A test for spatial disease clustering adjusted for multiple testing

Abstract

Tango (1995) proposed a test statistic C for detecting spatial disease clusters. However, like most other methods, it requires a value of the scale parameter which adjusts for the size of cluster to be detected in advance of its use. When an appropriate size of cluster cannot be predicted and many clustered areas are expected, we usually repeat the procedure by changing the size parameter, but this practice clearly faces a multiple testing problem. In this paper, to ameliorate this problem, we propose an extended test statistic which searches the minimum of the profile P-value of C for the parameter on cluster size which varies continuously from a small value near zero upwards until it reaches to about half the size of the whole study area. Monte Carlo simulation study shows that the power of the proposed method is shown to be reasonably high compared with the best power attained by C unadjusted for multiple testing in all the cluster models considered. The proposed procedure is illustrated with some disease maps simulated in the Tokyo Metropolitan Area.