黑料网

Abstract

Background: Imprecise point locations and missing geocodes are common in public health datasets. However, they are rarely accounted for in geospatial analyses and, as a result, their impact on disease cluster detection is largely unknown. Objective: The objective of this study was to assess the sensitivity of detecting geographic clusters of high incidence rates of female breast cancer under various levels of geocoding imprecision and missing data. Methods: We used geocoded data for invasive female breast cancer diagnosed among New Jersey residents from 2001-2007 (N=36,000). Using two statistical methods, spatial scan statistics and Bayesian geo-additive models, for locating areas higher than expected incidence rates we performed analysis utilizing three approaches for handling imprecise and missing data: excluding them from analysis, assigning them to a zip code centroid, and using geographical imputation to assign a location. Geographical imputation with iterative resampling was also used to evaluate the sensitivity of the methods in identifying clusters. Results: Both statistical methods located several areas with higher than expected breast cancer incidence and, in general, both methods identified comparable areas, however the total number of significant clusters, their location and geographic area varied by each of the approaches for handling imprecise and missing data. Overall geographical imputation with iterative resampling provided a more comprehensive map showing the locations of all possible clusters and estimates of how often they appear with each iteration. Conclusion: Results from spatial scan statistics and other spatial models can be sensitive to geocoding imprecision and missing data, and the inconsistency in cluster detection should be considered by public health researchers as data quality can result in flawed findings that can lead to poor public health decisions.

Biography