Performance of Residual-Based Control Charts Using Generalized Exponential Smoothing of Syndromic Data for Routine Health Surveillance

• Activity Number: 317
• Type: Invited
• Date/Time: Tuesday, August 8, 2006 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Howard S. Burkom*+ and Galit Shmueli and Sean Murphy
• Companies: Johns Hopkins University Applied Physics Laboratory and University of Maryland and Johns Hopkins University Applied Physics Laboratory
• Keywords: process control ; syndromic surveillance ; exponential smoothing ; regression control charts

Abstract

Syndromic surveillance depends on the monitoring of consumer data sources for early warning of unspecified disease outbreaks. These sources include clinical data, such as counts of emergency department visits, and nonclinical data, such as over-the-counter remedy sales. For sensitivity and timeliness at practical alert rates, developers have tried to adapt chart-based methods of statistical process control. Obstacles are evolving, often nonstationary input data streams, target signal uncertainty, and cyclic or seasonal background data behavior. Reliable performance often requires a combination of modeling and process control. This presentation applies generalized exponential smoothing to make forecasts and measures predictive accuracy and the performance of residual-based charts. This approach is compared to conventional methods for sensitivity and robustness using several data types.