Verify risk-set sampling in a nested case-control study

Reconstruct each case-time risk set

A nested case-control study samples controls from people at risk when a case occurs within a defined cohort. Control status is therefore a role at a particular time, not necessarily a permanent label. Begin with the full cohort entry, exit and event records needed to reconstruct eligibility at every sampled case time.

The primary methods paper Nested case-control sampling without replacement distinguishes the standard design from modifications to control sampling. In the standard design, a control can be selected again at a later case time and can subsequently become a case. Do not remove those records as duplicates without checking the protocol.

Preserve the sampling occasion

The analysis table needs both person identity and risk-set identity. Check that controls were event-free and under observation at the case time, met any matching requirements and were not selected using future information. Keep the random selection procedure, seed or reproducible sampling record, and any deviations from the intended number of controls.

Exposure should be measured using the information window appropriate to that case time. A control selected early and again later may legitimately have a different exposure value. A single final exposure column can therefore misrepresent the sampled comparisons.

Hypothetical example

In a fictional cohort, person Q is eligible as a control when a case occurs on day 40. Q later becomes a case on day 90. Keeping Q in the day-40 control set is consistent with standard risk-set sampling: Q was still at risk at that earlier time.

By contrast, excluding Q from the earlier control pool because the analyst already knows the later event uses future information. Selecting controls only from people who remain event-free through the study end would create a different sampling design.

Audit the resulting comparison

Tabulate cases, sampled controls, unique people and repeated selections separately. Check sets affected by unavailable specimens or missing covariates. Confirm that the estimator and uncertainty method match the actual design, including any nonstandard sampling without replacement across occasions.

Explain the control-sampling scheme when interpreting the exposure estimate; do not automatically use the rare-disease explanation associated with a different case-control design. The matched case-control analysis guide covers matched-set modelling, while this audit focuses on whether the sets were constructed correctly. Archive synthetic risk-set tests and a sampling flow record so the design can be independently reconstructed.

Source verification: 6 November 2026. Examples are hypothetical.