A Noninferiority Test for a Matched Pair Design with an Ordinal Outcome with a Majority of Zeros

• Activity Number: 500
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Wanjie Sun*+
• Companies: FDA
• Keywords: Non-inferiority ; Matched Pairs Design ; Skewed ; Ordinal

Abstract

A non-inferiority test is proposed for a matched pair design where the test and active control products are applied to different body parts on the same subject at a number of visits. Endpoint is a highly skewed ordinal outcome with a majority of zeros. For such a highly-skewed multinomial distribution with a hierarchical level of within-subject correlation structure, parametric methods need to be robust to severe departure from normality. Since median response is 0 for both groups, it is not meaningful to compare median using traditional non-parametric methods. A non-parametric method is proposed to test whether the median paired difference of cumulative mean responses between groups exceeds a pre-specified maximum clinically irrelevant amount, i.e. a non-inferiority margin, among the discordant pairs. This tail comparison eliminates the zero median problem, and uses an exact test for small sample size. When no historical placebo measurement is available, non-inferiority margins are proposed based on the power and maximum significance level simulated for different sample sizes, effect sizes, variances, and alpha level. The proposed method is applied to real clinical trial data.