A Simulation Study Comparing Methods for Adjusting Confounding in EMR Data with Real World Complex Heterogeneity in the Treatment Assignment

• Activity Number: 523
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Haibo Zhou*+ and Jichang Yu and Fei Zou and Xianchen (Jason) Liu and Richard J Willke
• Companies: The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and Pfizer Inc. and Pfizer Inc.
• Keywords: Electronic medical record ; Comparative effective research ; heterogeneity ; propensity score ; confounding

Abstract

The widespread implementation of electronic medical records (EMR) in clinical practice has created an enormous opportunity for conducting comparative effectiveness research (CER) for competing medical treatments and interventions. Unlike the randomized controlled trial (RCT) where the treatment is randomly assigned to ensure that there is no confounding of the baseline covariates when comparing the treatment effects in two groups, treatment selection in real world medical practice is a complex process that involves many factors that includes impacts of patients, doctors, and health systems. The current analytic approach relies on the assumption that no unmeasured confounders exist. It is unrealistic to expect EMR data, like insurance claims, to capture all these dynamics, as they are not designed to do so. Incompleteness in measuring the underlying treatment assignment process can create heterogeneity that is complex and potentially intractable, and could bias the estimate of the true treatment effects. We evaluated, through simulation study, the performance of several classes of estimators on the impact this heterogeneity.