The Bayesian aerosol release detector: an algorithm for detecting and characterizing outbreaks caused by an atmospheric release of Bacillus anthracis

Abstract

Early detection and characterization of outdoor aerosol releases of Bacillus anthracis is an important problem. As health departments and other government agencies address this problem with newer methods of surveillance such as environmental surveillance through the BioWatch program and enhanced medical surveillance, they increasingly have newer types of surveillance data available. However, existing methods for the statistical analysis of surveillance data do not account for recent meteorological conditions, as human analysts did in the case of the Sverdlovsk anthrax outbreak of 1979 to determine whether the locations of victims were consistent with meteorological conditions in the week preceding their onset of illness. This paper describes the Bayesian aerosol release detector (BARD), an algorithm that analyzes both medical surveillance data and meteorological data for early detection and characterization of outdoor releases of B. anthracis. It estimates a posterior distribution over the location, quantity, and date and time conditioned on a release having occurred. We report a proof-of-concept evaluation of BARD, which demonstrates that the approach shows promise and warrants further development and evaluation.