Incidence

Incidence measures how often a new case of a specific outcome begins — how many people in a defined population first develop it within a stated time window. It counts starting events, not existing ones.

The measure appears in two forms. Cumulative incidence expresses the proportion of a group that develops the outcome over a fixed period — a probability. Incidence rate divides that count by the total time the population was under observation, which allows comparisons across studies with different follow-up lengths.

Readers encounter incidence claims in clinical trials, cohort studies, and pharmacovigilance reports — in any source reporting how often something starts, rather than how widespread it currently is.

What this design can establish

When an exposed group and a comparison group are followed under the same conditions, a difference in incidence rate is the raw material for causal inference. It is not causation itself — it is the signal that something may be driving the difference. How far that inference can go depends on the study design carrying the figure.

Incidence estimates also support statements about temporal trends. If the annual incidence of an outcome is tracked across successive years in the same population, a rising or falling figure is real evidence that new cases are beginning at a different pace — which can motivate investigation into what changed.

In drug safety work, adverse-event incidence in a treated group provides the initial signal for harm review. A higher rate in one arm of a trial is a legitimate basis for flagging a pattern, even before mechanism or causation is established.

What it cannot

Incidence cannot, by itself, explain why a rate difference exists. Confounding — systematic differences between groups that also predict the outcome — is always a competing explanation, and the incidence count alone cannot rule it out. That task falls to the design carrying the study.

It cannot establish how long an outcome persists once it appears, nor how common it is in the population at any given moment. Those questions belong to prevalence. A substance with low incidence of dependence but slow recovery could carry substantial prevalence — and treating one figure as a proxy for the other is the error readers are most likely to make.

An incidence figure from a monitored trial population also does not automatically describe what happens outside the study. Selection effects, close surveillance, and the particular conditions of a clinical setting all shape the count. A rate observed in that context tells you something started at a certain frequency there — it does not establish the same frequency in unmonitored real-world use.