Controlling for time-varying population distributions in disease clustering studies.

Both the Knox test and the Mantel test for the presence of disease clustering have previously been shown to be sensitive to changes that occur in the relative geographic population distribution during the time period of study. Although these procedures do not require knowledge of the underlying population density, they implicitly assume it is constant with time. If this is not the case, one may detect clustering which is completely independent of the disease process and simply reflects time-space clustering of the population. Modified procedures are presented which incorporate information from appropriately selected controls to adjust for the effect of a shifting population. Results of a computer simulation study demonstrating the stability of the modified tests in the face of an expanding population are reported.