Causal machine learning for high-dimensional mediation analysis using interventional effects mapped to a target trial

2025-04-22

AI Summary

Traditional parametric methods for high-dimensional mediation analysis exhibit limited applicability under target trial mappings. To enhance identifiability and modeling flexibility along the exposure–mediator–outcome pathway, this paper proposes a causal machine learning estimation framework. It innovatively defines three classes of target trial intervention effects and constructs two estimators: a one-step estimator and a partially targeted minimum loss estimator (TMLE), both achieving √n-consistency, efficiency, and multiple robustness. The methodology integrates efficient influence functions, first-order bias correction, and partial TMLE techniques. Applied to the Longitudinal Study of Australian Children, the framework successfully quantifies the causal effect of adolescent hypertension mediated by inflammatory and metabolic biomarkers downstream of childhood overweight. Empirical results demonstrate its validity and robustness in high-dimensional real-world data.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Causal mediation analysis examines causal pathways linking exposures to disease. The estimation of interventional effects, which are mediation estimands that overcome certain identifiability problems of natural effects, has been advanced through causal machine learning methods, particularly for high-dimensional mediators. Recently, it has been proposed interventional effects can be defined in each study by mapping to a target trial assessing specific hypothetical mediator interventions. This provides an appealing framework to directly address real-world research questions about the extent to which such interventions might mitigate an increased disease risk in the exposed. However, existing estimators for interventional effects mapped to a target trial rely on singly-robust parametric approaches, limiting their applicability in high-dimensional settings. Building upon recent developments in causal machine learning for interventional effects, we address this gap by developing causal machine learning estimators for three interventional effect estimands, defined by target trials assessing hypothetical interventions inducing distinct shifts in joint mediator distributions. These estimands are motivated by a case study within the Longitudinal Study of Australian Children, used for illustration, which assessed how intervening on high inflammatory burden and other non-inflammatory adverse metabolomic markers might mitigate the adverse causal effect of overweight or obesity on high blood pressure in adolescence. We develop one-step and (partial) targeted minimum loss-based estimators based on efficient influence functions of those estimands, demonstrating they are root-n consistent, efficient, and multiply robust under certain conditions.

Problem

• Estimating interventional effects for high-dimensional mediation analysis
• Developing robust causal machine learning estimators for target trials
• Assessing mediator interventions to mitigate disease risk in exposed

Innovation

• Causal machine learning for mediation analysis
• Interventional effects mapped to target trials
• Multiply robust estimators for high-dimensional mediators