Conceptual and practical issues in the detection of local disease clusters: a study of mortality in Hamilton, Ontario

Recent advances in local spatial statistics and operational computing capacity have led to growing interest in the detection of disease clusters for public health surveillance and for improving understanding of disease pathogenesis. Although conceptual reviews and applied examples have appeared in the literature, few studies have addressed the connection between conceptual and practical issues that confront researchers interested in using local statistics to detect disease clusters. Here we review recent literature on the use of local statistics for cluster assessment and focus on the practical issue of assigning correct geographic coordinates. The process of assigning geographic coordinates to an address or postal code, known as 'geocoding', is a necessary step in conducting small-area health analyses. With a study of mortality data from Hamilton, Ontario, we illustrate inaccuracies that may be encountered when using Statistics Canada postal code conversion files. Using the Moran's I and Getis-Ord Gi and Gi* local spatial statistics to identify significant mortality clusters or 'hot spots, we demonstrate that small geocoding errors, even those that affect less than one percent of a total dataset, can have a discernible impact on analytic results. To assist other researchers, we supply guidelines to minimize error introduced by geocoding. These results emphasize the importance of accurate geocoding in local health analyses.