|
Activity Number:
|
305
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statisticians in Defense and National Security
|
| Abstract - #300856 |
|
Title:
|
Discrimination and Thresholding in Evaluation of Disease Cluster Detection Algorithms
|
|
Author(s):
|
Linda Moniz*+ and Howard S. Burkom and Jian Xing
|
|
Companies:
|
Johns Hopkins University and Johns Hopkins University and Centers for Disease Control and Prevention
|
|
Address:
|
11100 Johns Hopkins Road, Laurel, MD, 20723,
|
|
Keywords:
|
scan statistics ; cluster detection
|
|
Abstract:
|
A number of researchers have evaluated disease cluster detection algorithms based upon measurements of specificity and sensitivity, usually represented in ROC or AMOC curves. However, the ROC and AMOC curves are quite sensitive to the thresholds used in the clustering algorithms. The clustering algorithms, in turn, are sensitive to both cluster geometry and density of the background (baseline) distributions. In this study we vary both cluster geometry (via injected disease cases) and thresholds (via both with selection of geographic regions according to background density and with explicit thresholding for the ROC curves). We show the evolution of the ROC curves as a function of these components and indicate when the ROC curves can faithfully evaluate the algorithms as well as when they are a misrepresentation of the results of the algorithm.
|
