Utility of Regression Splines for Propensity Score Adjustment in Post Market Safety Analyzes with Multiple Treatments

• Activity Number: 447 - Recent Advances in Propensity Score Methods for Observational Studies with Multiple Treatments
• Type: Invited
• Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Elande Baro* and Yuxi Tian and Rongmei Zhang and Yuqin Wei and Mao Hu and Jiemin Liao and Sandia Akhtar and Michael Wernecke and Jeffrey Kelman and David Graham
• Companies: US Food and Drug Administration and University of California Los Angeles and Food and Drug Administration and Acumen LLC and Acumen LLC and Acumen LLC and Acumen LLC and Acumen LLC and Centers for Medicare & Medicaid Services and Acumen LLC
• Keywords: propensity score; observational; multiple treatments; splines; post market; confounding

Abstract

Investigation of post market safety frequently involves comparing several treatment options in observational settings. FDA has used inverse probability of treatment weighting (IPTW) to adjust for potential confounders in several studies. However, the method is susceptible to extreme weights, a problem that can be amplified with multiple treatments. Regression on a spline function of the propensity score (PS) has shown good performance in two-treatment settings. However, this method has rarely been used at FDA and its performance with three or more treatments is unknown. Thus, this research aims to assess the potential utility of spline functions of PS for post market safety analyses. First, we illustrate the motivation for our research and challenges in conducting analyses with multiple treatments using a FDA study that examined the comparative safety and effectiveness of four anticoagulant therapies in Medicare claims data. Then, we summarize the results of our simulation study comparing IPTW and spline functions of PS as adjustment methods in three-treatment settings. Last, we discuss about benefits, limitations, and potential uses of spline PS analyses in regulatory settings.