Exact Inference for 3-Treatment, 3-Period, 6-Sequence Crossover Design

• Activity Number: 372
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Ching-Ray Yu* and Michael Riggs and Sam Weerahandi
• Companies: Pfizer Inc. and Pfizer Inc. and Pfizer Inc.
• Keywords: REML ; ML ; Exact inference ; Mixed model

Abstract

The crossover design is a special class of repeated measures design wherein all or some of the subjects who are randomized to receive different sequences of treatments over a span of 2 or more time periods. In the early phase studies, crossover designs are employed primarily to produce more efficient analyses with smaller sample sizes (relative to parallel designs). Over the last 20 years, it has become standard practice to analyze crossover study data in a Mixed Model (MM) framework, with/ without repeated measures, within time periods. However, the widely used REML and ML based inferences underlying the MM frame work are based on asymptotic properties which may not hold in small sample situations. In this talk, we focus on the 3-treatment, 3-period, 6-sequence crossover design and present a theoretical framework for deriving the tests with generalized p-values and confidence intervals of treatment effects based on exact probability statements that are valid for any sample size. The relationship of the proposed exact parametric approach to more familiar nonparametric exact methods will be discussed, briefly. A simulation study has revealed that the exact test has greater power t