- Introduction
Since the outbreak of COVID-19, there have been many disruptions in everyday life, including sudden campus closures and transitions to remote learning. However, as we work towards returning to normalcy amid these changes, the need to make informed, data-driven decisions on how to safely reopen college campuses amid the COVID-19 pandemic is critical. In response to this need, researchers at the NSF Spatiotemporal Innovation Center and George Mason University (GMU) have developed an interactive web service, called SpatioteMporal cAmpus/school Reopen decision support system (S.M.A.R.T.), to model the spread of COVID-19 in a simulated setting to better inform reopening strategies and systems.
The school reopening simulation system is built to forecast and recommend policies for controlling the spread of COVID-19. It is targeted to serve the decision makers of college campus and county-level school system reopening. This project is envisioned to be a close collaboration between academia and communities to provide COVID-19 resilience decision support tools. It fully considered the community requirements and takes in historical and real time data to eventually research, develop, and operate an innovative service/product for counties and campuses nationwide to address the compelling challenges of whether or not we reopen schools and how to do so during the pandemic. This achieved through the following goals: 1) researched and developed Spatiotemporal campus/school Reopen monitoring and decision support system for COVID-19 to simulate potential outbreak and transmission within a school/system, and 2) deployed a highly available cloud-based service to college and K-12 school systems for school reopening decision support.
- Method
From considering factors like testing capability to the likelihood of asymptomatic spread among students and faculty, this tool allows users to customize conditions and create simulated scenarios on transmissions and mitigation techniques. The SMART utilizes a discrete-time, agent-based model that expands upon the base SEIR model by accounting for testing, masking, symptoms monitoring, and social isolation, among others.
There are 5 components in this project: 1) Historical and near real-time datasets are being collected and quality controlled by the team through past spatiotemporal I/UCRC investigation and NSF Rapid Response Project and shared with the community through GitHub and Harvard Dataverse. 2) A classic Susceptible-Exposed-Infective-Recovered (SEIR) model and Agent-Based Modeling (ABM) have been investigated based on our previous research to simulate the potential scenarios with different opening strategies for campus and school systems. 3) We validated, verified, and developed a user-friendly dashboard and report for decision makers in the civic communities who make decisions for school reopening. 4) These services have been deployed, tested, and verified by the communities on highly available zones of the Amazon Web Service (AWS) cloud computing platform based on our past experiences (Figure 1).
- Validation Study Case
To validate if the SMART system could accurately simulate the COVID-19 spreading on campus, the GMU campus was used as a test bed.13% of people (~5000) returned to campus, and simulations were performed under different control strategies: 1) no control; 2) wear mask; 3) social distancing ; 4) symptom monitoring; and 5) all policies combined with testing and contact tracing (Figure 2). This was being adopted for GMU campus reopening decision support.
The results represent the different infection curves under the simulated scenarios for GMU’s campus. By comparing predicted cases curves with the grey area for “No control,” the performance of each control strategy can be evaluated. When stricter policies were involved, the infection rate curve flattened. However, those policies cannot always be implemented perfectly. Policy enforcement levels are thus applied in the simulation, which result in the dynamics representing changes in different enforcement levels. In the meantime, uncertainty was monitored in this simulation. Each curve was generated by fifteen iterations of the simulation process, which represents model stability.
By comparing the simulated results with recorded positive cases on GMU’s campus, a weekly validation has been done for the first 4 weeks with the confirmed cases reported in the GMU COVID-19 information dashboard (Figure 3).
Results show most of the confirmed cases fall into the range of simulated results, which represent the max and min for the worst case and best case generated by our model. Some of the cases do not belong to this range, which may results from how 1) not all population on campus have been tested, which may results in the asymptomatic has not been counted, 2) this system replicated most daily activities of the people on campus, however, some privacy information may missed such as room share, off campus contact etc., 3) more accurate R0 and contact matrix may be adopted to enhance the accuracy of this service.
- Online App
An interactive web service that implements this model for public use is available (https://smartgateway.stcenter.net/); the software is user-friendly and has built-in assistance. This program allows the user to modify the values for each parameter, and a standardized output for the given scenario is produced. The user is given the choice to access interactive charts or to generate a numerical data report. The information derived from this online service can serve as an added information for campus/school administrators to prepare reopening prompt a response that will ensure the safety of as many individuals on campus as possible, and mitigate the detrimental effect of COVID-19 on each school’s campus.