Massive Data, Massive Parallelization, and the Bayesian Self-Controlled Case Series

• Activity Number: 155
• Type: Invited
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: IMS
• Author(s): Marc A. Suchard*+ and Trevor Shaddox and David Madigan
• Companies: UCLA and UCLA and Columbia University
• Keywords: MCMC ; parallelization ; GPU ; drug safety

Abstract

Following a series of high-profile drug safety disasters in recent years, many countries are redoubling their efforts to ensure the safety of licensed medical products. Large-scale observational databases such as claims databases or electronic health record systems are attracting particular attention in this regard, but present significant methodological and computational concerns. The self-controlled case series (SCCS) offers one successful approach to associating rare adverse events with medical product usage. However, no one has previously attempted fully Bayesian SCCS implementations that provide measures on uncertainty on these myriad associations, primarily owing to the outrageous computational cost involved in fitting the model to databases encompassing millions of patients using thousands of different products. We describe novel massive parallelization techniques using graphics processing units to simulate from the SCCS model posterior distribution. For this construction, we explore several different Markov chain Monte Carlo-based sampling procedures and discuss their relative merit in terms of efficient, scalable parallelization and application to massive data.