Biosurveillance systems use outbreak detection algorithms to analyze surveillance data and search for signs of an outbreak. As discussed in previous chapters, the surveillance data are typically formed into time series of daily or weekly counts prior to analysis. For each unit in the time series, an outbreak detection algorithm calculates a value that is a measure of how unusual one or more recent counts are, when compared to historical counts. If the degree of anomaly is above some threshold, an outbreak is considered to be present and the algorithm generates an alarm. Although for concreteness, we will discuss methods for evaluating outbreak detection algorithms for algorithms that analyze time series data, these methods also apply to algorithms, such as PANDA (described in Chapter 18) or WSARE (described in Chapter 15), that do not aggregate data into time series.
To evaluate an outbreak detection algorithm, the evaluator requires (1) surveillance data representing a sufficiently large number of outbreaks that he can measure the algorithm's sensitivity and average timeliness, and (2) surveillance data recorded during outbreak-free intervals with which to measure the algorithm's specificity (false alarm rate).
Was this article helpful?