Cohort Studies

1
Cohort Studies

• Principles of Epidemiology
• Lecture 9
• Dona Schneider, PhD, MPH, FACE

2
Cohort Studies

• Type of Analytic study
• Unit of observation and analysis Individual
  (not group)
• Also called follow-up studies, incidence studies,
  panel studies, longitudinal studies, or
  prospective studies

3
Assembling a Cohort

• Cohorts may be chosen because they represent
• The general population (i.e., the outcome of
  interest has a high incidence rate)
• Special exposure groups (e.g., smokers, uranium
  miners or asbestos workers with high levels of
  specific exposures)
• Special resource groups (e.g., alumni,
  physicians, nurses)
• Geographically or facility-defined groups (e.g.,
  Three Mile Island, hospitals with specialized
  maternity care)

4
Design

• At baseline (1st observation point)
• Subjects are all disease free
• Exposure is used to classify subjects into
  exposed or unexposed groups
• Subjects are followed to document incidence (2nd
  observation point)

5
Assembling the Cohort

• Before beginning the study, determine who is
  susceptible and who is immune to the outcome of
  interest
• You may need to do this with diagnostic tests or
  medical histories

6
Single Sample Cohort Study Design
TIME
Diseased
Disease-Free Cohort
Exposed
Not Diseased
Target Population
Diseased
Not Exposed
Not Diseased
7
The Framingham Study

• Since 1948, samples of residents of Framingham,
  Massachusetts, have been subjects of
  investigations of risk factors in relation to the
  occurrence of heart disease and other outcomes

8
The Framingham Study

• Hypotheses
• Persons with hypertension develop CHD at a
  greater rate than those who are normotensive.
• Elevated blood cholesterol levels are associated
  with an increased risk of CHD.
• Tobacco smoking and habitual use of alcohol are
  associated with an increased incidence of CHD.
• Increased physical activity is associated with a
  decrease in development of CHD.
• An increase in body weight predisposes a person
  to CHD.

9
The Framingham Study

• Study population consisted of 5,127 men and women
  between ages 30 and 62 years and were at the time
  of entry free of cardiovascular disease
  (1948-1952)
• Cohort was examined every 2 years and by daily
  surveillance of hospitalizations at Framingham
  Hospital

10
The Framingham Study

• Exposures included
• Smoking
• Alcohol use
• Obesity
• Elevated blood pressure
• Elevated cholesterol levels
• Low levels of physical activity, etc.

11
Comparison (Control) Groups

• With a one-sample (population-based) cohort,
  exposure is unknown until after the first period
  of observation
• Example
• Select the cohort (all residents of Framingham)
• All members of the cohort are given
  questionnaires, and/or clinical examinations,
  and/or testing to determine exposure status
• The cohort is then divided into exposure
  categories based on those results
• The nonexposed become the internal controls
• For continuous variables, such as caloric intake
  or amount of exercise, multiple levels of
  exposure are constructed
• It is common to break exposure into quantiles
  (equally ordered subgroups) and to use the
  extremes as the comparison (referent) group

12
Question

• How does the design of a cohort study change if
  everybody in the cohort is exposed (special
  exposure cohort)?
• Example All persons exposed to radiation from
  the Chernobyl accident.

13
Answer

• You need to select a separate control cohort
  people as similar as possible to the exposed
  cohort (income, age, gender, employment) but with
  no exposure
• If you cannot find a comparison group, you may
  use available population incidence rates under
  certain circumstances

14
Multi-Sample Cohort Study Design
TIME
Diseased
Study Cohort
Exposed
Not Diseased
Diseased
Control Cohort
Not Exposed
Not Diseased
15
Selecting Comparison (Control) Groups

• If the cohort is the general population, subjects
  are selected based on exposure and the comparison
  group is internal - from the same sample - who do
  not have the exposure
• If the cohort is based on a high risk population
  selected on the basis of a given exposure (e.g.,
  Chernobyl residents, asbestos workers), external
  controls must be sought
• Sometimes both comparison groups are sought
• This eliminates the healthy worker effect and
  confounding for etiologic agents other than the
  exposure of interest

16
Selecting Comparison Groups (cont.)

• If a comparison group cannot be assembled, known
  population rates for outcomes may be acceptable
  but only if they are adjusted for the exposure
• Lung cancer rates are based on the population but
  should not be used for comparison to compare to
  populations with high smoking rates, such as
  miners. WHY?
• Leukemia rates from the general population can be
  used to compare rates to Three Mile Island
  residents. WHY?

17
Determining Exposure

• Valid means of determining exposure include
• Questionnaires
• Laboratory tests
• Physical measurements
• Special procedures
• Medical records
• What if the exposure is chronic, such as radon or
  smoking?

18
Measuring Disease

• You must determine endpoints in a similar manner
  for both the exposed and the non-exposed
• That is, procedures for disease identification
  must be the same for the exposed and the
  non-exposed
• Define the outcomes of interest (set diagnostic
  criteria)
• If you are looking for multiple outcomes, each
  must be defined

19
Measuring Disease (cont.)

• Mortality may be ascertained from medical
  records, autopsy records, death certificates,
  physician records, or next-of-kin
• Using mortality records does not allow for
  multiple outcomes
• Hospital records can be scanned for specific
  types of admissions
• Health records of employers and schools can be
  monitored
• Reportable diseases may be ascertained from state
  registries
• Absenteeism may be monitored with work records,
  self reporting, school records or household
  surveys
• Common ailments that do not usually require
  medical care may be monitored through
  self-reports, telephone surveys or calendar sheets

20
Relative Risk (RR)

• A ratio that measures the risk of disease among
  the exposed to the risk among the unexposed
• RR Numerator Incidence rate in the exposed
• RR Denominator Incidence rate in the unexposed

21
Example Calculating the Relative Risk
Disease Status
No CHD (Controls)
CHD cases (Cases)
TOTAL
176
288
112
Smoker
Exposure Status
224
312
88
Non-smoker
A/(AB)
112 / 288
Relative Risk



1.38
B/(CD)
88 / 312
22
Example Interpreting the Relative Risk
The risk of developing CHD is 1.38 times higher
for a smoker than for a nonsmoker.
or
The risk of developing CHD is 38 higher for a
smoker than for a nonsmoker.
23
RRgt1
RR1
RRlt1
Risk for disease is higher in the exposed than in
the unexposed
Risk of disease is equal for exposed and unexposed
Risk for disease is lower in the exposed than in
the unexposed
Risk comparison between exposed and unexposed
Exposure increases disease risk(Risk factor)
Particular exposure is not a risk factor
Exposure as a risk factor for the disease?
Exposure reduces disease risk (Protectivefactor)
24
Types of Cohort Studies

• Prospective
• Exposure baseline in the present
• Follow-up period present to future
• Retrospective
• Exposure baseline in the past
• Follow-up period past to present
• Historical prospective or ambispective
• Exposure baseline in the past
• Follow-up period past to present to future

25
Cohort study data collection (pg. 221)
FUTURE
PRESENT
PAST
DESIGN
D
E
Prospective
D
E
Retrospective
Historical prospective
D
E
E
26
Types of Cohort Studies (cont.)

• You may also NEST a case-control study within a
  cohort study
• Example
• Begin with a cohort of 10,000 individuals without
  rheumatoid arthritis
• Test for the presence of RA antigen
• Assume those with RA antigen are the exposed and
  those without the controls
• Follow for 10 years and determine the incidence
  of disease among both cohorts
• This reduces the cost of testing

27
Outcome Measures

• Incidence in the exposed
• Incidence in the unexposed
• Relative risk
• Attributable risk (risk difference)
• Population attributable risk
• Attributable risk percent
• Population attributable risk percent
• Standardized mortality ratio

28
Advantages of Cohort Studies

• Temporality Exposure precedes outcome because
  the cohort is disease free at baseline
• Efficient for studying rare exposures
• May be used to study multiple outcomes
• Allows for calculation of incidence of diseases
  in exposed and unexposed individuals
• Minimizes recall bias

29
Disadvantages of Cohort Studies

• Tend to be expensive (large sample size) and time
  consuming (long follow-up period)
• Loss to follow-up
• When multiple outcomes or specific disease
  incidence is the outcome of interest, bias can be
  a serious problem
• Inefficient to study rare diseases

30
Disadvantages of Cohort Studies (cont.)

• Nonparticipation (selection bias) it cannot be
  assumed that those who chose to participate had
  the same prevalence of exposures nor incidence of
  disease as those who did not participate
• A difference in prevalence of exposure in
  nonparticipants will not bias the results
• A difference in rate of disease among
  nonparticipants will bias the results