Abstract

The authors propose a procedure for the detection of significant clusters of chronic diseases, with particular reference to cancer. The procedure allows for variations in population density and avoids the problem of "post hoc" formation of hypotheses or self-defined populations. This accounts for several of the principal problems of cluster evaluations. The techniques are practical but "computer-intensive." The procedure, termed the "cluster evaluation permutation procedure," is applied to leukemia incidence data for an Upstate New York region obtained from the New York State Cancer Registry and census files. Comparisons are made with two other recently proposed clustering methods, namely the U-statistic method of Whittemore et al. (Biometrika 1987;74:631-7) and the "geographical analysis machine" of Openshaw et al. (Lancet 1988;1:272-3). Routine examination of disease occurrence with the cluster evaluation permutation procedure would allow state health officials to prioritize case investigations and to respond in a timely and efficient manner to inquiries of reported clusters.