Cohort Study Designs

Christopher Whalen, M.D., M.S. Department of Epidemiology and Biostatistics

Goals of Research
Evaluate new therapies and diagnostics Evaluate screening procedures Determine cause effect relationships Describe a disease, prevalence and natural history Determine a prognosis Medical Review

Research Study Design

Method of ascertaining subjects and measuring states or events in them to answer specific scientific questions.

Spectrum of Study Designs

Case Report Case Series Cross-sectional study Case-Control study Cohort study Randomized clinical trial Meta-analysis

Select Study Design to Match the Research Goals

Objective Description of disease or spectrum Design Case series or report Cross-sectional study Determine operating characteristics Cross-sectional of a new diagnostic test Describe prognosis Cohort study Determine cause-effect Compare new interventions Summarize literature Cohort study Case-control study Randomized clinical trial Meta-analysis

Cohort Study Design

An epidemiologic design in which the incidence of a disease (or condition) is compared among exposed and unexposed individuals

Cohort Study Design

Rationale
Cohort study designs evolved because of the need for information on the length of survival and the natural history of disease
Clinical and public health interest Incorporates the passage of time and includes the complexity of dynamic populations and changing exposures

Cohort Study Design

History
Origins
Graunt (17th century) used cross-sectional mortality data to reconstruct life history using life-table methods Farr (19th century) advanced the use of lifetable methods a an indicator of population health Insurance industry study 1870 1899

Cohort Study Design

History
Tuberculosis studies (20th century)
Natural history of tuberculosis as treated in sanatoria, especially Trudeau Sanitarium, established age-specific mortality rates Compared life table from tuberculosis patients to life table for general population WH Frost is credited for the first use of cohort analysis to study the age- and birth-cohort effects of tuberculosis on mortality
Age, cohort and period effects

Cohort Study Design

History
Tuberculosis (20th century)
WH Frost performed the first retrospective cohort study in a cohort of 132 homes with tuberculosis
Used person-years to estimate attack rates

WH Frost initiated prospective cohort study of tuberculosis in Williamson county, Tennessee

Cohort Study Design

History
Prospective cohort studies
Chronic Disease Cohorts (20th Century) Framingham study of cardiovascular disease, 1948 Japanese atomic bomb survivors, 1946 British physician study, 1950s Colorado Plateau uranium miners, 1950s

Retrospective cohort studies

Aniline-dye occupational cohort, 1954

Cohort Study Design

History
Current Studies
Nurses Health study, 1976 to present British physician study Multi-center AIDS Cohort Study MACS, 1984 1999

Cohort Study Design

Exposed PAR S Study Group Unexposed Outcome PAR = S T = = Population at Risk Sampling design Elapsed time T No Outcome Outcome

Cohort Study Design

Types of Cohorts
Fixed Cohort
A group of individuals recruited and enrolled at a uniform point in the natural history of a disease or by some defining event Cohort does not take on new members after it is assembled Examples
Patients admitted to the ER with acute MI Survivors of Hiroshima bombings Children born to HIV-infected mothers

Cohort Study Design

Type of Cohorts
Open cohort
A group of individuals recruited and enrolled through a mechanism that allows for in and out migration of people Defined by characteristic other than disease, e.g., geographic location, administrative unit Dynamic population Examples
Framingham Study Kaiser Permanente

Fixed Cohort
Loss to follow-up Deaths

Open Cohort

Out -

Migrations

In Time

Migrations

Cohort Study Design

Directionality
Prospective
Direction of inquiry moves forward in time

Retrospective
Direction of inquiry moves back in time

Both are longitudinal! Direction of study moves forward with time.

Cohort Study Design

Directionality
Longitudinal

1990

2000 Retrospective Prospective

2010

Retrospective Cohort Study

Prospective Cohort Study

Study Population
Define Population at Risk using inclusion criteria Individuals with outcome of interest at time of screening and enrollment are not eligible for study Sub-clinical presentation of diseases may be present challenges in defining the cohort

Study Populations
Examples
Framingham study of cardiovascular disease
Individuals 30 62 years old in community at risk for disease Framingham, MA, 1948 to present

Framingham Study
Cohort Assembly
No. Men 3,074 2,024 No. Total Women 3,433 6,507 2,445 4,469

Random Sample Respondents

Volunteers
Respondents free of CHD

312
1975

428
2,418

740
4,393

Volunteers free of CHD

Total free of CHD

307
2,282

427
2,845

734
5,127

Study Populations
MACS
Multi-Centered AIDS Cohort Study
Goal to elucidate the natural history of HIV/AIDS 5000 gay men, volunteers 5 cities in US 1984 1999 Extensive evaluations
Questionnaire Physical examination Laboratory testing Repository

Cohort Types
Representative cohort
May have low level of exposure

Enriched cohort
Enrich cohort with exposed individuals

Occupational cohort
MD Health study, RN health study, businessmen

Administrative cohorts, e.g., HMOs, PPOs Infectious disease outbreaks Institutional cohorts Etc.

Measuring Exposure
Content - Nature of the exposure; biologic mechanisms Quality
Continuous - e.g., serum cholesterol Periodic - e.g., cigarettes, sexual contacts Singular - e.g., nuclear exposure

Quantity
Continuous and periodic exposures must be quantified Dose-response relationship

Measuring Exposure
Measurements
Chart review Interview Blood tests or other specimens
Biomarkers

Other laboratory tests Sample storage

Measuring Exposure
Measuring exposure is one of the fundamental activities of a cohort study Exposure measurement must be comparable for all members of the cohort Carefully defined in advance of study Specific attention should be given to the accuracy and precision of proposed measurements
Pilot studies often needed

Outcome Definition
Primary outcome - the main event that will be related to the exposure
Failure-time outcomes
Death Disease occurrence

Repeated measures

Secondary outcomes - other events that are of interest and may corroborate the findings of the main outcome

Follow-up
Completeness and non-participation
90% rule of thumb

All subjects must have an equal likelihood for detecting the outcome Disease ascertainment must be comparable between the exposed and unexposed subjects
Number of visits Reasons for additional evaluations

Follow-up mechanisms
Active Passive

Follow-up
Passive Surveillance
Hospitals Disease Registries Clinics or physician offices Surveillance systems, e.g., National Death Index, CDC reportable conditions

Active surveillance
Systematic evaluations for outcome of interest Regular time intervals In all study subjects

Regardless of active or passive surveillance, the persons evaluating subjects must be blinded to exposure status

Cohort Study Design

Potential Biases
Detection bias
Subjects with exposure are more (or less) closely evaluated for outcome Blinding to exposure status

Information bias
The quality of information is different between exposed and unexposed subjects

Non-response and loss to follow-up

Selective loss of exposed (or unexposed) persons

Cohort Study Design

Exposed PAR S Study Group Unexposed Outcome PAR = S T = = Population at Risk Sampling design Elapsed time T No Outcome Outcome

Relative Risk
Disease Exposed A No Disease B A+B

Unexposed

C
A+C

D
B+D

C+D
N

Incidenceexp= A/A + B

Incidenceunexp = C/C + D

Relative Risk
a Incidence exposed a b RR c Incidence unexposed cd

Risk Difference
RD Incidence Incidence unexposed exposed

c a c d a b

Analytic Methods
Life-table methods Failure-time methods
Cox proportional hazards

Longitudinal data analysis

Repeated measures analysis

Poisson regression analysis

Cohort Study Design

Challenges in Analysis
Type of comparison groups
Historical controls vs. concurrent controls Internal vs. external controls

Time dependent effects

Changing exposures over time

Confounding
A variable related to both the exposure and the outcome may interfere with the interpretation of the relative risk

Strengths and Weaknesses

Strengths
Useful for rare exposures Multiple effects of single exposure Temporal relationship between exposure and outcome Ascertainment bias minimized Direct measurement of incidence

Weaknesses
Inefficient for rare diseases Expensive Requires excellent followup Losses to follow-up can invalidate the study