Wavelet-Based Monitoring in Modern Biosurveillance

Abstract:     
Current biosurveillance relies on classical statistical control charts for detecting disease out-breaks. However, these are not always suitable in this context. Assumptions of normality, independence, and stationarity are typically violated in syndromic data. Furthermore, outbreak signatures in such data are of unknown patterns, and therefore call for general detectors. We propose wavelet-based methods, which make less assumptions and are suitable for detecting abnormalities of unknown form. Wavelets have been widely used for data denoising and compression, but little work exists on using them for monitoring. We discuss monitoring-based issues and illustrate them using data on military clinic visits.

Keywords: Early detection, autocorrelation, disease outbreak, syndromic data, discrete wavelet transform

Working Paper Series