Integrating epidemiological data and mathematical models to forecast COVID-19 spread in the United States

The rapid global outbreak of COVID-19 has raised interest in the computational forecast of the spread of infectious diseases, but the early projections in the current pandemic were limited in their ability to describe longer-term outcomes. This issue was partially due to the limited knowledge of the mechanisms of disease spread and development. Our study aims to integrate epidemiological time-series data into a mathematical model that can describe the fundamental mechanisms of COVID-19 spread, with the ultimate goal of utilizing model forecasts to determine early indicators of large outbreaks as well as assessing public health interventions to control their severity. We used publicly available data from the 10 most heavily impacted states in the US to calibrate a SEIRD-type model and obtain state-specific sets of epidemiological parameters. Our model was able to recapitulate the early observations of cumulative infections and deaths (CCC > 0.9, R2 > 0.9). We further explore the use of model parameters and forecasts as early indicators of subsequent large outbreaks. Finally, we argue that mechanistic models that describe infectious disease spread can help mitigate the human cost of pandemics by anticipating effective public health interventions and enabling the optimized allocation of key medical resources.