Cohort Studies

Issues of Analysis
Overview
 Analysis depends on
Type of cohort (closed vs. open)
 The investigator defines the cohort and when follow up begins, no new
members are added. Cohort is more likely to be static
 Open cohort is dynamic. Recruitment is open after follow up begins
and members are added over time.

The comparison group (internal vs. external)

 For common exposure, investigators enroll a general cohort and then
collect data on all and divide them into exposed and non
exposed(referred to as internal comparison group)
 For rare exposure, investigators select a special cohort based on their
exposure and select another cohort that is not exposed (external
comparison group)
Objectives and preferences of investigators
 Measures of disease occurrences and
associations in Cohorts: measuring risk
Overall Cumulative Incidence = 4 events / 1000
Cumulative Incidence in Exposed versus Unexposed

Association: relative and absolute

risk ratios (relative risk RR): cumulative incidence ratios of exposed versus
unexposed;
 risk difference: Difference in cumulative incidence : I exposed- I unexposed
/Attributable risk(AR), important from a public Health perspective
Internal Comparison: Closed Cohort
Vasectomy and prostate cancer (Health Professionals Cohort
Study; https://www.hsph.harvard.edu/hpfs/index.html)
PROSTATE CANCER
VASECTOMY Present Absent TOTAL
Present 90 9965 10055
Absent 210 37590 37 800
TOTAL 300 47555 47855
(Fixed)

Overall incidence (risk) of outcome in sample

=300/47855= 6.26 per 1000 population
What type of measures?
Incidence of CVD among exposed = 90/
Overall Measure? 10055= 8.95 per 1000

Relative measure / Measure of association ? Incidence of CVD among nonexposed =

210/37800 = 5.55 per 1000
How to evaluate measure of association?
Incidence risk ratio= relative risk= 1.61
 Evaluation of the Incidence Risk
Ratio(Relative Risk)

 Significance (p-values based on chi2 statistics or on

Confidence Intervals-CI): similar to x-sectional study designs

 X2 = 14.7 P < 0.0001

 95 % CI = 1.25, 2.06

 Interpret ?
Measures of disease occurrences and associations
in Cohorts II _ Rate
When you have an open cohort or a closed cohort but a very dynamic one
A realistic scenario is graph B
Rate: overall incidence density rate (IDR); rate in exposed versus
unexposed
Association: rate ratio: ratio of IDR in exposed versus IDR in unexposed

Graph A Died of Graph B

cancer

Loss to follow up
 Analysis of an open cohort study

osed  Time at risk for

exposed= 72.5
 Time at risk for
unexposed= 35.7

xposed
IRex=2/72.5
IRuexp
=1/35.7
IDRR = 0.985
 Tuberculosis among HIV-infected patients receiving HAART: long term
incidence and risk factors in a South African cohort

Objectives: To determine the long-term incidence of tuberculosis (TB) and

associated risk factors among individuals receiving HAART in South Africa.
Design: Prospective cohort study.
Methods: Microbiologically or histologically confirmed incident TB was
identified in a hospital-based cohort of 346 patients receiving HAART between
1996 and 2005 in Cape Town.
Results: The TB incidence density rate was 3.5/100 person-years in the first year
and significantly decreased during follow-up, reaching 1.01/100 person-years in
the fifth year (P = 0.002 for trend). TB incidence during the study was highest
among patients with baseline CD4 cell counts < 100 cells/μl and those with
World Health Organization (WHO) clinical stage 3 or 4 disease (5.71 and
3.88/100 person-years, respectively). Risk of TB was independently associated
with CD4 cell count < 100 cells/μl (adjusted risk ratio [ARR], 2.38; 95%
confidence interval (CI), 1.01–5.60; P = 0.04), WHO stage 3 or 4 disease (ARR,
3.60; 95% CI, 1.32–9.80; P = 0.01) and age < 33 years (ARR, 2.86; 95% CI,
1.29–6.34; P = 0.01). Risk of TB was not independently associated with plasma
viral load, previous history of TB, 
HAART: highly active antiretroviral therapy
 Table 1
Tuberculosis among HIV-infected patients receiving HAART: long term incidence and risk
factors in a South African cohort . Open cohort / internal comparison
Lawn, Stephen D; Badri, Motasim; Wood, Robin AIDS. 19(18):2109-2116, December 2, 2005.

Table 1. Tuberculosis incidence density rate stratified by baseline sociodemographic and clinical characteristics.

Copyright © 2016 AIDS. Published by Lippincott Williams & Wilkins. 10

External Comparison: the case where cohort is
assembled based on the exposure
Illustration: A Retrospective Cohort Study of Leukemia and
Other Cancers in Benzene Workers.
Benzene-exposed subjects were identified in work units
from 233 factories. External unexposed subjects were
selected from those employed in the manufacture of
mechanical instruments, textiles, and in clothing production
and were not exposed to benzene or any other known
occupational carcinogen.
Measures of disease occurrence and association similar to
closed cohorts except that we cannot have an overall
measure of risk or overall rate
11
For some outcomes(cancer, HIV conversion,
etc) and some research questions, time to event
is important
 Long-term survival after epilepsy surgery compared with matched epilepsy
controls and the general population

Abstract
This study evaluates if there was a difference in long-term survival between epilepsy
surgery patients, individually matched controls with intractable epilepsy, and
controls from the general population.
In a cohort study, we compared the survival of patients operated with epilepsy
surgery in Norway 1948–1988 with: (1) a control group with prolonged medical
treatment for intractable epilepsy individually matched for age, gender, and seizure
type (n = 139), and (2) expected mortality for matched individuals in historical
cohorts of the general population (n = 196). Survival was compared using Kaplan–
Meier curves and stratified proportional hazards analysis.
After on average 25 years of observation after surgery, there was no difference in
survival between the epilepsy surgery group and the controls with intractable
epilepsy (p = 0.18). The risk ratio for death after epilepsy surgery was 0.6 (95% CI
0.4–1.1;p = 0.08) compared with the control group. However, survival of epilepsy
surgery patients was lower than that of a matching general population (p < 0.001),
with a risk ratio for death of 6.2 (95% CI 3.1–12.6; p < 0.001).
 External comparison using population rates
 Does vasectomy increase the risk of prostate cancer?
 Design: Computerized record linkage study of cohort of men with vasectomy
and comparison of cancer rates with those in the whole Danish population;
manual check of all records of patients with testicular and prostate cancer
diagnosed within the first year of follow up.
 Subjects: Cohort of 73 917 men identified in hospital discharge and pathology
registers as having had a vasectomy for any reason during 1977-89.
# of men # of person years of follow up Observed # of cases STANDARDIZED
MORBIDITY RATIO
(95 % CI)
73917 482413 165 0.98 (0.8-1.4)
The incidence of cancer among cohort members was compared with the incidence in the
Danish population as a whole by using indirect standardisation for age and period, both in
five year intervals.  Observed cancer cases and person years at risk were counted from the
date of hospital discharge or pathological examination to the date of death or emigration or
31 December 1989, whichever occurred first

Measures of morbidity?
- Cumulative incidence?
- Incidence density rate?
http://www.bmj.com/content/309/6950/295.short
IN SUMMARY: INCIDENCE IN COHORT
STUDIES
• CUMULATIVE INCIDENCE:
All cases known to have occurred in the baseline cohort during
the duration of the study, divided by the number of individuals
enrolled in the study at baseline, per unit time. A risk measure.

• INCIDENCE DENSITY:
All cases known to have occurred in the total cohort during the
duration of the study, divided by the person years of observation
contributed by the total cohort per unit time. A rate measure.

In closed cohort, we can measure either cumulative incidence or

incidence density. In an open cohort, we can only measure
incidence density.
 IN SUMMARY: MEASURES OF
ASSOCIATION IN COHORT STUDIES

• RISK RATIO OR RELATIVE RISK

The ratio of two cumulative incidences

• RATE RATIO OR RELATIVE RISK

The ratio of two incidence densities

• RISK DIFFERENCES
The arithmetic difference between two cumulative incidences

• RATE DIFFERENCES
The arithmetic difference between two incidence densities