Geocoding Pipeline Development and Data Quality Assessment For Enhancing Precision Medicine Research

Abstract

Geographic location shapes health outcomes through environmental exposures, socioeconomic conditions, and access to care. However, the quality of geocoded address data to facilitate these inferences remains poorly characterized to understand the positional error from geocoding nationwide. This thesis presents a geocoding benchmarking study evaluating the DeGAUSS geocoding framework across a curated reference dataset of over 270,000 public locations spanning U.S. states and territories. Address preprocessing applied conformance and formatting checks aligned with USPS Publication 28 standards, including flags for embedded city or state information, malformed unit identifiers, missing spaces, and state-specific address patterns. Using verified GIS coordinates as reference coordinates, positional error between the reference coordinate and the geocode output was calculated via the Haversine formula and evaluated across five DeGAUSS match score thresholds (60%–100%), with territories analyzed separately due to substantially higher error and poor reference coordinate quality. Principled outlier removal filters out geocodes that fall outside the expected state boundary, distinguishing misclassification (geocoder error) from misannotation (reference coordinate error). Relaxing the match score threshold from 100% to 60% increased successfully geocoded records from 58.3% to 86.9% of the dataset, while producing only modest increases in median positional error nationally from 65.3m to 68.0m. In states with higher error, lower-scoring records showed meaningfully worse positional error. Geographic stratification revealed substantially lower performance in U.S. territories, where median positional error exceeded 450m even in Puerto Rico and surpassed 4,000m in Guam. Together, these findings suggest that geocoding quality standards may need to incorporate geographic context to support accurate residential linkage in integrative precision medicine research.