Mediation Analysis using Semi-parametric Shape-Restricted Regression with Applications

Area of Science:

• Environmental Health
• Biostatistics
• Reproductive Endocrinology

Background:

• Mediation analysis often assumes linear relationships, which may not hold true in biological systems.
• The exact functional form of relationships in biological processes can be unknown but scientifically hypothesized.
• Existing methods may not adequately capture complex, non-linear associations in mediation pathways.

Purpose of the Study:

• To develop a novel methodology for mediation analysis using shape-restricted inference, accommodating unknown non-linear relationships.
• To investigate the mediation effect of human chorionic gonadotropin (hCG) on the association between pesticide exposure and birth weight.
• To apply this new method to population-level data from a prenatal screening program.

Main Methods:

• Development of a shape-restricted inference framework for mediation analysis.
• Application of the methodology to population-level data on pesticide exposure, hCG levels, and birth weight.
• Estimation of natural direct and indirect effects under hypothesized non-linear relationships.

Main Results:

• The proposed shape-restricted inference methodology was successfully applied to mediation analysis.
• A positive association was suggested between pesticide application and birth weight through natural direct effects.
• Negative natural indirect effects were found for pesticide application on birth weight mediated by hCG.

Conclusions:

• Shape-restricted inference provides a robust approach for mediation analysis with hypothesized non-linear relationships.
• Pesticide exposure may have complex, non-linear effects on birth weight, with hCG acting as a mediator.
• The findings highlight the importance of considering non-linear pathways in environmental health research.