Propensity Score Analysis with Partially Missing Covariates

• Activity Number: 629
• Type: Invited
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Bo Lu*+ and Robert Ashmead
• Companies: Ohio State University and Ohio State University
• Keywords: Propensity score ; missing ; sensitivity ; multiple imputation

Abstract

Propensity score based adjustment has become popular for analyzing observational data. In practice, however, observational studies often suffer from missing covariates, which presents some challenges on how to use propensity score effectively. Current literature focuses on multiple imputation (MI) based methods, which first impute the missing covariate under certain models, then make propensity score adjustment using the imputed data. Following Rosenbaum and Rubin's work (1984), we propose a propensity score matching strategy that matches both on observed covariates and missing data pattern. Under MAR, the matched pairs with missing covariates are just like the ones with complete covariates. Under MNAR, we use a sensitivity analysis approach to gauge the potential impact of missing data on the study conclusion. An extensive simulation study compares our method with the MI based methods.