Cohort Studies
Jan 20, 2016
Cohort Studies
XX
X
X
Compare incidence
XX
XXX
XXX
Compare past exposures
Compare their past exposures.
Cohort Study
Identify non-diseased people;
group by risk factor status
Follow longitudinally to compare incidence.
Case –Control Study
Find cases with disease & non-disease controls.
• Follow up studies• Longitudinal studies• Incidence studies
Compare Incidence
X
XX X
Time passes
Not ExposedNot Exposed
ExposedExposed
Non-diseased initially
Key Features of Cohort Studies
• Start with non-diseased people. • Group by exposure status.• Compare incidence of disease.
In prospective cohort studies conception, design, & enrollment occur before anyone develops the outcome.
XX
X
XCompare incidence
Enroll non-diseased subjects; collect baseline exposure data
Follow up at intervals to get accurate outcome data.
Identify a cohort retrospectively, e.g. employees at a tire factory.
XX
X
XCompare incidence
Obese
Lean
Exposedto chemicals
Non-exposed office workers
Determine what then happened to them.
A randomized clinical trial is similar to a prospective cohort study.
The investigators assign subjects to a treatment or intervention.
X
XXX
Compare incidence over time
Aspirin
Placebo
Compare Incidence Start
of Study
Start of Study
Kiddy Pool
No Kiddy Pool
Club Members
Had factor
Didn’t have factor
During the short time course of this study fairly reliable data was collected by interviews with club members.
PastPast
Compare Incidence of Death Start
of Study
Start of Study
PastPast
Tire makers
Clerical staff
Employees of tiremanufacturer
Had factor
Didn’t have factor
During the long time course of this occupational study data on exposure was from employee records and there was little information on other confounding factors.
Compare Incidenceof Cancer Start
of Study
Start of Study
PastPast
X-Ray therapy
Other therapy
Patients with ankylosing spondylitis
Had factor
Didn’t have factor
During the long time course of this medical study accurate data on exposure was obtained from medical records, but accurate information about smoking, alcohol consumption, BMI, & occupational exposures was missing.
Data Sources for Retrospective Cohort Studies
• Questionnaires
• Employee records
• Medical records
However, accurate information on the main exposure and confounding factors is often missing because the data source was not designed for the purpose of conducting the study.
Retrospective cohort studies are useful for unusual exposures and they are efficient for diseases with long latent periods (time between exposure & disease).
Similar in Design & Analysis
Similar in Design & Analysis
• You can collect data in a spreadsheet.
• Organize findings in a 2x2 table (or R x C).
• Calculate RR, RD, AR%
• Test for significance with chi square.
• Compute 95% confidence interval for RR.
• Retrospective Cohort• Prospective Cohort• Clinical Trial
• Retrospective Cohort• Prospective Cohort• Clinical Trial
Recording the data for analysis:
A table with subjects in rows. Exposures and outcomes are listed in columns.
Analysis of Retrospective Cohort Studies
Given the assumption that exposure status is fixed and the lack of periodic follow up, it makes sense to compare cumulative incidence.
Yes No
Got Giardiasis
14 341 355
16 108 124 Yes
No
CumulativeIncidence
12.9%
3.9%
In kiddy pool
Not exposed
Relative Risk =12.9%
3.9%= 3.3 ( 1.7-6.5)
P value = 0.0003
117,000 Nurses without cancer or CVD
Compare Incidence of disease
obese
lean
Prospective Cohort Study
The CohortThe Cohort
Have risk factor
Don’t have it
Follow-up
Start of Study
Start of Study
FutureFuture
A major advantage of prospective cohort studies is that they are carefully designed with specific goals in mind. Therefore, information on baseline exposures can be carefully collected on all subjects before anyone has an “event”.
Potential Sources of Data in a Prospective Cohort Study
• Interviews
• Questionnaires
• Biological specimens (blood, urine, toenails, hair)
• Imaging studies (x-ray, ultrasound, CT, etc.)
• Medical records (verifying outcomes such as myocardial infarction, stroke, cancer, etc.)
Data collected specifically for a prospective cohort study will be of higher quality than that obtained from pre-existing records, but it is more expensive to do a large, prospective cohort study.
Potential Problems
• Exposure status may change. What if some women lost or gained a
substantial amount of weight during the period of study? What if smokers quit?
• Loss to follow up. If follow up is over many years, subjects
may loose interest and quit.
These problems can be overcome by taken steps to minimize loss to follow-up and by careful periodic follow up with re-assessment of exposure status.
Incidence Rate For Variable Follow-up in A Prospective Cohort Study
SubjectA-B-C-D-E-F-G-H-I-J-K-L-
x
x
x81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
Incidence (IR) = 3/107.7 = .028/ person-yr = 28/1000 p-yrs
Total time at risk =107.7 person-yrs
Timeat Risk
8.311.0
14.014.0
10.2 3.0
7.0
10.0
3.0
9.06.2
12.0
x= when theygot disease
Risk difference = 62.3 excess cases /100,000 P-Yrs
# MIs(non-fatal)
41
57
56
67
85
person-yearsof observation
177,356
194,243
155,717
148,541
99,573
MI Rate per100,000 P-Yrs
(incidence)
<21<21
21-<2321-<23
23-<2523-<25
25-<2925-<29
>29>29
BMI:
Nurses Health Study
Is obesity associated with an increased risk of heart disease?
Relative Risk
1.0
1.3
1.6
2.0
3.7
23.1
29.3
36.0
45.1
85.4
Another Unusual Risk Factor
Veterans’ Problems after Viet Nam:
• Skin rashes• Psychological problems• Infertility• Liver abnormalities• Cancers
Agent Orange
Goals of the Agent Orange Study
Look at a single, unusual exposure.
Look at multiple possible effects.
Prospective partRetrospective partIncidenceof cancer?
Incidence ofskin rash?
Start of StudyStart of Study
Exposed
Unexposed
An Ambi-directional Cohort Study
The “Ranch Hand” study looked at effects of Agent Orange
»1,264 pilots who sprayed dioxin in Vietnam
»1,264 pilots who flew only cargo missions in Vietnam
Why were pilots flying only cargo missions an appropriate comparison groups?
Subjects groupedby risk factor status
239 3
139
Yes No
Skin Disease
98
119
Yes No
Liver Abnormalities
138
169
Yes No
Psychological Problems
227 10,807
217 10,820
Yes No
CancerYes
No
AgentOrange
A cohort study can look at multiple effects of a single exposure.
Conclusions from the Ranch Hand Study
Exposure to dioxin was associated with a significant increase in risk of soft tissue sarcoma, non-Hodgkin's lymphoma, and Hodgkin's disease.
There was suggestive, but not statistically significant evidence of an association with respiratory cancers, prostate cancer, and multiple myeloma.
Common risk factors (obesity, HBP): • A cohort from the general population: (e.g., Framingham Heart Study)• A special study group: doctors, nurses (e.g. The Nurse’s Health Study)
Unusual risk factors:• A special exposure group: (e.g., tire manufacturing, dioxin, asbestos)
Selecting Subjects for a Cohort Study
1) As similar as possible with respect to other factors that could influence outcome.
2) Comparable & accurate information.
Ideally, from within the cohort (internal controls), but this is not always possible.
The Comparison Group
Where does the comparison group come from?
Compare Incidence of Death
Problems with the Comparison Group
Start of Study
Start of Study
PastPast
Tire makers
Clerical staff
Employees of tiremanufacturer
An Open Cohort
Had factor
Didn’t have factor
Subjects may switch exposure status or become lost to follow up (quit)
What are the strengths and weaknesses of using clerical employees as the comparison group?
Compare Incidence of Death
Problems with the Comparison Group
Start of Study
Start of Study
PastPast
Tire makers
Clerical staff
Employees of tiremanufacturer
An Open Cohort
Had factor
Didn’t have factor
Subjects may switch exposure status or become lost to follow up (quit)
The clerical staff may be much less exposed to the chemicals, but there may be important differences in confounding factors such as gender, age, socioeconomic status, education, diet, smoking, alcohol consumption, etc.
Internal Control Group
The best comparison group is an unexposed or “least exposed” subset of the study cohort.
Nurses
Leanest
Heaviest
Incidence of heart disease?
General Population
General Population Control
RubberWorkers
vs.
Mortality Rates?
The general population is often used in occupational studies of mortality, since data is readily available, and they are mostly unexposed. Some data sets allow adjustment for age, gender, SES, and race.
The main disadvantage is bias by the “healthy worker effect.” The employed work force (mostly healthy) generally has lower rates of mortality and disease than the general population (with healthy & ill people).
A Comparison Cohort
vs.
RayonFactoryWorkers
Exposed to disulfide
PaperMill
Workers
No disulfide exposure
Incidence of heart disease?
Both groups consist of blue collar workers, minimizing differences in SES, education, etc. The “healthy worker effect” is also not a factor. However, the paper mill workers may have other exposures that affect heart disease.
• Suspicion that occupational exposure to chemicals in tire manufacturing increase risk of death.
• Viet Nam veterans with a variety of health problems. Was Agent Orange responsible?
• Does x-ray treatment of ankylosing spondylitis (back problem) increase the risk of cancer?
• The need for a better understanding of behavior modifications that can prevent cancer & heart disease (e.g., smoking & obesity).
A strength is that cohort studies are efficient for unusual exposures.
However, cohort studies also provide high quality information for examining common exposures as well.
Example:True incidence of thromboembolism:
Subjects lost to follow up: 4,012 4,008
Subjects with TE lost to follow up:
Apparent incidence of TE:
OC Users Non-OC users
20/10,000 10/10,000
12 2
8/5,988 8/5,992
True RR = 2.0 Apparent RR = 1.0
Can occur in cohort studies & intervention trials.
Effects: over- or under- estimate of association.
Bias From Loss To Follow-up
If subjects in one of the exposure groups are more likely to be lost to follow up if they develop the outcome, then the estimate of association will be biased.
Maintain Follow UpMaintain Follow Up
• Choose subjects who are motivated• Choose subjects who are easy to track
(professionals)• Keep them interested: newsletters,
incentives, part of the “family”• Frequent phone calls• Make questionnaires easy to fill out
Retrospective cohort studies:
• Efficient for diseases with long latent period (time from exposure to development of disease).
• Temporal sequence between exposure & disease is clear (e.g., obesity preceded CHD)
• Can directly calculate incidence, RR, RD, AR%
• Efficient for looking at rare exposures or unusual risk factors (e.g. agent orange).
• Can evaluate multiple effects of a single exposure.
Advantages of Cohort Studies:
Prospective cohort studies:
• Provide better information on exposures.
• Are less vulnerable to bias because exposure status is determined before outcomes.
Bot
h R
etro
spec
tive
& P
rosp
ectiv
e:
XX
X
XCompare incidence
Identify a cohort retrospectively (e.g. tire manufacturing workers vs. desk employees. Look at what subsequently happened to them.
Enroll non-diseased subjects; collect baseline exposure data
Follow up at intervals to get accurate outcome data.
XX
X
XCompare incidence over time
Obese
Lean
Exposed
NotExposed
• Efficient with long latent period from exposure to development of disease).
• Better information on exposures.• Less vulnerable to bias because exposure status is determined before outcomes.
Retrospective Cohort Study
• Poor information on exposures & confounding factors.
• More vulnerable to bias.
Disadvantages of Cohort StudiesDisadvantages of Cohort Studies
Prospective Cohort Study
• May need large numbers of subjects for long periods of time.
• Can be expensive and time consuming.
• Inefficient for diseases with long latent periods.
• Loss to follow up can bias results.
• Inefficient for rare outcomes (both).
XX
X
XCompare incidence
Identify a cohort retrospectively (e.g. tire manufacturing workers vs. desk employees. Look at what subsequently happened to them.
Enroll non-diseased subjects; collect baseline exposure data
Follow up at intervals to get accurate outcome data.
XX
X
XCompare incidence over time
Obese
Lean
Exposed
NotExposed
• Poor information on exposures & confounding factors.
• More vulnerable to bias.
• Expensive and time consuming.• Inefficient for diseases with long latency.• Loss to follow up can bias results.
• How were the study groups selected or defined?
• Did they differ in other ways that could affect the outcome?
• Data collection: Accurate? Comparable for all groups?
• How complete was the follow-up?
When reading a cohort study, consider…