|
Activity Number:
|
392
|
|
Type:
|
Invited
|
|
Date/Time:
|
Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Government Statistics
|
| Abstract - #300172 |
|
Title:
|
Performance Assessment for Biosurveillance Alerting Algorithms
|
|
Author(s):
|
Howard Burkom*+
|
|
Companies:
|
Johns Hopkins University
|
|
Address:
|
11100 Johns Hopkins Rd., Laurel , MD, 20723-6099,
|
|
Keywords:
|
alerting algorithms ; biosurveillance ; receiver operating characteristic ; activity monitor operating characteristic ; outbreak detection
|
|
Abstract:
|
The objective of alerting algorithms in biosurveillance is to assist in the timely recognition of disease outbreaks by detecting their early effects on monitored data streams. These algorithms are applied both to clinical data such as physician office visits and to nonclinical data such as over-the-counter sales. The uncertainty in the effects of outbreaks on these datasets makes it difficult to evaluate detection methods. This talk presents a methodology to assess the utility of alerting algorithms using authentic background data and simulated outbreak effects based on theoretical epicurves of primary cases. For practical expected false alert rates, receiver operating characteristic (ROC) methods are used to measure the sensitivity of these algorithms, and activity monitor operating characteristic (AMOC) methods are used to measure their timeliness. The approach will be illustrated with methods of the Early Aberration Reporting System (EARS) developed at the U.S. Centers for Disease Control and Prevention. The detection performance of these algorithms will be analyzed for various data backgrounds and plausible signal effects. Optimal threshold selection will be discussed.
|
