ANALYSIS OF EPIDEMIOLOGIC

STUDIES: EXPLORING THE ROLE

OF ERRORS IN MEASUREMENT

DR. MUIA NDAVI

MBCHB,MMED O/G, MSc.EPID,DLSHTM,FHBR
OBJECTIVES

At the end of this session, the students will

be able to:

 Review criteria for study design choice in

epidemiological research
 Outline and describe the major classes of errors
in measurement
 Explore the general approaches to minimize
their effects.
Criteria for study design choice in
epidemiological research

• State of knowledge of the problem

• Type of research questions
• Type of study design
INTRODUCTION I
 Errors
in measurement are
basically grouped into two major
categories:

 Bias

 Confounding.
INTRODUCTION II
Bias and confounding may be minimized
by ensuring that groups of study subjects
are comparably:
• Selected
• Interviewed (or otherwise provide data), and
• Influenced by risk factors other than the one
being studied.
 BIAS I

 Definition: Bias may be defined as any

systematic error in an epidemiologic study that
results in an incorrect estimate of the
association between EXPOSURE and RISK OF
DISEASE/OUTCOME.
 BIAS II:

2 Types
 2.1 SELECTION BIAS:

 Is any error arising from the process of

identifying the study populations.
 Gives rise to non-comparability between
 a) cases and controls in case-control study and
 b) exposed and unexposed in cohort and

experimental studies
 BIAS III:

 2.1 SELECTION BIAS cont…

 Primary cause of selection bias varies with the
study design viz:

 In cohort studies selection bias occurs when

factors associated with the health outcome
influence study participation
 BIAS IV:

 2.1 SELECTION BIAS cont…

 Primary cause of selection bias varies with the
study design viz:

 In case-control studies, selection bias occurs

when factors associated with exposure influence
study participation
 BIAS V:

 2.1 Selection bias cont….

 Selection bias is a particular problem in case-

control and retrospective cohort studies, where
both the exposure and outcome have occurred
at the time individuals are selected into the
study.
 BIAS VI:

 2.1 Selection bias cont….

 Toprevent selection bias, all study subjects

(cases and controls or exposed and unexposed
persons) must be selected in a comparable
manner from the same population
 BIAS VII

2 Types cont…
 2.2 INFORMATION (OBSERVATION,
CLASSIFICATION, OR MEASUREMENT)
BIAS:
 This group of biases concerns the way in which
information is obtained from or about study subjects.
 BIAS VIII

 .2.2 Information bias:

 Results from systematic differences in the way

information on exposure and or outcome is
obtained from various study groups. The
distinguishing feature of information bias is that
it occurs after the study subjects have been
selected.
 BIAS IX

 2.2Types of information bias cont…

 Recall Bias: Arises when individuals with a

particular adverse health outcome remember
and report their previous exposure experience
differently from those who are not similarly
affected
 BIAS X

 .Types of information bias cont…

 Interviewer Bias: Concerns the way

information is obtained from or about study
subjects.
 BIAS XI

 .Types of information bias cont…

 Loss of subjects to follow-up: a major source

of bias in cohort and or intervention studies due
to the necessity of following individuals for a
period of time after exposure and or treatment
to ascertain whether they develop the outcome
of interest
 BIAS XII

 .Types of information bias cont…

 Ascertainment Bias: occurs if the outcome of

interest is obtained by using different methods
for each comparison group
 Missing data bias

 Prevarication bias: study subjects or data

collection personnel intentionally distort data.
 BIAS XIII: CONTROL OF BIAS:

. The prevention and control of potential biases

must be accomplished largely through careful
study design: selection bias may not be rectified
once it has occurred.
 Design features that can minimize the potential
for bias in any study include choice of the study
population, the sources of data to be utilized
and data collection process.
 CONFOUNDING I

. Definition:

 Confounding bias occurs when the effects of

two risk factors or exposures are mixed in the
occurrence of the health problem under study.
 It is a function of the complex relationship
between various exposures and the disease in
the population.
 CONFOUNDING II

 Characteristics:

 Unlike bias, which is primarily introduced by the

investigator or study participants, confounding
is a function of the complex interrelationships
between various exposures and disease.
 CONFOUNDING III

 . Example 1:

 In a study that showed a relationship between increased level of

physical activity and decreased risk of myocardial infarction,
additional variables that might affect the observed magnitude of
this association are age and sex. People who exercise heavily
tend to be younger than those who do not exercise. Moreover,
independent of exercise, younger individuals have a lower risk of
MI than older people: thus those who exercise could have a
lower risk of MI quite apart from any effect of this habit simply
because of the greater proportion of younger individuals in this
group. Hence, age would confound the observed association
between exercise and MI and result in an overestimate of any
inverse relationship.
 CONFOUNDING IV

 .In most general terms, for a variable to be a

confounder, it must be:
 1. Associated with the disease (cause disease)
(D)
 2. Associated with the exposure under study (E)
 CONFOUNDING V

 E D

 CF
 Fig. 1. Interrelationship between an exposure
(E), confounding factor (CF), and disease (D).
 CONFOUNDING VI

 .Other characteristics of a confounder:

 First, while the potential confounding factor must, by

definition, be predictive of the occurrence of disease,
the association need not be causal:
 Second, the potential confounding factor must be

predictive of disease independently of its association

with exposure under study
 CONFOUNDING VII

 Finally, the potential confounder cannot merely

be an intermediate link/step in the causal
chain/pain between the exposure and disease
under study.
 CONFOUNDING VIII

 E ?CF D

 Fig. 2: Interrelationship between an exposure E, disease

D, and a potential confounding factor? CF which is in the
causal pathway and thus NOT a confounder.
 CONFOUNDING IX

 As shown in fig 2, if one mechanism of

action of the exposure is to alter the level
of the potential confounder, which in turn
itself affects disease risk, then the factor is
not a confounder but rather an
intermediate step in the causal chain
between the exposure and disease.
 CONFOUNDING X

 Example 2:

 Alcohol raises high-density lipoprotein which

HDL, in turn, is associated with reduced risk of
MI independent of alcohol consumption. Hence,
HDL is not a potential confounder when
considering the effect of moderate alcohol
consumption on risk of MI. In this case HDL is
known as an effect modifier.
 Confounding XI: testing for
confounding
 A practical means to determine whether a
given factor is in fact an actual confounder
in a study is to analyze the data, obtain the
crude overall estimate of the association,
control for the effect of the variable, and
observe whether the estimate of the
association between the exposure and
disease is altered.
 Confounding XII: testing for
confounding
 Example 3: All staff in a firm were
interviewed following an outbreak of
gastroenteritis (GE) in the canteen.
 Confounding XIII: testing for
confounding

% developed GE-crude
Exposure variable
attack rate

Ate fish-No. 87 76

Did not eat

130 57
fish-No.
 Confounding XIV: testing for
confounding
 Give examples of potential confounders
 Confounding XV: testing for
confounding

Mayonnaise as a confounder
EXPOSURE
% GE & Ate % GE & No
Mayonnaise Mayonnaise

Ate fish 63/72=88 3/15=20

Did not eat fish 62/76=82 12/54=22

 Confounding XVI: testing for
confounding
 Do you think the fish caused the outbreak?

 What in your opinion is the best way to
summarize these results?
 Confounding XVII: how to select
confounding variables at design stage
 Making a judicious choice about potential
confounders depends on knowledge of the
disease, previous evaluations of the same
or related questions, and most importantly,
the best judgement of the investigator at
the time the study is initiated. Since
uncontrolled confounding is a major threat
to the validity of results, it is imperative
that the design features of the study permit
the collection of data to address the issue.
 Confounding XVIII: controlling for
confounding
 Randomization:
 Restriction:
 Matching: