When is the epidemic warning cut-off point exceeded?

The concept of excess mortality from influenza and pneumonia was introduced in 1963 by Serffing [1]. His method of detection of epidemics is based on a simple non-linear regression model which allows the prediction of the number of expected cases in a given week and the definition of the epidemic cut-off point. An epidemic is declared when this cut-off point is exceeded for 2 consecutive weeks. This method has been used routinely by the Centers for Disease Control (CDC) in the context of the influenza and pneumonia mortality surveillance. In 1981, Choi and Thacker [2] proposed a method of detection based on the ARIMA techniques, valuable in this context because they take into account the time-dependence of the data. The use of mortality data to monitor influenza syndromes raises several problems: the problem of notification delay, the interval between the onset of the epidemic and the increased mortality associated with this epidemic, the impossibility of detecting epidemics causing little or no excess mortality when, for example, few elderly subjects are affected. Similarly, hospital morbidity only reflects the most severe cases. The influenza case data (147,810 cases defined according to the WHO criteria (sudden onset of fever exceeding 39 °C, myalgia, respiratory symptoms) between November 1984 and August 1993) obtained from general practitioners in the context of the RNTMT, therefore offered an excellent opportunity to test, for the first time, the application of the methods available for rapid detection of influenza epidemics based on morbidity data [3]. For simplicity, we have adopted Serffing's approach which we applied to the estimation of the mean number of cases per doctor for a given week. The principle of the method consists of adjusting a regression model to the data from non-epidemic weeks and predicting the number of cases expected in the absence of an epidemic. Let us consider the example of the prediction of the number of nonepidemic cases for the winter of 1992-1993. We decided to consider as being epidemic those weeks for which the number of cases per doctor exceeded 3 (selected by trial and error). After elimination of these weeks, we adjusted the following regression equation to the non-epidemic data for November 1984 to August 1992 to predict the number of expected non-epidemic cases for the following winter (1992-1993).