Masters candidate in Biostatistics, Jingxin Lei, will present:

“Bayesian Multisource Exchangeability Models for Combining Supplementary Weibull-Distributed Survival Data”

Plan B Adviser: Joe Koopmeiners

Abstract: Clinical trials on new medical treatments and therapies are expensive and time consuming. Incorporating information from supplemental and historical data sources has the potential to shorten the trial duration and reduce the costs. Multisource exchangeability models (MEMs) are dynamic Bayesian models for integrating supplemental sources which have been shown strong theoretical and small-sample properties for Gaussian distributed data. In this manuscript, we introduce MEMs for survival data that were assumed to follow the Weibull distribution. Besides the standard parameterization of Weibull distribution, we also demonstrate how MEMs can be used to facilitate borrowing on other functionals of survival (i.e. median survival time and the survival rate) without borrowing for nuisance parameters. Our simulation with one supplementary source showed that the mean square error (MSE) of MEMs was reduced to around 40% when compared with only using the primary data in the case that the two sources were assumed to be exchangeable, and was increased to 2-3 times when the sources were similar but not exactly the same. When only considering borrowing on functionals of survival and ignoring of the nuisance parameter, MEMs reduced around 50% of the MSE when the sources were exchangeable and doubled the MSE at the worst case. We also illustrate the application of MEMs using two colorectal cancer trials, which resulted in a 30% improvement in efficiency compared with only using the primary data.