Incorporating Parcel-Based Housing Conditions to Increase the Precision of Identifying Children With Elevated Blood Lead.

Context: Area-level predictive models are commonly used to screen children for blood lead levels (BLLs) greater than the Center for Disease Control and Prevention (CDC) blood lead reference value (BLRV) of 3.5 µg/dL.

Objectives: To increase screening accuracy and precision by creating a parcel-level model incorporating housing characteristics to predict parcels where children are at high risk.

Design: Residential addresses associated with child blood lead tests were linked to neighborhood- and parcel-level characteristics and proximity to lead sources. Regression forests were fit using different predictor combinations and compared using cross-validated accuracy and decile-based agreement across all residential parcels.

Setting: Hamilton County, Ohio, United States.

Participants: Children less than 6 years of age with blood lead tests between January 2020 and April 2023.

Main outcome measure: Cross-validated model accuracy and decile-based agreement across residential parcels.

Results: 27,782 tests were matched to a residential parcel. Regression forests using Parcel + Source (70.8% AUC) and Neighborhood + Parcel + Source predictors (70.3% AUC) had the highest cross-validated accuracy for predicting BLLs >3.5 µg/dL. Parcel-level predictions revealed heterogeneity of risk across parcels within the same tract.

Conclusions: Parcel characteristics improved the accuracy of predicting locations of children with BLLs >3.5 µg/dL and can help identify children at high risk living in low-risk areas. A parcel-level identification of housing-based lead hazards could guide and support action to prevent pediatric lead exposure.