Automated detection of epidemics from the usage logs of a physicians' reference database

Abstract

Epidemics of infectious diseases are usually recognized by an observation of an abnormal cluster of cases. Usually, the recognition is not automated, and relies on the alertness of human health care workers. This can lead to significant delays in detection. Since real-time data from the physicians' offices is not available. However, in Finland a Web-based collection of guidelines for primary care exists, and increases in queries concerning certain disease have been shown to correlate to epidemics. We introduce a simple method for automated online mining of probable epidemics from the log of this database. The method is based on deriving a smoothed time series from the data, on using a flexible selection of data for comparison, and on applying randomization statistics to estimate the significance of findings. Experimental results on simulated and real data show that the method can provide accurate and early detection of epidemics.