All Times ET
Virtual
Sample Size Re-Estimation Using Bayesian Predictive Power Based on Weighted Test Statistic in Adaptive Design (305318)
Abhijoy Saha, Eli Lilly and Company
Victoria Soldatenkova, Eli Lilly and Company
*Meng Summer Xia, Eli Lilly and Company
Keywords: Sample size re-estimation, Clinical trial, Adaptive study design, Bayesian
The sample size of a clinical trial is determined by the magnitude of treatment effect. If there is uncertainty of this effect or if it is mis-specified at the design stage, an adaptive design with unblinded sample size re-estimation (SSR) can weigh in and effectively save a trial. Adaptive designs with SSR are supported by regulatory agencies when proper statistical methods are used and type I error is strictly controlled (FDA, 2019). Cui, Hung and Wang (CHW, 1999) proposed a test procedure by modifying the weights in traditional test statistics, so that the type I error is preserved at the target level. Wang (2007) applied Bayesian predictive power to incorporate variability in treatment effects without type I error adjustment, and is therefore used only for in-house exploratory studies. We propose a novel SSR method for time-to-event analysis using Bayesian predictive power based on the CHW test to incorporate both uncertainty and variability, while additionally controlling for the type I error. We demonstrate improvement of type I error control compared to Wang (2007) via simulations, and discuss performance measures to tune design parameters to select the optimal design.