6th Annual Public Health Information Network Conference: Scenario-Based Evaluation of Cluster Detection and Tracking Capability

Wednesday, August 27, 2008: 10:40 AM

Atlanta EFG

Howard S. Burkom, PhD , Johns Hopkins University Applied Physics Laboratory, Laurel, MD

Linda Moniz, PhD , NSTD-STJ, Johns Hopkins University Applied Physics Laboratory, Laurel, MD

Jian Xing, Phd, MS , Centers for Disease Control and Prevention

Jerome I. Tokars, MD, MPH , National Center for Public Health Informatics, Centers for Disease Control and Prevention, Atlanta, GA

Audio File Slides Recorded presentation

Situational awareness of plausible public health problem scenarios is driving current research in automated disease surveillance. In this presentation, we describe a disease cluster detection and tracking method that may also be used as a potential early warning. We also describe methods for evaluation of current surveillance tools. Focused evaluation and comparison are essential for the standardization of automated detection systems.

Few prospective detections of localized outbreaks have been documented; most published studies have been retrospective and do not include spatially detailed data from known outbreaks. This study presents tools for projecting the likely course of an outbreak scenario onto an authentic dataset. These tools use location information at the same spatial resolution as the location fields in the original data records. The tools implement basic adjustable geometries that reflect disease spread constrained by geography, weather, and travel patterns. Users may tune the model to reflect changes in spatial and temporal spread.

We apply this outbreak simulation tool to inject stochastically drawn case counts into the original data set. Using repeated trials with the cluster detection tools, we measure performance of a scan statistics implementation with regard to sensitivity and timeliness for several scenarios. We evaluate the ability to detect and track disease clusters as functions of outbreak size, concentration and growth pattern.

The study dataset is a 3-year collection of daily outpatient visit clinic counts derived from the Department of Defense data monitored in the BioSense program. Practical guidance suggested by study results will be presented. For example, given a localized outbreak with a modest spatial spread over the course of 10 days, we found that the number of cases required for a detection with 90% probability at nominal alerting levels roughly doubles for an urban region relative to a rural one.

See more of: Evaluation of Surveillance Systems
See more of: Abstracts

<< Previous Abstract | Next Abstract