Power Calculations Conditioned on a Given Effect Size: Can We Do Better?

• Activity Number: 342
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Peter Hu*+ and Ih Chang
• Companies: Bristol-Myers Squibb and Bristol-Myers Squibb
• Keywords: effect size ; expected power ; prior ; posterior ; joint density ; probability of success

Abstract

Uncertainty of the effect size estimated from previous small studies, for the underlying true effect size characterized by both the location and scale parameters, may markedly inflate the estimated probability of success (PoS) in rejecting the null hypothesis at the design phase for the next trial. Further to the approach suggested by Spiegelhalter et al (2004), we propose a PoS assessment framework based on the expected posterior power given the observed data, in the sense that the uncertainties on the estimates for both the location and scale parameters for effect size are taking into considerations simultaneously. Comparisons of this PoS assessment framework with the one assuming only the location parameter unknown will be made. Several scenarios, in which a study may be under- or over-powered due to relying on power calculation conditioned on the effects estimated from 1 or 2 previously observed small data, will be demonstrated using simulation. How PoS may be helpful in these scenarios for making better informed decisions for the next trial will be illustrated.