Using the Cuscore technique in the surveillance of rare health events

An adaptation from Cuscore control charts for normal distributions to the problem of monitoring the rate of occurrence of a rare event is considered. The presented scheme defines an alarm in terms of the sum of ‘scored distances' between successive events of interest. A procedure is given for using this scheme in an ‘optimal’ manner which minimizes the out of control expected delay for a given rate of false alarms. Comparisons over a number of examples, after matching the rate of false alarms, show that the present technique is more efficient than those of Shewhart and Sets, and also than Cusum under conditions of practical relevance.