Home/Insights/Methodology reference/Cohort study methodology
Cohort study methodology
A cohort follows people forward from exposure to outcome. It is the observational design that gets closest to a causal claim, and the one most often undone by who left.
The structure
A group defined by exposure status is followed over time, and outcomes are counted as they occur. Because exposure is recorded before the outcome happens, a cohort establishes temporality — the thing a cross-sectional study cannot do.
Prospective cohorts recruit and then follow, so the investigator controls what is measured. Retrospective cohorts assemble the cohort from records already created, which is faster and cheaper but limited to whatever was recorded at the time.
What it measures
Cohorts yield incidence: new cases per person-time at risk. From that come relative risk, hazard ratios and absolute risk differences. Absolute measures matter for interpretation — a doubling of a rare risk is still a small absolute change, and reporting only the relative figure overstates it.
The threats that matter most
Loss to follow-up. The central threat. If those lost differ in both exposure and outcome, the result is biased in an unpredictable direction. Report how many were lost, and compare them with those retained.
Confounding. Exposure was not randomly assigned, so exposed and unexposed differ systematically. Adjustment handles measured confounders only; residual confounding by unmeasured factors always remains possible and should be said.
Immortal time bias. A period during which the outcome could not by definition occur is misclassified as exposed time, manufacturing an apparent benefit. It arises when exposure is defined by something that happens after follow-up begins.
Competing risks. Where another event prevents the outcome, standard survival methods overestimate incidence.
In practice
Fix, flag or query.
The distinction that keeps an edit defensible: what an editor may change silently, what must be shown to the author, and what only the author can answer.
- Report: numbers lost to follow-up, and how they differed from those retained.
- Report: absolute risks alongside relative ones.
- Acknowledge: residual confounding — adjustment is not control.
Reporting standards
- STROBE Statement, cohort study checklist
- ROBINS-I, for risk of bias in non-randomised studies of interventions
Reporting guidelines named here are hosted by the EQUATOR Network. Writing to the checklist for your design before submission is the cheapest available improvement to a methods section.
Related