ITR: Bayesian Modeling for Biosurveillance

University Of Pittsburgh, Pittsburgh PA

Investigators

Abstract

Early, reliable detection of outbreaks of disease, whether natural (e.g., West Nile virus) or bioterrorist-induced (e.g., anthrax and smallpox), is a critical problem today. It is important to detect outbreaks early in order to provide the best possible medical response and treatment, as well as to improve the chances of identifying the source. A primary goal of this project is to develop basic new Bayesian probabilistic inference algorithms that monitor electronically available healthcare data to achieve early, reliable detection of outbreaks. In particular, inference will take place on Bayesian networks that model the links between available data and possible causes of outbreaks. The scientific challenge of monitoring for outbreaks within an entire population create major computational challenges in building and applying Bayesian models that are orders of magnitude larger than those developed previously. The project will apply and extend state-of-the-art probabilistic inference methods to achieve efficient inference. If inference indicates that an outbreak is likely, an alert will be raised automatically. Appropriately, however, public health officials are unlikely to blindly trust an outbreak alert, unless there is an explanatory justification. Automated explanation of Bayesian inference is therefore another key project goal. The scientific contributions of this project will follow from developing, investigating, and evaluating new modeling and algorithmic techniques that make Bayesian biosurveillance practical for monitoring and diagnosing (in real time) the disease-outbreak status of an entire population. In investigating these issues, this project is anticipated to make both specific scientific contributions to computer science, statistics, and public health, as well as broader contributions to public safety.