Cohort construction · Implementation

Cohort Attrition Funnel

Step-by-step record of how the source population is filtered to the final analytic cohort, with the count remaining and the reason removed at each step.

Download SVG Open code and full worked example

When to use it

In every database study, as the transparency backbone of cohort definition. It is the RWE analogue of a CONSORT flow diagram and is expected by STaRT-RWE and most journals and HTA reviewers.

How to read it

Each tier shows N remaining; the side annotation shows N removed and why. Large, unexplained drops or a tiny final cohort relative to the source flag selection that may threaten generalizability.

Worked example

A claims study filters a 1,000,000-member source population down to the analytic cohort through five successive criteria, recording the count retained and the count excluded (with reason) at each step.

Source 1,000,000 → ≥1 diagnosis 142,000 → new initiators 38,500 → ≥12mo enrollment 24,100 → no exclusions 19,700 → final 18,900.

Result: The analytic cohort is 18,900 (1.9% of source). The largest drop is the diagnosis requirement (858,000 removed); the new-user washout removes a further 103,500 prevalent users — the step most relevant to avoiding prevalent-user bias.

Produced by

Database feasibility & attrition funnel New-user (incident-user) design

Reference: Schneeweiss S, Rassen JA, Brown JS, et al. Graphical Depiction of Longitudinal Study Designs in Health Care Databases. Ann Intern Med. 2019;170(6):398-406.