Sample-size Calculations for Noninferiority Clinical Trials Based on Poisson Regression Models

• Activity Number: 119
• Type: Contributed
• Date/Time: Monday, August 9, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Patrick Darken*+ and Peter Lane
• Companies: ALTANA Pharma U.S. and GlaxoSmithKline
• Keywords: equivalence trial ; noninferiority trial ; Poisson regression ; sample-size calculation ; generalized linear model ; power

Abstract

Poisson regression is commonly used for the analysis of count data, especially when mainly low counts and zeroes are anticipated. In clinical trials to be analyzed by Poisson regression, a complication for determining sample-size requirements is that when comparing the rates of an event between treatments, differences are usually considered in relative rather than absolute terms. An additional complication is that the occurrence of events from the same subject may be correlated: there is then a need to allow for overdispersion in the calculation of variance estimates, as the data can no longer be treated as arising from a true Poisson process. This can occur, for example, when counting the number of exacerbations of a disease-like asthma experienced by patients during a trial. The generalized linear model provides a convenient framework for these analyses, but care is needed in choosing between Pearson or deviance estimates of overdispersion or using GEE. We present formulae for the calculation of sample sizes for noninferiority clinical trials whose primary outcome will be analyzed using Poisson regression, along with details of the derivation and validation using simulation.