|
Abstract:
|
Double-blind, randomized controlled trials with continuous, normally-distributed treatment response ascertained at serial follow-up visits often specify the primary outcome as response at a single follow-up visit, typically the final study visit. Yet if early treatment benefit exists, and if anticipated, an average outcome over multiple timepoints may be a more powerful test. Allocation-blind procedures under an assumed family of alternative hypotheses can estimate (post-hoc) the potential treatment effect on outcome at each timepoint or for an outcome averaged over multiple timepoints, as well as estimate within-group pooled variance of those outcomes. We propose an adaptive analysis strategy to increase robustness of power compared to the traditional option of choosing an analysis a priori. Under a range of simulations and adaptive strategies, we confirm Type I error is controlled at the nominal level and evaluate power characteristics. Adaptive analysis strategies for a randomized trial of intervention for persons with aortic stenosis are considered, with the intentionally vague hypothesis that intervention reduces progression of stenosis assessed at 3 annual follow-up visits.
|
