Population-average mediation analysis for zero-inflated count outcomes.

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Mediation analysis is crucial for understanding causal pathways in interventions.
• Existing methods for count outcomes with excess zeros are often computationally intensive, biased, or difficult to interpret.

Purpose of the Study:

• To propose a novel mediation methodology for zero-inflated count outcomes.
• To provide population-average mediation effects with rapid variance estimation.
• To extend the methodology to include exposure-mediator interactions.

Main Methods:

• Utilized the marginalized zero-inflated Poisson (MZIP) model.
• Employed the counterfactual approach to mediation.
• Applied the delta method for rapid variance estimation.
• Extended the model to handle exposure-mediator interactions.

Main Results:

• The proposed MZIP method demonstrated minimized bias and computation time compared to existing methods.
• The methodology allows for straightforward interpretation of mediation effects.
• Simulations confirmed the model's performance against alternative zero-inflated and Poisson methods.

Conclusions:

• The novel MZIP-based mediation approach offers a robust and efficient solution for zero-inflated count data.
• This method enhances the interpretability of causal pathways in public health and medical research.
• The approach is applicable to complex scenarios, including those with exposure-mediator interactions.