Howard S. Burkom, PhD¹, Eugene Elbert, MS², Andrew Feldman, PhD¹, Jeffrey Lin, MS¹, and Sean Patrick Murphy, MS¹. (1) National Security Technology Department, Johns Hopkins Applied Physics Laboratory, 11100 Johns Hopkins Road, Laurel, MD 20723, 240-228-4361, Howard.Burkom@jhuapl.edu, (2) Walter Reed Army Institute of Research, DoD Global Emerging Infections System (DoD-GEIS), 503 Robert Grant Avenue, Silver Spring, MD 20910

The Electronic Surveillance System for the Early Notification of Community-based Epidemics (ESSENCE) has been enhanced by the Department of Defense Global Emerging Infection System and the Johns Hopkins University Applied Physics Laboratory to monitor military and civilian data streams for disease outbreaks. This system monitors time series of multiple data sources stratified by covariates such as patient location and diagnosis type. Detection algorithms flag anomalies represented as alerts for follow-up by epidemiologists. The screening occurs on multiple jurisdictional levels. Systems growing in complexity must retain sensitivity, specificity, and timeliness. We have implemented combinations of regression modeling and statistical process control to address the alerting performance issues. Multivariate, multiple univariate, and hybrid strategies are used. Modifications of multivariate methods avoid oversensitivity to irrelevant changes in covariance among the data streams. Bayes belief nets (BBNs) are among several strategies for fusing outputs of algorithms applied to separate covariates. Our BBN approach fuses outputs into a continuous-valued alert probability by assigning probabilities to detected anomalies in individual data streams. These alerting strategies proved sensitive and robust in a 5-city retrospective exercise including several authentic data streams. Both univariate and multivariate approaches gave timely alerts for a set of outbreaks identified by a group of medical epidemiologists. The application of a BBN to separate algorithm outputs improved alerting timeliness in some contexts. As these systems grow, a hybrid suite of process control algorithms in a Bayes net framework is a viable model for the versatility that will be needed.

Learning Objectives:

• Modify appropriate multivariate anomaly detection algorithms for required detection performance.
• Combine univariate algorithms to retain sensitivity and avoid multiple testing problems.
• Assess system detection performance.

Keywords: Surveillance, Statistics

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.