Comparisons of Noninferiority Methods When the Control Group Rate Is Low (= 10%)

• Activity Number: 245
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Alison Pedley*+ and Joseph Massaro and Robert W. Tipping
• Companies: Merck and Boston University and Merck
• Keywords: noninferiority ; clinical trials ; binary ; farrington-manning ; confidence intervals ; proportions

Abstract

Confidence intervals (CIs) for the difference between two independent proportions are commonly used in randomized clinical trials to assess noninferiority of an experimental treatment to a control when the primary endpoint is binary. A number of methods are available for this CI calculation, such as the Wald Interval (normal approximation to the binomial). Given that the Wald and other methods are based on approximations, their performance when endpoint rates are extreme, yet realistic, is of interest. In this presentation, the results of a simulation study conducted to evaluate the performance of commonly used methods for assessing noninferiority (including the Wald Interval, the normal approximation with the Hauck Anderson continuity correction, and the Farrington-Manning method) in settings where the underlying probability of the event in the control group was low (= 10%) with reasonably small noninferiority margins (also = 10%) is reviewed. The type I error rates and power of each method were compared. Practical guidance is provided which generally recommends the Farrington-Manning approach due to its stronger control of the type I error rate and reasonable power.