DETECTING CLUSTERING--APPLICATION TO REPRODUCTION

Evaluation of clustering of disease and the associated problem of disease transmission are of interest in many fields of medical research including reproductive outcome, mental health and cancer. In previous published work, and work submitted for publication, the research team on the proposal has either developed or applied several statistics for time and space-time clustering. We propose to develop new procedures to detect time and space-time clustering. We propose to unify and generalize previous work with a view toward either simplifying the computation and interpretation of the statistics, providing measures of effect sizes, or suggesting methods that would enable the investigator to increase the power of the test statistics to detect clustering. The CUSUM statistic based on the sum of the number of excess outcomes, will be applied to time clustering based on the number of live births between congenital anomalies rather than chronological times between events. The statistic will also be generalized to space-time clustering. The scan statistic based on the maximum number of cases of disease in a 'window' of fixed length, will be applied to space-time clustering, and a measure of effect size will be developed. Knox's statistic based ont he number of pairs close in space and close in time will be modified for the common case where the space data consists of geographic units. Measures of attributable risk and effect size, and related confidence intervals will be developed. We will develop methods that will facilitate the computation of critical values for Mantel's statistic based on the times and distances between pairs of events.