Introduction: In recent years, many syndromic surveillance systems have been deployed around the United States for the early detection of biologic terrorism-related and naturally occurring outbreaks. These systems and the associated aberration detection methods need to be evaluated.
Objective: This study evaluated several detection methods of the Early Aberration Reporting System (EARS) under serially correlated syndromic data and to demonstrated the need for calibrating these methods.
Methods: In an initial evaluation of the Syndromic Tracking and Reporting System in Hillsborough County, Florida, serially correlated syndromic data were simulated using statistical models in conjunction with real syndromic data. The detection methods were tested against two patterns of simulated outbreaks. They were compared using a conditional average run length and a receiver operating characteristic curve under defined patterns of detection.
Results: Increasing serial correlation inflates the false alarm rate and elevates sensitivity. Among the detection methods in EARS, C2 and P-chart have the best overall receiver operating characteristic curve within the context of the simulations. C2 is least affected by the serial correlation, the outbreak type, and the defined patterns of detection signal.
Conclusion: Evaluation of the detection methods needs to be adaptable to the constantly changing nature of syndromic surveillance. Deployment of EARS and other methods requires adjusting the false alarm rate and sensitivity in accordance with the syndromic data, the operating resources, and the objectives of the local system. For timely detection, C2 is superior to other methods, including C3, under the simulation conditions. P-chart is the most sensitive when the serial correlation is negligible.