Interventional Effects for Mediation Analysis with Multiple Mediators.

Area of Science:

• Causal Inference
• Epidemiology
• Biostatistics

Background:

• Traditional mediation formulas rely on strong assumptions, often violated when confounders are exposure-dependent.
• This limits counterfactual-based mediation analysis in complex settings like multiple or repeatedly measured mediators.

Purpose of the Study:

• To adapt and extend interventional (in)direct effects for an exact decomposition of total effect.
• To enable mediation analysis in multiple mediator settings with unknown structural dependence.

Main Methods:

• Adapting VanderWeele, Vansteelandt, and Robins' interventional (in)direct effects.
• Developing methods for exact total effect decomposition in multiple mediator models.
• Addressing scenarios with unknown causal relationships or unmeasured common causes between mediators.

Main Results:

• The proposed method achieves an exact decomposition of the total effect.
• It successfully extends mediation analysis to multiple, potentially correlated, mediators.
• The approach identifies path-specific effects even when the structural dependence between mediators is unknown.

Conclusions:

• The adapted interventional effects provide a robust framework for mediation analysis under weaker conditions.
• This extends causal inference capabilities to complex multiple mediator scenarios, enhancing understanding of exposure-outcome pathways.