Recurrent outbreaks of influenza and other respiratory viruses continue to affect human health adversely. A number of intervention strategies exist to mitigate the progression of these pathogens, including vaccination, anti-viral therapeutics, public awareness campaigns, face masks, school closure, and quarantine. Public health agency use of these control strategies is guided by their historical effectiveness and implemented in light of the latest estimates of infection incidence, severity, and transmissibility; however, public health officials would be afforded more time to allocate their intervention measures if local outbreak characteristics, e.g., incidence timing, magnitude and duration, could be accurately and reliably forecast. Recent work has shown that some characteristics of seasonal influenza outbreaks can be predicted accurately with lead times of up to 9 weeks. These predictions are generated with a mathematical model of influenza transmission dynamics that has been recursively optimized using an ensemble data assimilation technique and real-time observations of infection incidence. In practice, the data assimilation process entrains the observational estimates of infection incidence into evolving mathematical simulations of pathogen transmission dynamics, and in so doing trains those model simulations, through state space estimation and parameter optimization, to better match the observed unfolding outbreak. Those trained simulations, having been optimized with the most recent observations, are then integrated into the future to generate a distribution of potential disease outcomes. This forecasting framework has been validated for accuracy and reliability, and during the 2012-2013 influenza season was used to generate weekly real-time predictions of influenza peak timing for 108 cities throughout the United States. For this project, we will build on and expand these forecast efforts. Specifically, we will: 1) Work to improve influenza forecast accuracy and reliability through development of multi-model forecast approaches, such as have been used in weather prediction; 2) Develop, test and analyze analogous forecast frameworks for other recurrent respiratory pathogens, such as rotavirus and respiratory syncytial virus; 3) Establish a dedicated operation center for maintaining, running and disseminating real-time weekly forecasts of influenza and other respiratory viruses; and 4) Work with public health officials in New York City, and, using their more detailed syndromic surveillance, explore the potential for more granular, borough or neighborhood-scale forecast of influenza and other viruses. These efforts will lead to an improved understanding of the benefits and limits of respiratory disease prediction, and the intelligent interpretation and incorporation of real-time forecasts in health response decision-making.
In the last few years, new methods have been introduced to epidemiological modeling that enable the accurate forecasting of infectious disease. For this project, we will expand these disease prediction capabilities and generate operational real-time forecasts of influenza and other respiratory pathogens. Specifically we will: 1) develop multi-model ensemble forecast methods; 2) establish a dedicated operational center for maintaining, running and disseminating the forecasts in real time through a free, user- friendly web portal; and 3) work with pubic health agencies and officials to ensure that the forecasts and web portal meet their needs.
Showing the most recent 10 out of 39 publications
