The ORNL ADDRESS Geocoder: Deployment, Quality Assessment, and Lessons Learned from Multi-State Cancer Registry and Pennsylvania Radon Residential Address Geocoding.

Abstract

The Automated Determination of Detailed Regional and Exact Spatial Segments geocoder, developed at ORNL, was applied to standardize geocoding of multi-state cancer registry residential history data for the NCI Cancer Outcomes and Neighborhood Environmental Correlations Tracker (CONNECT) project and the PA geocoding assessment of residential addresses associated with indoor radon measurements. This report documents the system architecture, deployment parameters, data filtering decisions, geocoding quality assessment, and identified failure scenarios. Of 2,982,472 retained records from four registries, 44.9% achieved geocoding coordinates within 1,000 meters of NCI reference coordinates. For the PA radon dataset with ∼1.5 million residential address-level records, ADDRESS achieved house-level matches for approximately 45% of addresses from 2014 onward. Across all years, we have 96.3% of house-level results and 74.3% of residential-level results validated within the address ZIP’s buffered ZIP Code Tabulation Area (ZCTA) proxy. Systematic issues were identified including city-name ambiguity in the fallback matching logic and ZIP code instability for historical addresses. This report consolidates lessons learned to support future deployments of ADDRESS for health research geocoding.