Toward Effective Anomaly Detection in Public Health Surveillance

• Activity Number: 143
• Type: Contributed
• Date/Time: Monday, August 7, 2006 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statisticians in Defense and National Security
• Author(s): Colin Goodall*+ and Sylvia Halasz and Arnold Lent and Simon Tse and John Allegra and Dennis Cochrane
• Companies: AT&T Labs-Research and AT&T Labs-Research and AT&T Labs - Research and AT&T Labs - Research and Emergency Medical Associates Research Foundation and Emergency Medical Associates Research Foundation
• Keywords: anomaly detection ; public health ; biosurveillance ; emergency department data ; alert

Abstract

In biosurveillance' use of statistical anomaly detection, multiple streams of data---from hospital emergency departments, testing laboratories, pharmacies, etc. - are continually examined for abnormal patterns. The detected anomalies, with attached measures of significance and severity, are queued for investigation, leading to possible action. A partially-adaptive anomaly detection method has been applied to daily feeds of emergency department data using a statistically-principled system of generated outbreaks. Alerts were generated for a variety of parameter settings, giving state public health officials a tradeoff among number of false alerts, the smallest outbreak to be detected, the outbreak profile, and the timeliness of detection.