58

Chapter 4

Study Designs

The term study design is used to describe the combination of ways in which study groups are
formed, and the timing of measurements of the variables. Choosing a study design appropriate for the
research question is important. No amount of statistical adjustment can compensate for not having a
firm idea about the research question and how it is to be developed.
The Three Axes of Epidemiologic Research Design
Once the research objective for a study has been adequately specified in terms of the target
population, exposure (or interventions as the case may be), and outcomes, the next step is to consider
the design of the study.
The design of a study can be considered from three aspects related to the recording of exposure
and outcome, as follows:
i). the direction e.g. looking forward (prospective) from exposure to outcome, or backwards
(retrospective) from outcome to exposure, or mixed (cross-sectional) when exposure and
outcome get measured together.
ii). sample selection e.g. by exposure, by outcome, or by other criteria.
iii). Historical, concurrent or mixed, depending on the calendar timing of measurement of
exposure and outcome, and the actual time of conducting the study.
Direction
Direction refers to the order in which exposure and outcome are investigated. Forward from
exposure to outcome; backward from outcome to exposure; or simultaneously where exposure and
outcome are determined at the same time.
Sample selection
This refers to the criteria used to choose study subjects; it can be based on exposure or outcome
or other criteria.
Timing
Timing refers to the relation between the time of the study and the calendar time of exposure
and outcome. Thus, historical means that both exposure and outcome occurred before the study;
concurrent means that exposure and outcome are occurring at the same time as the study; or mixed
timing.

Oxford Unveristy Press

59

The fundamentals of a study

All studies require the following:
1. Firmly established study objectives or hypothesis.
2. Carefully defined methods of assembling study subjects, including case definitions and avoidance
of bias in selection.
3. Making valid and reliable observations, which includes avoiding biased surveillance, variability
among observers, and "blinding" of observers or subjects where necessary. (In the context of
research blinding means that the observer or subjects or both would be kept unaware of the
research question (or hypothesis) or treatment so as to avoid bias).
4. Handling of incomplete observations, including how to handle those subjects who are lost to
follow-up, are non-responders, or change status during the study e.g. from being "controls" to
"cases".
5. Selecting appropriate comparison groups, including identifying beforehand important factors that
may influence the study hypothesis, and controlling for them.
When planning a study it is always useful to consider how the following attributes may apply to
the research question:
Is the aim of the study to describe or to compare?
Will the study be observational or is an intervention planned?
The orientation of the study in time.
The orientation with regard to the process? For example, prospective or retrospective?
If determining a cause is the aim, then what strength of causation is to be looked for. Will it be
speculation or causation?

In most studies the researcher is attempting to demonstrate that a process begins with an agent
acting and an outcome or event happens, as shown in the diagram below:
Agent
Acts

Outcome
Observed

Time . . . . . .

If the observation is made at a single point in time, the agent and the outcome as well as other factors
are measured simultaneously. One can only hypothesize which came first. But by observing the process
at two or more points in time one is better able to determine the sequence of events.

Oxford Unveristy Press

60

The standard approach is to move along in time with the process. This is called concurrent research. On
the other hand the researcher may know that an exposure or intervention has occurred in the past, and
is observing the results now. In this case the process has been going on non-concurrently.
__________________________________________________________________________________
Descriptive

Document experience,
observations, unusual
events, programmes,
treatments

Observational

Seek causes, predictors,

risk factors. Researcher
observes phenomenon
without intervention.

Explanatory

Experimental

Examine aetiology,
efficacy or cause using
strategy of comparison

Evaluate efficacy of
therapeutic and other
interventions.

Begin search for

explanations.
Examples:
- Case Reports or Series
- Prevalence Studies
- Surveys

Examples:

Examples:

- Case-control studies
- Cohort studies

Clinical trials
Educational
interventions
Health care
interventions.

_________________________________________________________________________________

Figure 4.1: Basic Study Designs.

Prospective/Retrospective research relates to the way subjects are selected. In prospective

research groups of individuals are assembled who would have been exposed to a risk factor or
intervention. The researcher then waits for the outcome to happen. In the case of retrospective research,
a group of individuals has been assembled because they have already experienced the exposure or
intervention in question. Thus, prospective/retrospective research differ from each other in their
orientation in time as well as what observation is being made. In prospective studies the research
commences in the present and moves forward in time collecting data about a population whose
outcome lies in the future.

Oxford Unveristy Press

61

In retrospective research the study begins and ends in the present but involves a major backward glance
to collect information about past events. A great deal of confusion has arisen in the past by these terms
being applied loosely to describe study designs. It is more informative to use prospective and
retrospective to refer to the temporal relationship between the initiation of the study and the occurrence
of outcome. Thinking of prospective and retrospective as clock-watchers rather than forms of study
designs can avoid much of the unnecessary confusion.
Descriptive Studies
Descriptive studies are undertaken to note the characteristics of a group ofsubjects. There is no
testing of a causal hypothesis or any comparisons with other groups. But descriptive studies can help
support a hypothesis by demonstrating a phenomenon that the hypothesisstates.
Descriptive studies document facts of clinical or theoretical significance. Comparative research
draws contrasts between two or more groups with the intent that the comparison will test a particular
hypothesis. Both descriptive and comparative research have scientific value when used to address the
appropriate question. In certain situations one type of research is more informative than the other. For
example, if we are looking at the effectiveness of a new treatment there are pitfalls in concluding from
descriptions alone. Most diseases have an unpredictable outcome. Some resolve spontaneously. It
would be better to compare the outcome in a group of patients given the new treatment with that in a
similar group given the conventional treatment or placebo. This would be a comparative study in its
simplest form. As we would see later both descriptive and comparative studies have their uses. It is not
unusual for researchers to conduct a descriptive study first, and follow it up with a comparative study
later.
All studies involve some description. Even those which are undertaken primarily to compare
two groups of subjects contain an element of description of multiple groups or of one group at multiple
points in time. The special use of descriptive studies is in providing much of the information on which
future more complex study designs would rely. Researchers generally follow a sequential development
of knowledge about a topic under study by sequential use of different research designs. If very little is
known about a subject a general sequence of research designs proceeding from descriptive to
correlational to analytical design will be efficient in building up reliable information. The concept of
progression within research designs is a general one, and there will be significant variation from one
topic to another depending on the nature of the subject and what is known about it.
Descriptive studies have three goals:
(i). What characteristics are present in members of the group.
(ii). How are these characteristics distributed in the members of the group.
(iii). How may these characteristics be expressed in a summary form that aids comprehension
and allows for later comparisons.
The simplest form of descriptive studies are Case Reports and Case Series.

Oxford Unveristy Press

62

The latter describe a small group of patients thought to share one or more important attributes e.g. sideeffects of drugs.

Cross-sectional studies form a bridge between simple descriptive studies like case reports and
case series and those that can be used to test hypotheses. They can suggest the presence of
relationships. Their major flaw lies in the lack of measurement over time. On the other hand they have
the advantage of being economical (see figure 4.2a).
Cross-sectional studies are also called prevalence studies. They are used to document
comparisons among various attributes of a group. But at a single point in time. They take a slice across
a process flowing from a point of onset towards an outcome. In such a slice it is not possible to say that
change in variable A occurred because of change in variable B (or vice versa), even though the subjects
may have the two variables in different levels of magnitude. However, a cross sectional study can say
that if it was true that factor A came before B then a cause-effect relationship might be present. Further
designs can then prove or refute this point. For example, a study of first time mothers in a squatter
settlement in Brazil revealed that one in ten mothers were less than 15 years old.
(J.Trop.Ped.1990;36:14-19). The authors speculated that occurrence of pregnancy was responsible for
leaving school and explained the low literacy rates among the young mothers. A second study in the
same settlement was undertaken to compare the characteristics of teenage mothers with those of the
older ones. (J.Trop.Ped.1991;37:194-198). This study revealed that a majority had already dropped out
of school before they became pregnant.
Children's growth charts are often made up of cross-sectional observations. The popularity of
cross-sectional studies besides economy is due to two factors viz .

Cross - Sectional Study

CASES
A cross - sectional study is a survey of the
frequency of disease, risk factors, or other
characteristics in a defined population at
one particular time.
NON - CASES

Oxford Unveristy Press

63

Figure 4.2a: Cross-Sectional Study.

i). One avoids the difficulty of tracking the subjects over long periods of time.
ii). A great deal of data exists in hospital files, which can be analyzed in a cross-sectional manner. (e.g.
diabetics attending a clinic. Patients with different stages of the disease may be found some with renal,
some with ocular and others with vascular complications). But a word of caution is needed. When
archival data has been used it is necessary to prove their validity and accuracy, especially when data
has come from sources set up initially not for research purposes. Most hospitals and clinical
departments have computerized systems for storing clinical information about the patients. All such
systems are primarily set up as health information systems. They are useful for audit and for analyzing
patient load, types of diagnoses made, laboratory and other investigations, and response to treatment.
But they cannot be expected to answer a research question which was not even remotely considered
when the systems were first set up.
Major weaknesses, which limit cross-sectional studies to the preliminary and early stages of
exploring hypotheses, are:

If subjects are of different ages or in different stages of advancement of a disease false conclusions
may be derived. The apparently homogeneous group may have been exposed to different risks or
subjected to different treatments.
It is not possible to demonstrate a temporal sequence in terms of causality. The researcher must not
only prove that the study results fit the hypothesis regarding causality but also that they do not fit
any competing hypotheses. And the correlations between the cause and effect variables also apply
to other variables that seem logically related.

In short, researchers performing cross-sectional studies must be careful to show that they have
not simply sifted the data until they found a relationship, which fitted their need.
Ecologic studies
When data are gathered which describe what happens in a group rather than the individual, for
example, exposure to radiation or fluoridation of water supply, the studies are called ecological.
Ecologic variables of this nature are used in a variety of study designs and not just in cross-sectional
studies. More frequently they are found in time series analyses e.g. time since establishment of
intensive care units and perinatal mortality trends.
Ecologic studies take populations rather than individuals as the unit of measurement. At their
simplest they explore differences in disease rates by geographical regions, or by differences in time in
the same region to identify disease patterns that suggest a hypothesis about the cause of disease. (A
good example is cigarette smoking and lung cancer rates). The weakness of ecological approach is that
exposure and prevalence are measured at the population level. So there is assumption that all members
of a given population share the same exposure and same risk of disease. This is a major assumption
since there may well be large within-group differences. Confounding variables like social status, habit
and life-style are not taken into account. Hence caution must be exercised in drawing conclusions for

Oxford Unveristy Press

64

the individual based on the results of an ecological study. The term ecological fallacy is applied to
such conclusions, since the true relation between exposure and disease is distorted by the inability to
control for other confounding variables.

Comparative studies
Case-control studies are the next level up from cross-sectional studies in attempts to show
causal relationships (see figure 4.2b). In a case-control study the observer starts with a group who have
already experienced an outcome and one or more groups who have not. The investigator determines the
proportion in each group who have experienced an exposure of interest. The hypothesis of causation is
supported if the proportion of subjects exposed is greater in the "cases" than in the "controls".
A variant of the case-control study uses pairs i.e. one "case" with one or more "controls". The
controls are matched to the case on one or more attributes. In such a design the outcome is determined
by the proportion of pairs for which the exposure status of the case and control are different.

Case - Control Study

Characteristic

Characteristic

Cases

Controls

A case - control study is an

observational study in
which characteristics of a
group with disease (cases)
are compared with those of
a selected sample without
the disease (controls).

Figure 4.2b: Case-Control Study.

Cohort Studies
Prospective studies could be either concurrent (i.e. moving with the researcher on the axis of
time) or non-concurrent (constructed from exposure events in the past and the outcome measured in
the present). Prospective studies may also be classified either as observational cohort studies or
experimental clinical trials.

Oxford Unveristy Press

65

Cohort Study

Study Group

Events

Controls

Events

A cohort study is an
observational study
of a group of people
with specified
charactersitics or
disease who are
followed over a
period of time to
detect events.

Figure 4.2c: Cohort Study.

Cohort studies can be carried out in four general ways:

(i). Single heterogeneous group followed from baseline to outcome.
The cohorts are homogeneous in one or more characteristics (e.g. birth date, residence, or
socioeconomic status), but heterogeneous for other characteristics that are related to the outcome of
interest (e.g. occupation or exposure to a risk factor). At the time of analysis, the researcher attempts to
find a relationship among the heterogeneous characteristics and the outcome. This type of study is
comparative since it compares outcome among subjects, and observational in that it does not actively
allocate groups to exposure. An example of a cohort study is a group of infants followed since birth to
age one year, some exclusively breastfed and others bottle fed. The number of episodes of diarrhoea
and their severity are compared between the two groups.
(ii). Two homogeneous groups followed from baseline to outcome.
The researcher assembles a cohort of subjects who all have a condition or characteristic of
interest. e.g. failure to thrive in infancy. A control cohort is then assembled with normal growth in
infancy, but who are otherwise similar to the other cohort in all possible ways. The two cohorts are
followed to a specified outcome e.g. school entry. At this time comparison is made e.g. by
developmental tests.
iii). Purely descriptive cohort study.
Only one group is followed and attributes of interest are documented as they change over time.
e.g. a longitudinal study of blood pressure or linear growth. If the study is purely descriptive no attempt

Oxford Unveristy Press

66

would be made to identify patient characteristics responsible for the pattern of change over time, or to
draw contrasts between subgroups who have different patterns of change.
iv). Epidemiologic cohorts.
The three types of cohort studies described all follow individual cases, and assess their status at
one or more points in time. In epidemiologic cohorts aggregate characteristics of groups are measured
at one or more points in time. For example, measuring fertility rates of women born in 1950, as
compared to those born in 1970. No individual is singled out for study.
Cohort and Case-control studies are the two basic types of observational analytic studies. In
theory it is possible to test a hypothesis using either design. However, each design offers certain unique
advantages and disadvantages. The decision to use a particular design is based on features of exposure
and disease, current state of knowledge, and the logistics of time and resource.

Observation vs. experimental

An intervention can occur in two ways:

A naturally occurring situation where intervention has been differentially applied to the members of
a population. For example, two neighbouring towns with one receiving its water supply from a
source of surface water, and the other from deep tube wells. Studies in such situation are called
observational. They simply document processes without the researcher's intervention.
Purposefully manipulating events. Interventions are deliberately applied, and other circumstances
altered in a way as to clarify the impact of intervention (see figure 4.2d).

In between the two are the quasi-experimental studies. These studies may be described as
nature's experiments. For example, one community in which a particular intervention has been applied,
and is being compared with another community where the intervention programme has not yet reached.
A good example is that of a nationwide programme of immunization.

Experimental Studies
Clinical trials are the common forms of experimental studies undertaken in clinical practice.
They have five key elements:
1. They are concurrent, prospective comparisons of two or more groups.
2. One or more of the groups is deliberately exposed to an intervention, usually a treatment,
while at least one group (the controls) is not exposed or receives a standard treatment or
placebo.

Oxford Unveristy Press

67

3. The groups are recruited from one single, homogeneous pool of subjects. Assignment of
individuals to study or control group is determined by random allocation, and without any form
of consideration as to who gets assigned to which group.
4. All study participants (clinicians and all care providers, subjects, outcome evaluators) are
unaware of which subject belongs to which group. The "blinding" may be extended to any
knowledge about the study hypothesis or outcome measures.
5. Control subjects receive an intervention, which is indistinguishable in appearance, taste and
other ways from the standard intervention. This includes attention to psychological factors like
the placebo effect or Hawthorne effect, so that subjects do not change behavior because they
know they are being treated or know they are being observed. Experimental and control
subjects experience same amount of intervention and observation.

Clinical Trial
Intervention

Outcome

Cases

No
Intervention

A clinical trial is an intervention study in which an intervention is applied to one

group of people and the outcome is compared with a similar group without the
intervention.

Figure 4.2d: Clinical trial.

Oxford Unveristy Press

Outcome

68

What differentiates cohort studies from clinical trials is that in the case of the former the intervention
has not been applie by the researcher. Intervention studies are in effect a form of prospective cohort.
Participants are identified on the basis of their exposure to intervention and followed up to determine
the outcome. The distinguishing feature is the researcher determines that exposure. Two major tasks
facing the researcher in conducting clinical trials are:
i). to make sure that no biases were operating at the time the subjects were being exposed to the
agent. Randomization helps to control for all recognized factors that may influence the outcome as well
as the more subtle and unrecognized ones.
ii). exposure did indeed take place. In the case of treatment with adrug it means full
compliance.
Clinical trials are suited best for trying out new treatments for known diseases. This is so
because the subjects have in common some easily defined characteristics that would be expected to
change rapidly and visibly in response to treatment. It is easy for the researcher to find a uniform pool
of subjects, and to conduct the study over a manageable period of time.
Clinical trials are less suited to studies of preventive measures, because the duration of time for
observation is prolonged and it is not possible to prevent the subjects from being exposed to other
treatments or exposures. Nevertheless, they are a powerful research strategy. When well designed and
conducted they provide the most direct evidence for judging whether an exposure ameliorates and
cures a disease.

Quasi-experimental studies.
In the social and behavioral sciences it is often not possible to follow the ground rules of classic
experimental studies. For example, two or more clinics may need to be compared or two or more
geographical communities are to be compared such that one is receiving an intervention and the other
is not. Or there may be only one group in which an intervention is being made, in which case a time
series design may be feasible.

Which study design to use?

The foregoing account of descriptive and analytic studies may have given the reader an
impression of clear-cut distinction between the two types of studies. In reality, this is not so. There is
invariably an overlap. Most descriptive studies tend to analyze and compare. Similarly, most analytical
studies provide descriptions of the attributes of the subjects. In general, descriptive studies are useful
for describing patterns of disease occurrence, and for allowing the formulation of hypotheses. The chief
contribution of analytical studies is in testing hypotheses. A given research question may be addressed

Oxford Unveristy Press

69

using different approaches. Each has its specific advantages and disadvantages.(See Tables 4.1 and 4.2)
The choice of study design is largely influenced by specific features of exposure and disease, logistics
of time and resources, results from previous studies and gaps in the knowledge to be filled. For most
situations both strategies need to be employed.
Table 4.1. Advantages and Disadvantages of common study designs.
__________________________________________________________________________________
Study design
Advantages
Disadvantages
Case reports

Cheap and easy for

generating hypotheses

Cannot be used for

testing hypotheses.

Case series

Provide descriptive
data on disease
characteristics

No control group. Cannot

be used for hypotheses
testing.

Cross-sectional

Can assess prevalence

easy; can generate
hypotheses.

Cannot evaluate timing

of exposure

Ecological

Rapid answers. Can

generate hypotheses

Difficult to control
for confounding.

Case-control

Can study multiple

exposures and
uncommon diseases
Requires few subjects
logistics are easy,
less expensive.

Selection of controls
difficult; possibly biased
exposure data; incidence
cannot be measured.

Cohort

Can study multiple

outcomes and uncommon
exposures, selection
bias and biased
exposure data less
likely, incidence can
be assessed

Possibly biased outcome

data, expensive, long
duration when done
prospectively, not suited to
rare diseases, can study few
exposures, loss of subjects
by drop-out

Randomized
Clinical
Trial.

Most convincing
design, controls
for known and
unknown confounders

Expensive, logistics
difficult, ethical
problems.

Oxford Unveristy Press

70

Table 4.2. Choice between observational and experimental studies in medical research
__________________________________________________________________________________
Observational studies
Experimental studies
ADVANTAGES
1. Practicality,
greater variety and larger scale
of tests of hypothesis, as
compared with experiments

1. Safety
relatively few
individuals exposed to
unpredicted risks in
untried interventions

2. Few ethical objections since

interventions are not imposed
by pre-planned designs

2. Vigour
capable of critical tests
of limits of applicability.

3. Relatively large number of

participants can be recruited

3. Precision
theorized causal factor
can be defined and limited
exposure of test factor
under control of
researcher.

4. Duration is relatively short

when observations are from
existing records
5. relatively inexpensive in time
and personnel

4 Efficiency
relatively few observa
tions needed to refute
some hypotheses.
5. Assumptions
random allotment of
treatments is assured
by the design.

DISADVANTAGES
1. Documentation
exposure to causal factor
less certain

1. Impracticality
inborn attributes cannot
be manipulated

2. Specificity
isolating causal factor
is difficult

2. Predicted risk of
intervention is great.
long term observation
is difficult

Oxford Unveristy Press

71

3. Remoteness
causal factor and outcome tend
to be separated in time.

3. Large number of subjects

need to detect small
differences

4. Time order
direction of relationship
between cause and effect
less certain

4. Reductionism.
focus on one independent
variable excludes others
from attention.

5. Systematic selection
availability of patients for
treatments
allotment of treatments
6. Efficiency of statistical tests
lower than in experimental
designs.

5. Representativeness
difficult to recruit a
truly random sample from
'universe'
6. Expense
relatively great in staff
and logistics

7. Public acceptance
undue suspicion of
outcomes
__________________________________________________________________________________

The same research question may be answered by several different ways depending upon the
available resources. For example, let us consider the following:

DOES SMOKING CAUSE CORONARY ARTERY DISEASE ?

The question may be answered by any of the seven basic designs (A to G) shown in Table 4.3
below in accordance with directionality and sample selection:

Oxford Unveristy Press

72

Table 4.3 Research design defined by directionality and sample selection

__________________________________________________________________________________
Directionality

Cohort

Case-control

Cross-sectional

__________________________________________________________________________________
Sample
Selection by
Exposure

Outcome

Other

__________________________________________________________________________________

A:. Cohort study with sample selection by exposure, e.g. smokers and non-smokers.
B:. Cohort study with "other" sample selection e.g. instead of sampling exposed (i.e. smokers)
and unexposed (i.e. non-smokers) we may select all subjects in a target population, determine
their smoking habits and then follow them up for a given period of time.
C:. Case-control study with sample selection by outcome. Presence of coronary artery disease is
defined on the basis of symptoms, abnormality of E.C.G. at rest or after exercise, and so on.
Those diagnosed as suffering from coronary artery disease are compared with healthy controls
with regard to their smoking habits.
D:. Case control with "other" sample selection. This is similar to C but instead of choosing
those with coronary artery disease and comparing them with those without, all the subjects in
the target population are recruited into the study. This design turns out to be less robust then C,
since very few subjects would be expected to have coronary disease recognizable by noninvasive methods at a given point in time.
E: Cross-sectional study with sample selection by exposure. People who smoke and who do not
smoke are compared for the presence or absence of coronary disease. This design is as inefficient as D.

Oxford Unveristy Press

73

F:. Cross-sectional study with sample selection by outcome. The records of people attending a
hospital for chest pain with and without coronary disease are compared for exposure to
smoking. This design is more efficient than E but there may be problems with determining
causality.
G:. Cross-sectional study with "other" sample selection. All the subjects in a target population
are classified according to smoking habits and coronary artery disease status. The design is
inefficient like D and E, and shares the same problems with regard to causality as E and F.
Now that we have seen the strengths and weaknesses inherent in different study designs what
strategy should one follow? In general, research proceeds from case reports and case series, which
suggest an association to case-control studies for exploring the association further. Cohorts may be
studied to further establish the association. Finally, clinical trials may be used for therapeutic or
preventive measures. The ultimate aim of research is to explore the relationship between two or more
variables with a view to exploring why events happen. In general the more one can control the
circumstances of the study (i.e. influence of other factors; the way in which study intervention is
applied, the measurement of the outcome, and so on) the stronger is the ground for demonstrating a
cause-effect relationship. These remarks are best exemplified by the way research has proceeded with
regard to HIV infection. The first published reports were in the form of case series. As the epidemic
unfolded there were cross-sectional studies and other descriptive analyses of the prevalence in different
groups. Next, case-control and cohort studies were designed to identify risk factors and outcome. In the
meantime a great deal of knowledge was being assembled from virological studies and animal
experiments whilst simultaneously therapeutic measures were evaluated by means of clinical trials.
Hopefully in the near future there would be trials to test and evaluate vaccines.
This has been an introductory chapter wherein the broad characteristics of different varieties of
study designs are described. Some of their strengths and weaknesses have been mentioned. In the
chapters that follow each design is explored more fully.

Oxford Unveristy Press