Addressing Quality: Evaluating the Impact of Noise in Patients’ Addresses

Background:Residential addresses in Electronic Health Records (EHRs) are key for linking clinical data with environmental and socioeconomic factors. However, inconsistent or noisy address data can compromise geocoding accuracy and bias spatial analyses in health research.Objectives:This study aimed to (1) identify and characterize noise in patient addresses from French hospital EHRs, (2) assess its impact on geocoding and derived socioeconomic data, and (3) detect hidden missing addresses corresponding to non-residential or administrative locations.Methods:We analyzed over three million patient addresses from two French hospitals—Hôpital Européen Georges Pompidou (Paris) and Bordeaux University Hospital. Noise in address was detected by aligning raw and standardized geocoded outputs using a Needleman–Wunsch–based algorithm. The most frequent noise terms were categorized by position within the address. To evaluate their effects, we simulated 15,000 noisy variants of French public school addresses and measured changes in geocoding precision and socioeconomic classification. Spatial frequency analysis identified address clusters suggestive of administrative placeholders. We used non linear regression to model the expected number of patients by distance to the hospital and identified unexpected clusters of patients.Results:Frequent noise elements included apartment or building references (e.g., “APT,” “BAT,” “CHEZ”), which had minimal impact on geocoding accuracy or socioeconomic inference. We also identified clusters of non-residential addresses near hospital sites reflecting administrative or placeholder entries. These hidden missing addresses likely correspond to socially vulnerable patients, whose true residences are underrepresented in spatial analyses.Conclusions:Textual noise in hospital addresses exerts limited influence on geocoding precision, but hidden missing addresses introduce systematic bias, particularly for disadvantaged populations. Address quality assessment and detection of non-residential entries are essential to improve the reliability and equity of geospatial health research.