Statistical Performance of Spatial Systems Open Access

Wang, Yuemei (2009)

Permanent URL: https://etd.library.emory.edu/concern/etds/41687h61z?locale=en
Published

Abstract

Detection of disease outbreaks is a crucial issue in public health. Therefore, we want statistical methods to evaluate the accuracy and reliability of proposed detection systems. Furthermore, detecting outbreaks in space is very challenging as the shapes and locations of outbreak clusters of disease can be unpredictable. In this thesis, we use area under ROC curve to evaluate statistical performance of several proposed spatial detection systems. Specially, we assess spatial statistical performance of two spatial scan statistics, and their applications to cardiac birth defect data from Santa Clara County, California. The results reveal SaTScan performs better if the cluster is compact, and Upper Level Set approaches offer improved performance when clustering is irregularly shaped. We also investigate performance of cluster detection methods when adjusting for covariates via Generalized Additive Models (GAM). We apply GAMs to archaeolog- ical data from Black Mesa, Arizona to identify clusters of early versus late Anasazi settlement sites when adjusting for exposure to rivers around those sites. Further- more, we evaluate spatial variations in power to detect different levels of clustering when clusters are allowed to occur at different locations within this application. We compare the GAM results and performance of the GAM methodology with those based on kernel density estimation of the early-to-late relative risk surface. Finally, we assess spatial performance of detection systems using decision fusion theory for the situation where a detection system can be comprised of a few expensive, precise detectors and many inexpensive, imprecise detectors. The performance of a system depends not only on the total allowable cost for the system, but also the performance of each individual detector, as well as the balance between expensive, precise and inexpensive, imprecise detectors. We quantify how, if we improve the performance of imprecise detectors even slightly, the performance of the resulting system improves dramatically. In addition, we show that lower-cost systems can perform as well as or better than systems expending the full allowable cost. These results indicate the need for careful calculation and computation to identify an optimal system, especially for systems comprised of small numbers of components.

Table of Contents

1. Introduction and Background

2. Cluster Detection Methods

3. Data Sets

4. Spatial Measures of Performance: Regional Count Data

5. Spatial Measures of Performance: Regional Point Data for Cases and Controls

6. Spatial Performance for Outbreak Detection Systems

7. Summary and Future Research

About this Dissertation

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
School
Department
Degree
Submission
Language
  • English
Research Field
Keyword
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files