Quantitative Methods in Defense and National Security 2007

Evaluation of the DC Department of Healths Syndromic Surveillance System
Arvind K. Jain, (RAND Corporation), arvind_jain@rand.org,
Beth Ann Griffin, (RAND Corporation), Beth_Ann_Griffin@rand.org,
Michael Stoto, (Georgetown University), stotom@georgetown.edu,
John Davies-Cole, (Department of Health, Washington DC), john.davies-cole@dc.gov
Chevelle Glymph, (Department of Health, Washington DC), chevelle.glymph@dc.gov,
Garret Lum, (Department of Health, Washington DC), garret.lum@dc.gov,
Gebreyesus Kidane, (Department of Health, Washington DC), gebreyesus.kidane@dc.gov,
and Samuel C. Washington, (Department of Health, Washington DC), sam.washington@dc.gov


Immediately following September 11, 2001, the District of Columbia Department of Health (DOH) began a syndromic surveillance program based on emergency room (ER) visits. Syndromic surveillance involves the collection of data (e.g. sales volume of over-the-counter anti-nausea medications) and simultaneous analysis in order to monitor the health of a specific population. In our data, ER logs from nine hospitals are transmitted daily to the health department and categorized into mutually exclusive syndromes such as unspecified infection and gastrointestinal illness. The data are then analyzed daily using a variety of statistical detection algorithms. This paper characterizes the performance of these statistical detection algorithms in practical terms, and helps identify the optimal parameters for each algorithm given the DC DOH data as well as the most effective algorithms. Analyzes were conducted to improve the sensitivity of each algorithm to detecting simulated outbreaks by fine tuning key parameters used in the algorithms. Simulation studies using the data show that over a range of simulated outbreak types, the multivariate CUSUM algorithms performed more effectively than other algorithms. Performance of the algorithms is also examined by applying them to known outbreaks such as flu seasons and a previously undetected series of gastrointestinal illness outbreaks. Our analyzes appear to indicate that the DC DOH system may prove to be more valuable in identifying the beginning of the flu season than for bioterrorist attacks. The analysis also indicates that when researchers/analysts apply these algorithms to their own data, fine tuning of parameters is necessary to improve overall sensitivity.

Take me back to the main conference page.