Construction of Models To Predict the Effectiveness of E-Waste Control through Capture of Volatile Organic Compounds and Metals/Metalloids Exposure Fingerprints: A Six-Year Longitudinal Study

Machinelearning-based models were developed to accuratelypredict the effectiveness of e-waste control in reducing populationexposure risks and thereby strengthen the regulation of e-waste pollution. The significant health implications of e-waste toxicantshave triggeredthe global tightening of regulation on informal e-waste recyclingsites (ER) but with disparate governance that requires effective monitoring.Taking advantage of the opportunity to implement e-waste control inthe Guiyu ER since 2015, we investigated the temporal variations inlevels of oxidative DNA damage, 25 volatile organic compound metabolites(VOCs), and 16 metals/metalloids (MeTs) in urine in 918 children between2016 and 2021 to demonstrate the effectiveness of e-waste controlin reducing population exposure risks. The hazard quotients of mostMeTs and levels of 8-hydroxy-2 '-deoxyguanosine in childrendecreased significantly during this time, indicating that e-wastecontrol effectively reduces the noncarcinogenic risks of MeT exposureand levels of oxidative DNA damage. Using mVOC-derived indexes asa feature, a bagging-support vector machine algorithm-based machinelearning model was constructed to predict the extent of e-waste pollution(EWP). The model exhibited excellent performance with accuracies >97.0%in differentiating between slight and severe EWP. Five simple functionsestablished using mVOC-derived indexes also had high accuracy in predictingthe presence of EWP. These models and functions provide a novel humanexposure monitoring-based approach for assessing e-waste governanceor the presence of EWP in other ERs.