Efficiently transporting causal direct and indirect effects to new populations under intermediate confounding and...

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Interventions can yield varied effects across different sites.
• Current mediation estimators are limited to single, binary mediators.
• Understanding these variations is crucial for effective public health strategies.

Purpose of the Study:

• To propose novel nonparametric estimators for transported interventional (in)direct effects.
• To extend existing methods to accommodate multiple, high-dimensional mediators.
• To address limitations of current estimators in complex mediation analysis.

Main Methods:

• Developed multiply robust, efficient, and asymptotically normal nonparametric estimators.
• Incorporated data-adaptive estimation for nuisance parameters.
• Focused on scenarios with multiple mediators and a single binary intermediate variable.

Main Results:

• The proposed estimators effectively handle multiple, high-dimensional mediators.
• Demonstrated the ability to explain differences in treatment effects across sites.
• Showcased potential for predicting treatment effects in target sites.

Conclusions:

• Novel estimators provide a powerful tool for analyzing complex causal relationships.
• Facilitates a deeper understanding of treatment effect heterogeneity.
• Enables more accurate prediction of intervention outcomes in diverse settings.