Cohort study

Negussie Deyessa, MD, MPH

March 2008
 Learning objectives
1. Describe the characteristics of a cohort study.

2. List the types of bias most likely to affect a cohort study.

3. List the conditions under which a cohort study is an

appropriate choice to address a research question.

4. Describe the advantages and disadvantages of a cohort

study design.

5. Calculate appropriate measures of disease occurrence and

exposure-disease association for a cohort study.

6. Calculate and interpret well on measures of association and

impact of exposure on disease status.
Cohort study
 It is also called follow up, longitudinal, prospective
study.

 The word cohort is used to designate a group of

people who share a common experience.
 a Birth cohort,
 a Cohort of smokers,
 a Cohort of MPH graduates in 2007, etc.

 It is an observational study that measures incidence of

disease occurrence.
Cohort studies
 An investigator studies a group of exposed and
unexposed subjects and follows the study subjects
over a period of time and compares the incidence
of developing disease of interest in the 2 groups
PRESENT TIME FUTURE TIME
Diseased
Exposed
No disease

Diseased
Unexposed
No disease
1. Basic elements
 “Disease” free at entry

 Selected by exposure status rather than outcome

 Follow up is needed to determine the incidence

of the outcome in each exposure group
 For non communicable chronic diseases this may take years

 Compares incidence rates among exposed

against unexposed groups
a. Exposure status
 Study subjects should be disease free

 Define study subjects using inclusion and exclusion

criteria on the basis of exposure status
 Environmental factors: smoking, air pollution, pesticides

 Criteria can be specified by amount of exposure

 Eg. Cigarette Smocking (# of cigarette per day)
b. Possible outcomes in cohort
 No disease

 Diseased

 Lost to follow up
 Cohort studies
Incident
case

Exp+

Non-case
Select
Ascertain
exposed and caseness
unexposed
Incident
case
Exp-
Non-case
 2. Features of cohort study

1. It shows temporal sequence

Exposure Disease

2. Good to assess effect of rare exposures

3. Could assess multiple effects of a single exposure

3. Types of cohort study

 Two forms of cohort study

 The major difference is on the basis of

 Initiation of study and the occurrence of
disease.

 The two forms are similar, because selection of

study subjects is made on the basis of their
exposure status.
 Cont…

1. Classical (Prospective) cohort

 The exposure may or may not have occurred at
the time when the study begin, but the outcome
has certainly not yet occurred.

2. Historical (Retrospective) Cohort

 Both the exposure and outcome have already
occurred when the study is initiated.
 Cont….
1. Prospective cohort
 Exposure status determined at
Past Present Future
present Disease
 Study groups followed up and Exposure outcome
disease outcome will be
ascertained in the future

2. Retrospective cohort
 Exposure determined in the past
Disease
from records Exposure outcome
 Disease outcome ascertained at
present
 Cohort study design
Classical (prospective)

1. measure exposure 2. measure outcome

Historical (retrospective)

1. Record of exposure 2. measure outcome

Timing of case-control, prospective and retrospective
cohort study in relation to exposure and outcome.
1. Case control
Exposure Disease
?
?

2. Prospective Cohort
Exposure Disease
?
?

3. Retrospective Cohort
Exposure Disease
?
?

? To be determined Exposure, (+) or (-)

 Cont….
 Some cohorts become ambi-directional, both
retrospective and prospective forms.

 Such type of design is used when exposed people

have both short-term and long-term effects.
Eg Radiation, risks a birth defect within short time and
cancer later.

 If large sample size and follow up is complete,

data from prospective study is reliable and
informative when compared to retrospective.
Steps in a prospective cohort study
1. Define the population at risk (=cohort)

2. Determine exposure status to a factor of interest of all

subjects in the cohort

3. Make sure that study subjects are free of the disease of

interest at time of enrolment

4. Follow exposed and non-exposed forward in time to

ascertain whether they develop the outcome of interest

5. Compare the outcomes in the exposed and the

unexposed group with each other
 Retrospective cohort studies

1. Well defined population healthy for disease of interest

2. Selection of study subjects on the bases of their

exposure status from a record (exposure status in the
past)

3. Ascertain outcome (ill or not ill) at present

4. Compare cumulative incidence among exposed to non-

exposed (relative risk)
Other classification
Open and closed cohort design

1. Closed (fixed) cohort

 Members of the cohort will be followed without adding
others.
 E.g 1000 children followed for 2 or 3 years

2. Open (Dynamic) cohort study

 Other than initially chosen study group, others could
be added or be lost.
 People at risk are measured using person-time
(incidence density)
 Data may be analyzed using ‘Time tables’ and
Kaplan Meir/ Cox regression.
 Open (Dynamic) cohort study

Birth In-migrants

Death Out-migrants

E. g. Butajira Rural Health Program

 4. Issues in the design of cohort studies
1. Selection of exposed population
 It depends on a variety of scientific and other feasibility
considerations, including:

a) The frequency of exposure

(finding sufficient exposed individuals).

 Common exposures (Cigarette, coffee drinking etc)

 Sufficient exposed people could be found.

 Rare exposures (Particular occupational/environmental

factors)
 Choose specific groups
 Cont…

b. Completeness of follow up
 Recruiting professionals of certain profession
(Stable occupation).
Eg. Doctors, Nurses, Veterans, Union members etc.
In Ethiopia, who are people with stable
occupation?

c. Nature of research questions being evaluated.

 Presence of records within institution
 Retrospective cohort.
 Cont…
2. Selection of comparison (non-exposed) group

 Difficult as in the selection of controls in case-control

studies.

 Groups should be as similar as possible with respect to

all other factors except exposure status.

 General cohort, and an internal comparison.

Eg. Cigarette smocking, dose variation.

 Cont…
 Information obtained from non-exposed should be
adequately comparable with exposed.

 In use of rare occupational exposure, we can use

external comparison groups (general population.).

 Comparison from within the institution but not

having exposure eg office workers Vs working in
plants.

 Comparing with other factories but having no

contact with the exposure.
 Cont…
 Comparison with general rates (such as mortalities,
cancer incidents).

 Comparison with general population, (has limitation

‘Healthy worker effect’) (Workers are relatively
healthier than the general population).

 Compare exposure with other cohort, having similar

demography but not exposed.

 Use of multiple comparison groups may be other

option.
5. Source of data
 In general we need accurate and complete information on
both exposure and outcome of interest.

1. Exposure ascertainment
a. Pre-existing data:- records from institutions
Advantages;
 It is inexpensive.
 Provides hard and unbiased information.

Disadvantage;
 May not contain detail, adequate and sufficient
information (incomplete).
 Cont…

b. Interviewing individuals

 Able to find information that can’t be

recorded
(eg life style of individuals).

Disadvantage;
 Bias can be introduced, (recall bias).
 Desirability bias
 Cont…
2. Outcome data ascertaining:
 For diseases that are fatal, death certificate may be the
source.

 For cause specific deaths, it may be obtained from

autopsies, physician, verbal autopsies and hospital
records.

 For non-fatal end points, physician, hospital records and

even examination by physicians or screening can be
obtained.

NB: Any outcome measurement should be done EQUALLY

both to the exposed as well as non-exposed groups.
6. Approaches to follow up
 Follow up is major challenge both cost-wisely and
timely.

 Length of follow up depends on the latency period

of the outcome. (days, months or years)

 Unless follow up is nearly complete, it is difficult to

interpret. (it may be source of bias).

 Therefore, it is crucial to think and achieve

complete follow up.
(People are mobile, changing job, changing address etc).
Issues in Analysis
 Cont….
 Calculating incidence of outcomes among
exposed vs non-exposed groups (RR).

 Incidences could also be compared among

various levels of exposure and combinations.

 Denominators could be number of individuals

or person-time units. (RR, AR)
 Cont…
 Incidence risk (Cumulative incidence)
Denominator = # of individuals at risk at baseline

 Incidence (density) rate

 Denominator = ‘person time’

 Calculate rate ratio (relative risk)

 Attributable risk (risk difference)

Interpretation
 As in other epidemiological studies,
interpretation requires the role of Chance, bias,
and confounding as explanations of findings.

 Role of bias

 Selection bias is less concern in cohort (prospective)

studies, but if knowledge of disease affects selection
of exposed and non exposed (retrospective), selection
bias may result.
7. Role of bias
 Any observational study is unlikely to categorize all
individuals correctly.

 Misclassification of exposure status or outcome status

occurs in most cohort studies (two forms).

A. Random (non-differential) misclassification :-

inaccuracies occur in similar proportions in each of the
study group.
(eg. smocking by # of cigarette Vs quality, pattern of smocking)
 Increases similarity between exposed and non-
exposed

 This can dilute and could under- or over estimate

the results.
 Cont…
B. Non-random (differential) misclassification:-
when misclassification produces one sided
difference, and resulted in difference of accuracy
or quality of information.

 Smoking and bronchitis, (eg. Smokers get more

medical attention, more diagnosed than non-smokers)

 Giving care in ascertaining both exposure and

outcome is crucial).
8. Effect of loss to follow up
 Loss to follow up is the major source of bias in cohort
studies.

 Members of cohort may be lost to follow up, if this

proportion is large, > 20- 25 %, it becomes difficult to
validate.

Reduce loss to follow up to an absolute minimum.

 Cont…
Losses to follow up: (Solution)

1. Try to find outcome measures from all possible

sources.

2. Compare presence of difference in

demographic characteristics between loss to
follow up and follow up ascertained people,

If no difference then it is not much a problem.

 Cont…
3. Indirect estimation of their effect
a. inclusion of assumption of all lose to follow up
as if they developed the outcome,

b. inclusion of assumption of all loss to follow up

as if they didn’t develop the outcome

 This gives the range of association it could lie.

 If the estimate is within the range then it is

good estimation, else it shows that there is bias
9. Effects of non-participation
 Non-participants (non-response) are participants who
are eligible to participate but who did not be included
in the study (non-volunteers).

 These who tend to participate differ to those who did

not-participate on motivation, attitude to health and
knowledge of risk status.

 The problem encountered by non-response is not on

the internal validity but by difficulty to generalize
(external validity).
 Cont…
 Effect of non-response can be assessed by
comparing basic social and demographic
characteristics of respondents and non-
respondents.

 If there is no difference in major demographic

characteristics, then results are able to be
generalized.
Measures of association
 Types of measures:
 RATIOS: relative risk, rate ratio

 DIFFERENCE: attributable risk

 Ratio: indicates how much more likely one group is to

develop disease than another.

 Difference: indicates on an absolute scale how much greater

the frequency of disease is in one group compared with the
other

 Relative risk (ratio) and attributable risk (difference) are two

most frequently used in epidemiology.
 Measures of Association in
cohort studies
Relative Measure Ie
 Relative Risk (RR)
Rate ratio
Risk ratio
Iue

Effect (impact) measures

 Absolute measures


Risk difference (RD) Ie - Iue
Ie = incidence in exposed
Iue= incidence in unexposed
Summary
Strength
Is of particular value when the exposure is rare.

Can examine multiple effects of a single exposure.

Can elucidate temporal relationship between

exposure and disease.

Allows direct measurement of incidence of disease in

the exposed and non exposed.
Summary
Limitations
Is inefficient for the evaluation of rare diseases,

If prospective, can be extremely expensive and

time consuming.

If retrospective, it requires the availability of

adequate records.

Validity of results can be seriously affected by

losses to follow up.
 Summary
 Cohort studies allow measurement of risk

 Case-control studies are rapid, but not able to measure

risk; (only estimate RR)

 In the ideal world: Prefer cohort to case-control study

 In the real world: Case-control studies usually do the job

 Learning objectives
1. Describe the characteristics of a cohort study.
2. List the types of bias most likely to affect a cohort
study.
3. List the conditions under which a cohort study is an
appropriate choice to address a research question.
4. Describe the advantages and disadvantages of a
cohort study design.
5. Calculate appropriate measures of disease occurrence
and exposure-disease association for a cohort study.
6. Calculate and interpret well on measures of
association and impact of exposure on disease status.
Thank you