Variable Selection and Estimation in Causal Inference Using Bayesian Spike and Slab Priors

• Activity Number: 118 - Emerging Challenges in Precision Medicine
• Type: Topic Contributed
• Date/Time: Monday, July 29, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): David Michael Vock* and Brandon Koch and Julian Wolfson
• Companies: University of Minnesota and University of Nevada Reno and University of Minnesota
• Keywords: causal inference; variable selection; precision medicine

Abstract

Unbiased estimation of causal effects with observational data requires adjustment for confounding variables related to both the outcome and treatment assignment. Standard variable selection techniques aim to maximize predictive ability of the model but ignore covariate associations with treatment and may not adjust for confounders weakly associated to outcome. We propose two methods that simultaneously consider models for both outcome and treatment using a Bayesian formulation with spike and slab priors on each covariate coefficient; the Spike and Slab Causal Estimator (SSCE) aims to achieve minimum bias of the causal effect estimator while Bilevel SSCE (BSSCE) aims to minimize its mean squared error. Simulations show SSCE can greatly reduce bias over a similar approach that does not consider a model for treatment, while BSSCE can substantially reduce MSE over a popular method that also models the outcome and treatment simultaneously. Both methods perform well with large numbers of covariates, even greater than the sample size. We illustrate SSCE and BSSCE by estimating the causal effect of vasoactive therapy versus fluid resuscitation on hypotensive episode length.