A0378
Title: Contrasting and interpreting test decisions induced by information borrowing in hybrid-control clinical trial designs
Authors: Silvia Calderazzo - German Cancer Research Center (DKFZ) (Germany) [presenting]
Abstract: When designing a novel clinical trial, some external information about the control and/or treatment effect is typically available. Borrowing of such external information is often desired in order to improve the trials' efficiency, and Bayesian designs with informative prior distributions offer a natural framework to achieve this aim. One major concern in this context is the potential for heterogeneity between the current and external data sources, which can significantly increase both the chances of making erroneous test decisions and estimation error. Robust dynamic prior choices are thus often employed: Such priors gradually discount external information based on the observed heterogeneity between the two information sources. In this context, it is of interest to understand how external information borrowing in general, and a chosen robust approach in particular, affects the trial's final test and estimation decisions as well as the long-run performance of the design. The focus is on hypothesis testing procedures in the context of two-arm hybrid-control trial designs, and we analytically investigate and report on the connections between the Bayesian and the frequentist paradigm. Graphical and analytical tools are provided to compare test decisions induced by different approaches. Moreover, it is shown how test decisions can be adapted to satisfy frequentist constraints and/or data-adaptive costs of type I and type II errors under a fully Bayesian analysis.
