more apt a descriptor
of studies with populations as the units of observation than studies of the
13.5 Knowledge from the Evidence 141

other type notably cohort studies and casecontrol studies which

IARC characterizes as representing analytical study designs, in unjustifiable
deference to epidemiological tradition. (Those studies are not about analysis,
and the logic in them is synthetic rather than analytic.)
7. IARC seems to use interpretation as a synonym for evaluation in judge-
ments about epidemiological evidence, while for us interpretation is a challenge
in understanding the meaning of the report of a study (as to what the
result actually was, and how it actually came about, according to the report;
sect. 13.3). But we are fully comfortable with the idea that a study needs to be
judged in respect to the possible role of bias [descriptive], confounding and
chance (sect. 8.2).
8. Different from IARC, weve left behind the common idea that Potential
confounding : : : should have been dealt with either in the design of the study,
such as by matching, or in the analysis, by statistical adjustment. We now
think of study design as specifying everything that will intentionally go into
the genesis of the study result, including the statistical model and its fitting to
the data; and confounding is dealt with in the design stage of that model in
non-experimental etiogenetic research and, thereby, in the synthesis (sic) of the
study data (into the study result and measures of its precision), without the latter
being an alternative to the former (chs. 7, 8). Matching generally is a matter of
the sampling for the base series being discriminate/stratified (according to the
distribution of the case series). This stratification needs to be accounted for in
the model; but it is this modeling, and not the matching, that constitutes control
of confounding. (That modeling achieves the control even in the absence of the
matching.)
9. IARC evidently takes, quite exceptionally, the concept of cohort study to
encompass what we here term semi-cohort study (sect. 6.4); and despite the
primitive nature of this kind of study, IARC, surprisingly to us, seems to
take it as potentially providing quite meaningful evidence about oncogenesis.
We think that the healthy worker effect is symptomatic of unhealthy study.
(There is no sick patient effect in clinical research.)
10. IARC conducts detailed analyses in respect to temporal variables in terms
that are different from what we consider appropriate. Evidently in reference to
histories in etiogenetic studies, these IARC concerns include age at first expo-
sure, time since first exposure, duration of exposure, cumulative exposure, etc.,
with the idea that analysis of temporal relationships may be useful in making
causal inferences. Instead of any of these, our concern is the studys objects
design in respect to the index and reference histories on the scale of etiogenetic
time, across the entire relevant range of this, with the index-reference contrasts
based on these.
11. Based on the human evidence, a judgement is made of the strength of [this]
evidence that the agent in question is carcinogenic to humans as though the
concern always were with qualitative inference (cf. # 3 above). And in this, the
Working Groups are said to deploy the several criteria of causality enunciated
142 13 From Studies to Knowledge

(as considerations) but A.B. Hill despite the obvious untenability of many
of these (cf. the Smoking and Health example above).
12. The Conclusions, representing the knowledge formulated by any given Working
Group, like those assessments/evaluations of the evidence from the studies, also
are only qualitative in nature (cf. # 11 above, in contrast to # 3). Yet, only
quantitative knowledge is the appropriate basis for authorities to set threshold
limit values, for example (cf. # 3 and # 4 above).

13.6 The Needed Innovations

The prime example of the fruits of twentieth-century epidemiological research is,

arguably at least, the attained knowledge about the health effects adverse, on
risk of lung cancer most notably of smoking, cigarette smoking in particular. The
research effort was huge (sect. 2.7) and its translation into official knowledge very
elaborate (sect. 13.5 above).
Even more elaborate have been and are the periodic processes of translating
into knowledge the available evidence concerning possible carcinogenesis, the way
they are performed by the International Agency for Research on Cancer for the
production of the IARC Monographs (sect. 13.5 above). Might it be that awe of
these efforts at the definition of epidemiological knowledge is at the root of the
Epidemiologic Reviews typically addressing epidemiological studies rather than the
knowledge derived from these? (sect. 2.6.)
Very respectable about the IARC (of the WHO) is its heeding of the principle that
relevant scientific knowledge is to be seen to be only one category of the relevant
inputs into rational decisions in public-health practice. IARC understands that
evidence, or even knowledge from it, does not translate into justifiable recommen-
dations about decisions in practice, in disregard of all other relevant considerations,
generally peculiar to the particular contexts of the decisions (sect. 13.5). This point
is compellingly argued by R.A. Pielke, Jr. in his The Honest Broker: Making Sense
of Science in Policy and Politics (2007).
This feature of the IARC reviews, which merely is what one should expect,
gains its poignancy from its unwitting foils, from the various official producers
of evidence-based guidelines/recommendations for public-health policy about
preventive medicine. Particularly notable as such foils of the IARC are the Canadian
Task Force on Preventive Health Care (App. 1) and the Preventive Medicine Task
Force in the U.S., both programmatically producing recommendations/guidelines
for decisions in the practice of what they take to be preventive medicine. And
notable about these particular programs of recommendations-production is not
merely that this is what they do but also the level of seriousness of the work as
for what types of panel of people does it and how (Apps. 1, 2) how casual all of
this is, in contrast to the work of the IARC.
Those task forces produce policy recommendations most eminently in respect
to various types of screening for a cancer, which has become a common concern
13.6 The Needed Innovations 143

of epidemiological researchers and hence a concern also in the Epidemiologic

Reviews when a particular cancer is addressed (2001 issue) and even as such (2011
issue; sect. 2.6). This interest of epidemiologists in screening for a cancer as a
purported topic in community-level preventive medicine is an outgrowth of past
programs of mobile units of fluoroscopic screening for pulmonary tuberculosis,
such case-finding as an element in the prevention of the spread of the infection
in the community. In Japan there now are public-health programs of screening for
lung cancer as direct, seamless outgrowths of previous programs of screening for
tuberculosis. But cancer, different from tuberculosis, is not a communicable disease,
so that its detection and treatment in no way serves prevention of the spread of the
disease in the community.
So determined are those task forces in their production of recommendations
on for or against screenings for cancers (and other diseases too), based on
the mistaken idea that screening for an illness is a preventive intervention; so
misguided consequently is the common idea among epidemiologists as for the
nature of the research and knowledge needed for public policies for or against
the screenings (public funding of the initial tests in these); and so different
are the genuine clinical counterparts of these, that we deal with the topic
quite extensively (sects. 2.6, 7.2; ch. 11; App. 2), even though at issue is not
epidemiological but meta-epidemiological clinical research.
As translation of the evidence from epidemiological research into knowledge for
public-health practice is commonly absent (sect. 2.6), occasionally very elaborately
done (sect. 13.5 above), and in yet other instances done very flippantly but used as
the practically sole basis of guidelines/recommendations regarding practice (Apps.
1, 2), something is fundamentally amiss in the evidence-to-knowledge phase of the
genesis of epidemiological knowledge, serving to sustain various anomalies also in
the production of the evidence the single most notable example of the latter being
the research on screening for a cancer.
The fundamental anomaly is, we suggest, near-complete absence of that which,
in science, normally would produce those evidence-to-knowledge transmutations,
namely: public discourse among the members of the relevant scientific commu-
nities. This anomaly is rooted in the prevailing general culture of our medical
journals, which do not serve as open forums for this critically important element
in the normal process by which empirical science progresses. Today these journals
are mainly directed to practicing physicians, to whom they supply fresh evidence
and conclusions based solely on this! instead of what doctors actually need from
science, namely updated knowledge.
A distinction should be made between journals of medicine for practitioners
of it and those of medical science for medical scientists; and in the latter there
is a need to distinguish between journals of evidence and those of inference (from
evidence; sect. 13.4). Genuine journals of medical science would not merely freely
allow, but would actually encourage and invite, public discourse among scientists.
This would be so in respect to inference in particular, as it is public discourse about
144 13 From Studies to Knowledge

this, by the very nature of science, that has the capacity to converge into experts
shared belief, into genuine knowledge on a topic in medical (as in any other) science;
and difficulties in this would also serve to underscore the deficiencies in the available
evidence, inviting remedies for them.
As for quintessentially applied epidemiological and meta-epidemiological
clinical science specifically, a third kind of needed medical journalism concerns
its theory concepts and terminology for a start (Preface) and then principles
(sect. 1.3). This, too, is less than satisfactory at present, in that consensus-seeking
public discourse is missing in this area as well. And so it still is that textbooks of epi-
demiological research generally are mainly about study of the etiology/etiogenesis
of disease rather than illness (disease/defect/injury or morbus/vitium/trauma);
addressed is methodology of the research in total disregard of the studies objects
design; and in the studies themselves a major distinction still is routinely made
between cohort and case-control studies instead of proceeding from each of
these into the corresponding corrected conceptions, into that which is essential in
all etiologic/etiogenetic studies in this genre of them (chs. 5, 6). As another major
example, the textbooks still are prone not to make the profound distinction between
inquiries that are epidemiological research and ones that are part of epidemiological
practice (sect. 14.1).
More on the needed innovations in the translation of evidence into knowledge for
the practice of community medicine is given in Appendix 3.

As for the evidence itself in specifically epidemiologic journals, its publication bias
remains a serious problem; but this too is remediable: Reporting on population-
level epidemiological studies should be accepted for publication before any results
are at hand that is, upon the authors having completed the Introduction, Objects
Design, and Methods Design sections of the report and the sections having passed
peer review as well as review by the journals editorial board. No report should be
accepted belatedly, after the study plans implementation has already begun. Each
timely acceptance of the study plan for ultimate publication of the report should be
a matter of accessible record (incl. for the purposes of grant applications and their
peer reviews, in addition to the needs of derivative studies).
That early acceptance of the ultimate report on a study would not mean that
whatever gets to be written in it under Results and Discussion and in the
Summary/Abstract has advance approval for publication. The journals editorial
board has the responsibility (to the scientific community) to make sure that the
writing accords with the principles of this. Prime among these is the imperative
to clearly and correctly distinguish between particularistic facts about study results
and inferential ideas based on these. As for this, pervasive still is writing about study
results as though causality (or lack of it) had been observed and documented in a
study (cf. sect. 4.2). Quotes in this text illustrate this misleading writing, starting in
section 2.2 (showed no reduction in, did not affect).
With these innovations, epidemiologic journals would much better focus on
genuine contributions to evidence, minimizing obfuscating litter-ature. I coined
the term litter-ature to denote that too much of the medical literature is littered
13.6 The Needed Innovations 145

with misleading and false-positive findings, writes Eric Topol in his The Creative
Destruction of Medicine: How the Digital Revolution Will Create Better Health
Care (2012; pp. 31-2).
We add that even more influential as a source of publication bias generally is
the common opposite: cleaning out of the literature reports with negative results,
commonly meaning only relatively imprecise results, however valid.
Chapter 14
Fact-finding in Epidemiological Practice

Abstract Practice of community medicine involves, quite centrally, acquisition of

statistics (frequency numbers) on the cared-for population and making, on this basis,
inferences about frequencies of health hazards and illnesses (levels of morbidity) in
the population. The epitome of this is the investigation of an epidemic with a
view to inferences about it, descriptive for one and etiogenetic for another this
as the basis of controlling the epidemics extent and prevention of its recurrence.
Commonly involved are sample surveys and their sample-to-population inferences
something termed survey research by statisticians. There thus is a strong tendency
to suffer the delusion that practice of community medicine is science.
In this chapter we first outline the profound difference between fact-finding in
epidemiological practice and that in (population-level) epidemiological research,
supplementing this with description of fact-finding in the epidemiological practice
of the public-health agency of Canadas capital city.
We then describe fact-finding in a very different line of public-health practice
in Ottawa. Now that national health insurance has been introduced into Canada,
clinical medicine in this country is public healthcare, with its administration a
provincial responsibility. The main concerns of the provincial ministers of health
now are quality assurance and cost containment of clinical medicine, hospital-
based care in particular. We describe what fact-finding in this vein the Ontario
government now obligates to the hospitals in the province, and how Ottawa hospitals
are responding to these.
Finally, given the centrally important place of quality assurance economic as
well as medical quality of clinical care, and of hospital-based care in particular, at
present and ever-increasingly in the future, we devote considerable attention to how
this quality assurance could be substantially improved by future practitioners of
epidemiology.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 147

DOI 10.1007/978-94-007-4537-7 14, Springer ScienceCBusiness Media Dordrecht 2012
148 14 Fact-finding in Epidemiological Practice

14.1 Fact-finding vis-`a-vis Research

Fact-finding is not what a piece of epidemiological research is about. For, the

researchers are not concerned with the assembled facts per se, not individually nor
in the aggregate, not out of interest nor as a basis for action on the individuals or
the population contributing to the study base. Even the study result this aggregate
fact on the study base is but a window barely transparent into what a piece
of epidemiological research really is about: an abstract truth rather than anything
particularistically factual. If the truth at issue about the abstract be known, facts
about whatever population-time or whatever series of person-moments would be
irrelevant from the vantage of epidemiological research.
In epidemiological practice, by contrast, an epidemiological study (sic) is a
project of fact-finding. In it, the concern is to establish a fact, or a set of these, about
the cared-for population, relevant to the care for this population. Paradigmatic
about this is study in the meaning of investigation of an epidemic as to its pattern
and degree of spread and, especially, as to its etiogenesis. The findings from
this investigation are not results for publication in a scientific journal. They are,
instead, inputs into ad-hoc decisions about control of the epidemic at issue and
about future prevention of other similar epidemics in the cared-for population
(sect. 14.4).
This chapter is devoted to delineation of fact-finding in epidemiological practice,
with two aims: to illustrate the distinction between this and epidemiological
research, and to help make more concrete the concept of epidemiology as practice
of community medicine (sect. 1.1) the advancement of which is the raison detre
of epidemiology in the research meaning of the term (sect. 1.2).
The last section this chapter we regard as being particularly important in respect
to modern public health, in which quality assurance and cost containment of clinical
care are the central concerns, and in which epidemiologists can make a critically
important contribution to society.

14.2 Fact-finding by Use of Routine Sources

In fact-finding about the populations demographic composition, and state of health,

a community doctor can make use of data readily available from routine sources.
Thus, Ottawa Public Health can ascertain the city populations demographics from
the Provincial Health Planning Database (maintained by the Ontario Ministry of
Health and Long-term Care) and from Ottawa Facts and Research and Ottawa
Facts, Data Handbook (produced by the City of Ottawa); its overall rates of
mortality and hospitalization, as well as rates of mortality specific to leading causes
of death and hospitalization rates specific for age and gender, from the Provincial
Health Planning Database; its rates of morbidity and mortality from any given
cancer, specific for age and gender, from the Ontario Cancer Registry (maintained by
14.3 Fact-finding by Ad-hoc Surveys and Monitorings 149

Cancer Care Ontario) and the Provincial Health Planning Database. The Provincial
Health Planning Database contains information also on rates of injuries, poisonings,
suicide, etc.
Besides, Ottawa Public Health can obtain the information on the rates of
communicable-diseases morbidity from the Integrated Public Health System of the
Ontario Ministry of Health and Long-term Care.

14.3 Fact-finding by Ad-hoc Surveys and Monitorings

Fact-finding by the use of routine sources a community doctor commonly supple-

ments with data from ad-hoc surveys and monitorings, conducted by the doctors
agency (possibly commissioning for this purpose another agency or organization) or
by another agency or organization independently of the community doctors agency.
The City of Ottawa 2006 Health Status Report gives several examples of these.
As one example, it surveyed 800 mothers of 3 [sic] and 6-month-old infants
to determine health status, factors influencing health and the utilization of health
care services by parents of infants. The survey also identified trends of methods
parents used to care for their infants (p. 133). Another example is the Rapid
Risk Factor Surveillance System, conducted by the Institute for Social Research,
York University, on behalf of Ottawa Public Health. It is an ongoing random-
digit-dialed telephone survey of adults aged 18 years and over, : : : The first
wave of data collection for Ottawa began in April 2001. Households are randomly
selected from all households in the City of Ottawa and a sample of 100 residents
are surveyed every month regarding health risk behaviours, knowledge, attitudes,
and awareness about health related topics of importance to public health such as
smoking, immunization, sun safety etc (p. 134).
An example of extrinsic survey is the Canadian Community Health Survey
(CCHS), conducted by Statistics Canada, data from which are available from the
Ontario Ministry of Health and Long-term Care. This is a national population
household survey for all provinces and territories in Canada : : : . The survey
collects information on the health of the Canadian population aged 12 and older
as well as socio-economic data. The survey runs in a two-year collection cycle
and is comprised of two distinct parts: a health region-level survey in the first year
with a total sample of 130,000 and a provincial-level survey in the second year
with a total sample of 30,000. Data collection commenced in 2000. A broad range
of topics are examined in the survey on health status, determinants of health and
health system utilization. Data available for Ottawa includes 2000-01 and 2003, and
the sample size is approximately 1900. The CCHS is the data source for many of
the Health Indicators generated by Statistics Canada and the Canadian Institute for
Health Information (p. 133).
A program of successive, periodic surveys on the same topics on the same,
sampled population we think of as being a program of monitoring (over time) rather
than a survey (at a particular time).
150 14 Fact-finding in Epidemiological Practice

14.4 Investigations of Epidemics

Even if endemic morbidity from non-communicable, chronic diseases has become

a major concern in the practice of epidemiology, community diagnoses about
epidemics of communicable disease and control of these remains an important
component in this practice. As described in the City of Ottawa 2006 Health Status
Report, Ottawa Public Health investigates all potential respiratory outbreaks in
long-term care residences, hospitals, day care centres, schools and other institutions.
Investigations also occur for potential outbreaks reported from the community or
those detected by routine communicable disease surveillance. Outbreak investiga-
tions can include case-finding, contact tracing, infection control, risk assessment,
and immunization. : : : In 2004, there were 196 confirmed outbreaks in the City
of Ottawa. : : : There were 62 outbreaks related to respiratory infections and
134 outbreaks related to enteric infections (p. 111).

14.5 Quality Assessment of Hospital Care

Epidemiological practice of healthcare in the meaning of community-level preven-

tive medicine (sect. 1.1) informed by fact-findings that are an integral part of it
(sects. 14.2, 14.3, 14.4 above) as supplements to inputs from general epidemio-
logical knowledge (ch. 2; sect. 13.5) used to be the entirety of epidemiological
practice. This was the case when the domain of public health was coterminous with
that of community health in reference to professional healthcare (paramedical as
well as medical), when public health thus was extrinsic to the domain of clinical
healthcare. In this framework, epidemiological practice was expressly societal
healthcare, not only societally financed but also governed by public policies, while
clinical practice was essentially private and autonomous, essentially outside societal
concerns.
All of this was upended by the advent of national health insurance, initially in the
U.K. in 1848, followed by other countries, Canada included. In the U.S., outside the
Veterans Administration, it still is limited to Medicare for the elderly and Medicaid
for the indigent. National health insurance brought clinical medicineinto the domain
of public health. As a consequence, clinical medicine became the greater segment
of public-health medicine, by far, relative to community medicine.
While the aggregate cost of healthcare in a modern society already consumes
a huge proportion of the Gross Domestic Product, this cost is growing at a rate
that unquestionably is unsustainable. With a view to clinical medicine in particular,
the central mantra of modern public health and todays Minister of Health
thus is cost containment; and as the pursuit of this may compromise the quality
of healthcare, quality assurance has become an associated, added mantra.
Promulgators of public-health policy have adduced various schemes of pursuing
the hoped-for cost-containment in healthcare, including, in Canada, restrictions in
14.5 Quality Assessment of Hospital Care 151

numbers of doctors in particular disciplines of medicine. But an unintended conse-

quence has commonly been a reduction in individuals access to justifiable care.
Practicing epidemiologists, as public-health officials, are accustomed to moni-
toring health-related goings-on (sect. 14.3) and taking or recommending remedial
action when called for in the interest of public health. They could, very usefully,
bring this culture to clinical medicine, perhaps to hospitals first and foremost, where
half of the healthcare expenditure is incurred in countries like Canada.
They could and really should introduce continual, sampling-based monitoring
of the care processes (rather than outcomes) into hospitals within a given jurisdic-
tion. This would be a matter of retrospectively documenting (by records abstraction)
the care for individual patients throughout their sojourns in a given type of hospital.
The case-specific findings would be submitted for review by a suitable panel of
experts, for them to identify lapses of quality, economic and/or medical, in the
care. An elementary and cardinal lack of requisite quality is, of course, lack of the
requisite recordings for use in this monitoring (i.a.).
Reports of the untoward findings of the rates of these would be submitted to
the appropriate administrators, for possible remedial action (or laudation).
While science is supposed to be self-corrective, medicine is supposed to be self-
policing. The essence of self-policing in medicine is active and competent pursuit
of identification of physicians substandard behaviors and, upon their identification,
bringing these to the attention of the relevant officials for remedial or punitive
action. As we outlined above, practicing epidemiologists, when in the service of
modern public health, are uniquely qualified to do this in respect to clinical care;
and so: noblesse oblige.

Given that good healthcare is not only medically but also economically good not
only effective but cost-effective as well and given also the ever-deepening cost
crisis in modern, societally sponsored healthcare in modern public health, that
is the topic of quality assurance in clinical healthcare deserves more deliberate
consideration here than the little sketch above, ultimately with a view to the
role(s) we think practicing epidemiologists need to adopt as genuine clinical
epidemiologists, without the term being self-contradictory.
With screening for a cancer now a misguided exception (Apps. 1, 2), clinical
medicine has retained a good deal of autonomy within the society even after
the advent of national health insurance. Societies regulate industries as for the
marketing of medications and medical devices, but they still refrain from regulation
of the use of these products or other professional actions by clinicians.
Regarding the autonomous quality-assurance in clinical medicine, M.S. Liang
and P. Fortin wrote in 1991 (Ann Rheum Dis; 50: 522-5) that The American
experience with quality-assurance programmes prompted a Committee of the
Institute of Medicine of the National Academy of Sciences to conclude that for
Medicare recipients the current system is in general not very effective and may have
various unintended consequences [ref.]. The Canadian system has similar roots but
to date has avoided some of the excesses of the American system. This paper traces
the growth and contrasts the system of quality assurance in America and Canada.
152 14 Fact-finding in Epidemiological Practice

In the summary of their essay, those authors wrote that Americas system
[of quality-assurance programs] is pluralistic, administratively very complex (and
expensive!). In contrast, Canadas system is smaller in size, less expensive, and run
by doctors. Both are increasingly preoccupied with bottom line and with linking
quality to cost containment efforts. They pointed out that no one argues [i.e.,
questions] that health care can be improved and that a portion of the expenditure
devoted to health care should be used to evaluate how we are doing, adding that
the costs of the system and the responsibilities of the system must be examined
carefully : : :
In 1992, A survey of medical quality assurance programs in Ontario hospitals
was published by B. Barrable (CMAJ; 146:153-60). It was predicated on this: It
is generally believed that in recent years fewer hospital medical staffs have been
undertaking quality assurance activities. : : : Several reasons have been suggested
for this, including but not limited to the following five:
Confidentiality and fear of liability : : :
A lack of compensation for physicians who involve themselves in quality
assurance activities in addition to their regular clinical workload. : : :
A loss of uniqueness and ownership. : : : [The quality assurance standards of the
Canadian Council of Health Facilities Association, adduced in 1983] were very
prescriptive about how medical staffs should conduct quality assurance. Before
[1983] these activities were referred to as medical audits and were the exclusive
domain of physicians [ref.]
A dearth of expertise in conducting quality assurance activities. : : :
The absence of comprehensive and clear legislation to encourage and, indeed,
mandate practical approaches to medical quality assurance. : : :
The survey questionnaire was sent to The person deemed by the chief executive
officer : : : to be most responsible for medical administration, in All teaching,
community, chronic care, rehabilitation and psychiatric hospitals that were members
of the Ontario Hospital Association as of May 1992. Of the 245 member hospitals,
participants from 179 (73%) responded, indicating a wide variety of quality
assurance activities. Most common was that In-hospital deaths were reviewed in
157 (88%) of the hospitals.
The legislature of Ontario introduced, in 2010, An Act Respecting the Care
Provided by Health Care Organizations (Bill 46), amending or repealing various
previous laws. This law stipulates that Every health care organization shall
establish and maintain a quality committee for the health care organization. This
committee is to monitor the quality of the services and report on this to the respon-
sible body, with reference to appropriate data. It also is to make recommendations
and to ensure that suitable materials on best practices are supplied to the care
providers. Every healthcare organization also is to carry out surveys of consumer
and provider satisfaction (annually and biennially, respectively).
This law, which calls itself the Excellent Care for All Act, 2010, mandates
the quality committee To oversee the preparation of annual quality improvement
plans. These must contain, at a minimum, (a) annual performance improvement
14.5 Quality Assessment of Hospital Care 153

targets; (b) information concerning the manner in and the extent to which health
care organization executive compensation is linked to achievement of those targets;
and (c) anything else provided for in the regulations.
In the city of Ottawa (in Ontario) in 2011, one such plan was published
(as required) by the Montfort Hospital (http://www.hopitalmontfort.com/
QualityImprovementPlan.cfm). The plan is:
to ensure that at least 65% of employees use proper hand hygiene before initial
patient contact by March 31, 2012
By March 31, 2012, achieve a substantial reduction in post-operative infection
rate for abdominal hysterectomies : : :
In collaboration with our partners, reduce the number of days that patients must
wait for an alternate level of care (ALC) from 18% to 16.7%, : : :
Reduce the wait times in the Emergency Department by 10% for patients with
complex conditions and admitted patients : : :
Increase the percentage of patients who would recommend the hospital to their
friends and family from 73.9% to 85% : : :
[Etc.]
To each stated objective is attached the statement of how it will be achieved.
For the Ottawa Hospital, the corresponding QIP (http://www.hopitalottawa.
on.ca/wps/wcm/connect/3870600046545f60a990fd2940f23d1f/QIP short-e.pdf?
MOD=AJPERES) is orientationally described by this Overview:
The Ottawa Hospital (TOH) aims to become a top North American hospital in terms
of quality and safety of patient care. To this end, our Quality Improvement Plan (QIP)
enhances hand hygiene and decreases the rates of hospital acquired infections. It improves
effectiveness by avoiding preventable deaths and unnecessary readmissions, and continues
our record of a balanced budget. It ensures patients are placed in the appropriate care setting
when they no longer need acute care. Those who need care will see reduced wait times
in Emergency. These actions will increase patient satisfaction and the quality and safety
of care.

And its Ideas for Improvement are these:

Monitor adherence to established best practice guidelines for central line infections,
ventilator associated pneumonias, and medication reconciliation at discharge.
Reduce Emergency Room overload through initiatives geared to reducing inpatient
occupancy rates, including improving patient flow and reducing the number and length
of stay of alternate level of care patients.
Using champions and physician leads to promote hand hygiene as well as encouraging
patients to ask staff to wash their hands.
Use innovative ways to assess and improve patient satisfaction.

The foregoing, with its focus on Canadas Ontario province and the nations
capital city Ottawa in it, is instructive of the principal segment of public-health
practice in the modern framework in which clinical medicine is public healthcare
consequent to the advent of national health insurance.
154 14 Fact-finding in Epidemiological Practice

The very first lesson from those facts is this: The Ontarian society, as the third-
party payer of clinical healthcare, wishes to be assured that the societally sponsored
care is of good quality, not only medically but economically as well. An expression
of this is the very name of Ontarios Bill 46: Excellent [sic] Care for All [sic] Act
(cf. above). In the Preamble of that Act, the first one of the nine brief statements
about The people of Ontario and their Government is that they Believe in the
importance of our system of publicly funded health care services and the need to
ensure its future so that all Ontarians, today and tomorrow, can continue to receive
high quality health care.
Next to this belongs the lesson that the Ontarian people and their provincial
government actually do not believe that all Ontarians now have, or in the future will
have, high-quality healthcare: for, the government of the province is now mandating
Quality Improvement [sic] Plans to be periodically developed and published by all
of the provinces hospitals (cf. above).
Third and most important, the provinces government evidently presumes to be
competent to tell the hospitals personnel how the quality of care in them can and
must be improved: it specifies the generic nature, though not the genesis, of those
Quality Improvement Plans it requires.
It deserves note here that just before the Excellent Care for All Act was
introduced, the provinces Ministry of Health and Long-term Care called attention
to opportunities for improvement (http://www.health.gov.on.ca/en/ms/ecfa/pro/
legislation/ecfa presentation 20100503.pdf). In Ontario (population about 13 mil-
lion), it said,
Forty thousand patients were admitted to hospital last year for ambulatory care sensitive
conditions that could have been better managed in the community
Last year, there were 140,000 cases of patients readmitted to hospitals within 30 days of
original discharge
Over 5,000 x-rays and 49,000 electrocardiograms were performed last year for patients
about to undergo cataract surgery, when evidence shows these tests to have no clinical
benefit
Many Ontarians with diabetes and other chronic diseases are still not receiving all care
recommended by clinical guidelines

Related to this is an instructional point of particular note: None of these

substandard practices could have been identified let alone remedied by earlier
implementation of such Quality Improvement Plans as have now been adduced
(examples above) pursuant to the Excellent Care for All Act. The implication
is that, while the hospitals are failing in their endogenous quality assurance, the
government, in turn, is failing in its endeavor to provide support to help them plan
for and improve the quality of the care they deliver based on the best available
scientific evidence (quote, again, from the Acts Preamble).

The industry of hospital-based healthcare in Ottawa (and elsewhere) is quite

anomalous among service (and manufacturing) industries at large: the hospitals lack
14.5 Quality Assessment of Hospital Care 155

built-in, serious programs of industrial quality control, in which the foundation of

everything else is continual, comprehensive monitoring of the quality.
For any given hospital, we suggest, this would be a matter of continual, fair
sampling of patients stays in the hospital; abstraction of the records of the sample
stays to form corresponding narratives of these, suitable for quality assessment; and
periodic (monthly, say) evaluation of these representative narratives by the hospitals
quality review committee. The quality assessments would focus on the broad topics
of admission, diagnoses, treatments, adverse events, and discharge. They would
focus on identification of deviations from community standards (unspecified) and,
especially, from specified guidelines and best practices, apart from addressing
adverse events.
Periodically, typically annually, the evaluative findings would be summarized in
terms of rates for the various types of substandard practice and adverse health event.
These findings would be reported to the Chief Executive Officer (physician) of the
hospital, for consideration and possible remedial action.
Any given hospital would need, for its serious program of quality control
(normal in any commercial service-industry while missing in government-financed
hospitals), protocols for the translation of hospital records into the corresponding
narratives; for the documentation of the evaluative judgements; and for the reporting
of these in terms of rates all this with the understanding that these protocols need
to evolve over time, as experience with them accrues and practices change.
As the development of these protocols would be contemplated in a given hospital
in Ontario, it would be realized that the challenges, quite considerable, would
be the very same in other hospitals in the same category of hospitals in the
province, general hospitals, for example. Upon this realization, the interest in any
given hospital would be in province-level development of the needed protocols,
insofar as the hospital leadership and other professionals in the hospital indeed
are interested in serious quality assurance for the care provided by the hospital.
The interest would be in this protocol development by suitable representatives of
the relevant set of hospitals, not representatives of the government of the province.
(In Canada, healthcare is provincial, not federal, in its societal policy promulgation
and administration.)
Thinking about the province-level development of protocols for in-hospital
quality assessment, the planners in any given type of hospital would realize that
there is considerable overlap in the nature of the programs across the different types
of hospital in the province; that there really is a need for a single province-level
Agency for Quality Assessment, with separate departments within it for each of the
different types of hospital. Development of the quality assessment protocols would
be the initial and one continual mission of this AQA.
This AQA would work with the hospitals internal clinical epidemiology units.
The CEU of any given hospital would sample the hospital stays, produce the case
narratives from the records on each hospital stay in the samples, produce the periodic
(annual?) statistical QA reports based on the QA committees evaluations, and it
would submit these reports to the hospitals CEO (cf. above). Another recipient
156 14 Fact-finding in Epidemiological Practice

of these hospital-specific QA reports would be the provincial AQA, which would

collate them across all of the hospitals of a given type. The AQA would produce
for each hospital the counterpart of the hospitals own QA report, concerning the
aggregate of all of the other hospitals of the same type, providing this to the hospital.
In addition, the AQA would produce, from each cycle of QA, an overall report for
the Ministry of Health and Long-term Care.
The provincial AQA and the CEUs of the provinces hospitals would, of course,
be financed by the provincial Ministry of Health and Long-term Care, separately
from the rest of the disbursements to the hospitals.

No introduction into industrial quality assessment and the rest of a productions

quality assurance in this instance in the service industry of hospital-based
healthcare is complete without at least a passing reference to the statistician
who became the outstanding pioneer and authority on this: W. E. Deming. Quality
of production to him was the productions cost-effectiveness, just as it is to a
modern Minister of Health. We direct the reader to http://en.wikipedia.org/wiki/W.
Edwards Deming and through this to Demings writings on the subject.
Regarding the need for quality assessment and assurance specifically in modern
clinical healthcare, very instructive, in many ways, is John Wennbergs Tracking
Medicine: A Researchers Quest to Understand Health Care (2010). He notes that
even among the select few academic medical centers that reside at the very top in
terms of their national reputation for excellence, there is little evidence that clinical
practice is based on scientific consensus on the best way to practice medicine
(p. 170). In fact, he points to a variety of evidence to the contrary.
And finally, as for quality assurance in modern public-health practice, a supple-
ment to the major innovation we suggested in the foregoing (cf. Preface) is a critical
commentary from the wider industrial perspective, given in Appendix 5.
Appendix 1: Canadian Task Force on Preventive
Health Care

[Science] knows nothing of policy or utility,

of better or worse.
Torsten Vebler, 1906

The scientist [has] neither the moral competence nor the moral right
to use the lecture-room or the learned journal
to pronounce what ought to be done.
Max Weber, 1918

Science never tells a man how he should act;

it merely shows how a man must act
if he wants to attain definite ends.
Ludvig von Mises, 1963

The Issue Advocate seeks to compel a particular decision,

while an Honest Broker of Policy Alternatives seeks
to enable the freedom of choice by a decision-maker.
Roger A. Pielke, Jr., 2007

The first two of those here highly-relevant, orientational quotes above we drew from
an indirect source, The Scientific Life: A Moral History of a Late Modern Vocation
(2008) by Steven Shapin. He also cites three notable definitions of vocation right
up-front, none of them consistent with the self-proclaimed mission of the CFTPHC
(below).
As epidemiological research inherently is in the service of the practice of com-
munity medicine, and as this practice epidemiological is community-level
preventive medicine, it is educational for a student of this research to gain
familiarity with the Canadian Task Force on Preventive Health Care. Two sources
of information about it are (refs. 1, 2):

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 157

DOI 10.1007/978-94-007-4537-7, Springer ScienceCBusiness Media Dordrecht 2012
158 Appendix 1: Canadian Task Force on Preventive Health Care

References:
1. http://canadiantaskforce.ca
2. Canadian Task Force on Preventive Health Care. Procedure Manual.
http://www.ualberta.ca/mtonelli/manual.pdf.

The [CTFPHC], previously known as the Canadian Task Force on Periodic

Health Examination was established in : : : 1976 by the Conference of Deputy Min-
isters of Health of the ten Canadian provinces. : : : The particular characteristic that
distinguishes the Task Force methodology from traditional approaches in decision-
making on prevention issues is that evidence takes precedence over consensus. : : :
In the 1980s the Task Force methodology was adopted : : : by the United States
Preventive Services Task Force : : : [both of them] developing guidelines for clinical
practice and public health policy. : : : In 2005, the [CTFPHC] was disbanded.
In 2010 [it] has been established with the support of [Public Health Agency of
Canada] and a renewed commitment and vision to continue its 25-year tradition
of excellence. (Ref. 1; italics ours).
The CTFPHC is (ref. 2, p. 7) an independent panel composed primarily of
clinicians and methodologists that makes recommendations for clinical preventive
actions : : : but its work is also directly relevant to other health care professionals,
developers of preventive programs, policy makers and Canadian citizens. : : : The
services must be provided in primary care settings or available by primary care
referral.
The CTFPHC uses the same definition of primary care as the US Institute of
Medicine (ref. 2, p. 36):
Primary care is the provision of integrated, accessible health care services by clinicians
who are accountable for addressing a large majority of personal health care needs,
developing a sustained partnership with patients, and practicing in the context of family
and community. This definition acknowledges the importance of the patient clinician
relationship as facilitated and augmented by teams and integrated delivery systems.
The CTFPHC is an independent body of fifteen primary care and prevention
experts who recognize and support the need for evidence informed preventive
activities in primary care in Canada (ref. 1). This body of experts is now chaired by
a physician who is a nephrologist, clinician-scientist and Associate Professor [in a
Department of Medicine] (ref. 1).
Having just been re-established, The Task Force met in early 2010 to establish
topic priorities and have begun the guideline development process. Topics being
worked on in 2011 are:
Screening for breast cancer
Screening for hypertension
Screening for depression
Screening for diabetes
Screening for cervical cancer
Screening for obesity
Screening for child obesity
(ref. 1).
Appendix 1: Canadian Task Force on Preventive Health Care 159

The CTFPHC classifies its recommendations as either strong or weak.

In addition to the quality of supporting evidence, this classification is influ-
enced by
the balance between desirable and undesirable effects;
the variability or uncertainty in values and preferences of citizens; and
whether or not the intervention represents a wise use of resources
(ref. 1). A strong recommendation, either for or against an intervention, is one
in the context of which the Task Force is confident that its desirable effects
outweigh the undesirable ones or vice versa. Weak recommendation corresponds
to this outweighing being only probable in the judgement of the Task Force. (Ref. 1.)
It thus remains unclear whether individuals values and societys interest in cost-
effectiveness actually bear on the recommendations. The example in Appendix 2
strongly suggests that they dont.
So, the CTFPHC, continuing its tradition of excellence, still seems to be about
periodic health examinations rather than preventive health care, though with
the meaning of those examinations narrowed down to screening. And as will be
evident from Appendix 2 below, the meaning of screening, too, is narrower than it
might be expected to be: rather than all of that which is involved in the pursuit of
latent-stage diagnosis (rule-in), it is narrowed down to the initial test in this pursuit
(in the diagnostic algorithm) even though the recommendations are directed to
clinicians rather than community doctors. And most remarkably, that diagnostic
test is taken to represent intervention and thus to have desirable and undesirable
effects instead of being, merely, a source of information.
Regarding such a test for example, mammography on a woman with no overt
indication of having breast cancer, or weighing (?) a child with no overt indication
of being obese (!) the CTFPHC approaches decision-making about its use with a
methodology in which evidence takes precedence over consensus and evidence
presents itself with a given level of quality but not quantity. The essence of
this methodology, in its bypassing of consensus, remains a mystery, however. It
is illustrated in Appendix 2 below.
Appendix 2: CTFPHC on Screening for Breast
Cancer

Drawing further from reference 1 in Appendix 1 above, there is this piece of news:
November 21, 2011 The Canadian Task Force for Preventive Health Care has released
an updated guideline for breast cancer screening in average risk women aged 4074 : : :
The new guideline, which weighs the potential harms of false positives and unnecessary
biopsies against the potential benefits : : : updates prior guidelines by the Task Force from
1991 and 2001. (Italics ours; cf. App. 1 above.)

And there is also the guideline itself, separately for ages 4049, 5069, and
7074, specifically in respect to mammography (as distinct from MRI and clinical
breast examination). For this intervention the full specification is mammography
(film or digital) every 2 to 3 years. Everything about the guideline is contained on
a single page.
For those three ranges of age the respective Recommendations are: not routinely
screening, routinely screening, and routinely screening. Each of these is
characterized as weak, the first two on the ground of moderate quality evidence,
the third based on low quality evidence. (For the meaning of weak, see App. 1
above.)
The Basis of Recommendation is specified, separately, for each of those three
ranges of age. As an example, all that is said in reference to the age range 7074 is
this:
Women who do not place a high value on a small reduction in breast cancer mortality and
are concerned about false positive results of mammography and overdiagnosis may decline
screening. About 480 women aged 7074 die of breast cancer in Canada each year.

Associated with that Recommendation and that Basis of it is this declaration:

To save one life from breast cancer over about 11 years in this age group [7074],
about 450 women would need to be screened every 23 years
11 women would have an unnecessary breast biopsy
about 96 women will have a false positive mammogram leading to unnecessary anxiety
and follow-up testing.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 161

DOI 10.1007/978-94-007-4537-7, Springer ScienceCBusiness Media Dordrecht 2012
162 Appendix 2: CTFPHC on Screening for Breast Cancer

Besides, as for this range of age, For every 1,000 women screened for about
11 years, about 5 women will unnecessarily undergo surgery for breast cancer
(italics ours).
To us it is very unclear what it is that thus is being said and really meant in respect
to Canadian (sic; App. 1) women 7074 years of age (who are without personal
or family history of breast cancer, without known BRCA 1 or 2 mutation, or prior
chest wall radiation). How can the option to decline screening be a basis for
recommendation for screening? How can the annual number (sic) of breast-cancer
deaths within this lustrum of age be a basis for recommendation about screening
within this same, narrow range of age? or is it that at issue actually is initiation of
screening in the early 1970s of age? And when said is that To save one life from
breast cancer over about 11 years in this age group [of 5 years], about 450 women
would need to be screened every 2 to 3 years, is this about periodic screening in
the early 1970s or over about 11 years starting in the early 1970s? And is that
statement about one averted death over about 11 years a statement about the
duration of screening or about the time horizon for the averted death or both?
That statement about every 1,000 women screened for about 11 years presum-
ably is about women in whom the screening is initiated in the early 1970s of age.
Thus it presumably is about screening that could be continued up to age 86 or so.
But: No data from our review addresses the benefits of screening in women : : :
older than 74, implying that at issue actually is screening in the early 1970s only,
for up to 5 years, and not for about 11 years starting at that age.
While the obfuscation in this guideline/recommendation statement is severe to the
point of suggesting, to us, that it is intentional, science writers for newspapers
evidently captured a simple message:
Five days after the publication of these guidelines/recommendations, a national
newspaper of Canada (The Globe and Mail) announced, in a very prominent
headline, that Provinces re-evaluate breast screening. The subheading read:
Health-care providers are taking a fresh look at their rules and the costs of
administering them. The article proper, by Renata DAlesio, was about how those
guidelines from the CPTFHC have sparked a fiery national debate over which
women should receive x-rays and how often.
In that same issue of that paper, another eminent headline read, Cures for
cancer at any cost; and the associated subheading was, The benefits of breast
and prostate screening have been proved exaggerated, but we are no less invested in
them. The author, Margaret Wente, ascribed that exaggeration idea to a statement,
in 2009, by the chief medical officer of the American Cancer Society. Wente said
that The backlash [to the exaggeration idea] was ferocious. : : : The whole drama
was repeated in Canada this week, : : : . The value of mammography screening,
especially for younger women, has been decisively disproven. Many experts say it
is of no value. Period.
One week after these articles, a headline in the same newspaper read, When
emotion prevails over cold, hard science. The subheading was, Exceptionalism
helps explain why mass breast cancer screening persists despite evidence it does
Appendix 2: CTFPHC on Screening for Breast Cancer 163

more harm than good. The author, John Allemang, like his two predecessors
(above), wrote as though he fully understood what the Task Force was saying. So
he presented a very clear chart with the opening predicate: If 2,100 women, 4049,
at average risk of breast cancer were screened every 2 years for 11 years. In it he
presented the consequent numbers of cases of harm, while only 1 woman would
escape a breast-cancer death, all of this pictorially illustrated.
The harms obviously would be incurred in the course of those 11 years, but what
about the deaths that would be averted? By any reasonable presumption, the bulk
of them would have occurred after that 11-year period of screening. Allemang did
not indicate his understanding of what the Task Force meant in this regard, notably
whether its time horizon for deaths was limited to those 11 years of screening; nor
did he comment on his understanding of the relevance of the statistic that About
470 women aged 4049 die of breast cancer in Canada each year, being that
11-year screening initiated at age 49 presumably bears on deaths in the 1970s, even.
He wrote about what he saw as the implications of cold, hard science when the
Task Force itself based its recommendations on moderate quality evidence and
low quality evidence (cf. above).
As we, different from science writers for popular press, have difficulties in
understanding what the Task Force is saying about screening for breast cancer, and
as The particular characteristic that distinguishes the Task Force methodology from
traditional approaches in decision making on prevention issues is that evidence takes
precedence over consensus (App. 1 above), we took a look at the evidence that was
used. This the Task Force specifies in the Canadian Medical Association Journal
2011; 183: 1991-2001.
We found that two of the seven trials used as the sources of evidence enrolled
women in the 7074 range of age. In these, the total number of women of this age
assigned to the screening arm of the trials the CTFPHC reported to be 10,339, but
based on the trial reports themselves we found it to be 10,339 C 296 D 10,635. The
rest of the numbers are to the effect that the CTFPHC focused on the larger one of
those experiences.
In this trials report the duration of screening on women aged 7074 at its
initiation is not given in the original report; but according to the PDQ website of
the NCI (of the U.S.), its duration was 3 years, not about 11 years. The typical
duration of follow-up (from entry into the trial) was 13 years overall but unspecified
for those entering in the early 1970s of age.
The rates of death from breast cancer in the two subcohorts (N1 D 10,339,
N0 D 7,307) evidently were derived, simply, as 49/10,339 and 50/7,307, their
difference being 2.10/1,000, not 2.22/1,000. The number needed to screen, as
defined (App. 1), thus was 1,000/2.10 D 480, not 450, at issue being screening for
three (rather than about 11) years and death from breast cancer within 13 years
(rather than ever). This 480 is but the statistical point estimate from evidence so
imprecise as a matter of quantity rather than quality of the evidence that the
null P-value (one-sided) for the rate difference (and rate ratio) is as large as 0.04,
164 Appendix 2: CTFPHC on Screening for Breast Cancer

and derived from a study in which the Risk of bias, according to the CTFPHC,
was serious even with no regard for the incompleteness of the adherence to the
regimens in the trials two arms.
The Task Force drew the evidence for the three ranges of age solely from seven
randomized trials, evidently believing that these indeed represent the useful segment
of the research, of the entirety of it. At the core of this belief is the idea highly
aberrant that the initial test in screening is an intervention, together with the
associated idea common that an interventions intended effect is to be studied
by means of a randomized trial.
Such is the CTFPHCs reverence of randomization in the continuation of its
tradition of excellence that it seemingly takes no notice of the quality-relevant
particulars of this, while others see a sequence of embarrassments in this regard.
In his cancer-focused The Emperor of All Maladies (2010), Siddhartha Mukher-
jee relates some of the sad stories. In the Health Insurance Plan trial in New York,
the entries into the study cohorts, randomly separated as to intention-to-screen,
were initially without any determination/confirmation of the asymptomatic status,
the women in the control arm of the trial not even knowing of their involvement in
the trial; and later, women who were symptomatic on entry were removed from the
index cohort, but they could not be removed from the reference cohort (p. 297).
Edinburgh was a disaster. : : : [Various irregularities by both the doctors and
the patients] confounded any meaningful interpretation of the study as a whole
(p. 298). And the nature of the randomization completely undid the Canadian trial
(p. 2989).
So here we have gained some insight into the Task Force working with its renewed
commitment and vision to continue its 25-year tradition of excellence in which
a distinguishing characteristic is that evidence takes precedence over consensus
(App. 1). We see that this group makes authoritarian declarations of the form of
knowledge (present or future tense) but with content that in principle is formed
by mere results of studies, even very low quality evidence in this limited
meaning of evidence; and what is much more, those declarations can be grossly
counterfactual about the studies actual results. It is with this pseudo-knowledge
as the principal input that the CTFPHC engages in its main mission: the faux pas
(App. 1) of formulating authoritarian guidelines/recommendations for the practice
of healthcare clinical care.
Evidence per se, especially when misinterpreted (sect. 13.3) and seriously
misrepresented, should not take precedence over consensus-seeking deliberations
among genuine experts on a given topic of the knowledge-base of preventive
healthcare (sect. 13.5) of what is correctly understood to be preventive healthcare
(see below).
Given what the CTFPHC says about its work in a general way (App. 1) and what it
now says about screening for breast cancer specifically (above), and given also what
science writers say about those new recommendations/guidelines (above), some
additional critical comments are in order for the student to weigh and consider
(F. Bacon, sect. 1.4):
Appendix 2: CTFPHC on Screening for Breast Cancer 165

1. The CTFPHC classifies screening for a cancer (i.a.) as a matter of preventive

medicine, invoking the concept of secondary prevention and saying that this
is (by its own definition; ref. 2 in App. 1, p. 7) directed to asymptomatic
individuals with risk factors for a condition or preclinical disease (but not clin-
ically evident disease). But: our dictionary of medicine (Dorlands, 28th edn.)
defines preventive as serving to prevent the occurrence of, and preventive
medicine, accordingly, as that branch of study and practice which aims at the
prevention of disease and promotion of health. Secondary it defines as second
or inferior in order of time, place, or importance; derived from or consequent
to a primary event or thing.
2. The CTFPHC treats diagnostic testing, such as mammography, as an interven-
tion. But: according to our medical dictionary (above), intervention is: 1. the
act or fact of interfering so as to modify. 2. specifically, any measure whose
purpose is to improve health or alter the course of a disease. Mammography is
but a type of initial testing in the pursuit of early, latent-stage detection of
rule-in of diagnosis about breast cancer. Instead of an effect (in terms of
improvement in the course of the cancer), its yield is information (about the
latent presence of breast cancer). The associated intervention potential only
is the cancers early treatment (in lieu of its alternative, late treatment upon the
cancer already being overt, clinically manifest, symptomatic).
3. As the CTFPHC takes the purpose of mammography on asymptomatic women
to be well exemplified by the concern To save one life from breast cancer
over about 11 years and declares how many women need to be screened
every 2 to 3 years to accomplish this, it leaves altogether unspecified the
protocol/algorithm of the diagnostics-cum-treatment in each of those rounds
of mammography; and the rest of the meaning also is obscure, to say the least
(cf. above).
4. The meaning of this is not much clarified by the Procedure Manual of the
CTFPHC (ref. 2 in App. 1). Said there (p. 36) is this:
Evidence tables are also prepared, reporting information related to the key questions,
the grade of evidence and the results. The Number Needed to Screen (NNS) is also
calculated and added to the evidence table. NNS is calculated using the relative risk
method: first a weighted relative risk (RR) must be calculated and then the number of
lives saved per million ((1 RR) multiplied by control group event rate per million)
is calculated. Finally, the number needed to screen (1,000,000/lives saved per million)
is calculated. : : : All calculations and presentation of data in the evidence set are
rounded to four [sic] decimal places.

5. While we had great difficulty understanding what the CTFPHC is now saying
about screening for breast cancer and actually found the purportedly evidence-
based statements to be seriously misleading about the evidence, and while the
primary-care doctors receiving its Recommendations on screening for breast
cancer likely are equally uncertain about what is meant, women do understand
what a given round of screening for breast cancer is about: detecting and
treating, at this time, a latent case of breast cancer, should a detectable case
be present at this time this with a view to the thus-enhanced prospects of cure
166 Appendix 2: CTFPHC on Screening for Breast Cancer

of the cancer. The word cure is in the title of Ms Wentes column prompted
by the recent CTFPHC report (cf. above) while no reference to this concern is
made in that report.
6. The following should be agreeable by anyone seriously concerned with the
intended consequences of mammographic screening for breast cancer:
There is to be a definition, for a given round of the screening, of the
diagnostic algorithm in all relevant respects (all the way to the rule-in
diagnosis following biopsy); and this is to be supplemented with definition
of the treatment algorithm following the cancers rule-in diagnosis.
In reference to these regimens, the doctors, rational concern is to know,
specifically for the woman at issue, at the time: (a) the probability that a
single round of application of the defined diagnostic algorithm would result
in rule-in diagnosis of the cancers presence; (b) the probability that the
possibly detected cancer actually would be a life-threatening one; (c) the
probability that the (possibly detected) cancer, if indeed life-threatening,
would be curable by the defined early treatment while being incurable by
any available late treatment; and (d) the time to the death that thus might
be averted (ideally the probability distribution of this time, conditional on
otherwise surviving).
These concerns are in sharp contrast to the CTFPHCs key question:
Does screening with mammography : : : decrease mortality from breast cancer
: : : ?
7. As for the unintended, harmful consequences of such a defined single round of
diagnostics and potential early treatment, the concern is to know, again for the
woman at issue, at the time, the probabilities of: (a) positive result (as defined
by the algorithm) of the initial mammography (nothing false about this so
long as a correct algorithm is correctly followed); (b) biopsy resulting from
the work-up following positive mammography (nothing unnecessary about
this so long as, again, : : : ); and (c) being treated for a cancer that actually is
not life-threatening (cf. part b above, the treatment in these cases, too, being
necessary per the calculated risks accounted for in the protocol).
8. Even if all of this (nos. 6 and 7 above) be known from clinical (sic) research,
unknown from research would remain the valuations that the particular woman
at issue, at the time, associates with the probability of getting an already
extant, latent case of otherwise fatal cancer cured by early treatment and with
the probabilities of harms from the processes involved. All that the womans
realistic doctor can possibly hope to provide is that scientific knowledge input
to the decision. The woman herself brings the valuations and, then, integrates
these two sets of inputs informally into a decision about (the round of) the
screening at the time.
9. If the CTFPHC would understand the nature of the requisite knowledge-base
for rational decisions about mammographic screening for breast cancer, it
would realize that the relevant research remains essentially non-existent. While
it should understand its proper mission to be the formulation of the requisite
Appendix 2: CTFPHC on Screening for Breast Cancer 167

knowledge experts consensus alone (`a la IARC; sect. 13.5), in the prevailing
absence of the requisite knowledge it has a particularly compelling imperative
to refrain from putting forward purportedly research-based recommendations
for practice. And even if the scientific knowledge be there, it should understand
that [Science] knows nothing of policy or utility, of better or worse (App. 1).
10. The CTFPHC, after having studied this Appendix, should also study
Appendix 3 below and then reconsider its renewed commitment and vision to
continue its 25-year tradition of excellence (App. 1).
11. The Public Health Agency of Canada (of which the CTFPHC is part; App. 1)
should be clear on whether it still in this era of national health insurance
upholds its social contract in respect to medicine, continuing to respect the
professions autonomy and limiting its public policies about, for example,
screening for an illness to matters of cost reimbursement and its consequent
concern for cost containment consistent with medically good-quality care
(sect. 14.5).
Appendix 3: Needed Task Forces on Preventive
Healthcare

Healthcare aimed at prevention of illness should be maximally scientific not

in the meaning of science telling doctors what to do (App. 1) but as a matter
of having a rational theoretical framework together with knowledge-base from
science. Preventive healthcare, when scientific, is directed to removal of known
causes of illness (constitutional, environmental, and/or behavioral) or else to known
ways to protect against their adverse effects on health. Scientific healthcare is
knowledge-based rather than evidence-based healthcare, with the understand-
ing that the relevant knowledge is but part of the relevant inputs into decisions about
preventive action.
While prevention-oriented health research abounds, it tends not to naturally
translate into advancement of prevention-relevant knowledge (largely, we suggest,
because of the prevailing nature of the scientific journalism surrounding medicine;
sect. 13.6). There thus is a need for task forces with the missions to seek
to formulate the scientific knowledge for the multifarious sectors of preventive
healthcare.
Paradigmatic to these task forces can be taken to be the International Agency
for Cancer Research within the WHO. Even when focusing on a given substance
as a possible carcinogen, it makes elaborate ad-hoc arrangements in forming topic-
specific expert committees; and these, then, work very seriously in the pursuit of
consensus opinions (expressly refraining from making policy recommendations for
practice; sect. 13.5). Paradigmatic is not only the ad-hoc formulation of narrowly
focused expert panels, this by an agency that itself is no slouch in these matters. Of
major note also is the fact that the IARC does not form Canadian and U.S. panels, or
Scandinavian and Russian ones. It understands that scientific issues are not country-
specific but universal. It is, after all, an agency of the WHO. (All members of the
CTFPHC are Canadian.)
So, in the framework of the WHO or some other global organization concerned
with preventive healthcare there should be a number large number of
task forces in the meaning of differentiated standing committees for preventive

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 169

DOI 10.1007/978-94-007-4537-7, Springer ScienceCBusiness Media Dordrecht 2012
170 Appendix 3: Needed Task Forces on Preventive Healthcare

healthcare. These task forces would not formulate their own consensus opinions
on various topics; they would arrange suitable ad-hoc panels of even more
differentiated, ultimate experts to do this.
One rather obvious example of these task forces would be that focused on
radiation effects on health. It would arrange one expert committee to address the
health implications of the use of mobile phones; another one focusing on the safety
of the body scannings taking place at airports; and a third one on the health hazards
attendant to medical imagings (incl. in screenings for cancers); etc.
The very multifarious topic of nutrition as a determinant of health would
obviously require expertise very different from that having to do with radiation as
a health hazard. In this centrally important area, at least two separate task forces
would be needed, as the issues are so different between macro- and micronutrients.
In the formation of these, and other, task forces, one overarching principle is the
one already insinuated by these examples: they should be differentiated according to
the respective determinants they focus on, with no restriction on the health outcomes
of concern. There thus should not be a task force for preventive cardiology, another
for preventive oncology, etc. And there should not be separate task forces according
to the recipients of the knowledge general practitioners and community-level
educators, for example.
A` la IARC, the product from each expert panel should indeed be knowledge, not
guidelines/recommendations. For, Science never tells a man how he should act;
: : : (App. 1).
Related to all this, wed like to propose, for good measure, one task force
of a very different kind. The various expert panels focusing on their respective
areas of prevention-relevant research would have a shared problem: they would
need to pursue consensus essentially through their intra-panel discourse alone,
given the paucity of public discourse on the evidence per se for one, and on
inference from evidence for another (sect. 13.6). So, a useful purpose would be
served by an undifferentiated, single Task Force on Prevention-oriented Science,
making recommendations on how to improve the journalism environment of this
enormously important segment of health science (cf. sect. 13.6).
Appendix 4: On Introductory Teaching
of Scholars

K.S. Miettinen

An adequate introduction into a scholarly field might prepare the student for further
studies, and perhaps for initial steps in practice; but an excellent introduction is more
ambitious: it establishes the foundations needed for a career, perhaps half a century
in duration, and it specifically plants, at the beginning of that career, those seeds
which may, with suitable experience and reflection, mature into wisdom by the time
of retirement. A sign of an excellent introductory text is that the student repeatedly
returns to it, cherishes it, and continues to deepen his understanding of it throughout
his career. As DAlembert put it, To completely understand the elements [of a
science] requires more than simply knowing what they contain. One must also know
their use, applications and consequences. : : : The distinctive trait of a good book
about elements is leaving much to think about.
In reviewing numerous cherished introductions on my technical bookshelf, a
number of common characteristics stand out: the best introductions motivate the
field, clarify concepts, establish methodology (perhaps abstracted from progression
of method), identify points of divergence from and convergence with related
fields, warn the student about logical and conceptual problems, and recommend
approaches for prudently navigating those problems.
An example of excellent introductory texts is Linus Paulings General Chem-
istry, a text which has stood the test of time (since 1947) and changed how
chemistry is taught, specifically in introductory courses. The text motivates chem-
istry by depicting the universe as being composed of matter and radiant energy;
weaves an exposition of scientific methodology into the sequence of chapters in an
incremental fashion; explains difficult concepts such as temperature with precision;
and identifies points of divergence from and convergence with physics.
The present introduction into epidemiological research has these same qualities,
as it moves from grappling with the essence of the field to the knowledge that has
been achieved; clarifies important concepts such as the study base; and identifies
points of convergence with and divergence from related fields such as statistics and
sociology.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 171

DOI 10.1007/978-94-007-4537-7, Springer ScienceCBusiness Media Dordrecht 2012
172 Appendix 4: On Introductory Teaching of Scholars

Points of divergence and convergence are often unknown to those who learn
their field after it has reached steady state. For instance, a student searches in vain
through modern introductions into logic for the reason why logic was reformulated
in early twentieth century, or where modern logic diverges from classical logic.
But fortunately, Alfred Tarskis Introduction to Logic and to the Methodology
of Deductive Sciences is still available. Tarski explains the revolution in logic in
terms of establishing the foundations of mathematics (and deductive science more
generally), and throughout the text he points out the corresponding simplifications
(e.g., neglect of the logic of properties a matter of meanings as distinct from
the logic of classes, properties being unnecessary for establishing foundations of
mathematics). Tarski is a case in point of what W. Edwards Deming repetitively
emphasized, that teaching of beginners should be done by a master, not by a hack,
because so much of excellent introductory teaching consists of establishing nuanced
understanding of fundamentals. An implication of masters teaching introductory
courses is that these courses should generally have the highest students-to-teacher
ratios, as beginners are the most common kind of student and masters the least
common kind of teacher.
Tarskis inside knowledge also includes recognition of something that Deming
never pointed out, namely that among masters we should prefer those of that
generation which participated in the most recent revolution in the field.
This introduction into epidemiological research presents the insights of a seminal
participant in the most recent revolution in the field, the theory innovations of the
1970s. In this transition to developing introductory material the senior author fol-
lows in the footsteps of scholars such as Andrei Kolmogorov, who famously toward
the end of his career chose to focus on publishing introductory texts, developing a
new state program of education in mathematics, and teaching introductory courses
to young elite students in the belief that cultivating mastery of fundamentals in
the next generation of mathematicians was the most valuable contribution he could
make.
All fields of scholarly study have logical and conceptual problems; only the
most dogmatic (and least enlightened) teachers would present their field as being
problem-free. For instance, among the conceptual problems in mathematics are
the semantic problem of what particular numbers (such as 3) mean, and the
ontological problem of what infinity is. The solution of this in modern mathematics,
via deployment of the abstract method, is to banish meanings of numbers as
leading to too many fallacies; numbers are defined by the rules they follow without
otherwise having any specific meaning. A good introduction (such as that of
Timothy Gowers) tells the student that for every number A apart from 0 there
is a number C such that AC D 1 is a rule (the rule of multiplicative inverses), and
that this is a rule for which zero is not an exception; it has no exceptions. Since the
rules of arithmetic have no exceptions (else following the rules wouldnt sufficiently
define numbers) and infinity does not conform to the rules of arithmetic, infinity
is not a number. Such are among the first conceptual problems that a student of
mathematics should be introduced to; there are many others.
Appendix 4: On Introductory Teaching of Scholars 173

Logical problems in a science generally are of two kinds: problems within the
field, and problems about the field. In mathematics, for instance, a good introduction
would familiarize the student with Goldbachs conjecture (a simple unproven
apparent truth) and Wiles proof of Fermats last theorem (how can we know it
is a proof?), as well as Skolems paradox (do we understand what a real number
is?). Such problems do not need to be solved for the student to have a successful
career in a field, but understanding their difficulty is effective inoculation against
professional hubris.
Poor introductions, by contrast, misrepresent the field at issue through dangerous
simplifications that may make delivering the content to the student easier, but at
the risk of misleading the student as to truth. For instance, in one introduction
into statistics I read that the objective of statistics is to make an inference about a
population based on information contained in a sample and to provide an associated
measure of goodness for the inference. This is, of course, a simplification that
allows development of a mathematical body of statistical theory, but as a statement
of objectives it is false in general, though perhaps true in special cases (such as
opinion polling).
A better introduction into statistics would warn the student that this simplifica-
tion, which undergirds the theory, is generally false, and that the general objective
is to make inferences based on samples that can be validly applied beyond the
population of which the sample is representative. For instance, the objective of
the Framingham Heart Study surely was not limited to learning about the 1948
population of the town of Framingham. Going as far as is reasonable to validate
inference from past data to future experience (a special form of the problem of
inference to unsampled populations) without dogmatically reifying a metaphysical
state of nature is the topic of Walter Shewharts parvum opus Statistical Method
from the Viewpoint of Quality Control, a cherished introduction that I routinely
reread. (Shewhart also addresses the more ambitious question of what evidence for
the existence of a state of nature would look like; and as both Shewhart and his
most prominent apostle Deming point out, all rigorous attempts to develop evidence
indicating the existence of a state of nature have failed.)
Whereas all scholarly fields have logical problems, a good introduction explains
those problems while demonstrating valid ways to navigate them. For example,
Athanasios Papoulis classic introduction into Probability, Random Variables, and
Stochastic Processes opens with a chapter on the logical difficulties with the very
concept of probability (independence of the axioms and the various definitions),
which leads to demonstration of how comparing the quantum mechanical theories
of Maxwell/Boltzmann, Bose/Einstein, and Fermi/Dirac requires using three defi-
nitions of probability in a three-stage process. The classical (a priori) definition is
used to develop quantum mechanical models, the axiomatic definition is implicit in
the theory of probability with which the models are manipulated, and the relative
frequency definition is used in empirical testing of the model predictions. The
logical problem of developing the theory of probability is exposed to the student
174 Appendix 4: On Introductory Teaching of Scholars

rather than concealed, and prudent handling of the resulting difficulties is then
demonstrated.
The present introduction to epidemiological research gives the student similar
exposure to logical difficulties with what now are held as standard epidemiological
methods (such as cohort and case-control studies), and presents their needed
corrections.
Should any child of mine choose a career in epidemiological or related research
I would look forward to their being taught from this text, by one of the authors or
some other master, with the hope that the youngster would be a fertile receptor and
career-long cultivator of the wisdom sown here.
Appendix 5: Quality Assurance for Modern
Public-health Practice

K.S. Miettinen

I welcome the request (see Acknowledgements) to bring the perspective of industrial

quality control at large to bear on what in this text is suggested for the industry of
hospital-based healthcare (sect. 14.5). While agreeing with the suggestions in this
text, I add some philosophically relevant background to motivate and augment the
call to action.
The statistical and epistemological methods of W. Edwards Deming (originally
of Walter Shewhart, Demings teacher) are central to modern industrial quality
control, service industries included. The first two of his famous 14 points for
management are: Create constancy of purpose toward improvement and Adopt
the new philosophy.
Deming explained in his writings that while various stakeholders in an enterprise
may have different interests a labor representative may be concerned with
employee morale and job security, investors and taxpayers with costs, consumer
advocates with customer satisfaction, etc. all of these concerns are interrelated,
and regardless of which one is valued, the appropriate plan of action focuses, solely
and constantly, on improvement of quality.
Thus, the pursuit of quality is an indirect means of pursuing myriad other
things, which may or may not be publicly stated objectives. This is reflected in
the observation in the present text that good healthcare is not only medically good
but also economically good, that improving the medical quality of care is a way to
indirectly pursue also other societal objectives related to healthcare.
In nominally pursuing improved quality, as Deming explained, we should really
pursue greater uniformity of the goods produced or of the services provided. This
point is rooted, first, in his insight that the causes of poor quality are also causes of
variation in quality, so that effective pursuit of better quality and effective pursuit
of more uniform quality are identical. And besides, he pointed out that while a
rigorous epistemological basis for defining the limit to how far we can go exists for
uniformity, it often does not exist for excellence. Thus, quality improvement should
be pursued indirectly in terms of reduction in variation.

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 175

DOI 10.1007/978-94-007-4537-7, Springer ScienceCBusiness Media Dordrecht 2012
176 Appendix 5: Quality Assurance for Modern Public-health Practice

Finally, in nominally pursuing greater uniformity (lesser variation) of quality we

should really pursue greater randomness of variation in quality. This point is rooted
in his insight that a system whose variation of quality satisfies rigorous tests for
randomness is unimprovable, no matter how broad the variation in quality happens
to be; such a system can be replaced or redesigned, but its operation as designed is
as good as can be achieved. On the other hand, a system whose variation of quality
exhibits signs of nonrandomness can be improved by locating and removing the
causes of nonrandom variation, no matter how narrow the range of variation already
is. Greater uniformity too must be pursued indirectly, by eliminating nonrandomness
rather than by directly addressing the variation.
Thus, industrial quality improvement a` la Deming consists of two activities:
continuous effort to identify assignable causes of variation in systems whose results
show statistical evidence of nonrandomness, and regular redesign or replacement
of systems apparently operating with only random variation. A notable implication
of this is that while there may be acceptable levels of absolute quality and/or of
variation of quality for some purposes, there is no acceptable level of either of these
for the purposes of quality control. This is because, as I noted above, industrial
quality control is an indirect means to achieving numerous other objectives (e.g.,
cost reduction), rather than just quality per se.
This insight of Demings is the basis of his exhortation to eliminate management
by objective: there is no rational basis for setting a standard for what is acceptable.
What is needed is constant pursuit of improvement, whatever be the level of
performance already achieved.
Thus, in the quality improvement plan for Montfort Hospital cited in the text,
the problem with the plan to ensure 65% compliance with handwashing protocols
is not the absurdly low level of the goal; it is the very idea that an acceptable level
can be rationally assigned. The plan should be to identify and eliminate assignable
causes of variation in handwashing practices; and when that has been driven to
the extreme of apparently no remaining assignable causes, then the follow-on plan
should be to rethink and redesign workflows so that further improvements in the
rate of handwashing can be made. Redesign of a system that is not yet in control
is contraindicated; much of the information and insight that is needed to design a
better system is not available until the existing system truly is operating as it was
designed to operate.
The approach outlined in the text, involving sampling-based monitoring of
care processes (rather than outcomes), is directly analogous to the sampling-based
monitoring of processes currently used in other industrial programs of continuous
improvement. I would add that in the industrial programs in general the information
from samples is not used to identify causes of variation; it is used, instead, to
indicate when it would be economical to initiate inquiry into causes of the variation.
For the process of searching for assignable causes of variation costs time, money,
lost productivity, and often also pride and workplace harmony. For these reasons,
such inquiries are kept to a minimum. Statistical sampling of processes (not merely
records; the sampling should produce data not otherwise recorded) provides the
basis for deciding when to search, and it limits the scope of where to search and
Appendix 5: Quality Assurance for Modern Public-health Practice 177

what to search for; but the identification of an assignable cause of variation (an
object of repair, retraining, or disciplinary action) is never made on the basis of
the sampled data, it is made only on the basis of a thorough inquiry triggered by
statistical anomalies in the sample data.
There are details of Demings perspective that would lead to augmentations of
the quality assessment program sketched in section 14.5 beyond sampling of the
records by sampling of the processes to produce new data. Some of these may
be worth mentioning. In the text the authors proposed review, by an in-hospital
panel, of narratives of the care of individual patients. This is, no doubt, salutary,
and representative of the activities of development programs in all professions
But the Deming perspective is not to focus on the sequence of processes that a
unit (e.g., a patient) passes through, but rather focus on the sequence of units that
pass through an identifiable process (e.g., colonoscopy). Both perspectives provide
valuable information, but it is the latter perspective that enables the statistical search
for evidence indicating nonrandom variation.
The rates of untoward findings should not be the basis for remedial action
as suggested in the text: Deming often made the point that rates themselves
do not communicate whether a system is operating at capacity. Instead, they
provide raw material which epidemiologists would be well qualified to critically
examine in searching for evidence of performance levels outside of system control
limits. Epidemiologists would provide this service by combining rates of untoward
findings with other facility-specific information, for example on demographics of
the served population, caseload during the period in question, experience level of
staff, etc., to assess whether an observed rate of untoward findings deviates enough
from experience across other facilities to warrant an inquiry into the existence of
assignable causes of variation at that facility.
These details of a program of quality improvement can be worked out once hos-
pital administrators become acquainted with and adopt the conceptual framework of
continuous improvement. There will, however, be institutional obstacles to adopting
this framework, which need to be thought through. Demings industrial quality
improvement philosophy requires close collaboration among the parties involved
in ways that differ from models of confidentiality and proprietary information
currently fashionable in medicine. In manufacturing-related industrial settings, part
of the means to avoid expensive inspections of purchased goods (e.g., parts pur-
chased from a supplier for further assembly) is the requirement that vendors provide
evidence of quality control of their processes along with their deliveries, so that
purchasers can confirm that a lot of goods was produced under suitably controlled
circumstances; the goods themselves are not inspected; instead, information about
the vendors internal processes is disclosed along with the delivery. In a hospital
setting, the analogue would be providing to those concerned, patients included, a
report on the degree of uniformity achieved in the processes relevant to the quality
of the provided care.
Even though medical institutions are supposed to be self-policing, it is not
enough to supply quality-assessment reports to the hospitals quality-assurance
boards and government regulators, for these entities are not interested in the quality
178 Appendix 5: Quality Assurance for Modern Public-health Practice

in the classical sense of the term. They do not have a stake in the quality of care,
as they do not suffer the consequences of poor-quality care. While the persons in
those functions may well have the best interests of patients and society at heart,
management practices are generally anchored to the interests associated with the
positions involved, rather than to presumptive virtues of the people assigned to those
positions. Society at large, and patients in particular, have a stake in the quality of
healthcare, so they are the genuinely interested parties to whom evidence of internal
quality control should ultimately be disclosed.
Administrators and regulators have a two-tiered responsibility in quality control:
first, to drive the continuous pursuit of quality within systems operating as designed,
and second, to rationally redesign systems operating at the limits of their capacity
for better performance. It is with respect to the first of these two that assurance of
performance should be provided to patients and to society at large, and it deserves
note that this is not information of a professionally esoteric kind; information
about the state of control of internal processes is largely non-technical and thereby
comprehensible to the general public. This information does not tell patients
whether their specific care will be (or was) good or not, but it can assure them
that the systems that care for them are operating at capacity, that is, that the
care being provided is statistically as good as can be reasonably expected of the
system. On the other hand, information supporting redesign of workflows should be
shared within competent professional communities, notably among administrators
of similar facilities across the jurisdiction involved.
The application to medicine of the quality control methods of industry at large
is complicated by the fact that medicine is a service industry (not unlike, e.g.,
the hospitality industry). Even though the methods pioneered in manufacturing
have been successfully applied in service industries as well, banking and hotels
being outstanding examples, many problems specific to service industry have been
identified in these applications.
One of the most difficult problems in service industries in general is that
enterprises tend to serve captive audiences; there is little pressure from distant
competitors, and therefore considerable difficulty in motivating the necessary
constancy of purpose necessary for continuous improvement. In line with this, there
are rarely two general hospitals operating across the street from one another, and
hospitals at a distance are only slightly in competition with one another.
Another problem of service industries is that they do not face elastic market
demand; a service enterprise cannot expand the overall volume of business in its
segment (e.g., by productivity breakthroughs that lower costs); it can at best take
market share from competitors. Thus, a hospital that lowers the cost of delivering
babies may take business away from other hospitals, but it is unlikely to affect the
overall number of births in a given population. Improvements in the cost of child
birth are not rewarded with a baby boom.
In service industries there is a tendency to sample users of the of service
regarding their satisfaction, while really needed for quality control is sampling of
the processes that go into providing service. Just as quality control in manufacturing
cannot be based on examining finished products, so also quality control in medicine
Appendix 5: Quality Assurance for Modern Public-health Practice 179

cannot be based on interviews of served recipients of care. Medical care in larger

institutional settings provides opportunities for sampling of care processes, while
the service processes of solo practitioners are difficult or impossible to document.
A final problem typical of service industries, and shared by medicine, is the
problem of one-of-a-kind in service environments. No two hospitals have the same
layout, or serve demographically identical populations, or have identical average
ambulance transportation times from residential neighborhoods. As a consequence,
each facility is likely to implement distinct processes, and therefore lessons learned
at one facility may not be transferrable to another, and rates of error will be difficult
to compare between facilities. Some service enterprises, chain restaurants and some
hotel chains being prominent examples, have made great efforts to standardize
facility layout and location (e.g., relative to road intersections) in order to mitigate
the one-of-a-kind effect, and the results have been truly remarkable. Many diners
may not think of chain restaurant food as being of high quality, but when quality is
understood in terms of eliminating nonrandom sources of variation, then it is indeed
of very high quality it is very consistently what management specifies it should
be, the diners wishes notwithstanding. The degree to which such standardization is
possible (or politically acceptable) in medicine remains to be seen.
As medicine shares with certain other service industries (e.g., education) the
peculiar structure of having third-party payers cover some or most of the expense,
this structure poses certain further challenges for quality assurance, notably agency
conflicts and data quality (objectivity) challenges. Whereas third-party payer sys-
tems are inherently prone to shift costs, that is, charge some customers more than
the true cost of their service in order to subsidize similar service to others (e.g.,
cost transfers between urban and rural populations, or between wealthy and poor
populations, etc.), third-party payer systems tend not to generate objective cost data
on the services rendered the subsidies are hidden in the cost data, making the data
not truly representative of costs under management. Without reliable data, quality
assurance is greatly complicated, if not precluded outright.
Where possible, the remedy would be to separate the subsidy function from the
function of rendering service, for example by having the government restrict its
role to paying subsidies (and monitoring regulatory compliance) and allowing the
private sector to render service (and collect geographically varying market fees).
Similar separation could be accomplished between national and state/provincial
governments, to a degree. Such separation of duties is called for in all aspects
of management practice where the integrity of data is a concern, and it is the
management analogue of the principle of separation of powers in government,
whereby preventing concentration of responsibility in one branch makes collusion,
concealment, and deception more difficult. The argument for separation of duties
is only indirectly an efficiency argument; it is directly a transparency and integrity
argument. Separation of duties is an aspect of system design intended to expose
information that might otherwise be concealed or distorted, by documenting the
information in the public interactions of the separated parties. This information
is then available for better management toward whatever objective is socially
desirable, be that efficiency, equity, availability, or quality.
180 Appendix 5: Quality Assurance for Modern Public-health Practice

Major cultural changes are required if programs of improvement in hospital-

based healthcare were to be brought to the standards of service industry in general,
but this should not deter taking on the mission. One reason for setting out to meet the
challenges is the enormous improvement not only in manufacturing but in service
industries as well consequent to the introduction of the quality control ideas in
the 1920s (the time of the original work of Shewhart, Dodge, and others). And
another, burning reason is the well-known and ever worsening cost crisis of modern
healthcare. The authors gambit deserves both attention and follow-on.
Author Index

A E
Allemang, J., 163 Einstein, A., 173
Armstrong, B., 19
Atwood, K.C., 128
F
Ferguson, N., 2, 3, 9
B Fermi, E., 173
Bacon, F., 13, 27, 125, 164 Feynman, R., 17
Barrable, B., 152 Fletcher, S.W., 30, 31
Beresford, S.A.A., 76 Fortin, P., 151
Berkman, L., 26, 27 Franklin, B., 2
Berlin, I., 134
Bismarck, O., 23, 24
Blackburn, H., 32 G
Bohr, N., viii Gibran, K., 13
Boltzmann, L.E., 173 Goldbach, C., 173
Bose, S.N., 173 Gorbachev, M., 10
Boyle, R., 13, 124 Gowers, T., 172
Bronowski, J., 17, 22 Graham, E., 33
Greenland, S., 7, 39

C H
Caballero, B., 29 Hansheer, R., 134
Cochran, W.G., 137 Hardy, H., 134
Cohen, B., 2 Harris, R., 30, 31
Cooper, T., 23 Hayes, R.B., 76
He, F.J., 24
Herlihy, D., 2
D Hill, A.B., 137139, 142
DAlembert, J., 171 Hippocrates, 6, 47
DAlesio, R., 162 Hofman, A., vi, ix
Deming, W.E., 156, 172, 173, Hume, D., 47
175177
Demissie, K., 80, 82, 83
Deutsch, D., 9 I
Dirac, P.A.M., 173 Ibrahim, M., 25, 26
Doll, R., 18, 19 Ipsen, J., 7

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 181

DOI 10.1007/978-94-007-4537-7, Springer ScienceCBusiness Media Dordrecht 2012
182 Author Index

J Pielke, R.A., 142

Jenner, E., 2, 5, 6 Plato, 47
Johnson, P., 47 Popper, K., 9
Porta, M., 7, 39
Proctor, R., 4, 33, 66
K Pugh, T.F., 7
Kant, I., 9, 47, 54, 73
Kawachi, I., 26
Kelishadi, R., 29, 30 R
Keys, A., 17, 23 Rose, J., 109
Koch, R., 6 Rothman, K.J., 7, 9
Kolmogorov, A., 172 Russell, B., viii
Krieger, N., 26
Krist-Etherton, P.M., 18
Kuhn, T., 123125
S
Saracci, R., viii
L Sartwell, P.E., 7
Lambert, M., 30 Schilling, R.S.F., 4, 6
Last, J.M., 7, 39 Schweninger, E., 23, 24
Liang, M.S., 151 Shewhart, W., 173, 175, 180
Lilienfeld, A.M., 7 Shuurman, A.G., 76
Lilienfeld, D., 6 Snow, J., 6
Lippman, S.M., 76 Socrates, 47
Stamler, J., 17
Stare, F., 23
M Steinberg, J., 23
MacGregor, G.A., 24 Stone, J.F., 17
MacMahon, B., 7, 9
Marshall, B.J., 128, 129
Maxcy, K.F., 7 T
Maxwell, J.C., 84, 173 Tarski, A., 172
McGovern, G., 23 Taubes, G., 1624, 32, 126, 127
Miettinen, K.S., ix, 171, 175 Topol, E., 145
Miettinen, O.S., viii, 5 Trichopoulos, D., viii, 9
Morris, J.N., 4
Mukherjee, S., 9, 164
Muller, F., 33, 65, 66 V
Van der Weyden, M., 128
Vebler, T., 157
N Von Mises, L., 157
Newhouse, M.L., 4

W
O Wald, A., 60
Ockham, W., 47 Warren, J.R., 128, 129
Olsen, J., viii Weber, M., 157
Wennberg, J., 108, 156
Wente, M., 162, 166
P Wynder, E., 33
Papoulis, A., 173
Pasteur, L., 2, 6
Pauling, L., 171 Y
Peto, R., 18 Yoshizawa, K., 76
Subject Index

A Initiation of c., 77, 78, 81

A Dictionary of Epidemiology, 7, 39, 65, 67 Oncogenesis, 1821, 74, 141
Ageing, 11, 2122 Promotion of c., 77, 78
American Heart Association, 17 Prostate c., 19, 20, 25, 2730, 7679
American Institute for Cancer Research, 19 Carcinogenesis. See Oncogenesis under Cancer
Ancillary Case-control study. See Study
A. contrasts, 99 Causal contrast, 36, 37, 78, 79, 84
A. outcomes, 99 Causation/cause, 3640, 47, 48
Antioxidants, 20, 21, 7480, 98 Cohort study. See Study
Atherogenesis, 1618 Community diagnosis/etiognosis/ prognosis,
3638
Community medicine. See Medicine
B Confounder score, 61
Base Confounding
B. series, 5761, 6571, 82, 88, 89, 9194, C. bias, 49, 66, 71, 81, 89, 90, 92, 98, 100,
99, 101, 102, 120, 121 101, 105, 121, 122, 138, 139, 141
Source b., 5658, 65, 67, 69, 71, 8991, C. factor, 49, 51, 54
121, 122 Control of c., 49
Study b., 48, 50, 5658, 61, 6567, 70, 71, Prevention of c., 66, 89
75, 76, 80, 8994, 98, 99, 101, 120, Cost containment, 150, 152
122, 132, 133, 135, 148
Bias
Confounding b., 49, 66, 67, 71, 79, 89, 90, D
92, 98, 100, 101, 105, 121, 122, 138, Data analysis, 84
139, 141 Data matrix, 58
Documentation b., 90, 91 Design
Selection b., 90, 132 Methods d., 12, 13, 71, 8795, 100103,
Biostatisticians vs. epidemiologists, 85 110
Black death, 2 Objects d., 10, 7385, 99100, 110
Blinding, 89, 101 Disease
Chronic d. as misnomer, 32
D. of civilization, 17, 18, 127
C D. vs. illness, 143, 144
Canadian Task Force on Preventive Health
Care. See Task Force
Cancer E
Breast c., 19, 30, 31, 80, 83, 114, 120, 158, Efficiency
159, 161167 Optimization of e., 92

O.S. Miettinen and I. Karp, Epidemiological Research: An Introduction, 183

DOI 10.1007/978-94-007-4537-7, Springer ScienceCBusiness Media Dordrecht 2012
184 Subject Index

Epidemics, 2, 4, 68, 22, 119 I

Epidemiologic Reviews, 2533, 142 Indicator variate, 52
Epidemiology Informativeness, 91, 93, 94, 110, 111
Analytic e., 27, 28 Informed consent, 104
Clinical e., 10, 11, 151, 155 Inoculation, 2
Concept of e., 3 Institute of Medicine, 151
Descriptive e., 27, 28 Insulin, 19, 21, 23, 24, 126, 127
E. as healthcare, 24 International Agency for Research on Cancer,
E. as research, 58 136, 138142, 169
E. as research discipline, 810 International Epidemiological Association,
E. as subject of study, 1013 25
Ecological e., 27 Intervention
Journalism for e., 144 I. as type of cause, 36
Social e., 11, 2527 I. misunderstood, 65, 159, 161, 164
Uses of e., 3, 4 I. prognosis, 99, 102
Ethics I. study (see Study)
Deontological e., 103, 104
Teleological e., 103
Etiogenesis. See Etiology K
Etiogenetic proportion Knowledge, 1533, 135142, 169, 170
Factor-conditional e. p., 39
Overall e. p., 38, 39
Etiognosis, 36, 37
Etiology, 36, 3841, 4354 L
Evidence Linear compound, 53, 102
Burden of e., 95
Essence of e., 133
Inference from e., 134135 M
Interpretation of e., 133134 Manual of operations, 94
Reporting on e., 144 Matching, 66, 92, 139, 141
Expert committees, 136, 169, 170 Medicine
Clinical m., 4, 3940
Community m., 3, 4, 3839
F Preventive m., 4, 108, 109
Fact-finding, 148, 149 Public-health m., 108, 119
Framingham Heart Study, 17, 32, 64, 88, 93, Metabolic syndrome, 1719, 21, 22, 30
173 Micronutrients, 20, 46
Model
Logistic m., 58, 102
G Log-linear m., 5254
Guidelines. See Recommendations M. fitting, 5860
Operationalized m., 88
Statistical m., 52
H
Hazard ratio, 79, 125, 126
Healthcare N
Evidence-based h., 169 National Cancer Institute, 19, 76, 78
Knowledge-based h., 169 National health insurance, 108, 114, 150, 151,
Scientific h., 169 153
Health education, 16, 25, 37, 39, 45, 46, 109, National Heart and Lung Institute, 32
115, 119 National Lung Screening Trial, 81
Hospital care, 150156, 175, 177 Nurses Health Study, 88
Hypertension, 2425 Nutrition, 1625, 126
Subject Index 185

O Rate ratio
Obesity, 2224, 25, 2930, 126, 127 Adjusted r. r., 50, 59
Occupational health, 4, 68, 69 Crude r. r., 48, 49
Occurrence relation, 8, 9, 51, 52, 54, 5659, Standardized r. r., 4952
61, 99, 100 Recommendations, 138, 142, 158, 161
Omega-3 fatty acids, 17, 18 Research
Oncogenesis. See Cancer Basic r., 5
Oncology, 5, 6, 8, 9 Epidemiological r., 5
Ottawa Public Health, 3, 4, 148150 Oncological r., 5, 8
Overmatching, 66 Risk
R. assessment, 109, 110
Relative r., 28, 134
P R. factor, 4, 38, 44
Paradigm, 124129 R. function, 102, 110
Parameter of Nature, 52 R. indicator, 110
Parsimony, 47 R. period, 109, 110
Person-moment, 5658, 99 R. profile, 109
Placebo, 89, 101 R. ratio, 125
Policy, 44, 45, 114, 115, 126, 136, 142, 158
Population
Catchment p., 56, 67, 71 S
Closed p., 56, 63 Salt, 24
Cohort-type p., 56, 64, 65 Sampling (for base series)
Dynamic p., 56, 57, 64, 65 Discriminate/stratified s., 71, 92, 141
Open p., 56, 64 Indiscriminate/unstratified s., 71, 91, 141
Source p., 56, 57, 64, 65, 68, 69, 71 Matched s., 94, 141
Study p., 56, 57, 64, 65, 67, 68, 71 S. by extraneous outcomes, 91
Precision, 5961, 93, 95 Science
Prevention/preventive, 25, 7, 16, 19, 20, 22, Applied s., 5
2427, 29, 30, 3740, 76, 78, 98, 105, Basic s., 5
108110, 114, 118, 126, 136, 138, 140, Epidemiology as s., 2, 5
148, 157159, 163165, 169170 Normal s., 124, 143
Preventive Medicine Task Force. See Task Paradigms in s., 124
Force Philosophy of s., 124
Public health, 3, 4, 148, 150, 151, 153, 156, Public discourse in s., 143
175180 S. writers, 162
Screening, 25, 80, 114122, 158, 161
Series
Q Base s., 5761, 6371, 75, 88, 89, 99, 101,
Quality assurance/control, 148, 150152, 120
154156, 175180 Case s., 47, 5760, 6468
Smallpox, 2, 5
Smoking, 32, 45, 65, 136
R Smoking and Health, 32, 136138
Rate Source base, 56
Adjusted r., 38, 50, 51 Statistical methods, 85
Crude r., 4852 Statistical variate, 52
Index r., 48 Statistics, 12, 171
Overall r., 48, 49 Study
R. as emergent phenomenon, 8 Case-control s., 57, 65
Reference r., 48 Cohort s., 63
R. ratio, 48, 49 Cross-sectional s., 67
Specific r., 49 Derivative s., 132
Standardized r., 49 Etiogenetic/etiologic s., 47, 51, 5660
186 Subject Index

Study (cont.) Time

Intervention s., 98105 Cohort t., 63, 99, 118
Object of s., 7, 48, 51, 59, 70, 7484 Etiogenetic t., 99
Original s., 132 Intervention-prognostic t., 99
Prospective s., 64, 137 Scientific t., 89, 99
Quasi-experimental s., 104 Study t., 99
Retrospective s., 64, 137 Trial
S. base (see Base) Before-after t., 99
S. design (see Design) Parallel t., 102
Semi-cohort s., 68 Randomized t., 20, 31, 98, 115118, 125,
S. protocol, 90, 94 164
S. result, 51, 58, 75, 94, 135 Screening t., 81, 118
S. size, 90, 95

T V
Task force Vaccination, 2, 108
Canadian TF on Preventive Health Care, Validity. See also Bias
142, 157, 161 V. assurance, 91, 122
Needed TFs, 169 Variola. See Smallpox
Preventive Medicine TF, 31, 142
Teaching
Introductory t., 10, 171174 W
Temporal relation, 75, 77, 100, 101, 137 Wald statistic, 60
The Epidemiologic method, 7, 137 World Cancer Research Fund, 19
Theory of research, 9, 10, 74, 144 World Health Organization, 2, 136, 169