Optimizing Biosurveillance Systems that Use Threshold-based Event Detection Methods

Abstract

a b s t r a c tWe describe a methodology for optimizing a threshold detection-based biosurveillance system. The goal is to maximize the system-wide probability of detecting an ''event of interest" against a noisy background, subject to a constraint on the expected number of false signals. We use nonlinear programming to appropriately set detection thresholds taking into account the probability of an event of interest occurring somewhere in the coverage area. Using this approach, public health officials can ''tune" their biosurveillance systems to optimally detect various threats, thereby allowing practitioners to focus their public health surveillance activities. Given some distributional assumptions, we derive a one-dimensional optimization methodology that allows for the efficient optimization of very large systems. We demonstrate that optimizing a syndromic surveillance system can improve its performance by 20-40%.Published by Elsevier B.V.

Some methodological issues in biosurveillance

“…Furthermore, if the main purpose of biosurveillance is to guard against bioterrorism, then one should design algorithms that can incorporate information about the adversary, particularly information about likely modes and locations of attack. For example, Fricker and Banschbach 35 have developed a methodology for optimizing a (very simple) threshold detection‐based biosurveillance system that uses information about the probability of attack by location. In so doing, they maximize the system‐wide probability of detecting an ‘event of interest’ subject to a constraint on the expected number of false signals by differentially setting detection thresholds at each location in accordance with the probability that each location is attacked.…”

“…In an example, Fricker and Banschbach 35 show how one might optimally set thresholds on a hypothetical biosurveillance system designed to simultaneously monitor all 3141 counties in the United States. For the purposes of illustration, they used the proportion of the total population in a county as a surrogate for the probability that the county is attacked.…”