Misspecification of Event Rates and Sample Size Re-Evaluations in Noninferiority Trials

• Activity Number: 355
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Hwasoon Kim*+ and Jeff M Szychowski
• Companies: University of Alabama, Birmingham and University of Alabama at Birmingham
• Keywords: non-inferiority trial ; sample size ; power ; simulation

Abstract

In clinical trials with binary outcomes, an experimental treatment is non-inferior to the control if its probability of a successful outcome is lower than the control arm by no more than a predetermined margin. Sample sizes are selected to achieve desired power to reject the null (inferiority) when examining the lower bound of the confidence interval for treatment differences. However, underestimation (overestimation) of the true event rates may lead to an overpowered (underpowered) study. Consider, for example, a test of two treatments evaluated at the lower 95% confidence interval bound, assumed 75% event rate and 10% non-inferiority margin. With 300 in each group, the study has 80% power to detect non-inferiority. However, if the actual event rates are >85%, the study quickly exceeds 90% power. Lower event rates similarly result in lower power. Thorough understanding of consequences in these cases is essential and early detection is necessary for corrective actions. We detail the issues related to misspecification of event rates in non-inferiority trials and investigate interim monitoring strategies and sample size re-evaluations to control type I error and power.