A mediation analysis for a nonrare dichotomous outcome with sequentially ordered multiple mediators.

Area of Science:

• Biostatistics
• Reproductive Medicine
• Epidemiology

Background:

• Single mediator analyses are insufficient for complex biological mechanisms with interconnected mediators.
• Understanding intricate exposure-outcome pathways requires advanced mediation analysis techniques.
• Dichotomous outcomes present unique challenges in mediation analysis.

Purpose of the Study:

• To develop and validate a method for analyzing multiple sequentially ordered mediators for dichotomous outcomes.
• To present identifiability assumptions for path-specific effects in complex mediation models.
• To provide a computationally efficient estimator for path-specific effects.

Main Methods:

• Proposed a closed-form estimator for path-specific effects using a probit model for dichotomous outcomes.
• Derived asymptotic variance and explored approximation methods (delta method, bootstrapping).
• Conducted simulations to assess the method's validity and efficiency.

Main Results:

• The proposed method accurately captures path-specific effects in simulations when counterfactual outcome probabilities are not small.
• A computationally efficient variance approximation alternative to bootstrapping was demonstrated.
• The method was successfully applied to analyze polycystic ovarian syndrome effects on live birth rates via estradiol and oocyte retrieval.

Conclusions:

• The developed method provides a robust approach for mediation analysis with multiple sequential mediators and dichotomous outcomes.
• The findings offer valuable insights into the biological mechanisms underlying reproductive outcomes.
• An R package (SOMM) is available for implementing this novel statistical methodology.