Power Analyses Based on Negative Binomial Models for Two Period Cross-Over Trials Involving Overdispersed Count Data

• Activity Number: 31
• Type: Contributed
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Mallik Rettiganti*+ and Haikady Nagaraja
• Companies: University of Arkansas for Medical Sciences and Ohio State University
• Keywords: parametric tests ; count data ; crossover trials ; negative binomial

Abstract

Cross-over trials are quite common in clinical trials and methods for analyzing count data from cross-over trials are well documented assuming the Poisson distribution provides a reasonable fit. However, such count data are generally overdispersed and the negative binomial (NB) distribution is used as an alternative. We model count data from cross-over trials using a NB distribution and examine likelihood based parametric tests such as likelihood ratio test, Rao's score test, Wald tests as well as nonparametric alternatives to test for the treatment effect. We compare the Type I error rates and power of these tests using simulation and propose the best tests under different parametric scenarios. We propose sample size estimates for crossover trials using these tests and compare them to sample sizes obtained using nonparametric alternatives.