A0358
Title: Measuring racial disparities in emergency general surgery via approximate balancing weights
Authors: Luke Keele - University of Pennsylvania (United States) [presenting]
Abstract: The basic research design for the study of racial disparities in surgical care uses statistical methods to compare black and white patients that are similar in terms of baseline characteristics. Differences in outcomes are interpreted as due to differences in care as a function of race. We develop a new form of approximate balancing weights for this purpose. Approximate balancing weights are generalizations of inverse propensity score weights that are designed to directly target covariate balance in the estimation process. This class of weighting methods solve a convex optimization problem to find a set of weights that target a specific loss function. The approximate balancing weights we develop rely on a hyper-parameter that governs the bias-variance trade-off in weighting. We also develop a data-driven method for hyper-parameter selection and review how outcome modeling can be applied for additional bias reduction. We conduct a series of simulation studies to understand bias reduction properties. We apply this method to study racial disparities in emergency general surgery. In one comparison, we only compare patients in terms of risk factors. In a second analysis, we compare patients on risk factors within the same hospital. We find that racial disparities in outcomes persist when we compare whites and blacks with similar risk factors. Racial disparities are eliminated when we compare similar patients within the same hospital.
