Predictive Posterior Power Estimation for Sample Determination

• Activity Number: 414
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Marc Sobel*+ and Ibrahim Turkoz
• Companies: Temple University and Janssen
• Keywords: power ; sample size determination ; treatment effects ; particle filter

Abstract

When designing a confirmatory clinical trial, it is often the case that necessary information (e.g., assumed population variance) is not fully available and information that is used is often subject to a high degree of uncertainty. At interim points of the trial, re-evaluation of the assumptions may be beneficial. We evaluate uncertainty in single stage clinical trials by designing sample size adjustments before unblinding. Sample size adjustments frequently improve the chance that the trial will reach a definitive conclusion. We design sample size adjustments by estimating the power of hypothesis tests designed to distinguish the presence of treatment effects. Former power estimation using future samples have made use of the EM algorithm to estimate parameters needed for this purpose; we use a completely Bayesian framework. Using this framework we construct HPD intervals for power estimation and demonstrate their accuracy. We also evaluate uncertainty in multiple stage medical trials in which further adjustments are required based on newly acquired observations. It is shown that Bayesian particle filter models can be advantageously used in this setting.