Conditional Type I Error and Conditional Power of Historical Control Studies

• Activity Number: 550
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Bingzhi Zhang* and Hui Quan
• Companies: Sanofi and Sanofi
• Keywords: historical control ; sample size ; conditional power ; conditional type I error

Abstract

Historical control is one of the alternative approaches adopted when a randomized controlled trial is not feasible and when one or more trials are previously conducted using the same control therapy. Several methods have been applied to calculate the sample size of the experimental group in the new trial. Zhang et al. [2010] compared these methods regarding the conditional power and type I error given historical data, and proposed a new test statistic and sample size calculation method. This method controls the conditional type I error and power in a way such that, given any p and q, the conditional type I error is less than the nominal level ? with probability p and the conditional power is greater than a fixed ? with probability q. However, only marginal properties of conditional type I error and power were discussed separately, while it's of more interests to reveal the characteristics of the joint distribution of the two. In this presentation, we provide formulas and simulation results that compare different methods with respect to the likelihood of achieving a desired power while having the type I error controlled, and propose an approach to achieve the specified likelihood.