Frequentist Conditional Power, Bayesian Predictive Power, and Their Applications

• Activity Number: 408
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Aijun Gao*+ and Fanni Natanegara and Karen Price
• Companies: PharmaNet/i3 and Eli Lilly and Company and Eli Lilly and Company

Abstract

In clinical trial monitoring, Frequentist conditional power is often used to make a decision whether to stop or continue a trial in the planned interim analysis time point. Similarly Bayesian predictive power can be used to make a decision whether to stop a clinical trial or quantifying what is going to happen in a trial from any time point on, given the currently available data information. Both conditional power and predictive power can be used for sample size re-estimation in adaptive trials. A binary variable (any event of interest) was assumed as the main outcome for a randomized controlled clinical trial with two treatment groups. Both conditional power and predictive power were calculated at the interim analysis for the trial with available observation. The beta-binomial distribution was used in Bayesian predictive power calculation. The conditional power and the predictive power were also used for the sample size re-estimation. All the computations were implemented using R. The R code could be used for general cases. In addition, the equivalence between three different conditional power formulae from Dmitrienko, Denne and Proschan was also addressed.