Comparison of Generalized Models for Analysis of Standardized Incidence Rates of Recurrent Events

• Activity Number: 565
• Type: Topic Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): John C Evans*+ and Vivek Pradhan and Jian Huang and Wen Ding
• Companies: Boston Scientific Corporation and Boston Scientific Corporation and Boston Scientific Corporation and Boston Scientific Corporation
• Keywords: Recurrent events ; Standardized incidence rates ; Patient years ; Varying rate parameter

Abstract

FDA has requested the standardized incidence rate of recurrent events per 100 patient-years for recent annual progress reports on mandated post-market studies. We consider two case studies, one involving a single device, and the other comparing two devices. Evans et al (2009) considered the distribution of the number of recurrent events per subject as either Poisson or Negative Binomial. They assumed that either events were randomly distributed over time or the rate parameter was constant over time. However, when the recurrence of events depends on the number and times of previous events, this can lead to a varying rate parameter. We will consider departures from these assumptions and compare the goodness of fit of counts to the Poisson and Negative Binomial distributions with other models, including the Generalized Poisson, Double Poisson, Zero-Inflated Poisson, Zero-Inflated Negative Binomial, Poisson Inverse Gaussian, and Generalized Negative Binomial. We will compare results of analyses of standardized incidence rates based on these models.