Cohort

A cohort study follows a defined group of people — the cohort — who share a common characteristic or exposure, tracking what outcomes develop over time. In a prospective cohort, participants are enrolled and then followed forward; in a retrospective cohort, researchers reconstruct the same structure from records that already exist.

Cohort studies sit between case reports and randomised trials in the evidence landscape. Unlike a case report, they can generate incidence rates — how often an outcome occurs in a population — and can follow large numbers of people long enough to detect effects that take months or years to appear.

What this design can establish

A well-conducted cohort study can establish that an exposure and an outcome occur together and, critically, that the exposure came first. Temporal sequence — knowing the exposure preceded the outcome — is something a snapshot survey cannot provide.

When a cohort is large and the follow-up long, the design supports estimates of risk magnitude: how much more often an outcome appears in exposed people than in unexposed ones. It is also well-suited to detecting rare but serious outcomes that trials cannot afford to wait for.

Cohort data is especially useful for studying populations that clinical trials routinely exclude — people with multiple conditions, people using substances in uncontrolled real-world settings, or people whose use patterns span years rather than weeks.

What it cannot

A cohort study cannot establish causation, and the inference readers most often draw — that because heavy users had worse outcomes, heavy use caused those outcomes — is not warranted by the design alone.

The problem is confounding. People who are exposed differ from those who are not in ways researchers may not have measured: prior health, other substances used, socioeconomic circumstances, reasons for use. An apparent association between a substance and an outcome may reflect those background differences rather than the substance itself.

Analytic methods can adjust for known confounders but cannot correct for ones that were never recorded. The design is also limited when the exposure is rare, the follow-up is short relative to the condition's natural history, or participation patterns make the cohort unrepresentative of the broader population it is meant to describe.