Explanatory Machine Learning Model to Associate All-Cause Mortality with PM_2.5 by Sources and Components: A...

Area of Science:

• Environmental Health
• Epidemiology
• Toxicology

Background:

• Long-term exposure to fine particulate matter (PM2.5) poses significant health risks.
• The diverse chemical components and sources of PM2.5 are often underestimated in health risk assessments.
• Understanding source- and component-specific effects is crucial for effective public health interventions.

Purpose of the Study:

• To investigate the association between long-term PM2.5 exposure from various sources and chemical components with all-cause mortality in China.
• To identify the most hazardous PM2.5 sources and components contributing to mortality.
• To evaluate the performance of different statistical models in assessing PM2.5 health effects.

Main Methods:

• Utilized county-level census data from China (1990, 2000, 2010) with a difference-in-differences (DID) study design.
• Applied log-linear models and advanced mixed exposure models including weighted quantile sum (WQS), ridge regression, random forest, XGBoost, and an ensemble model.
• Employed cross-validation for model performance assessment and Shapley additive explanations (SHAP) for interpretability.

Main Results:

• A 10 μg/m³ increase in PM2.5 was linked to a 3.16% rise in all-cause mortality.
• Industrial (3.46%), power (2.32%), and residential (1.46%) sources were most hazardous.
• Ammonium (6.46%), nitrate (4.23%), and black carbon (3.93%) were the most potent components.
• Industrial-sourced black carbon was identified as the primary contributor to PM2.5-related mortality by superior predictive models (random forest, ensemble).

Conclusions:

• The health impacts of PM2.5 are significantly influenced by its source and chemical composition.
• Industrial black carbon demonstrates higher toxicity and warrants priority in air pollution control strategies.
• Tailored interventions focusing on specific PM2.5 sources and components are essential for mitigating health risks.