## Introduction

As illustrated by the postincident report from the Glasgow Health Department (summarized in the previous chapter), uncertainty about whether there was a threat to human health was a key topic in the deliberations of the Incident Management Team. In particular, the team had difficulty estimating the probability that people might become infected based on the available water surveillance data. The problem that Glasgow faced is increasingly common owing to the widespread adoption of new methods of biosurveillance.

In this chapter, we continue our example of a decision analysis of a boil-water advisory decision, focusing on techniques for estimating the probability of an outbreak from surveillance data. Figure 30.1 (reproduced from the previous chapter) is an example of a spike in surveillance data that raises questions of whether there is an outbreak, and, if so,

what is the biological agent, how big is it, how long has it been ongoing, how much longer will it last, and how aggressively should it be investigated.

We generally cannot answer questions about the existence and nature of an outbreak with certainty from imprecise surveillance data, such as sales of diarrhea remedies. However, we can interpret the surveillance data probabilistically; that is, we can express our uncertainty about the presence of an outbreak as a probability, which we can then use in combination with cost and benefit considerations to make a decision. For example, we may estimate that the probability that there is a Cryptosporidium outbreak is 0.80, which strongly suggests that some action is in order. Alternatively, we may estimate that the probability is only 0.001, and therefore, no action is required.

In fact, there are a small number of algorithms that do compute the probability of an outbreak directly. We call them Bayesian detection algorithms because they use Bayes' theorem. PANDA and BARD (discussed in Chapters 18 and 19, respectively) are two examples of Bayesian detection algorithms that are capable of computing the probability of an outbreak. The far more common situation in current practice is different, however.

## Post a comment