Analysis of Two-Period Crossover Design with Binary Response Data

• Activity Number: 547
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): James Lee and John D.S. Hwang*+ and Chyi-Hung Hsu
• Companies: Daiichi Sankyo Pharma Development and B.R.S.I. and Janssen Research & Development
• Keywords: Two-period crossover ; Binary data ; Exact McNemar Test ; Multinomial distribution

Abstract

Most of the papers in the analysis of a two-period cross-over design considered continuous response data only. We shall consider the binary response data case. Gart (1969, 1971) proposed an analysis based on Cox (1958 c). They all employed logistic modeling and assume all the random variables to be independent. In reality, the two observations of a pair are often correlated. The appropriate test to compare the two treatments of a matched pair is the Exact McNemar Test, which involves the joint probabilities P01, P10 of discordant pairs. Our approach will be modeling these joint probabilities. No assumption of independence is imposed. Gart and Cox considered type 1 error only. We shall provide power and sample size for study planning purpose based on Multinomial to Binomial transformation of Hwang and Lee (2009), Hwang, Lee, and Hsu (2012).