Comparing causal inference methods for point exposures with missing confounders: a simulation study

Causal inference methods based on electronic health record (EHR) databases must simultaneously handle confounding and missing data. In practice, when faced with partially missing confounders, analysts may proceed by first imputing missing data and subsequently using outcome regression or inverse-probability weighting (IPW) to address confounding. However, little is known about the theoretical performance of such reasonable, but ad hoc methods. Though vast literature exists on each of these two challenges separately, relatively few works attempt to address missing data and confounding in a formal manner simultaneously. In a recent paper Levis et al. (Can J Stat e11832, 2024) outlined a robust framework for tackling these problems together under certain identifying conditions, and introduced a pair of estimators for the average treatment effect (ATE), one of which is non-parametric efficient. In this work we present a series of simulations, motivated by a published EHR based study (Arterburn et al., Ann Surg 274:e1269-e1276, 2020) of the long-term effects of bariatric surgery on weight outcomes, to investigate these new estimators and compare them to existing ad hoc methods. While methods based on ad hoc combinations of imputation and confounding adjustment perform well in certain scenarios, no single estimator is uniformly best. We conclude with recommendations for good practice in the face of partially missing confounders.