You are here:
COMPARISON OF GEOCODING METHODS USED IN CASE-CONTROL STUDY OF AIR QUALITY AND BIRTH DEFECTS
Citation:
Gilboa, S., P Mendola, Olshan. AF, C. Harness, A. H. Herring, AND P. Langlois. COMPARISON OF GEOCODING METHODS USED IN CASE-CONTROL STUDY OF AIR QUALITY AND BIRTH DEFECTS. Presented at International Society for Environmental Epidemiology, New York, NY, August 1-4, 2004.
Description:
Introduction: Accurate geocoding of maternal residence is critical to the success of an ongoing case-control study of exposure to five criteria air pollutants and the risk of selected birth defects in seven Texas counties between 1997 and 2000. The geocoded maternal residence at delivery will be used to assign exposure during the critical window of gestational weeks three through eight using ambient pollution concentrations measured at the monitor closest to the residence. We compared the use of automated and manual methods to geocode addresses that were initially unmatched. Methods: Geocoding was performed in ArcGIS 8.3 using ESRI ? StreetMap as the geocoding service for 5338 cases and 4574 controls. We geocoded the address reported on the live birth or fetal death certificate provided by the Texas Bureau of Vital Records, or that reported in the medical record if available (cases only). The matching options were 60 (minimum match score), 10 (minimum candidate score) and 80 (spelling sensitivity). In an effort to increase the percentage of successful matches, two strategies were then undertaken: (1) re-match automatically with lower standards of 40 (minimum match score), 10 (minimum candidate score) and 70 (spelling sensitivity; and (2) interactively (manually) attempt to match those that were ungeocodeable. Results: The first geocoding procedure yielded an overall matching rate of 73.3%. The unmatched observations (2557 cases and controls combined) were then geocoded using the lower standards yielding 361 (14%) matches. The manual process, designed to check the validity of the lower standards, yielded 912 (36%) matches. Considering the manual matches as the gold standard, the sensitivity and specificity of matching using the automated procedure with lower standards were 26% and 92% respectively. Among the 237 observations that matched using the two additional strategies, only 2% were matched to the exact same latitude and longitude coordinates. In total we successfully geocoded 4582 cases (85.8%) and 3669 (80.2%) controls. Discussion: The accurate assignment of exposure based on maternal residence relies on the quality of the geocoded data. Our data suggest that automatically geocoding using substandard matching criteria does not yield comparable quality data to that which results from interactively matching addresses to their best available candidate. The effort to manually match is preferable to restricting geocoding results to those yielded via automated procedures. This is an abstract of a proposed presentation and does not necessarily reflect EPA policy.