Testing Noninferiority Through Generalized p-values

Abstract

Testing for noninferiority of a new experimental drug compared to a standard reference drug has generally been carried out through statistical tests assuming equal variances. In this article, we present a generalized p-value approach for assessing noninferiority that gives exact tail probabilities when variances are not equal between treatments. Two generalized test functions are constructed corresponding to the fixed-margin and synthesis approaches of retaining a fraction of the reference treatment’s effect relative to placebo in historical trials. Monte Carlo simulations are used to compute the p-value. Performance of these test functions are evaluated and results indicate that the generalized test function for the fixed-margin retention performs well in terms of maintaining the Type I error probabilities. Its power also increases and approaches 1 as both the sample size and the parameter denoting the fraction of the effect of the reference drug with respect to placebo increase. Furthermore, confidence intervals (CI’s) can be easily constructed using duality between CI’s and accepting or rejecting noninferiority.

Keywords: :

• Generalized test function
• Noninferiority
• Sample size
• Type I error