Biostatistics — structured clinical viva

NNT. Absolute benefit for response = 0.50 − 0.35 = 0.15 → NNT = 1/0.15 ≈ 7 patients treated for 12 weeks for one extra responder. Prefer this language to relative slogans.^[1]

Relative framing. Non-response rates are 50% vs 65%; RRR of non-response = (0.65 − 0.50)/0.65 ≈ 23%, not 43%. The press line may be mixing relative increase in response (0.15/0.35 ≈ 43%) with "risk reduction" wording — demand absolute rates.^[1]

p-value. p = 0.008 is compatibility with H0 under the model, not P(drug useless) and not clinical importance by itself.^[2]

HR CI. HR 0.72 with CI 0.48–1.08 includes 1 → not conventionally significant; possible benefit remains but so does null — do not claim proven relapse prevention from this secondary alone; absence of significance is not proof of no effect if underpowered.^[3][4][6]

Biomarker. Sn/Sp look good in a high-prevalence mood clinic; PPV will fall in low-prevalence community screening — recompute with Bayes or 2×2 at local base rate before adoption.^[5]