Cohort Studies

1
Cohort Studies

• October 14, 1999
• Iris A. Granek, M.D., M.S.

2
Cohort Study Design

• Compares incidence (risk) of developing the
  disease (outcome)
• Exposed (factor under study)
• vs.
• Unexposed (without factor under study)

3
Selection of Study Population

• Community based
• sample of residents of a geographic region
• Framingham Heart Study
• Special exposure group
• occupational groups
• Motivated identifiable group
• Nurses, physicians
• Nurses Health Study, Health Professionals
  Follow-up Study

4
Selection of Study Population

• Exclude those not at risk for the disease
• those already with the disease
• those who cannot get the disease
• Ascertainment of exposure
• for main exposure of interest
• other potential confounding factors

5
Ascertainment of Outcome

• Vital records (death certificates)
• Hospital medical records, disease registries
• Follow-up questionnaires exams of the subjects
• Ideally those collecting outcomes data should be
  blind to subjects exposure status

6
Cohort Study Design
Cohort
Outcome
Incidence of disease in the exposed group
Group of subjects to be studied. None of them
have the disease outcome. They differ in their
exposure status.
Exposed
Incidence of disease in the unexposed group
Unexposed
7
Prospective Cohort Studies

• Select cohort without disease
• Identify those with without the factor (exposed
  not exposed)
• Follow-up over time determine disease incidence
  in those exposed not exposed

8
Radiation Leak in Nuclear Plant
Cohort Workers 1969
Workers disease status 1999
12 workers on site at time of the leak 24
workers off duty at home at time of the leak
Exposed
4 with leukemia 8 no leukemia
Unexposed
4 with leukemia 20 not leukemia
9
Retrospective Cohort Studies

• Cohort is assembled retrospectively
• Exposure data is determined retrospectively
• Outcomes are determined currently

10
Radiation Leak in Nuclear Plant
Cohort of Workers 1999 collect info from1969
Workers disease status 1999
12 workers on site at time of the leak 24
workers off duty at home at time of the leak
Exposed
4 with leukemia 8 no leukemia
Unexposed
4 with leukemia 20 not leukemia
11
Analysis of Cohort Studies

• Calculate incidence (risk) in exposed group
• Calculate incidence in not exposed group
• Calculate relative risk between the incidence in
  the exposed and the non exposed

12
Calculation of Relative Risk

• Disease No Disease
  Totals
• Exposed a b a b
• Not Exposed c d c d
• I exposed a I not exposed c
• a b c d
• Relative Risk RR I exposed
• I not exposed

13
Calculation of Relative Risk

• Disease No Disease
  Totals
• Exposed 4 8 12
• Not Exposed 4 20 24
• I exposed 4 I not exposed 4
• 12 24
• Relative Risk RR 0.33 1.9
• 0.17

14
Relative Risks

• RR gt 1 indicates the factor increases the risk of
  developing the disease
• The risk of developing leukemia was approximately
  twice that of the unexposed workers
• RR lt 1 indicates the factor decreases the risk of
  developing the disease
• RR 1 indicates factor has no effect

15
95 Confidence Limits

• 95 probability that the true value lies within
  the confidence interval or between the confidence
  limits
• Relative risks are statistically significant if
  they do not include 1
• RR 7 (0.5 - 15.0) not statistically significant
• RR 7 (3.0 - 12.0) is statistically significant

16
Relative Risks

• RR 1.3 (1.1 - 3.2)
• The risk of developing the disease is 1.3 times
  greater in the exposed group (those with the
  factor) than in the non exposed group.
• 30 increased risk
• RR 0.7 (0.3 - 0.9)
• The risk of developing the disease in those with
  the exposure is 70 of those without it. There
  is a 30 reduction in risk.

17
Analysis of Cohort Studies

• Attributable Risk
• Since some of those not exposed to the factor
  also get the disease, attributable risk gives the
  disease risk due solely to the exposure.
• AR I exposed - I not exposed
• AR 0.33 - 0.17 0.16

18
Analysis of Cohort Studies

• Attributable Risk Percent
• AR Attributable Risk X 100
• Incidence in exposed
• AR I exposed - I not exposed X 100
• Incidence in exposed
• AR 0.16 / 0.33 48.5

19
Bias in Cohort Studies

• Bias is a systematic error in the study that
  distorts the results limits the validity of the
  conclusions.
• Confounding
• need to control for in analysis (adjustment,
  stratum specific results)
• Ascertainment bias (not blinded)
• Exposure status in retrospective studies subject
  to bias

20
Advantages of Cohort Studies

• Gives direct estimate of the relative risk from
  the exposure
• Can calculate incidence rates in exposed and non
  exposed
• can calculate absolute difference in the
  incidence rates (attributable risk)
• can calculate the ratio of these rates (relative
  risk)

21
Advantages of Cohort Studies

• Can correlate a specific exposure with a disease
• temporal relationship (cause effect)
• frequently can examine dose response
  relationships
• Efficient for study of rare exposures
• Can provide information on multiple outcomes or
  diseases from an exposure

22
Disadvantages of Cohort Studies

• Large, expensive long-term studies
• Problems of lost to follow-up can bias results
• Subjects can change their exposures over time
• smoking, exercise, hormone use, etc.

23
Disadvantages of Cohort Studies

• Not efficient for rare diseases
• Changes in diagnostic methods may bias the
  determination of outcomes (surveillance bias)

24
Nurses Health Study Example

• Prospective Cohort Study
• Cohort assembled in 1976
• 121,701 nurses who returned questionnaire
• 11 US states
• aged 30 - 55 yrs
• Answer questionnaire every 2 yrs
• risk factors, exposures, health outcomes

25
Abdominal Adiposity CHD

• JAMA 12/2/98
• 1986 questionnaire included hip waist
  measurements - height weight
• 102,252 responded to questionnaire
• 89,921 free of heart disease
• 44,702 remaining after eliminate missing data

26
Person - years of follow-up

• Accounts for varying lengths of follow-up from
  start of study until gets the outcome
• no longer at risk so no longer part of the
  denominator for calculating incidence
• 100 person-years
• 10 people followed for 10 yrs
• 100 people followed for 1 yr
• 20 people followed for 5 yrs

27
NHS Abdominal Adiposity CHD

• Person-years of follow-up calculated as
• time from completion of 1986 questionnaire to
  June 1994
• OR
• time of diagnosis of endpoint (fatal or non fatal
  MI)

28
NHS Abdominal Adiposity CHD

• Hip to waist ratio of lt0.72 chosen as the
  reference value (no exposure)
• 17 outcomes for 55,542 person-years
• Hip to waist ratio of 0.80-lt0.84
• 73 outcomes for 61,073 person-years
• Hip to waist ratio of gt0.88
• 56 outcomes for 24,642 person-years

29
NHS Abdominal Adiposity CHD

• 0.80lt0.84
• gt0.88

• Crude RR 3.9
• Age Adjusted 2.85 (1.68 - 4.84)
• Additional
• adjustment 2.02 (1.14 - 3.59)
• Crude RR 7.4
• Age Adjusted 5.04 (2.92 - 8.71)
• Additional
• adjustment 2.43 (1.32 - 4.48)