Application of marginalized zero-inflated models when mediators have excess zeroes

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Mediation analysis is crucial for understanding intervention mechanisms.
• Existing methods for zero-inflated count data in mediation analysis lack unbiased population average effects.

Purpose of the Study:

• To extend the counterfactual mediation framework for zero-inflated count mediators.
• To derive unbiased direct and indirect effects for population average interpretation.
• To enable mediation analysis with complex count data, like smoking or alcohol consumption.

Main Methods:

• Utilized a Marginalized Zero-Inflated Poisson (MZIP) model for the mediator.
• Derived direct and indirect effects for continuous, binary, and count outcomes.
• Extended the framework to include mediator-exposure interactions.
• Applied the methodology to alcohol consumption, sex differences, and cholesterol levels.

Main Results:

• The proposed MZIP mediator framework provides unbiased population average direct and indirect effects.
• The method was validated through a simulation study comparing it to existing approaches.
• Demonstrated the utility of the MZIP model in a real-world application.

Conclusions:

• The novel MZIP approach offers a robust solution for mediation analysis with zero-inflated count mediators.
• This methodology enhances the ability to generalize mediation effect findings to the population.
• Facilitates a deeper understanding of mechanistic pathways in diverse research areas.