Towards Real Time Epidemiology: Data Assimilation, Modeling And Anomaly Detection Of Health Surveillance Data Streams

An integrated quantitative approach to data assimilation, prediction and anomaly detection over real-time public health surveillance data streams is introduced. The importance of creating dynamical probabilistic models of disease dynamics capable of predicting future new cases from past and present disease incidence data is emphasized. Methods for real-time data assimilation, which rely on probabilistic formulations and on Bayes' theorem to translate between probability densities for new cases and for model parameters are developed. This formulation creates future outlook with quantified uncertainty, and leads to natural anomaly detection schemes that quantify and detect disease evolution or population structure changes. Finally, the implementation of these methods and accompanying intervention tools in real time public health situations is realized through their embedding in state of the art information technology and interactive visualization environments.