Iterative Random Aggregation of Small Units Using Regional Measures of Spatial Autocorrelation for Cluster Localization

A method for localization of spatial disease clusters which uses a regional measure of spatial autocorrelation (RSAC) was recently developed by Munasinghe and Morris. They found this method to be an effective tool for the identification of regional disease clusters. In order to reduce the spurious variability of the estimated relative risks, the smallest geographic units were aggregated into analytic areas, consisting of a predefined minimum number of persons at risk (PAR). We found RSAC to be a valuable method and will propose some improvements. The present study illustrates that RSAC is quite sensitive both to the choice of PAR and the aggregation algorithm. Moreover, it does not utilize all the geographic details provided by the data sets, for instance the disease rates of the geographic units within the analytic areas. In order to overcome, at least to some extent, these problems, a modified version of the RSAC, called IRARSAC, is proposed. This method uses information, provided by the RSAC, from many different levels of aggregation. The performance of IRARSAC was shown to be more stable as compared to the RSAC, and it also seems to localize a greater proportion of the true clustering areas.