MODEL-ASSISTED INFERENCE FOR COVARIATE-SPECIFIC TREATMENT EFFECTS WITH HIGH-DIMENSIONAL DATA

Covariate-specific treatment effects (CSTEs) are heterogeneous treatment effects across subpopulations defined by certain selected covariates. In this study, we consider marginal structural models in which CSTEs are represented linearly using a set of basis functions of the selected covariates. We develop a new approach for high-dimensional settings to obtain not only doubly robust point estimators of CSTEs, but also model-assisted confidence intervals, which are valid when the propensity score model is specified correctly, but the outcome regression model may be misspecified. With a linear outcome model and subpopulations defined by discrete covariates, both the point estimators and the confidence intervals are doubly robust for CSTEs. In contrast, the confidence intervals from existing highdimensional methods are valid only when both the propensity score and the outcome models are specified correctly. We also establish several asymptotic properties of the proposed point estimators and the associated confidence intervals. The results of our simulation studies demonstrate the advantages of the proposed method over existing methods. Lastly, we apply the proposed method to a large clinical data set on psoriasis from a national registry in China, the Psoriasis Center Data Platform, to explore the effects of biologics versus those of conventional therapies across different subpopulations.