To evaluate an outbreak detection algorithm, an evaluator uses methods similar to the methods we just discussed for evaluating case detection algorithms. There are several technical and practical differences however. A technical difference is that timeliness of detection of an outbreak correlates to sensitivity and specificity and is a fundamental property of outbreak detection algorithms. As the evaluator changes the threshold setting for an outbreak detection algorithm, not only do sensitivity and specificity change, but so does time of detection.This difference requires evaluators to use a form of ROC curve analysis that is generalized to a third dimension—time—which we will discuss. A key practicality is that it is extremely difficult for an evaluator to assemble a set of 100 outbreaks for study. This difference induces evaluators to resort to a variety of creative approaches to using simulated or partly simulated data in the evaluation of detection algorithms.
Was this article helpful?