A Bayesian approach to the detection of aberrations in public health surveillance data.

The accurate and timely detection of unusual patterns in data from public health surveillance systems presents an important challenge to health workers interested in early identification of epidemics or clues to important risk factors. We apply the Kalman filter, a Bayesian method, to public health surveillance data collected in the United States to illustrate a methodology used to detect sudden, sustained changes in reported disease occurrence, changes in the rate of change of health event occurrence, as well as unusual reports or outliers. The method allows use of information external to reported data in forecasting expected numbers, information such as expert judgment, changes in case definition or reporting practices, or changes in the health event process. Results show good agreement with epidemiologically established patterns beginning early in the data series and demonstrated usefulness on a relatively short series. Because the method is unfamiliar to most practicing epidemiologists in public health, it should be compared with other techniques and made more "user-friendly" before general application.