Recurrent Event Analyses Illustrated in the Pivotal Exacerbation Study SPARK in the Respiratory Area

• Activity Number: 546
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Hua Li* and Paul Gallo and Richard Cook
• Companies: Novartis and Novartis Pharmaceuticals and University of Waterloo
• Keywords: Recurrent event data analysis ; marginal Cox model ; rate-based model ; semiparametric negative binomial model

Abstract

For many clinical trials in respiratory, the primary response is based on potentially recurrent exacerbations observed during a treatment period. Conventionally, logrank tests and Cox proportional hazards models have been used to compare treatments with regard to the time to the first exacerbation and a parametric time-homogeneous negative binomial model has been used to estimate the exacerbation rate. These analyses do not make use of potentially important information, namely the times of event occurrence, which can improve understanding of the exacerbation process and enhance understanding of the treatment benefits; one can also base analyses on methods which are robust to model misspecification. In this paper, we review a number of approaches reflecting recent advances in recurrent event methodology, including the marginal Cox model (Wei, Lin, and Weissfeld, 1989), rate-based models (Anderson and Gill, 1982; Lawless and Nadeau, 1995; Lin et al, 2000) and the semiparametric negative binomial model (Therneau and Grambsch, 2000). We demonstrate the application of these methods to the pivotal SPARK study conducted for approval of a drug for COPD. We discuss the strengths and limitations of the competing methods and discuss the interpretation of the findings from the various analyses before making recommendations on approaches for the design and analysis of future trials in COPD.