A1026
Title: Propensity score matching and stratification for multiple and ordinal treatments: Application to an EHR-derived study
Authors: Stacia DeSantis - University of Texas Health Science Center at Houston (United States) [presenting]
Abstract: Currently, methods for conducting multiple or ordinal treatments propensity scoring in the presence of high-dimensional covariate spaces that result from big data are lacking. The most prominent method relies on inverse probability treatment weighting (IPTW), which has limitations. We present a novel propensity scoring framework that uses the entire propensity score vector for multiple or ordinal treatments to establish a scalar balancing score that can achieve covariate balance in the presence of high-dimensional covariates. Specifically, we fit a one-parameter power function to the cumulative distribution function of the propensity score vector, resulting in a scalar balancing score that is used for matching and/or stratification. We present simulation results that show excellent performance in achieving covariate balance and estimating average treatment effects in the presence of multiple treatments. We then apply the approach to a study derived from electronic health records to determine the causal relationship between three different vasopressors and mortality in patients with non-traumatic aneurysmal subarachnoid hemorrhage. Results suggest that the method performs well when applied to large observational studies with multiple treatments that have large covariate spaces.