Propensity Score Analysis for Subgroup Effects with Correlated Covariates

• Activity Number: 124 - Causal Inference and Observational Health Policy Studies
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Shan-Yu Liu* and Bo Lu and Chunyan Liu and Edward Nehus and Maurizio Macaluso and Mi-Ok Kim
• Companies: UCSF and The Ohio State University and CCHMC and CCHMC and CCHMC and University of California San Francisco
• Keywords: effect modification; propensity score; causal inference; doubly robust estimation

Abstract

In large observational studies, individuals may respond differently to the treatment with some showing the intended effect but others showing little. Estimating subgroup effects is important to identify characteristics of individuals who may benefit from the treatment. In practice, factors defining subgroups may be correlated with one another. Consequently, the observed subgroup effect may be spurious due to the correlation with other variables. We examine propensity score approaches to the subgroup effect estimation with correlated covariates through an extensive simulation study. Confounding effects are adjusted through either matching or doubly robust estimation. We compare marginal analyses that evaluate the subgroup one by one and conditional analyses that include all potential subgroup variables simultaneously. When the number of subgroup variables is big, we also explore using variable selection strategy to automate the process. We use a kidney transplant study as a real life example to illustrate different methods being compared.