Euclidean distances and ZIP code imputations in healthcare research: Is good enough really good enough?

Background: As the use of GIS and spatially oriented data increase in healthcare research, it is important to understand implications in using various methodologies surrounding the process of geocoding and calculating distance measurements. Research Objective: To determine the effects of using straight-line Euclidean measurements and ZIP code centroid geo-imputation compared to more precise spatial analytical techniques (i.e. drive distance and street-level residential geocoding) in healthcare research Methods: Members with an inpatient claim during October 2005- September 2006 were extracted for study. Distance from admitting inpatient facility to member's home and ZIP code centroid (geographic placement) was compared using Euclidean straight-line and shortest-path drive distances (measurement technique). Using non-parametric Wilcoxon signed-rank tests, linear differences between geographic placement and measurement techniques were examined. Principal Findings: Measurement technique had a greater impact (p<0.05) on distance values compared to geographic placement of the member. Distances to admitting facilities were statistically higher when members were placed at their centroid versus residential address (p<0.05) but actual median values were low (0.8 miles or less). Distance values using the most precise method were highly correlated (r=0.99) to values using the least precise method. Conclusions & Implications: Researchers without capabilities to produce drive distance measurements and/or address geocoding techniques could rely on simple linear regressions to estimate correction factors with a high degree of confidence. For the majority of healthcare research projects utilizing linear distance measurements, it is unlikely using less precise methodology would significantly influence the overall outcome. This is particularly useful in healthcare research because a member/patient's actual address is often either unknown because of incomplete data or needs masking for privacy protection. Lastly, health plan organizations with robust administrative databases often work with very large sample sizes and can benefit greatly from ZIP code imputation and Euclidean measurements because these methods inherently consume fewer resources.