ABSTRACT
Timely detection of adverse health events in conjunction with public health policies to rectify the situation or prevent repeated occurrences is beneficial to the affected individuals and society. This involves systematic collection, analysis, and interpretation of large amounts of outcomespecific data by national health programs in different countries and international networks; see Sonesson and Bock (2003, pp.5-6) who also point out that there are many situations in which the sequentially accumulated data can be used prospectively to quickly detect an increased incidence of a disease so that timely rectifying actions can be taken. They note that while much of the research on statistical surveillance originated from engineering applications, “the context of public health surveillance implies specific problems that are not generally present in the case of industrial production control.” These include problems of seasonal effects and reporting delays, inherent differences among diseases (such as chronic versus infections), monitoring not only cases of disease but also risk factors. Kulldorff (2011) points out that when surveillance is carried out repeatedly over time, “sequential statistical methods should be used”. He distinguishes between sequential testing methods “to quickly detect (an AE) problem that has been there from the beginning of the analysis”, and sequential detection methods “to monitor a process for a sudden shift or unknown shift that occurs at some unknown time,” which are commonly used in an industrial setting to quickly detect a suddenly malfunctioning manufacturing process.” After a comprehensive overview of sequential testing methods for pharmacovigilance in Section 8.1, we focus on a number of statistical issues and recent developments to address them in Sections 8.2 and 8.3. The supplements and problems in Section 8.4 provide additional details for sequential testing and detection.
