B1178
Title: Leveraging external data in Bayesian adaptive platform designs
Authors: Alejandra Avalos Pacheco - Harvard Medical School (Mexico) [presenting]
David Rossell - Universitat Pompeu Fabra (Spain)
Steffen Ventz - Dana Farber Cancer Institute (United States)
Lorenzo Trippa - Dana Farber Cancer Institute (United States)
Abstract: There is growing interest in trial designs that incorporate data from real-world observational studies or from previously completed trials with the goal of increasing power and reducing the sample size of clinical studies in comparison with Randomized Controlled Trials (RCT). However, if the outcome distributions of the external and internal data differ, the integration of external data may lead to biased treatment effects estimates, reduced power or increased type I error rates. We introduce a novel design that leverages external data via a Bayesian model averaging approach. The design adjusts for confounding and satisfies a set of constraints on the study's operating characteristics required by regulators. We compare two methods to perform the final analyses of the trial: i) a test based on weighted averages of p-values; ii) a non-parametric test based on permutations. We illustrate the performance of our proposed hybrid design in simulation studies based on data from real phase II and III trials.
