Sequential Rematched Randomization and Adaptive Monitoring with the Second-Generation p-Value to increase the efficiency and efficacy of Randomized Clinical Trials

Abstract

Randomized Trials are considered the gold standard in establishing the benefit of an intervention with the overall societal benefit of knowledge gained outweighing the collective costs in time, money, human resources, etc. (Johnston, et al. 2006). However, trial costs may limit sample sizes and trials that end pre-maturely – before providing a clear clinical conclusion – leave important clinical questions unanswered. Sequential Rematched Randomization extends Sequential Matched Randomization (Kapelner and Krieger 2014) and provides greater covariate balance and increases the effective sample size. Sequential Rematched Randomization is able to achieve without model based adjustments nearly the same efficiency as fully-adjusting for baseline covariates using the form of the true model generating the outcomes. Better and finer baseline covariate balance aids in exploration of personalized medicine by providing more balance subgroups. Adaptive monitoring design that utilizes the Second Generation p-value (SGPV, Blume, et al. 2018) follows studies until either ruling out trivial effects or ruling out actionable effects. By requiring effects to be non-trivial, as opposed to simply non-null, it controls the probability of making a point-null type I error and reduces the false discovery probability. SGPV adaptive monitoring is very easy to implement; however, estimating the operating characteristics for a given study design is challenging. Thus, we introduce easy to use software to estimate the operating characteristics of the adaptive monitoring design and explore the impact of various design choices. Based on extensive simulations, we provide practical recommendations to the adaptive monitoring design to control the type I error and reduce the risk of an inconclusive study when outcomes are not observed immediately relative to enrollment.