Enhancing a university community COVID-19 model with Bayesian model-data fusion

During the COVID-19 pandemic, universities experienced unique challenges, and current research has used mathematical models that leveraged model parameters calibrated to local contexts, such as hospitals, to investigate these challenges. However, key questions remain regarding the impacts of a formal model calibration that uses data from a university, on model performance and parameter inference. We use an adapted Susceptible-Exposed-Infected-Recovered (SEIR) compartment model, that represents a semi-enclosed campus population in tandem with surveillance testing data from a medium-sized private university in the northeastern United States’ Fall 2020 semester. We leverage a formal Bayesian model calibration to quantify uncertainty in model parameters and diagnose model shortcomings by comparing positive tests and the isolation population between the model and the data. We introduce new model developments to address the model’s inaccuracy in representing real-world data. The introduction of new model developments and model calibration leads to a 12% and 84% decrease in error for the positive tests and isolation population, respectively, and an improved qualitative representation of the data. This noticeable decrease in error values demonstrates the importance of fuzing real-world data with mathematical models to produce accurate and informed models that can be leveraged for public health efforts.