Cohort Study

Cohort study (from the Latin cohors, denoting a military unit of Roman legionaries marching together) is a longitudinal observational study design in epidemiology and clinical biostatistics in which a defined group of individuals free of the outcome of interest at baseline is classified according to exposure to a suspected risk factor, protective intervention, or biomarker, and followed forward through time to compare the incidence of new disease events across exposure strata. Because exposure status is measured before disease onset, cohort studies establish clear temporal sequence—the primary requirement among the Bradford Hill criteria for causal inference—and enable the direct calculation of absolute incidence rates, cumulative risk, and multiple health outcomes from a single exposure without relying on the retrospective exposure recall required in a case–control study.

Temporal Relationship Between Investigator and Follow-Up

Epidemiologists classify cohort studies according to when the baseline exposure assessment and longitudinal follow-up occur relative to the investigator's initiation of the study, a distinction that can be visualized on a two-dimensional age-by-calendar-time Lexis diagram:

1. Prospective (Concurrent) Cohort Study: The investigator defines the cohort in the present, collects standardized baseline measurements (questionnaires, physical examinations, and biobanked blood or tissue specimens), and follows participants into the future for years or decades as incident disease outcomes occur.
2. Retrospective (Historical or Non-Concurrent) Cohort Study: Both the baseline exposure and the subsequent follow-up period have already taken place in the past before the study begins, reconstructed from pre-existing occupational, military, or electronic health registries. Despite looking at historical records, the analytical direction remains forward in time from baseline exposure to subsequent disease incidence—distinguishing a retrospective cohort from a case–control study, which samples by outcome status.
3. Ambispective Cohort Study: Combines historical registry follow-up from a past exposure date up to the present with ongoing prospective clinical surveillance into the future (used extensively in occupational radiation and chemical toxicology cohorts).

Landmark Epidemiological Cohort Benchmarks

Cumulative Incidence (Risk) vs. Incidence Density (Person-Time Rate)

In a closed cohort followed over a fixed time interval Δt with minimal loss to follow-up, the cumulative incidence (CI or Risk) in exposure group e is the proportion of baseline at-risk individuals (N_e) who develop the outcome (A_e):

In dynamic cohorts where participants enter at staggered calendar dates, migrate, or die from competing causes, simple proportions are biased by unequal observation durations. Investigators instead compute the incidence rate (IR, or incidence density) by dividing incident cases A_e by the exact summed person-time at risk (PT_e = ∑ Δt_i) contributed by each participant up to disease onset, censoring, or study termination:

Relative and Attributable Risk Metrics

Comparing exposed (e = 1) and unexposed (e = 0) cohorts yields both multiplicative (etiological) and additive (public health impact) effect measures:

• Risk Ratio (Relative Risk, RR): RR = CI₁ / CI₀ = (A₁ / N₁) / (A₀ / N₀).
• Incidence Rate Ratio (IRR): IRR = IR₁ / IR₀ = (A₁ / PT₁) / (A₀ / PT₀).
• Risk Difference (Attributable Risk, RD): RD = CI₁ − CI₀ (or IR₁ − IR₀), whose reciprocal 1 / |RD| gives the Number Needed to Harm (NNH) or Number Needed to Treat (NNT).
• Population Attributable Fraction (PAF): PAF = [ P_e (RR − 1) ] / [ 1 + P_e (RR − 1) ], where P_e is the population prevalence of exposure, quantifying the fraction of total disease incidence in the population that would be prevented if the exposure were eliminated.

When exact time-to-event data with right-censoring are available, unadjusted event-free survival trajectories are estimated non-parametrically using the Kaplan–Meier estimator, and covariate-adjusted hazard ratios are estimated semi-parametrically via the proportional hazards model or Poisson/negative-binomial generalized linear models.

Loss to Follow-Up (Attrition Bias) and the Healthy Worker Effect

Although cohort studies avoid retrospective recall bias, their internal validity is threatened by differential loss to follow-up (attrition bias): if participants who drop out of the exposed group have a systematically different risk of developing the outcome than those lost from the unexposed group, both RR and IRR become biased. Additionally, occupational cohorts that compare employed industrial workers against the general census population frequently exhibit the healthy worker effect—a form of selection bias in which active workers display 15%–30% lower all-cause mortality because chronically ill individuals are excluded from employment—requiring internal dose-stratified comparisons among workers within the same cohort.

What is the difference between a retrospective cohort study and a case–control study?

Both designs examine historical data that already exist when the investigator begins the study, but they differ in how participants are sampled: a retrospective cohort study selects disease-free subjects based on their past exposure status and follows their records forward to measure absolute disease incidence and relative risk, whereas a case–control study selects subjects based on their final disease status (cases vs. controls) and looks backward at prior exposure odds.

How does immortal time bias arise in observational cohort studies?

Immortal time bias occurs when a cohort study defines exposure based on an event that happens after baseline enrollment (such as filling a prescription 60 days after hospital discharge) and misclassifies the waiting period prior to that prescription as "exposed person-time." Because a patient had to survive outcome-free to fill the prescription, those 60 days are "immortal," artificially depressing the event rate in the exposed group unless exposure is modeled properly as a time-varying covariate in a proportional hazards model.

1. Doll, R., & Hill, A. B. (1954). "The mortality of doctors in relation to their smoking habits: a preliminary report." British Medical Journal, 1(4877), 1451–1455. https://doi.org/10.1136/bmj.1.4877.1451
2. Kannel, W. B., Dawber, T. R., Kagan, A., Revotskie, N., & Stokes, J. (1961). "Factors of risk in the development of coronary heart disease—six year follow-up experience: the Framingham Study." Annals of Internal Medicine, 55(1), 33–50. https://doi.org/10.7326/0003-4819-55-1-33
3. Colditz, G. A., Manson, J. E., & Hankinson, S. E. (1997). "The Nurses' Health Study: 20-year contribution to the understanding of health among women." Journal of Women's Health, 6(1), 49–62. https://doi.org/10.1089/jwh.1997.6.49
4. Doll, R., Peto, R., Boreham, J., & Sutherland, I. (2004). "Mortality in relation to smoking: 50 years' observations on male British doctors." BMJ, 328(7455), 1519. https://doi.org/10.1136/bmj.38142.554479.AE
5. Rothman, K. J., Greenland, S., & Lash, T. L. (2008). Modern Epidemiology (3rd ed.). Lippincott Williams & Wilkins. https://books.google.com/books?isbn=9780781755641