|
Abstract:
|
Two adaptive designs (ADs) for vaccine trials aimed at immune response titers are discussed: a Bayesian Emax model-based design, & one based on T-statistics from isotonic regression. These ADs are compared via simulation to non-adaptive designs for a vaccine Ph2 dose-finding trial. The simulated ADs had 4 cohorts of 50 or 150 patients each - 1st cohort randomized 0:1:0:1 or 1:1:1:1 to doses 1:2:3:4; other cohorts assigned per AD to target a low & a high response. Non-ADs randomized 200 or 600 patients equally to doses. Distribution of dose assignments & target dose estimates, SEs, & bias of estimated titer response at TRUE target doses were summarized. Results: (1) ADs assign more patients at or near doses with TRUE target responses - useful to provide better safety data; (2) ADs require more time than non-ADs & stratified analysis to account for inter-assay variability; (3) ADs estimate target doses & titers at target doses with similar precision as non-ADs; (4) design choice depends on TRUE dose-response curve & priority of objectives; (5) if range of shapes & locations of responses over dose-choices can be pre-identified, best design may be identifiable via simulation.
|
