1.16—Study design for health risk factors

Judge a study by asking who was sampled, how exposure and outcome were measured, whether comparison groups are appropriate, and whether the time sequence can support the claim. A large sample cannot rescue a biased design.

Check representativeness and sample size; define variables consistently; control or stratify confounders; use blinding or randomisation where possible; and report drop-outs and measurement uncertainty.

A questionnaire finding that diet and CVD are associated may suffer recall bias and self-selection. A prospective cohort with repeated measurements gives stronger temporal evidence but still cannot control every confounder.

Validity concerns whether the design supports the intended inference; reliability concerns consistency. Ethical limits may make a randomised exposure impossible, so conclusions should match the design.

An observational study can support an association without proving causation. Do not treat “peer reviewed” or “statistically significant” as substitutes for examining the method.