Designing a Time-to-Event Bayesian Clinical Trial That Incorporates Historical Data

• Activity Number: 127
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Barry Eggleston*+ and Catellier Diane and Joseph Ibrahim
• Companies: RTI International and RTI International and University of North Carolina
• Keywords: Clinical Trial Design ; Bayesian Statistics ; Hierarchical Bayesian Models ; Piecwise Exponential Models ; Bayesian Clinical Trials

Abstract

Piecewise exponential models can be used to easily approximate semi-parametric time-to-event Cox models. As a consequence, piecewise exponential models can be extended easily to Bayesian modeling. Added benefits of Bayesian modeling of time-to-event data using piecewise exponential models is the possibility of smoothing out the piecewise exponential parameters and the ease of ncorporating historical data. In the context of clinical trials, the incorporation of historical data will allow for smaller sample sizes, yet the design of a time-to-event Bayesian trial that incorporates historical data must consider the benefit gained by the available data relative to the risk of compatibility issues between available data and future trial data. In this presentation, I will use simple simulations to illustrate the process of designing a Bayesian time-to-event clinical trial that studies Type I error, power, and MSE to aid in making choices about future trial sample size and prior distributions as well as assess the risk of historical data incorporation. For simulation, I created hypothetical clinical trial datasets and analyzed each dataset using ten-piece piecewise exponential Bayesia