New temporal methods for detecting disease outbreaks

A new procedure for detecting disease outbreaks using multiple data streams is presented. Three popular methods for identifying temporal clusters in statistical process control (SPC) – the Shewhart chart, the CUSUM, and the EWMA – are extended for detecting disease outbreaks in multiple geographic regions.  The proposed method has some features that are more advantageous for disease surveillance than most conventional SPC methods. The proposed method does not assume the disease counts to follow a specific distribution, as disease count data tend to violate assumptions of conventional SPC methods. A data-adaptive approach is used with the proposed method to regularly update baseline data to account for population changes over time, unlike conventional methods that rely on fixed baselines. The relatively low false alarm rate is a highlight of the proposed method, as excessive false alarms are a common problem with conventional SPC methods. The proposed method uses p-values and controls the false discovery rate enabling the utilization of more powerful multiple comparison procedure as opposed to conventional SPC methods that rely on less powerful family-wise error rate control methods. Through extensive simulation studies, it is shown that the EWMA and CUSUM methods have superior performance over the long-established Shewhart charts. The rapid detection ability of the method is illustrated with the 2011 E.Coli outbreak in Germany.

Audience Take Away:

• The presentation provides novel methodologies to design disease surveillance systems
• Other faculty with research interests in areas such as outbreak detection, biostatistics, etc., could use this work to expand their own research