Statistical methods for real-time forecasts of infectious disease: expanding dynamic time-series and machine learning approaches for pandemic scenarios

The emergence and global expansion of SARS-CoV-2 as a human pathogen over the last four monthsrepresents a nearly unprecedented challenge for the infectious disease modelling community. This pandemichas benefitted from huge volumes of data being generated, but the rate of dissemination of these data hasoften outpaced existing data pipelines. While the last decade has seen significant advances in real-timeinfectious disease forecasting - spurred by rapid growth in data and computational methods - thesemethods have primarily focused on seasonal endemic diseases based, are based on historical data, and sodo not apply easily to this novel pathogen, or to pandemic scenarios. New methods are needed to leveragethe wealth of surveillance data at fine spatial granularity, together with associated information about policyinterventions and environmental conditions over space and time, to reason directly about the mechanisms toforecast and understand the transmission dynamics of SARS-CoV-2 transmission. These methods must usesound statistical and epidemiological principles and be flexible and computationally efficient to provide real-time forecasts to guide public health decision-making and respond to changing aspects of this global crisis.The central research activities of this project are (1) to develop scalable, computationally efficient Bayesianhierarchical compartmental models to flexibly respond to state-level public health forecasting needs, and (2)to design models and conduct analyses to draw robust inference about the effectiveness of interventions inimpacting the reproductive rate of SARS-CoV-2 infections within the US to build an evidence-base forcontinued responses to COVID-19 and future pandemics.