The focus of this project is to develop accurate and reliable forecasts of influenza outbreak characteristics. We use data assimilation methods, as commonly employed in numerical weather prediction, in conjunction with real-time observations of influenza incidence to train and optimize model simulations of influenza transmission dynamics on the fly and then use those optimized models to generate real-time forecasts of influenza outcomes. Our forecasts have been shown to be accurate with lead times of up 10 or mores weeks. As part of this work we test and develop a range of model forms and data assimilation methodologies.
In 2014, using the methods we developed in this project, Jeffrey Shaman, Wan Yang, Alicia Karspeck, and Marc Lipsitch won the ‘CDC Predict the Flu Challenge’. The prize announcement can be found here:
Through this project we have also begun to address more fundamental questions related to the prediction of nonlinear systems and, in particular, the data needed to facilitate such predictions. The objectives are to understand and quantify the trade-off between model complexity and data richness, the levels of model complexity that could feasibly produce accurate and reliable forecasts, and the abundance, frequency and quality of data needed to make those forecasts possible.
Funding for this work was first provided by the NIH (NIGMS)/NSF (DMS) joint initiative to support research at the interface of the biological and mathematical sciences.
Additional funding comes from the Biomedical Advanced Research and Development Authority of the Department of Health and Human Services, as well as the Models of Infectious Disease Agent Study (MIDAS) program of the the NIH and the Defense Threat Reduction Agency.