Zip Codes or Street Addresses? Comparing Ambient Ozone Exposures for Alternative Spatial Resolutions of Cohort Subjects

Abstract

SM5-PD-03 Introduction: Accurate estimation of exposures is of the outmost importance for drawing valid inferences about the spatial relationship of risk factors with health outcomes of concern. In air pollution epidemiology, estimates are often derived from monitoring data. With GIS technology, we can map individuals to their street address rather than to a central location by zip code. Assuming the validity of the spatial interpolation method chosen and the representativeness of exposure data, an independent question remains: Does enhanced positional accuracy of subjects result in reduced exposure misclassification? We compared the estimated ozone exposures assigned to our cohort subjects when their locations are resolved alternatively by zip code centroids and by street addresses. Methods: Monitors with representative data for at least 9 months in 3 years or more from 1996 to 1999 were selected and the average annual concentration calculated. We then identified cohort subjects with valid, geo-coded residence information during 1996 to 1999 (n = 584). Differences in exposure with respect to positional accuracy were assessed by implementing 2 spatial interpolation methods: Inverse Distance Weighting (IDW) and Ordinary Kriging (OK). OK- and IDW-derived ozone concentrations were then assigned to each subject's mapped zip centroid and street address. Results: Ozone estimates for both spatial resolutions were highly correlated regardless of the interpolation method used (RIDW = 0.978 and ROK = 0.987). The estimated kriging standard errors were also highly correlated for both sets of predictions (R = 0.94). The average kriging standard error was virtually the same for predictions resolved at the street level (0.0705 ppm) and those for zips (0.0700). The average difference between exposures assigned to zip and street addresses was positive but of small magnitude. For OK predictions, this value reached 0.0001 ppm (SD = 0.00096 ppm), while that for IDW estimates was even smaller, 0.00006 ppm (SD = 0.0012 ppm). Discussion and Conclusions: Refining the locational resolution of subjects did not result in a large change in estimated exposures. Both interpolation methods predicted greater exposures for zip centroids. Kriging generated greater differences between predictions for zips and street addresses. Thus, greater spatial accuracy is of less concern when we use IDW for estimating ozone exposures. Using zip centroids seems to be a reasonable approach for modeling ozone exposures. This notion is yet to be corroborated for other types of small-area units. Our results lend validity to previous findings from our study in which subject locations were mapped to zip centroids.