Dagu workshop at Gondar University May 14-18, 2018
Advanced Epidemiology
Lars Åke Persson
COHORT STUDY
Design and inferences
Suppose that we want to study
the association between breast-feeding
and diarrhoeal disease
Three alternative approaches:
1) A cross-sectional study
2) A cohort study
3) A case-referent study
IMCH, Lars Åke Persson
The Cohort Design
The cohort study
• We follow two groups DD DD To-
of children, one of Yes No tal
which has been
breast fed for 6 BF 50 150 200
months, the other Yes
group not
• We register if BF 150 50 200
diarrhoea occurred No
during a defined
period, i.e. 6-11.99
mo.
The cohort study
IBF = 50/200 = 0.25 DD DD To-
INon BF = 150/200 = 0.75 Yes No tal
BF 50 150 200
RR = 0.75/0.25 = 3 Yes
BF 150 50 200
Interpretation? No
Cohort: stratified analysis
Suppose that we have to take water quality into account
BF Water DD DD Total Incidence
Yes No
Yes Good 5 45 50 0.10
0.25
Bad 45 105 150 0.30
No Good 75 45 120 0.625
0.75
Bad 75 5 80 0.938
Total Good 80 90 170 0.47
0.50
Bad 120 110 230 0.52
What do you want to show?
Objective Design
Prevalence Cross-sectional
Incidence Cohort
Risk Cohort; case-referent
Prognosis Cohort
Treatment effect RCT
Study base
The “study base” is a population of
individuals, each carrying the
burden of personal and group risk
factors
individuals
time
Cohort study (prospective or retrospective)
Disease +
Exposed
Disease -
Study population
Disease +
Non-exposed
Disease -
1 Why is this study done
Example cohort study
for critical reading 2 What is main exposure?
3 What is “dynamic”
cohort?
4 Cluster sampling, what
is that?
5 How was confounding
managed?
6 Effect modification?
7 Comment result
1. Why is this study done?
• Need to go from “efficacy” to program
“effectiveness”
• Influence of contextual factors
• Effects in subgroups
• Cost and effectiveness
2. What is main exposure?
3. What is “dynamic” cohort?
• Closed cohort: no one entering after start
• Open or dynamic cohort: may enter after
start
4. Cluster sampling, what is that?
• Here 18 randomly selected communities within four
sentinel districts
• Feasible sampling strategy to get a representative
sample (e.g. DHS)
• Consequences for analysis: variation within and
between clusters
• More about this in cluster course (third semester)
5. How was confounding managed?
• Multivariable analyses with potential
confounding factors
• A factor may be a confounder and/or effect
modifier
6. Effect modification?
7. Comment result
• Size of effect
• Bias?
• Money well spent
Cohort study
• To study incidence, natural history
• To analyse risk factors, relative risk
• Cause and effect, temporal sequence
• Retrospective cohorts often cheaper
• May look at several outcomes
• Confounding major problem
• Selection bias, loss to follow up may
cause bias
Cohort studies
• Starts with the population at risk
• Measures characteristics at baseline
• Follow-up over time to measure incidence
of disease
• Compare event rates in the cohort (among
those who have and those who don’t have
the characteristics of interest)
• Confounding may be a problem
Confounding
• An alternative explanation for observed
association between exposure and
disease or
• A mixing of effects – the association
between exposure and outcome is
distorted
• May distort the true association toward the
null or away from the null
Confounding
• A variable can not be a confounding factor if it is located
on the causal chain from exposure to outcome
• Confounding can be controlled in the design (matching,
restriction) or in the analysis (stratified analysis,
adjustment)
• Confounding factors are (other) risk factors for disease
• If comparing crude and adjusted measures of
association you may judge whether confounding was
present or not
Confounding
• Randomisation in trials – to control for
known and unknown confounding factors
• There may be residual confounding –
addressed in the discussion
Quiz 2
1. Which of the following measures is used frequently as a denominator to calculate the incidence rate of a disease?
a) number of cases observed
b) number of new cases observed
c) number of asymptomatic cases
d) person-years of observation
e) persons lost to follow-up
2. When a new treatment is developed that prevents death but does not produce recovery from the disease, the following will occur:
a) prevalence of the disease will decrease
b) incidence of the disease will increase
c) prevalence of the disease will increase
d) incidence of the disease will decrease
e) incidence and prevalence of the disease will decrease
3. Which of the following is an advantage of a retrospective study?
a) There is little of no bias in assessment of exposure to the factor
b) Multiple disease outcomes following a selected exposure can readily be studied
c) Dependence on recall by subjects in the study is minimized
d) It is possible to determine the true incidence rate of the disease
e) It may be used to study etiology of a rare disease
4. True statements concerning cohort studies include which of the following:
a) Cohort studies are longitudinal in design
b) Subjects are selected on the basis of characteristics present before the onset of the condition being studied
c) Subjects are observed over time to determine the frequency of occurrence of the condition under study
d) They are primarily descriptive, rather than analytic
