practice, which in turn have impact on health care planning decisions.

Biostatistics and Epidemiology In the following section, we will consider selected epidemiologic
concepts.
Chapter One: Principles and Methods of Epidemiology
“Medicine to produce health has to examine disease; and music to
create harmony, must investigate discord” (Plutarch A.D. 46–120).

1.1 The Uses of Epidemiology

Epidemiology may be defined as the study of the distribution of health
and disease in groups of people and the study of the factors that
influence this distribution. Modern epidemiology also encompasses the
evaluation of diagnostic and therapeutic modalities and the delivery of
health care services. There is a progression in the scientific process
(along the dimension of increasing credibility of evidence), from casual
observation, to hypothesis formation, to controlled observation, to
experimental studies. Figure 1.1 is a schematic representation of the
uses of epidemiology. The tools used in this endeavor are in the
province of epidemiology and biostatistics. The techniques used in
these disciplines enable “medical detectives” both to uncover a medical
problem, to evaluate the evidence about its causality or etiology, and
to evaluate therapeutic interventions to combat the problem.

Descriptive epidemiology provides information on the pattern of

Figure 1.1
diseases, on “who has what and how much of it,” information that is
essential for health care planning and rational allocation of resources. 1.2 Some Epidemiologic Concepts: Mortality Rates
Such information may often uncover patterns of occurrence suggesting
etiologic relationships and can lead to preventive strategies. Such The comparison of defined rates among different subgroups of
studies are usually of the cross-sectional type and lead to the formation individuals may yield clues to the existence of a health problem and
of hypotheses that can then be tested in case-control, prospective, and may lead to the specification of conditions under which this identified
experimental studies. Clinical trials and other types of controlled health problem is likely to appear and flourish.
studies serve to evaluate therapeutic modalities and other means of In using rates, the following points must be remembered:
interventions and thus ultimately determine standards of medical
  Biostatistics and Epidemiology

 A rate is a proportion involving a numerator and a the midyear population is an estimate of the average number during the
denominator. year. One can, however, speak of death rates over a five year Period
 Both the numerator and the denominator must be clearly rather than a one-year period, and one can define the population at risk
as all those alive at the beginning of the period.
defined so that you know to which group (denominator) your
rate refers.
1.3 Age-Adjusted Rates
 The numerator is contained in (is a subset of) the denominator.
This is in contrast to a ratio where the numerator refers to a
When comparing death rates between two populations, the age
different group from the denominator.
composition of the populations must be taken into account. Since older
Mortality rates pertain to the number of deaths occurring in a particular people have a higher number of deaths per 1,000 people, if a population
population subgroup and often provide one of the first indications of a is heavily weighted by older people, the crude mortality rate would be
health problem. The following definitions are necessary before we higher than in a younger population and a comparison between the two
continue our discussion: groups might just reflect the age discrepancy rather than an intrinsic
difference in mortality experience. One way to deal with this problem
The Crude Anual Mortality Rate (or Death Rate) is to compare age-specific death rates, death rates specific to a
The total Number of Death During a Year in the Population at Risk particular age group. Another way that is useful when an overall
 summary figure is required is to use age-adjusted rates. These are rates
The Population at Risk
adjusted to what they would be if the two populations being compared
(Usually Taken as the Population at Mid Year)
had the same age distributions as some arbitrarily selected standard
The Cause - Specific Anual Mortality Rate population.
Number of deaths occurring due to a particular cause
For example, the table below shows the crude and age-adjusted
during the year in the population at risk
 mortality rates for the United States at five time periods.15,7 The
population at risk (all those alive at midyear)
adjustment is made to the age distribution of the population in 1940 as
well as the age distribution of the population in 2000. Thus, we see that
The Age - Specific Annual Mortality Rate
in 1991 the age-adjusted rate was 5.1/1000 when adjusted to 1940
number of deaths occurring in the given age group standard, but the crude mortality rate was 8.6/1000. This means that if
during the year in the population at risk in 1991the age distribution of the population were the same as it was

population at risk(those in that age group alive at midyear) in 1940,then the death rate would have been only 5.1/1000 people. The
crude and age-adjusted rates for 1940 are the same because the 1940
A reason for taking the population at midyear as the denominator is population serves as the “standard” population whose age distribution
that a population may grow or shrink during the year in question and
is used as the basis for adjustment.

 Page 2
 Biostatistics and Epidemiology

When adjusted to the year 2000 standard, the age-adjusted rate was 9.3. Number of persons with a disease
Prevalence rate of a disease 
If in 1991the age distribution were the same as in 2000, then the death Total number of persons in population
rate would have been 9.3/1000 people. So age-adjusted rates depend at risk at a particular point in time
on the standard population being used for the adjustment.

This is also known as point prevalence, but more generally referred to

just as “prevalence.”

For example, the prevalence of hypertension in 1973 among 15,190

black males, ages 30–69, defined as a diastolic blood pressure (DBP)
of 95 mm Hg or more (4,268) at a blood pressure screening program
conducted by the Hypertension Detection and Follow-Up
Program(HDFP), was calculated to be:
Although both crude and age-adjusted rates have decreased from 1940,
4 2,68 with DBP  95 mmHg
the decrease in the age-adjusted rate is much greater. The percent
15,190 black men aged 30 - 69 screened
change in crude mortality between 1940 and 1991 was (10.8 –8.6)/10.8
= 20.4%, whereas the percent change in the age-adjusted rate was (10.8
Several points are to be noted about this definition:
– 5.1)/10.8 = .528 or 52.8%.
1) The risk group (denominator)
The age-adjusted rates are fictitious numbers they do not tell you how
2) The point in time
many people actually died per 1,000, but how many would have died
3) The definition of the disease
if the age compositions were the same in the two populations.
4) The (numerator) subset of individuals in the reference group
However, they are appropriate for comparison purposes.
who satisfy the definition of the disease.

1.4 Incidence and Prevalence Rates Number of new cases of a disease per unit of time
The incidence rate 
Total number at risk in beginning of this time period
The prevalence rate and the incidence rate are two measures of
morbidity (illness). For example, studies have found that the ten-year incidence of a major
coronary event (such as heart attack) among white men, ages 30–59,
with diastolic blood pressure 105 mm Hg or above at the time they
were first seen, was found to be 183 per 1,000. This means that among

 Page 3
  Biostatistics and Epidemiology

1,000 white men, ages 30–59, who had diastolic blood pressure above observe them over a period of time, and determine how many develop
105 mm Hg at the beginning of the ten-year period of observation, 183 the disease over that time period. The implementation of such a process
of them had a major coronary event (heart attack or sudden death) is difficult and costly.
during the next ten years. Among white men with diastolic blood
pressure of <75 mm Hg, the ten-year incidence of a coronary event was 1.4 Measures of Relative Risk: Inferences from Prospective Studies
found to be 55/1,000.Thus comparison of these two incidence rates, In epidemiologic studies we are often interested in knowing how much
183/1,000for those with high blood pressure versus 55/1,000 for those more likely an individual is to develop a disease if he or she is exposed
with low blood pressure, may lead to the inference that elevated blood to a particular factor than the individual who is not so exposed. A
pressure is a risk factor for coronary disease. simple measure of such likelihood is called relative risk (RR). It is the
Often one may hear the word “incidence” used when what is really ratio of two incidence rates: the rate of development of the disease for
meant is prevalence. You should beware of such incorrect usage. For people with the exposure factor, divided by the rate of development of
example, you might hear or even read in a medical journal that the the disease for people without the exposure factor.
incidence of diabetes in 1973 was 42.6 per 1,000 individuals, ages 45– Example: Calculation of Relative Risk from Prospective Studies
64, when what is really meant is that the prevalence was 42.6/1,000.
The thing to remember is that prevalence refers to the existence of a Relative risk can be determined by constructing a 2 ×2 table as follows:
disease at a specific period in time, whereas incidence refers to new
cases of a disease developing within a specified period of time. RISK DISEASE (developing in the Total
FACTOR(determined specified period)
Note that mortality rate is an incidence rate, whereas morbidity may be at beginning of study Yes No
expressed as an incidence or prevalence rate. In a chronic disease the period ):High Blood
Pressure
prevalence rate is greater than the incidence rate because prevalence
Present a=90 b=403 a+b= 493
includes both new cases and existing cases that may have first occurred Absent c=70 d=1201 c+d= 1271
a long time ago, but the afflicted patients continued to live with the
condition. For a disease that is either rapidly fatal or quickly cured, incidence of disease among those with high BP
Relative Risk 
incidence and prevalence may be similar. Prevalence can be incidence of disease among those with normal BP
established by doing a survey or a screening of a target population and a
a b  90
counting the cases of disease existing at the time of the survey. This is   493
 3.31
c 70
a cross-sectional study. Incidence figures are harder to obtain than c  d  1271
prevalence figures since to ascertain incidence one must identify a
group of people free of the disease in question (known as a cohort),

 Page 4
 Biostatistics and Epidemiology

 The incidence of the disease among individuals with high obtained at time of sampling, in contrast to prospective studies where
blood pressure is 3.31 times that of with normal blood pressure. the independent variables are measured at time of sampling and the
dependent variable is measured at some future time (i.e., when the
Relative risk, or hazard ratio, can be calculated from Cox proportional disease develops).The real distinction between case-control or
hazards regression models (which allow for adjustment for other retrospective studies and prospective studies has to do with selecting
variables). individuals for the study—those with and without the disease in case
control/retrospective studies, and those with and without the factor of
1.5 Odds Ratio: Estimate of Relative Risk from Case-Control Studies interest in prospective studies.

A case-control study is one in which the investigator seeks to establish Since in prospective studies we sample the people with the
an association between the presence of a characteristic (a risk factor) characteristic of interest and the people without the characteristic, we
and the occurrence of a disease by starting out with a sample of can obtain the relative risk directly by calculating the incidence rates
diseased persons and a control group of non-diseased persons and by of disease in these two groups. In case-control studies, however, we
noting the presence or absence of the characteristic in each of these two sample patients with and without the disease, and note the presence or
groups. It can be illustrated by constructing a 2 ×2 table as follows: absence of the characteristic of interest in these two groups; we do not
have a direct measure of incidence of disease. Nevertheless, making
RISK DISEASE
FACTOR certain assumptions, we can make some approximations to what the
Present Absent
relative risk would be if we could measure incidence rates through a
Present a b prospective study. These approximations hold best for diseases of low
Absent c d
incidence.
Total a+c b+d
(number of persons (number of persons
To estimate relative risk from case-control studies note that
With disease) Without disease)
a a  b  a c  d 
The objective is to determine if the proportion of persons with the 
disease who have the factor is greater than the proportion of persons c c  d  c a  b 
without the disease who have the factor. In other words, it is desired to
know whether a/(a+ c) is greater than b/(b+ d). Now assume that since the disease is not all that common, c is
negligible in relation to d (in other words among people without the
Case-control studies are often referred to as retrospective studies risk factor there aren't all that many who will get the disease, relative
because the investigator must gather data on the independent variables to the number of people who will not get it). Assume also that, in the
retrospectively. The dependent variable—the presence of disease—is population, a is negligible relative to b, since even among people with

 Page 5
  Biostatistics and Epidemiology

the risk factor not all that many will get the disease in comparison to 475 * 61
OR   9.60  estimate of relative risk
the number who won't get it. Then the terms in the parentheses become 431* 7
d in the numerator and b in the denominator so that
This means that smokers of 15–24 cigarettes daily are9.6 times more
a c  d  ad likely to get lung cancer than are nonsmokers.
reduces to OR 
c a  b  bc
One more thing about the odds ratio: it is the ratio of odds of lung
This is known as the odds ratio (OR) and is a good estimate of relative cancer for those who smoke 15–24 cigarettes a day, relative to odds of
risk when the disease is rare. lung cancer for those who don't smoke. In the example above, we get

475
An example of how the odds ratio is calculated is shown below. In a for smokers : odds of lung cancer are
case-control study of lung cancer the table below was obtained. Note 431
that we are not sampling smokers and nonsmokers here. Rather we are
7
for nonsmokers : odds of lung cancer are
sampling those with and without the disease. So although in the 61
475 431
population at large a is small relative to b, in this sample it is not so. ratio of odds 
7 61

So the point is, the odds ratio is the odds ratio, whether the disease is
rare or not. It is always the ratio of odds of disease for those with the
exposure versus the odds of disease for those without the exposure. But
when the disease is rare, it is also a good estimate of the relative risk.
Odds ratios can be calculated from logistic regression (which allow for
adjustment for other variables).

1.6 Response Bias

The odds ratio, as an estimate of the relative risk of developing lung There are many different types of bias that might lead you to either
cancer for people who smoke 15–24 cigarettes a day, compared with underestimate or overestimate the size of a relative risk of odds ratio,
nonsmokers is and it is important to try to anticipate potential sources of bias and
avoid them. The following illustration shows the impact of one kind of
possible bias: ascertainment or response bias.

 Page 6
 Biostatistics and Epidemiology

Assume that you have the following situation. Of 100 people exposed
to a risk factor, 20% develop the disease and of a 100 people
unexposed, 16% develop the disease, yielding a relative risk of 1.25, as
shown in the illustration.

Now imagine that only 60% of the exposed respond to follow-up, or

are ascertained as having or not having the disease, a 60% response rate
among the exposed. Assume further that all of the ones who don't
respond happen to be among the ones who don't develop disease. The
relative risk would be calculated as 2.06.

Now imagine that only60% of the non-exposed reply, a 60% response

rate among the non-exposed, and all of the non-exposed who don't
respond happen to be among the ones who don't have the disease. Now
the relative risk estimate is 0.75.

To summarize, you can get conflicting estimates of the relative risk if

you have differential response rates. Therefore, it is very important to
get as complete a response or ascertainment as possible. The tables
showing these calculations follow.

1.7 Confounding Variables

A confounding variable is one that is closely associated with both the

independent variable and the outcome of interest in those unexposed.
For example, a confounding variable in studies of coffee and heart

 Page 7
  Biostatistics and Epidemiology

disease may be smoking. Since some coffee drinkers are also smokers, Whenever we enrolled a coffee drinker into the study, we would
if a study found a relationship between coffee drinking (the determine if that person was a smoker. If the patient was a smoker, the
independent variable) and development of heart disease (the dependent next patient who would be enrolled who was not a coffee drinker (i.e.,
a member of the comparison group), would also have to be a smoker.
variable), it could really mean that it is the smoking that causes heart
For each coffee-drinking nonsmoker, a non–coffee-drinking
disease, rather than the coffee. In this example, smoking is the nonsmoker would be enrolled. In this way we would have the same
confounding variable. number of smokers in the two groups. This is known as one -to-one
matching.
If both the confounding variable and the independent variable of
interest are closely associated with the dependent variable, then the
observed relationship between the independent and dependent
variables may really be a reflection of the true relationship between the
confounding variable and the dependent variable. An intuitive way to
look at this is to imagine that if a confounding variable were perfectly
associated with an independent variable, it could be substituted for it.
It is important to account or adjust for confounding variables in the
design and statistical analysis of studies in order to avoid wrong
inferences.

There are several approaches to dealing with potential confounders.

One approach is to deal with it in the study design by matching; another
way of controlling for confounding variables is in the data analysis
phase, by using multiple-variable analysis (eg., Multiple logistic
Regression , Cox-Proportional Hazards Model …).

1.8 Matching

One approach to dealing with potential confounders is to match

subjects in the two groups on the confounding variable. In the example
discussed above concerning studies of coffee and heart disease, we
might match subjects on their smoking history, since smoking may be
a confounder of the relationship between coffee and heart disease.

 Page 8
 Biostatistics and Epidemiology

 Cross -Sectional Study

Chapter Two: Design of Studies
In a cross -sectional study the measurements are taken at one
point in time. For example, in a cross-sectional study of high
2.1 Introduction blood pressure and coronary heart disease the investigators
determine the blood pressure and the presence of heart disease at
While the generation of hypotheses may come from anecdotal the same time. If they find an association, they would not be able
observations, the testing of those hypotheses must be done by to tell which came first. Does heart disease result in high blood
making controlled observations, free of systematic bias. pressure or does high blood pressure cause heart disease, or are
Statistical techniques, to be valid, must be applied to data both high blood pressure and heart disease the result of some
obtained from well-designed studies. Otherwise, solid other common cause?
knowledge is not advanced.
 Case-Control Study
2.2 Categories of Research Design/Studies
In a case-control study of smoking and lung cancer, for example,
There are two types of studies: the investigator starts with lung cancer cases and with controls,
and through examination of the records or through interviews
 Observational Studies
determines the presence or the absence of the factor in which he
 Experimental Studies
or she is interested (smoking). A case-control study is sometimes
2.2.1 Observational Studies referred to as a retrospective study because data on the factor of
interest are collected retrospectively, and thus may be subject to
These are studies where “Nature” determines who is exposed to various inaccuracies.
the factor of interest and who is not exposed. These studies
demonstrate association. Association may imply causation or it  Prospective (Or Cohort) Study
may not.
In a prospective (or cohort) study the investigator starts with a
Observational studies may be of three different study designs: cohort of non-diseased persons with that factor (i.e., those who
cross-sectional, case-control, or prospective. smoke) and persons without that factor (non-smokers), and goes
forward into some future time to determine the frequency of

 Page 9
  Biostatistics and Epidemiology

development of the disease in the two groups. A prospective Time of Measurement/Observation

study is also known as a longitudinal study. Past Present Future
Exposure
The distinction between case-control studies and prospective Cross-Sectional Outcome
studies lies in the sampling. In the case-control study we sample
from among the diseased and non-diseased, whereas in a Case-Control Exposure Outcome
prospective study we sample from among those with the factor
Prospective Exposure Outcome
and those without the factor. Prospective studies provide stronger
evidence of causality than retrospective studies but are often
more difficult, more costly, and sometimes impossible to 2.2.2 Experimental Studies
conduct, for example if the disease under study takes decades to
These are studies, where the investigator determines who is
develop or if it is very rare.
exposed. These may prove causation.
In summary, the objective in observational studies is to assess the In the health field, an experimental study to test an intervention
relationship between some factor of interest (the independent of some sort is called a clinical trial. In a clinical trial the
variable), which we will sometimes call exposure, and an investigator assigns patients or participants to one group or
outcome variable (the dependent variable). The observational another, usually randomly, while trying to keep all other factors
studies are distinguished by the point in time when measurements constant or controlled for, and compares the outcome of interest
are made on the dependent and independent variables, as in the two (or more) groups. More about clinical trials is in
illustrated below. Chapter 5.
In cross-sectional studies, both the dependent and independent Now, let us summarize study designs that we have seen so far in
(outcome and exposure) variables are measured at the same time, ascending order of strength in terms of demonstrating causality:
in the present. In case-control studies, the outcome is measured
now and exposure is estimated from the past. In prospective  Cross-sectional studies: useful in showing associations,
studies, exposure (the independent variable) is measured now in providing early clues to etiology.
and the outcome is measured in the future.  Case-control studies: useful for rare diseases or
conditions, or when the disease takes a very long time to

 Page 10
 Biostatistics and Epidemiology

become manifest (synonymous name: retrospective

studies). The hypothesis we test statistically is called the null hypothesis.
 Cohort studies: useful for providing stronger evidence of Let us take a conceptually simple example. Suppose we are
causality, and less subject to biases due to errors of recall testing the efficacy of a new drug on patients with myocardial
or measurement (synonymous names: prospective infarction (heart attack). We divide the patients into two
studies, longitudinal studies). groups—drug and no drug—according to good design
procedures (randomization), and use as our criterion measure
 Clinical trials: prospective, experimental studies that provide
the most rigorous evidence of causality. mortality in the two groups. It is our hope that the drug lowers
mortality, but to test the hypothesis statistically, we have to set it
All in all, whatever the degree of causality demonstration of a up in a sort of backward way. We say our hypothesis is that the
study is, its design relates with the hypothesis to be tested as well. drug makes no difference, and what we hope to do is to reject the
“no difference” hypothesis, based on evidence from our sample
2.3 General Concepts of Hypothesis Testing of patients. This is known as the null hypothesis. We specify our
test hypothesis as follows:
 How Do We Test A Hypothesis?
HO: (death rate in group A) – (death rate in group B) = 0.
First of all, we must set up the hypothesis in a quantitative
manner. Our criterion measure must be a number of some sort. We test this against an alternate hypothesis, known as HA, that
For example, how many patients died in a drug group compared the difference in death rates between the two groups does not
with how many of the patients died who did not receive the drug, equal 0.
or what is the mean blood pressure of patients on a certain anti-
hypertensive drug compared with the mean blood pressure of We then gather data and note the observed difference in mortality
patients not on this drug. Sometimes variables are difficult to between group A and group B. If this observed difference is
quantify. For instance, if you are evaluating the quality of care in sufficiently greater than zero, we reject the null hypothesis. If we
a clinic in one hospital compared with the clinic of another reject the null hypothesis of no difference, we accept the alternate
hospital, it may sometimes be difficult to find a quantitative hypothesis, which is that the drug does make a difference.
measure that is representative of quality of care, but nevertheless
it can be done and it must be done if one is to test the hypothesis.

 Page 11
  Biostatistics and Epidemiology

2.3.1 Error Minimization

 Why Do We Test the Null Hypothesis? The important point is that we can never be certain that we are
right in either accepting or rejecting a hypothesis. In fact, we run
Suppose we believe that drug A is better than drug B in the risk of making one of two kinds of errors: We can reject the
preventing death from a heart attack. Why don't we test that belief null or test hypothesis incorrectly, that is, we can conclude that
directly and see which drug is better, rather than testing the the drug does reduce mortality when in fact it has no effect. This
hypothesis that drug A is equal to drug B? The reason is that there is known as a type I error. Or we can fail to reject the null or test
is an infinite number of ways in which drug A can be better than hypothesis incorrectly, that is, we can conclude that the drug does
drug B, so we would have to test an infinite number of not have an effect when in fact it does reduce mortality. This is
hypotheses. If drug A causes 10% fewer deaths than drug B, it is known as a type II error. Each of these errors carries with it
better. So first we would have to see if drug A causes 10% fewer certain consequences. In some cases a type I error may be more
deaths. If it doesn't cause 10%fewer deaths, but if it causes 9% serious; in other cases a type II error may be more serious.
fewer deaths, it is also better. Then we would have to test whether
our observations are consistent with a 9% difference in mortality We cannot eliminate the risk of making one of these kinds of
between the two drugs. Then we would have to test whether there errors, but we can lower the probabilities that we will make these
is an 8% difference, and so on. errors. The probability of making a type I error is known as the
significance level of a statistical test. When you read in the
On the other hand, when we test the null hypothesis of no literature that a result was significant at the .05 level it means that
difference, we only have to test one value—a 0% difference and in this experiment the results are such that the probability of
we ask whether our observations are consistent with the making a type I error is less than or equal to .05. Mostly in
hypothesis that there is no difference in mortality between the experiments and surveys people are very concerned about having
two drugs. If the observations are consistent with a null a low probability of making a type I error and often ignore the
difference, then we cannot state that one drug is better than the type II error. This may be a mistake since in some cases a type II
other. If it is unlikely that they are consistent with a null error is a more serious one than a type I error. In designing a
difference, then we can reject that hypothesis and conclude there study, if you aim to lower the type I error you automatically raise
is a difference. the type II error probability. To lower the probabilities of both
the type I and type II error in a study, it is necessary to increase
the number of observations.

 Page 12
 Biostatistics and Epidemiology

2.4 Research Ethics and Statistics  Protection of Human Research Subjects

At first glance it may seem that statistics and research ethics have Human subjects in medical research contribute greatly to
nothing to do with each other. Not so! Consider why so many improving the health of people. These volunteers must be
people volunteer for medical research studies. In many cases it is protected from harm as much as possible.
because there is an expected benefit. For example in cancer
clinical trials often the investigational drug is a last hope and may There are Ethical Principles and Guidelines for the Protection of
not be available outside of the trial. In many cardiovascular Human Subjects of Research. The guidelines are based on three
disease studies, participants appreciate the additional care and basic principles: respect for persons, beneficence, and justice.
attention and are willing to try a new drug, for example, for
hypertension. And in fact, it has been shown that often, clinical  Respect for persons recognizes that people are
trial participants live longer and do better than the general autonomous beings and can make their own informed
population even if they are treated with a placebo. But what is choices about participating in research, free of coercion.
perhaps not sufficiently appreciated is that many, many people The informed consent process is predicated on this
participate in studies out of altruism to advance scientific principle. Participants who are not able to make their own
knowledge. Scientific knowledge is not advanced when a study choices, such as comatose patients, or mentally
is poorly designed, or carried out without sufficient rigor, or not incapacitated persons, or young children, must have
large enough to give an answer. Proper statistics are a special protections.
determinant of the ethics of a study.
 Beneficence, or the principle of non-malfeasance, means
So the point is that in order for a study to be “ethical” it must be that the risks of the research must be kept to a minimum,
designed and powered well enough so that it can answer the the benefits maximized, and the researcher is responsible
questions it poses. Otherwise, people who consent to participate for protecting the participant.
in the expectation that they will contribute to knowledge may
actually not be contributing because the study is poorly designed,  Justice in this context refers to a fair distribution of the
powered or executed, and may be needlessly exposed to risk. risks and benefits of research. One group of people should
not be exposed to research risks for the benefit of another
group of people.

 Page 13
  Biostatistics and Epidemiology

o Informed Consent participant’s native language. All this must be approved by the
medical institutions IRB (or Institutional Review Board), which
One of the most important elements in protection of human is generally a committee of experts and lay people who review
subjects is the principle of informed consent. The study subject and must approve all research protocols before the research is
must freely consent to be part of the study after being fully started, and who monitor adverse events as the research
informed of the potential risks and benefits. progresses. Different IRBs have different specific requirements
that are usually posted on their websites. Informed consent is an
There are certain elements that must be in a written consent form.
ongoing process. It is not just signing a form at the beginning of
The purpose of the research must be stated; a 24-hour contact
a study. The researcher has an obligation to keep the participant
person must be listed; there must be a description of the study
informed of relevant new research that may affect his or her
procedures: what is expected of the participant, the duration of
decision to continue participating.
the study, and how much of the participant’s time it will take.
The potential risks and discomforts, potential benefits, Note that, when is it ethical to begin a clinical trial of a new
inconvenience to the participants, all must be clearly stated. treatment? When there is equipoise. Equipoise means that there
There must be a statement that participation is voluntary and that is about equal evidence that the treatment may provide benefit as
the participant has the right to withdraw at any time and that this there is that it will not provide benefit.
will not prejudice the care of the participant. If the research may
result in need for further care or diagnostic procedures, the  Research Integrity
participant must be told to what extent he or she is responsible
for further care and what the study will pay for. If there is any For research conclusions to be valid, data collection procedures
compensation to the participants, either for expenses incurred in must be rigorously and uniformly administered. No data may be
participating or time spent, they must be informed of the amount. altered without documentation. If there is a clerical error, the
(The amount should not be excessive, as that may appear change and reason for it must be documented. Enrollment must
coercive.) A statement assuring confidentiality and how it will be be according to strict and pre-planned standards. Sometimes
maintained must be included. (fortunately, rarely) there is a great pressure to enroll subjects in
a given time frame, or the researcher (in violation of the principle
Most important, the participant must understand what he or she of equipoise) really believes the treatment can help his or her
is agreeing to and the consent form must be phrased in language patients, and so “bends” the enrollment rules. This may
that is understandable, and if appropriate, translated into the invalidate the research and so is unethical.

 Page 14
 Biostatistics and Epidemiology

 Authorship Policies In summary, the ethical conduct of research has many

components. New and difficult ethical questions arise as science
In medical research most original research articles have multiple advances and new technologies become available. This brief
authors, since medical research is a collaborative effort. Most chapter just begins to give you an idea of some of the issues
medical journals, and research institutions, have specific and involved.
strict authorship policies (published in journals and/or on
websites) many of which embody the following elements: (1) co-
authors must make an intellectual contribution to the paper (e.g.
conceive the research, perform analyses, write sections of the
paper, or make editorial contributions); (2) all co-authors must Reading Assignment
bear responsibility for its contents; (3) co-authors must disclose
potential conflicts of interest (e.g. relevant support from industry,  Studies of change over time
lectureships, stockownership). Order of authorship may  Choosing a study design
sometimes be a point of contention and should be discussed by
the coauthors early in the process.

 Data and Safety Monitoring Boards

Generally, clinical trials have a Data and Safety Monitoring
Board (DSMB) to oversee the trial. These are independent groups
of experts in the relevant disciplines who are in an advisory
capacity. Their job is to monitor the trial and to assure the safety
of participants. In a blinded trial they are the only ones who see
the unblinded data at regular, pre-specified intervals. If they find
excessive benefit or harm in one arm of the trial, they would
advise to stop the trial. Usually the criteria for stopping a trial due
to harm in the treatment group are more stringent than stopping
for benefit.

 Page 15