This chapter focuses on the detection of spatial clusters of disease, with the goal of rapidly detecting emerging outbreaks by prospective surveillance. Spatial cluster detection has two main goals: identifying the locations, shapes and sizes of potential disease clusters, and determining whether each of these potential clusters is due to a genuine outbreak or due to chance fluctuations in case counts. In other words, we want to know whether anything unexpected is going on, and if so, where? This task can be broken down into two parts: first, figuring out what we expect to see, based on populations or on expected counts inferred from historical data, and then determining which regions deviate significantly from our expectations. In this chapter, we present an overview of spatial cluster detection, and then discuss a number of cluster detection methods, focusing on the spatial scan statistic. In addition to presenting the standard spatial scan framework, we consider a number of extensions to this framework, including generalization of the scan statistic to situations in which baselines must be inferred from historical data, computational methods for fast spatial scanning, and extensions to spatio-temporal cluster detection. This chapter does three things. First, it provides the basic statistical and computational tools to make the spatial scan applicable and useful for analyzing large real-world data sets. Second, it motivates cluster detection approaches and third, it compares them to other outbreak detection methods.
Epidemiologists have been analyzing biosurveillance data spatially since the seminal work of John Snow on cholera (Snow, 1855). In the 1970s, researchers automated the map creation aspect of spatial analysis. The results of this work— geographic information systems—are now in widespread use in health departments. Over the past decade, researchers have developed spatial scan statistics, which automate the pattern recognition component of spatial analysis. This advance enables a biosurveillance organization to analyze spatial data far more exhaustively than ever before.
Was this article helpful?