University of Washington - Department of Statistics
Advisor: Jon Wakefield
We consider the problem of detecting clusters of non-infectious and rare diseases. Cluster detection is the routine surveillance over a large expanse of small administrative regions to identify individual 'hot-spots' of elevated residual spatial risk without any preconceptions about their locations. A class of cluster detection procedures known as moving-window methods superimpose a large number of circular regions onto the study area. For each of these circles, the significance of the observed data are determined with respect to the hypothesis of no unmeasured risk inside the circle, as compared to outside the circle. The SaTScan software provides a popular implementation. However, many such methods suffer from two drawbacks. First, they rely on tests of fixed size alpha regardless of sample size. Hence the setting of fixed alpha rules effectively ignore power. Second, multiple testing issues arise due to the large number of circular regions being tested. As a solution, we propose a Bayesian model that incorporates prior knowledge while accounting for multiple clusters. All posterior probabilities are estimated using Markov chain Monte Carlo and identify individual areas with high posterior probability of being in a cluster. As an example we use the Upstate New York Leukemia dataset from Turnbull (1990) which has been examined extensively in previous cluster detection endeavors. After presenting a simulation study to compare the sensitivity and specificity of the Bayesian and SaTScan methods, we apply the methodology to the Surveillance, Epidemiology and End Results program database of cancers for 13 counties in Western Washington in the years 1995â€“2005.