Statistical surveillance - Exponentially weighted moving average methods and public health monitoring

Abstract

The need for statistical surveillance has been noticed in many different areas and examples of applications include the detection of an increased incidence of a disease, the detection of an increased radiation level and the detection of a turning point in a leading index for a business cycle. In all cases, preventive actions are possible if the alarm is made early. If the change is detected too late, this can have severe consequences both at a personal level for affected individuals and to society as a whole. In these important situations we must evaluate the evidence value of the information we have about the process in order to guide us in the choice of making an alarm or not. The aim is to detect the change as quickly as possible and at the same time control the rate of false alarms. To do this efficiently we construct alarm systems using the available observations of the process, which are taken sequentially in time. (Note the difference from a test of one hypothesis.) The theory of statistical surveillance deals with the construction of alarm systems and the evaluation of such systems. This licentiate thesis consists of two papers with this common subject.

The first paper (1) deals with the properties of a special type of surveillance methods called EWMA methods. One attractive feature of EWMA methods is the easily interpretable alarms statistic, which is an exponentially weighted moving average of all available observations of the process. Several ways of constructing alarm limits to this statistic have previously been suggested in the literature. In this paper new types of evaluations of the performance of suggested variants are made and the results cast new light on both the merits of the variants and the optimality criteria commonly used. Methodological issues of general interest in the area of statistical surveillance are also treated, such as the definition of comparability between methods.

The second paper (2) deals with statistical surveillance in the area of public health. A critical review with emphasis on the inferential issues is made. The merits of different approaches are discussed and a new method is derived. Especially noticeable from the review is the lack of methods of surveillance of a spatial pattern, an area that includes many important applications, not only in public health.