Epidemiologic methods for psychiatry — structured clinical viva

Newspaper claim. Admission counts are service-use rates, not community incidence or prevalence. They depend on bed supply, thresholds, pathways, and help-seeking. Community burden needs population-based surveys (period prevalence for current caseload; incidence cohorts for new-onset rates) with clear case definitions.[1]

Case-control OR 2.2 → “causes 50%”. Case-control designs efficiently study associations for psychosis onset but face selection and information bias and cannot alone prove causation.[2][3] An OR is not a PAF. A PAF near 50% would require a valid causal RR, correct exposure prevalence, and minimal bias/confounding — assumptions that must be stated and often fail.[4] Reverse causation and shared vulnerability (confounding) are live alternatives; temporality and triangulation matter (Hill weight-of-evidence).[5][2]

Confounding vs effect modification. Confounding: common cause distorts the exposure–outcome association — control by design/analysis if measured.[2][8] Effect modification: effect differs across strata — report strata rather than “adjust away.” DAGs help avoid adjusting mediators or colliders.[8]

School screening. At prevalence under 1%, even Sn/Sp of 85% generate many more false positives than true positives among screen-positives → low PPV, stigma and assessment burden.[6] Screening needs programme criteria and pathways, not a questionnaire alone.

Follow-ups. Person-time denominators for incidence rates with unequal follow-up.[1] STROBE improves transparent reporting of observational methods — it does not certify causal truth.[7]