Estimating Group Average Treatment Effects in High-Risk Populations with Unobserved Target Membership
Abstract
Estimating treatment effects within a prespecified high-risk population is challenging. When risk status is unobserved or the target population is small, analyses restricted to high-risk individuals can be unstable or inefficient. We propose a risk-score-weighted soft-targeting (RSW) estimator for group average treatment effects in high-risk populations. It accommodates two settings: when high-risk status is observed, the membership probability can be estimated within the analysis sample; when unobserved, it can be obtained from a reference model using covariates in the analysis sample. RSW uses these probabilities to represent the target population, learns a doubly robust treatment effect in the full sample, and aggregates a low-dimensional treatment-effect surface over the risk-weighted target distribution. We establish pointwise asymptotic normality and an influence-function-based variance estimator. In simulations with continuous and binary outcomes, RSW had similar bias and lower variance than two hard-targeting alternatives. Sensitivity analyses showed that the gain persisted when treatment-effect sufficiency was violated, although RSW bias increased under strong violations. We applied the method to the All of Us study to estimate effects of non-GLP antidiabetic and antihypertensive medication use on mild cognitive impairment (MCI)-free time restricted at 48 months within comorbidity-defined high-risk subgroups. We estimated the risk score in a nonoverlapping internal reference sample and applied it to the outcome-analysis sample. This illustrates how soft targeting can improve precision while retaining an interpretable high-risk target population.
Related
- Estimating Average Treatment Effects with Projected Instrumental Variables
- Imputation is all you need: double robustness, semiparametric efficiency, and automatic covariate balance for estimating the average treatment effect
- Estimating the average treatment effect under limited overlap via Polynomial Approximation and Extrapolation
- On the Impossibility of Estimating the Average Mixing Time
- Sharp Bounds on Continuous Treatment Effects under Unobserved Confounding
BibTeX
@misc{zhang2026estimating,
title = {{Estimating Group Average Treatment Effects in High-Risk Populations with Unobserved Target Membership}},
author = {Shubo Zhang and Yinjun Zhao and Donglin Zeng and Yuanjia Wang},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.30709},
eprint = {2610.30709},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.30709},
}
RIS
TY - UNPB
TI - Estimating Group Average Treatment Effects in High-Risk Populations with Unobserved Target Membership
AU - Shubo Zhang
AU - Yinjun Zhao
AU - Donglin Zeng
AU - Yuanjia Wang
PY - 2026
DA - 2026/10/22
PB - arXiv
DO - 10.48550/arxiv.2610.30709
UR - https://arxiv.org/abs/2610.30709
AB - Estimating treatment effects within a prespecified high-risk population is challenging. When risk status is unobserved or the target population is small, analyses restricted to high-risk individuals can be unstable or inefficient. We propose a risk-score-weighted soft-targeting (RSW) estimator for group average treatment effects in high-risk populations. It accommodates two settings: when high-risk status is observed, the membership probability can be estimated within the analysis sample; when unobserved, it can be obtained from a reference model using covariates in the analysis sample. RSW uses these probabilities to represent the target population, learns a doubly robust treatment effect in the full sample, and aggregates a low-dimensional treatment-effect surface over the risk-weighted target distribution. We establish pointwise asymptotic normality and an influence-function-based variance estimator. In simulations with continuous and binary outcomes, RSW had similar bias and lower variance than two hard-targeting alternatives. Sensitivity analyses showed that the gain persisted when treatment-effect sufficiency was violated, although RSW bias increased under strong violations. We applied the method to the All of Us study to estimate effects of non-GLP antidiabetic and antihypertensive medication use on mild cognitive impairment (MCI)-free time restricted at 48 months within comorbidity-defined high-risk subgroups. We estimated the risk score in a nonoverlapping internal reference sample and applied it to the outcome-analysis sample. This illustrates how soft targeting can improve precision while retaining an interpretable high-risk target population.
ER -