When Measurement Mediates the Effect of Interest

2025-06-06

AI Summary

Health interventions—such as screening and prevention programs—often simultaneously affect both intervention coverage and health outcomes, yet outcomes are only observable among covered individuals, rendering the total causal effect nonidentifiable under standard assumptions. To address this, we propose the *counterfactual stratified effect* as a novel causal estimand that relaxes the conventional assumption of full observability. We develop a two-stage Targeted Minimum Loss-Based Estimation (TMLE) framework that jointly integrates counterfactual modeling, missing-data handling, and sensitivity analysis. This approach is the first to systematically resolve mediator-confounding and selection bias arising from coverage-dependent measurement. Simulation studies demonstrate that our method substantially improves unbiasedness and robustness in estimating the total causal effect compared with existing approaches, and reveals systematic bias in mainstream methods when coverage and outcome share a joint generative mechanism.

Technology Category

Machine Learning: Causal LearningReasoning under Uncertainty: CausalityIntelligent Robots: State Estimation

Application Category

User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systemsSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingSecurity and Privacy: Large-scale security measurements

Abstract

Many health promotion strategies aim to improve reach into the target population and outcomes among those reached. For example, an HIV prevention strategy could expand the reach of risk screening and the delivery of biomedical prevention to persons with HIV risk. This setting creates a complex missing data problem: the strategy improves health outcomes directly and indirectly through expanded reach, while outcomes are only measured among those reached. To formally define the total causal effect in such settings, we use Counterfactual Strata Effects: causal estimands where the outcome is only relevant for a group whose membership is subject to missingness and/or impacted by the exposure. To identify and estimate the corresponding statistical estimand, we propose a novel extension of Two-Stage targeted minimum loss-based estimation (TMLE). Simulations demonstrate the practical performance of our approach as well as the limitations of existing approaches.

Problem

• Address missing data in health promotion strategies
• Define total causal effect with counterfactual strata
• Estimate effects using extended TMLE method

Innovation

• Counterfactual Strata Effects for causal estimands
• Two-Stage targeted minimum loss-based estimation
• Handling missing data in health promotion strategies