Weighted Normal Spatial Scan Statistic for Heterogeneous Population Data

In geographical spatial epidemiology and disease surveillance, all the existing spatial scan methods for cluster detection using continuous data are designed for evaluating clusters of individuals and analyzing individual-level data. Motivated by growing demands to study the spatial heterogeneity of continuous measures in population data, such as mortality rates, survival rates, average body mass indexes and pollution at state, county, and census tract levels, we propose a weighted normal scan statistic for investigating the clusters of the cells (geographic units such as counties) with unusual high/low continuous regional measures, where the weights reflect the uncertainty of the regional measures or sample size (number of observed cases) in the cells. Power, precision, the effect of the weights, and the sensitivity of the proposed test statistic to data from various distributions are investigated through intensive simulation. The method is applied to 1988–2002 stage I and II lung cancer survival data in Los Angeles County in order to search for clusters of geographic units with high/low survival rates in a short-term/long-term survival after diagnosis, and to 1999–2003 breast cancer age-adjusted mortality rate data in the U.S. collected by the Surveillance, Epidemiology and End Results (SEER) program in order to evaluate the clustering pattern of counties with high mortality rate. The proposed method is included in the latest release of the SaTScan software (www.satscan.org).