Efficient, Scalable Multisource Surveillance Algorithms for BioSense

Carnegie-Mellon University, Pittsburgh PA

Investigators

Linked publications & trials

Abstract

We propose new research into detection algorithms for BioSense that better exploit the unprecedented[unreadable] multiplicity of data sources in the Biosense program. The proposal describes how many public health threats[unreadable] can be detected earlier and more accurately through carefully designed methods for combining data. These[unreadable] threats include natural outbreaks (e.g. new forms of influenza, bacterial outbreaks in water supplies) and[unreadable] intentional biological attacks (e.g. airborne anthrax, food supply tampering). We have seven specific aims.[unreadable] First, multivariate Bayesian spatial scanning for regions in which multiple data sources 'agree' on evidence of[unreadable] such a threat. Second, algorithms that infer the state of a city's health from dozens of pieces of tiny evidence[unreadable] over many weeks of data. Third, algorithms that search for meaningful similarites in detailed case summaries[unreadable] that might be currently overlooked. Fourth, underlying data structures to make it tractable for other[unreadable] algorithms to search over millions of time series in seconds. Fifth, methods to generate a thousand[unreadable] sufficiently realistic synthetic datasets for algorithm testing. Sixth, collaboration with six external[unreadable] biosurveillance enterprises (identified in the proposal) for real-world evaluation of and improvements to new[unreadable] algorithms, and seventh, deployment of the new algorithms within BioSense. These will be achieved using a[unreadable] combination of new algorithmic approaches, recent developments in probabilistic reasoning systems and[unreadable] recent accelerations of spatial scan. Our research group has experience in basic research and in systems[unreadable] deployment in these three areas. We will also use domain experts from whom we will elicit the probabilistic[unreadable] knowledge for some of the algorithms. All algorithms will be published, implemented, evaluated, delivered to[unreadable] BioSense and also distributed freely for general use in Public Health surveillance.[unreadable] This research aids Public Health by developing better ways for computers to combine the wide variety of[unreadable] data sources in CDC's BioSense program. We will discover the extent to which smart algorithms that[unreadable] continuously search through millions of potential hidden threats can efficiently identify times, places and atrisk[unreadable] populations for which multiple independent signals indicate there is evidence of a threat.[unreadable]