Cohort studies
Example
0
0.5
1
1.5
2
2.5
3
3.5
%withpneumoniaover2yrs
Parents smoke at home
Parents don’t smoke at home
Some characteristics of cohort
studies
- Longitudinal study: typically decades of follow-up in a large
sample (5,000 – 100,000)
- Defined population: community, birth cohort, occupational
groups
- High response rate = representative sample
- Measure level of exposure to risk factors
- Observe deaths, development of disease or some other
condition e.g. high blood cholesterol
Cohort studies can serve several
purposes
- Identify new cases of disease
- Provide direct measurement of risk of developing
disease
- Compare disease risk in the groups over time
- Analytic studies, study aetiology (causation)
- Examine wide range of outcomes
- Record the life histories of sections of the
population
- Tell us what circumstances predict development of
disease or health improvement e.g. social position,
disease risk score
time
direction of enquiry
Advantages of cohort study
- Temporal sequence is clear (exposure before disease)
- Less prone to ‘reverse causality’
- Allows calculation of disease incidence
- Allows calculation of absolute and relative rates of
disease
- Can examine many exposures simultaneously
- Multiple outcomes can be examined
- Less possibility for bias compared with case-control
study
Disadvantages of cohort study
- Exposure may change over time
- Some diseases take years/decades to develop so may not
be suitable
- Findings might not be relevant at end of study
- High costs because large sample and long duration
- Participant burden
- Loss to follow-up usually depends on outcome of interest
(selection bias)
- Assessment of causality problematic in observational
setting (although less problematic in cohort than other
types of observational studies)
Cohort vs cross-sectional design
Cohort Cross-sectional
Investigate rare
disease
- Investigate
rare
exposure
++ Study
multiple
exposures
+++ ++
Assess
temporality
++ Direct
measure
of incidence
+++ Adapted
from “Basic Epidemiology”, Bonita et al. WHO 2006.
Some well-known cohort studies
 British Birth Cohorts
◦ Millennium Cohort Study
◦ 1970 British Cohort Study (BCS70)
◦ 1958 National Child Development Study
◦ 1946 National Survey of Health and Development
 Studies of specific diseases (e.g. cardiovascular disease):
◦ Whitehall II study
◦ Framingham Study
◦ HAPIEE (Health,Alcohol and Psychosocial
Indicators in Eastern Europe)
 Studies of specific exposures/groups of population
◦ War veterans
◦ Nurses Health Study
Prospective cohort study
• Identify a group of individuals and follow them
over time
• Usually to assess whether exposure affects
incidence of outcome/disease.
Historical/retrospective cohort study
• Identify a group and obtain records/information
from earlier time
• The aim is still to compare exposed and
unexposed
• The exposure and development of disease
already happened
Advantages & disadvantages of
retrospective cohort studies
Advantages
 Quick
Disadvantages
 Measurement error
from poor quality
records
 Exposure not measured
exactly as wish
Open cohorts
• People move in and out of the study
Closed cohorts
• Participant population is fixed at baseline
• People can only exit study (withdrawal, death)
Closed cohort study Open cohort study
Year
1 2 3 4 5 6 7 8 9 10
alive
withdrawn
died
alive
withdrawn
alive
withdrawn
died
alive
alive
Year
1 2 3 4 5 6 7 8 9 10
Representativeness in cohort studies
Validity of estimates rests on sample being
representative.This is influenced by:
 Selection of study sample & response rate
 Poor measurement of exposure & outcome
 Loss to follow-up
◦ A significant challenge for longitudinal studies
The 1970 British Cohort Study,
dates of contact & sample size
Year Age (yr) Target sample Achieved
sample
1970 Birth 17 287 16 571
1975 5 16 810 13 071
1980 10 17 275 14 874
1986 16 17 529 11 621
1996 26 17 329 9003
2000 30 17 050 11 261
2004 34 13 107 9656
Cohort profile: 1970 British Birth Cohort (BCS70). Elliott & Shepherd. Int J Epidemiol.
2006;35(4):836-43.
Some reasons why some people drop
out of longitudinal studies
 People who drop out more likely to live alone, have
lower SES, engage in fewer social activities, be
cognitively impaired and have poorer physical
functioning
 Study too time-consuming
 Contact too frequent
 Questionnaires too difficult, repetitive
 Travel to screening clinic difficult
 Dislike of medical tests
 Tests not seen as relevant
Summary of cohort studies
 Exposure measured usually in healthy
individuals
 Follow up
 Incidence
 Time consuming & expensive
 Temporality clear
 Possibly the “best” observational design
Case-control studies
Example
0
5
10
15
20
25
30
35
%withsmokingparents
Pneumonia in last year
Healthy children
CohortStart
Unexposed
Exposed
All healthy Follow-up (wait)
Disease
assessment
Controls
Cases
Start
Look back
Case-Control
Case-control studies are
 Ideal for rare diseases
 Usually “retrospective” in design
 Relatively quick
 Relatively cheap
Basic steps in a case-control study:
1. Cases
 Definition of a case (symptoms; duration…)
 Selection of cases (patients with certain disease
condition)
◦ Source: Hospital / outpatient clinic / etc
◦ Prevalent cases / Incident cases
Basic steps in a case-control study
2. Controls
 Definition of controls (subjects without the condition)
 Selection of controls (hospital, community...)
 Hospital controls:
◦ Feasible
◦ Willing to participate
◦ Might be of the same social and geographical
background as the cases
◦ Hospitalized people differ from the general
population (might have a higher or lower level of
exposure to the risk factor under study
compared to the general population)
 Community controls:
◦ May reduce selection bias
◦ Low participation rates
◦ Time consuming and costly
◦ Recall bias
Basic steps in a case-control study
 Measurement of exposure
 Comparing frequency of exposure in cases and controls
Time
“Now”
Cases
Controls
Time
“Now”
Cases
Controls
How do we quantify the association
in a case control study?
 Remember from earlier: Relative risk =
[a/(a+b)] / [c/(c+d)]
 If the disease is rare, then a would be very
small compared to b therefore:
 [a/b] / [c/d], the odds ratio, would be
approximately close to relative risk [a/(a+b)] /
[c/(c+d)]
Disease
+
(cases)
Disease
-
(controls)
Exposure + a b
Exposure - c d
Example from a cohort study that
shows odds ratio approximates
estimates of relative risk:
Disease + Disease Exposed
20 980 1000
Not Exposed 10 990 1000
• Relative risk = [20/1000] / [10/1000] = 2.00
• Odds ratio = [20/980] / [10/990] = 2.02
Relative risk CANNOT be estimated
from case-control studies.
Only odds ratio can be calculated
Why OR and not RR?
Cases Controls TOTAL Cases Controls TOTAL
Exp+ 30 100 130 30 300 330
Exp- 10 100 110 10 300 310
RR=(30/130)/(10/110)=2.54 RR=(30/330)/(10/310)=2.82
OR=(30/100)/(10/100)=3.00 OR=(30/300)/(10/300)=3.00
• Different sampling fraction among cases and controls
• RR is influenced by sampling fraction among controls
while OR is same (and is unbiased)
Matched case-control studies
 Cases and controls often differ in
important aspects (age, sex, ethnicity,
behaviours...)
 These can confound the study
 One way to eliminate such differences is
matching controls to cases on these factors
 More than 1 control per case can be used
Example: matching in the study of hip
fracture
 Risk of hip fracture depends on age and sex; men and
older people are more likely to suffer; these factors
have to be controlled
 Matching cases and controls on age and sex will
eliminate the confounding by these factors
 For each case [male; age 74] recruit one or more
controls [male; age 74]
 For each case [female; age 81] recruit one or more
controls [female; age 81] etc
Other ways to control confounding
 Matching may be impractical (if there are
many strata, it is difficult to find controls)
 Adjustment in analysis
◦ stratified analysis (eg within drinkers and non-
drinkers)
◦ multi-variable analysis (“adjusted” odds ratios)
Nested case-control study
 Using an existing cohort study
 Cases: subjects who developed the disease
 Controls: a random sample of subjects who
did not develop the disease
 Rationale: to reduce cost with lab
measurements
 Advantage: no reporting / measurement
bias
Strengths of case-control studies
 Quick (cases already exist, no need to wait)
 Cheap (not necessary to examine large
number of people)
 Can examine many exposures
 Suitable to study rare diseases
 Suitable to study stable exposures (eg
genetic markers)
Weaknesses of case-control studies
 Not suitable for rare exposure
 Cannot calculate incidence risk or death
rates
 Prone to selection bias
 Prone to misclassification of exposure
 Prone to reverse causation (people with
disease may have changed their behaviour)
Summary of case-control studies
 Cases vs. controls (current status)
 No follow up
 Good for rare outcomes
 Asking about exposure in past
 No incidence or prevalence
 No need to wait for cases  quick
 Temporality may be a problem
 Good for exposures stable over time