Mediation analysis with time varying exposures and mediators

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Causal mediation analysis is crucial for understanding complex health relationships.
• Existing methods often struggle with time-varying exposures and mediators.
• Time-varying confounders present significant challenges in identifying causal pathways.

Purpose of the Study:

• To develop a general framework for causal mediation analysis with time-varying exposures and mediators.
• To address the identification of direct and indirect effects in the presence of time-varying confounders.
• To introduce the mediational g-formula as a unified approach.

Main Methods:

• Non-parametric identification of causal effects.
• Parametric implementation strategies.
• Weighting approach using marginal structural models.
• Development of the mediational g-formula.

Main Results:

• The mediational g-formula accommodates time-varying exposures, mediators, and confounders.
• It provides identified causal effects analogous to natural direct and indirect effects under specific conditions.
• The formula generalizes existing methods like Robins' g-formula and Pearl's mediation formula.

Conclusions:

• The mediational g-formula offers a robust and generalizable method for longitudinal causal mediation analysis.
• This approach enhances the causal interpretation of effect estimates from longitudinal models.
• It provides a unified framework for complex mediation scenarios in research.