Methods for detecting disease clustering, with consideration of childhood leukaemia

In trying to interpret reports of disease clusters in specific localities, it is valuable to know whether the disease in question has a general tendency to cluster spatially. Methods for investigating localized disease clustering were the subject of a comparative study organized by the International Agency for Research on Cancer some years ago. This paper addresses some further aspects of one of the methods used in this exercise, namely the Potthoff-Whittinghill (P-W) test. Particular consideration is given to methodology for estimating the magnitude of overdispersion and for detecting whether one area, in particular, has an undue influence on the evidence for overdispersion using the P-W test, the extent to which is possible to detect clustering over regions of differing sizes using a components-of-variance approach and how adjustment for overdispersion might affect tests for raised disease rates in specific locations. These points are illustrated using data on childhood leukaemia incidence and reference is made to other analyses of the geographical distribution of childhood leukaemia that are based on this approach.