Analysis of Number of Recurrent Events with Early Withdrawals

• Activity Number: 669
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Kim Hung Lo*+ and Bin Zou and Jiandong Lu
• Companies: Johnson & Johnson and Johnson & Johnson and Johnson & Johnson
• Keywords: recurrent events ; Poisson regression ; negative binomial regression

Abstract

The number of recurrent events is often used in evaluating the efficacy of a treatment for diseases manifested by non-fatal events. It is typically analyzed as count data, for which Poisson regression and CMH row mean score test are commonly used. Negative binomial regression was also employed to allow varying event rate across patients. This article evaluates the performance of these methods in comparing treatment groups with respect to the event rate when early withdrawal occurs due to various reasons. Simulation studies are carried out to evaluate the statistical properties of various imputation methods for these analyses and to compare the performance of these methods under the assumption of MAR or MNAR. Poisson regression with over-dispersion is recommended because it is the most robust method with respect to the parameter estimate and hypothesis testing in the scenarios under study