Efficient Debiased Estimation of Heterogeneous Treatment Effect in Observational Studies

• Activity Number: 438 - Modern Statistical Learning Methods for High-Dimensional Biomedical Data: Treatment Heterogeneity and Data Integration
• Type: Topic-Contributed
• Date/Time: Thursday, August 12, 2021 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Waverly Wei* and Chong Wu and Jingshen Wang
• Companies: University of California, Berkeley and Florida State University and UC Berkeley
• Keywords: semiparametrics; high-dimension; causal inference; statistical methodology ; heterogeneous treatment effect; genomics

Abstract

This talk concerns the efficient debiased estimation for heterogeneous treatment effect given high-dimensional observational data. In this challenging environment, the conventional doubly robust estimator may generate biased and unstable estimates due to the finite sample constraint. We propose an efficient debiased estimator which has strong finite sample control through a novel bias correction term, and it attains the semiparametric efficiency bound by identifying a constrained class of likelihood. This method also consists of a one-step updating procedure which reduces the computational cost. We demonstrate the merit of this method with UK Biobank data to estimate the heterogeneous treatment effect of lifestyle interventions on cardiovascular disease risk, and the treatment subgroups are defined by high-dimensional covariates. Our method is able to detect treatment heterogeneity among subgroups and generate unbiased treatment effect estimates when the sample size is limited. Based on the real data analysis, we demonstrate the potential of using the efficient debiased estimator with high-dimensional observational data to inform precision medicine decisions.