| This dissertation addresses whether syndromic surveillance is a feasible component of an integrated biodetection system. Rapid detection of disease outbreaks is extremely desirable. Existing methods to do so, either through environmental sampling (biosensors) or through clinical detection, are challenged in terms of reliability and timeliness. An additional form of detection, syndromic surveillance, has been proposed to statistically monitor anomalies in preexisting data streams of pre-diagnostic health seeking behaviors for early signs of disease outbreaks. However, questions have been raised about the sensitivity, specificity and timeliness of such methods for disease detection. This dissertation outlines a methodology to show that such surveillance can indeed perform early detection of outbreaks of non-specific, flu-like symptoms that would be typical of seasonal respiratory illness or certain select agents. It further describes a large scale data experiment using de-identified historical medical encounter records to detect outbreaks "virtually prospectively," that is, calculating a probability of outbreak for each day's worth of data as they are processed. The dissertation then goes on to identify a methodology to simulate an anthrax release, and use it to augment de-identified historical encounter data to test typical syndromic surveillance detection algorithms. In general, syndromic surveillance was found to have good sensitivity, specificity and timeliness to serve as part of a biodetection system that includes environmental sampling and clinicians. |
