Bayesian Biosurveillance Using Multiple Data Streams

Introduction: Emergency Department (ED) records and over-the-counter (OTC) sales data are two of the most commonly used data sources for syndromic surveillance. Most current detection algorithms monitor these data sources separately, and either do not combine them, or combine them in an ad hoc fashion. This paper introduces a causal model that coherently combines the two data sources in order to perform outbreak detection. Objectives: This paper presents a Bayesian biosurveillance algorithm called PANDA that combines information from multiple data streams. We describe the model, along with an explication of assumptions and techniques used to make this approach scalable for real-time surveillance of a large population. Methods: We extend the causal Bayesian network model used in (1) to incorporate evidence from daily OTC sales data. We model, at the level of individual people, the actions that result in the purchase of OTC products, as well as admission to an ED. Results: The aim of this paper is to describe a detection model for monitoring both ED and OTC data. This paper provides preliminary support that despite the complexities of this model, the running time is tractable. Conclusion: This paper introduces a new Bayesian biosurveillance algorithm that models the interaction between ED and OTC data. It also provides preliminary results that are positive regarding the run time of the algorithm.