Environmental Research: Health · Published 2026-07-01 · DOI 10.1088/2752-5309/ae84f3
Oluwatobi O Oke, Sheryl Magzamen, Geoffrey S Siemering, Sheri P Johnson, Jeremy D Auerbach, Ellison M Carter
Blood lead levels (BLL) in children in the United States have decreased in recent decades; yet, lead is toxic at any concentration, and disparities in lead poisoning persist across population groups. As a result, lead exposure continues to be a major environmental public health concern. Because the relative contributions of different lead exposure sources are rarely evaluated together, mitigation efforts may be fragmented, with limited resources not always directed toward the highest-impact interventions. The objective of this study was to investigate the combined impacts of housing characteristics, soil lead concentrations, and water service line materials on childhood BLLs using tree-based machine-learning methods. Milwaukee, Wisconsin, was selected as a case study due to its high prevalence of elevated pediatric BLLs and large population at risk from older housing stock. Using existing data, we applied extreme gradient boosting model with SHapley Additive exPlanation (SHAP) to quantify the relative contributions of multiple lead exposure sources among children aged 1–5 years and to assess whether exposure profiles interact to produce higher BLLs. Our results indicate that housing age and property values, among other housing characteristics, were more strongly associated with elevated childhood BLLs than soil lead concentrations or the presence of lead service lines. Children living in homes older than 90 years with lead service lines exhibited increased exposure risk, whereas similar homes with copper service lines showed substantially reduced risk. Overall, our findings demonstrate that interpretable machine-learning methods can provide cost-effective insights to guide more targeted and impactful pediatric lead mitigation strategies.
Abstract from DOAJ. Public domain (CC0 1.0).
Read the article at the publisher →
Oke, O., Magzamen, S., Siemering, G., et al. (2026). Investigating patterns and sources of variability in children’s blood lead levels in Milwaukee, Wisconsin using a machine learning model. Environmental Research: Health. https://doi.org/10.1088/2752-5309/ae84f3