Simulating COVID-19 in a university environment

Abstract

Residential colleges and universities face unique challenges in providing in-person instruction during the COVID-19 pandemic. Administrators are currently faced with decisions about whether to open during the pandemic and what modifications of their normal operations might be necessary to protect students, faculty and staff. There is little information, however, on what measures are likely to be most effective and whether existing interventions could contain the spread of an outbreak on campus. We develop a full-scale stochastic agent-based model to determine whether in-person instruction could safely continue during the pandemic and evaluate the necessity of various interventions. Simulation results indicate that large scale randomized testing, contact-tracing, and quarantining are important components of a successful strategy for containing campus outbreaks. High test specificity is critical for keeping the size of the quarantine population manageable. Moving the largest classes online is also crucial for controlling both the size of outbreaks and the number of students in quarantine. Increased residential exposure can significantly impact the size of an outbreak, but it is likely more important to control non-residential social exposure among students. Finally, necessarily high quarantine rates even in controlled outbreaks imply significant absenteeism, indicating a need to plan for remote instruction of quarantined students.

Keywords: Agent-based modeling; COVID-19; Computational epidemiology; Coronavirus; Epidemics; Higher education; Post-secondary education; SARS-CoV-2.

Fig. 1

Distribution of class sizes (before splitting very large classes into smaller sections) averaged over 100 randomly-generated universities. Here the proportion indicates the fraction of all $\sim$$3,750$ courses which fall into the given size bin. The largest class in the university has approximately 800 students before subdivision.

Fig. 2

Cumulative infections over a 100-day semester. The black curve is the median daily value for 500 independent runs of the simulation and the dark and light regions indicate quantiles around the median containing 50% and 90% of outcomes, respectively. Under no intervention the median cumulative number of new infections reaches $20,126$ (falling between $20,026$ and $20,212$ in 90% of simulations), or 89.4% of the total campus population.

Fig. 3

Median number of total community members infected over the course of the semester and peak number of students in quarantine by intervention bundle, plotted on a log scale. Bars indicate quantiles around the median containing 50% and 90% of outcomes, respectively.

Fig. 4

Total infections and peak quarantine population as a function of the online class size threshold. The black curve is the median value for 500 independent runs of the simulation and the dark and light regions indicate the 50% and 90% quantiles, respectively.

Fig. 5

Number of days that an average class experiences $\ge 10\text{\%}$ of students in quarantine as a function of the online class size threshold. The black curve is the median value for 500 independent runs of the simulation and the dark and light regions indicate the 50% and 90% quantiles, respectively.