Source-resolved PM_2.5 oxidative potential predicts global mortality burden: A machine learning approach

Fine particulate matter (PM_2.5) oxidative potential (OP_v) is a crucial health metric, varying significantly by emission source. While concentration-response functions (CRFs) exist linking PM_2.5 mass to OP_v (OP_v^PM2.5), the lack of source-resolved OP_v (OP_v^source) CRFs limits accurate risk assessment globally. We developed global OP_v^source CRFs by integrating machine learning, statistical techniques, and comprehensive datasets of PM_2.5 concentrations, source apportionment (50 countries), and measured OP_v (>10,000 measurements in 29 countries). Our analysis revealed distinct intrinsic OP per unit mass (OP_m) for major sectors, ranking: energy (ENE) > transportation (TRO) > industry (IND) > agriculture and residential combustion (AGS + RCO). Applying these CRFs to 2017 PM_2.5 source data across 203 countries, we estimated the global average OP_v^source at 0.78 nmol min^-1 m^-3 (95 % CI: [0.39, 1.2]). ENE (33 %) and AGS + RCO (31 %) were predominant contributors globally, with substantial national variations. Poisson regression demonstrated that OP_v^source is a significantly stronger predictor of PM_2.5-attributable mortality than PM_2.5 concentration or OP_v^PM2.5. These findings establish source-resolved OP_v as a critical, more accurate metric for assessing mortality risks, supporting targeted air quality policies and interventions to mitigate PM_2.5 health burden.