Key Facts

Cohort studies are powerful observational designs that follow groups of people over time to determine whether exposure is associated with outcome. They are the strongest observational design for establishing causality because they establish temporal sequence.

Design Types

• Prospective cohort: recruits participants, measures exposure, follows forward in time to outcome; highest quality observational evidence
• Retrospective cohort: identifies cohort from historical records; exposure and outcomes already occurred; faster but dependent on data quality
• Ambidirectional: combines prospective and retrospective elements

Key Measures

• Incidence rate: events per person-time at risk
• Relative risk (RR): incidence in exposed / incidence in unexposed
• Attributable risk (AR): incidence in exposed − incidence in unexposed
• Population attributable risk (PAR): proportion of disease in population attributable to exposure

Strengths and Limitations

• Can study multiple outcomes from single exposure
• Can measure incidence directly (unlike case-control)
• Can assess temporal relationship
• Suitable for common exposures
• Limitations: expensive, time-consuming (prospective), attrition bias, confounding (cannot randomise)

Bias in Cohort Studies

• Attrition bias: differential loss to follow-up between exposed and unexposed
• Information bias: differential accuracy of outcome ascertainment
• Healthy worker effect: occupational cohorts may be healthier than general population
• Confounding: addressed through matching, stratification, or multivariable regression

Landmark Cohort Studies

• Framingham Heart Study (1948-present): identified hypertension, hypercholesterolaemia, smoking, diabetes, obesity as CVD risk factors
• British Doctors Study (Doll & Hill, 1951-2001): established smoking-lung cancer causal link
• Whitehall Studies (1967-present): social gradient in health and mortality
• Nurses' Health Study (1976-present): women's health, HRT, diet, lifestyle
• UK Biobank (2006-present): 500,000 participants; genetics, lifestyle, health outcomes
• Millennium Cohort Study: child development and health in UK-born children

Clinical Application

• Identifying risk factors for disease (cardiovascular risk factors, cancer aetiology)
• Evaluating prognosis (outcomes after diagnosis)
• Assessing harm (drug side effects, occupational exposures)
• Informing screening programme design

Critical Appraisal (CASP Cohort Checklist)

1. Did the study address a clearly focused question?
2. Was the cohort recruited in an acceptable way?
3. Was exposure accurately measured?
4. Was outcome accurately measured?
5. Were important confounders identified and controlled for?
6. Was follow-up long enough and complete?
7. What are the results (RR, CI)?
8. How precise are the results (confidence intervals)?
9. Do you believe the results?
10. Can results be applied to your population?

Addressing Confounding in Cohort Studies

• At design stage: restriction (limit to subgroups), matching
• At analysis stage: stratification (Mantel-Haenszel method), multivariable regression (logistic, Cox)
• Propensity score matching: statistical method to create comparable groups from observational data
• Sensitivity analysis: assess how robust findings are to unmeasured confounding

Cohort Study vs Other Designs

• Choose cohort when: exposure is common, multiple outcomes of interest, need to measure incidence, need temporal sequence
• Choose case-control when: disease is rare, limited resources/time
• Choose RCT when: evaluating treatment, ethical to randomise, need strongest evidence

• Cohort studies are essential for establishing long-term prognosis and natural history of disease
• They identified the major modifiable risk factors for cardiovascular disease, cancer, and diabetes
• Loss to follow-up >20% significantly threatens internal validity
• Large cohort studies (UK Biobank) increasingly used for pharmacoepidemiology and genetic epidemiology
• Mendelian randomisation uses genetic variants as 'natural experiments' within cohort data to strengthen causal inference

Relative Risk Interpretation

Attributable Risk Calculations