Mediation analysis for common binary outcomes

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Mediation analysis decomposes exposure effects into direct and indirect pathways.
• Existing logistic regression methods for mediation analysis assume rare binary outcomes, limiting their application to common diseases.
• Accurate causal inference is crucial for understanding disease mechanisms and developing targeted interventions.

Purpose of the Study:

• To develop novel mediation analysis methods applicable to common binary outcomes, relaxing the rare disease assumption.
• To provide closed-form expressions for natural direct and indirect effects on the odds ratio scale.
• To extend mediation analysis to models with both continuous and binary mediators.

Main Methods:

• Exploited the relationship between logit and probit models to extend mediation analysis for common binary outcomes.
• Derived closed-form expressions for natural direct and indirect effects on the odds ratio scale.
• Evaluated method performance through simulations and applied it to a real-world epidemiological study.

Main Results:

• The proposed method demonstrates good performance for common binary outcomes in simulations.
• The methods successfully identified potential DNA methylation mediators of smoking behavior on obstructed airway function in the Normative Aging Study.
• Closed-form expressions provide a computationally efficient way to estimate effects.

Conclusions:

• The developed mediation analysis methods effectively address the limitations of previous approaches for common binary outcomes.
• This work enhances causal inference capabilities in epidemiology, particularly for prevalent diseases.
• The application to the Normative Aging Study highlights the utility of these methods in identifying disease pathways.