A Tutorial on Applying Propensity Score Methods for Characterization of Treatment Effects on Patient Outcomes Using a Medical Claims Database

• Activity Number: 259 - SPEED: Missing Data and Causal Inference Methods, Part 2
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 3:05 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Ryan Ross* and Megan Caram and Paul Lin and Min Zhang and Bhramar Mukherjee
• Companies: University of Michigan and Institute for Health Policy and Innovation, University of Michigan Medical School and Institute for Health Policy and Innovation, University of Michigan Medical School and University of Michigan and University of Michigan
• Keywords: claims data; causal inference; propensity score; prostate cancer

Abstract

Medical insurance claims are becoming an increasingly common data sources to answer a variety of questions in biomedical research. While comprehensive in terms of longitudinal characterization of disease on potentially large number of patients, these datasets need to be repurposed for conducting research, as they are not originally designed for population-based research. Along with complex selection bias and missing data issues, these studies are purely observational, which limits effective understanding of therapeutic or non-therapeutic interventions and characterization of the treatment differences between groups being compared. Several methods have been developed to better estimate causal treatment effects, often utilizing the propensity score. This paper offers some practical guidance to researchers in using propensity methods for estimating causal treatment effects on several types of outcomes common to medical studies, such as binary, count, time to event and time varying outcomes. We provide a R-Markdown version of the paper with readily implementable code so that the paper can serve as a guided tutorial for practitioners.