Analysis of data associated with seemingly temporal clustering of a rare disease

Three statistical tests aimed at detecting temporal clustering within a given short series of diagnoses are presented. These tests are based on a standardized time interval between consecutive diagnoses. Two of the tests (the Cuscore and the Sets tests) are derived from sequential monitoring techniques which are sensitive to temporal clustering within the data set. The third test (R test) is not sequential and its sensitivity is focused on the average increase in the overall rate of the disease rather than on clustering within the series. Power curves are presented for conditions related to the intensity level of the subtle epidemic, the cluster size and the number of diagnoses. None of the techniques showed highest efficiency over all the specified conditions. The R test is the most efficient when the relative risk is 2 or less, and the Cuscore test is the most efficient method when the relative risk is > or = 2.5.