Propensity score weighting for causal subgroup analysis

Abstract

A common goal in comparative effectiveness research is to estimate treatment effects on prespecified subpopulations of patients. Though widely used in medical research, causal inference methods for such subgroup analysis (SGA) remain underdeveloped, particularly in observational studies. In this article, we develop a suite of analytical methods and visualization tools for causal SGA.First, we introduce the estimand of subgroup weighted average treatment effect and provide the corresponding propensity score weighting estimator. We show that balancing covariates within a subgroup bounds the bias of the estimator of subgroup causal effects. Second, we propose to use the overlap weighting (OW) method to achieve exact balance within subgroups. We further propose a method that combines OW and LASSO, to balance the bias-variance tradeoff in SGA. Finally, we design a new diagnostic graph-the Connect-S plot-for visualizing the subgroup covariate balance. Extensive simulation studies are presented to compare the proposed method with several existing methods. We apply the proposed methods to the patient-centered results for uterine fibroids (COMPARE-UF) registry data to evaluate alternative management options for uterine fibroids for relief of symptoms and quality of life.

Covariate adjustment in subgroup analyses of randomized clinical trials: A propensity score approach

“…However, explicitly writing out S in this definition helps clarify the role of propensity scores for subgroup analysis. 29,30 In RCTs, the treatment allocation process is completely controlled and the true propensity score equals the randomization probability of a specific trial (i.e. e(X, S) = 0:5 for all adjustment and subgrouping variables under a balanced treatment allocation design).…”

“…Similar to the causal subgroup analysis in observational studies, 30 we use the Ha´jek estimator, weighted difference-in-means in subgroup r, for estimating subgroup average treatment effect in RCTŝ…”

“…When S i is a subset of X i , then (X i S i ) should include the full interactions between S i and the rest of the covariates in the adjustment set. Yang et al 30 showed that overlap weighting exactly balances the mean of the covariates with interaction terms in each subgroup as well as the overall sample…”

“…For propensity score weighting estimators, the propensity score model specification is key. Because including subgroup-covariate interactions in the propensity score model leads to improved balance of covariates within subgroups, and is necessary to achieve exact subgroup covariate balance via overlap weighting, 30 we consider two types of logistic propensity score models. The first is the main-effects logistic model, which includes the subgroup indicator and the covariate main effects; the corresponding design matrix is (X, S).…”

“…24 We evaluated a cohort of 1724 patients with the outcome of 6-min walk distance measured at 3 months. The investigators pre-specified subgroups based on 10 baseline covariates.Figure 4 shows a Connect-S plot 30 of representative subgroups that have imbalanced covariates between treatment arms. Each row represents the subgroups of interest, and each column represents the adjustment variables.…”