Illustration: =0.4%, =1.2% n =35 per-arm per-stage Do all - PowerPoint PPT Presentation

Illustration: δ =0.4%, σ =1.2% n =35 per-arm per-stage Do all experimental treatments share a common effect? E.g experimental arm estimate y i = µ + ǫ i where Var( ǫ i ) is variance exp arm 5 of y i exp arm 4 Use Cochran’s Q statistic to test null hypothesis. exp arm 3 exp arm 2 Here, p-value for Q is 0.5: No evidence to reject common effect exp arm 1 ( µ ) hypothesis control arm Pooled ‘fixed effect’ estimate for µ Fixed effect estimate ● justified −0.25 0.00 0.25 0.50 0.75 1.00 Compare pooled estimate’s % change in Hba1c Confidence intervals to that of control group and declare class effective if no overlap 1 / 3

Heterogeneity between experimental treatments Perhaps treatments don’t share a common effect? E.g experimental arm estimate y i = µ i + ǫ i where Var( µ i ) is exp arm 5 between arm variation exp arm 4 Between arm variation as a prop n of total variation: I 2 = 50% exp arm 3 exp arm 2 Q statistic p -value=0.09 exp arm 1 I 2 =50% Pooled ‘random-effects’ estimate for control arm µ (= mean of µ i s) arguably justified ● Fixed effect estimate likely to be very similar to fixed effect estimate Random effects estimate ● −0.25 0.00 0.25 0.50 0.75 1.00 Wider confidence interval to % change in Hba1c acknowledge extra uncertainty. Lower power, but arguably right model 2 / 3

Issues surrounding dropped treatments Assume arm 1 & 4 dropped at interim (after 35 patients) Dropped trials have less precise exp arm 5 estimates than kept trials Dropped at stage 1 exp arm 4 Should we exclude dropped arms exp arm 3 when estimating pooled effect? exp arm 2 Exclude : PROS: Remaining arm exp arm 1 results more homogeneous. Likely I 2 (all arms) =62% to opt for a fixed effect estimate. I 2 (kept arms) =17% control arm CONS: Remaining arms potentially ● all arms biased, throwing away information kept arms only ● Include : PROS: Using all available −0.25 0.00 0.25 0.50 0.75 1.00 information. CONS: Confidence % change in Hba1c interval may still be wider due to use of random effects model. 3 / 3