Med4way: a Stata command to investigate mediating and interactive mechanisms using the four-way effect decomposition

Area of Science:

• Causal inference
• Biostatistics
• Epidemiology

Background:

• Understanding the total effect of an exposure on an outcome is complex when mediation and interaction are present.
• Existing methods often assess mediation or interaction separately, potentially missing nuanced relationships.

Purpose of the Study:

• To introduce a novel four-way decomposition of causal effects into components of mediation and interaction.
• To present the Stata command 'med4way' for estimating these decomposed causal effects.
• To unify methods for attributing effects to interactions and assessing mediation.

Main Methods:

• Developed a four-way decomposition framework for causal effects.
• Introduced the 'med4way' Stata command for estimating causal contrasts.
• Utilized parametric regression models accommodating various outcome and mediator types.

Main Results:

• The 'med4way' command enables estimation of effects due to neither mediation nor interaction, just interaction, both mediation and interaction, and just mediation.
• The command supports continuous, dichotomous, count, and survival time outcomes, with continuous or binary mediators.
• Applicable to cohort and case-control study designs.

Conclusions:

• The four-way decomposition provides a comprehensive approach to understanding complex causal relationships in public health.
• The 'med4way' command offers a practical tool for researchers to apply this decomposition.
• Demonstrated utility in analyzing manganese exposure and child neurodevelopment, and colorectal cancer survival disparities.