Causal mediation analysis with multiple mediators.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Empirical research often aims to decompose exposure effects into multiple pathways.
• Existing causal inference methods primarily address single mediators or mediators analyzed collectively.
• Complex biological systems frequently involve numerous sequential mediators, posing analytical challenges.

Purpose of the Study:

• To provide counterfactual definitions for path-specific effects in multiple mediator settings, accounting for sequential mediation.
• To discuss identification assumptions and propose sensitivity analysis for complex mediation pathways.
• To bridge the gap between single mediator theory and multiple mediator practice in empirical research.

Main Methods:

• Developed counterfactual definitions for path-specific estimands in multiple, sequential mediator models.
• Outlined identification assumptions required for estimating these effects.
• Illustrated methods with data on alcohol consumption, systolic blood pressure, body mass index, and gamma-glutamyl transpeptidase.

Main Results:

• Demonstrated that multiple decomposition strategies exist for path-specific effects in multiple mediator models.
• Highlighted the strong assumptions necessary for identification.
• Proposed sensitivity analysis as a tool when assumptions are not fully met.

Conclusions:

• Decomposing effects through multiple, sequential mediators is complex but feasible with appropriate methods.
• Researchers should be aware of the strong assumptions required and consider sensitivity analyses.
• The study offers practical approaches for analyzing complex mediation in biological and empirical research.