Detecting Disease Outbreak Regions Using Multiple Data Streams

A novel approach for biosurveillance using multiple data streams is presented. The proposed method is computationally simple, has rapid detection ability, and produces few false alarms. The proposed algorithm is applied to three popular statistical process control (SPC) charts: the Shewhart Chart, the EWMA, and the CUSUM. The proposed method collects disease counts from multiple data streams, computes charting statistics, and then compares these to empirical in-control distributions generated using bootstrap methods to decide whether to signal an alarm. As bootstrap methods are used, no assumption is made about the in-control distributions corresponding to a specific parametric distribution—an assumption that is common with most conventional SPC methods. The proposed method relies on p-values and controls the false discovery rate; which distinguishes it from traditional SPC methods. The relatively low false alarm rate is a highlight of the proposed method, as higher false alarm rates are a common problem with conventional SPC charts. Through extensive simulations, the EWMA and CUSUM methods are shown to have superior performance over the widely-used Shewhart charts, with the EWMA having a slight advantage over the CUSUM. The proposed method is applied to the 2011 Escherichia coli outbreak in Germany.