Generalized causal mediation and path analysis: Extensions and practical considerations.

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Genetics

Background:

• Causal mediation analysis decomposes treatment effects via intermediate variables.
• Existing methods struggle with sequential or multiple mediators and computational challenges.

Purpose of the Study:

• To present an extended causal mediation and path analysis methodology.
• To introduce the gmediation R package for advanced mediation analysis.
• To provide practical guidelines for mediation analysis and Monte Carlo error reduction.

Main Methods:

• Developed a novel methodology accommodating sequential and multiple mediators at each stage.
• Extended to handle generalized linear models for various outcome types.
• Incorporated support for unsaturated models and clustered data.
• Provided guidelines for decomposition and reference group multiplier selection.

Main Results:

• The gmediation R package implements the extended methodology.
• Demonstrated application in a cohort study analyzing socioeconomic status and dental caries.
• Illuminated the contribution of biological and behavioral paths in the observed effect.

Conclusions:

• The new methodology and gmediation package offer a robust tool for complex causal mediation analysis.
• Provides enhanced capabilities for understanding indirect effects in various research settings.
• Facilitates more accurate path-specific effect estimation and interpretation.