The Bayesian Bootstrap in a Predictive Power Analysis

Abstract

In the planning and design of new clinical trials, calculation of the required sample size and power is a critical part of the process. Power calculations are usually based on quantities estimated from analysis of historical data and are therefore subject to uncertainty. In many cases this is addressed by sensitivity analysis, but simple sensitivity analysis gives an incomplete picture of the uncertainty involved in estimates of power. Here we describe an analysis of historical clinical trial data using the Bayesian Bootstrap, which gives - by generation of the predictive power distribution - a fully probabilistic description of the uncertainty in a power calculation.