Beyond Bells and Whistles in Bayesian Sample Size Computation

• Activity Number: 350
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Mani Lakshminarayanan*+ and Xin Zhao
• Companies: Merck & Co., Inc. and Merck & Co., Inc.
• Keywords: Bayesian Power ; Decision Analysis ; Predictive Probability ; Conditional Power

Abstract

Computing sample size is one of the key efforts that need to take place at the design stage of any statistical experiment including clinical trial development. Suppose our primary interest is to compare two treatments based on a difference (D) in a pre-specified endpoint. Frequentist or classical statistical approach to sample size determination starts with finding the number of patients required with a prespecified power (for a fixed Type I error rate) to detect the difference D. All these assumptions that need to be made before estimating a sample size have contributed to the general agreement that a formal sample size computation in clinical trials (or other statistical experiments) is very subjective. A Bayesian approach provides a natural framework for incorporating any subjective information including the prior distributions for all the parameters. A Bayesian sample size computation provides multitude of options that are easily adaptable to underlying formulation of the problem. In this presentation, we will properties of approaches for both Bayesian sample size computation and reestimation and contrast them with frequentist alternatives.