As the new academic year opens across the country, one key unanswered question is how to reopen college campuses to in-person classes and yet still manage the spread of disease.
Some schools have opted for exclusively online classes, others have invited students back under strict isolation and social distancing rules, and yet others have adopted a hybrid approach. So far, the results have been mixed. Some universities opened only to backpedal when student illnesses spiked. And at least one school had to expel students for too much social interaction.
Ellen Kuhl, chair of Stanford’s mechanical engineering department, is taking a proactive approach to solving the reopening question: a computer model that combines machine learning and physics-based modeling to play out various reopening scenarios. Now, she’s putting that model in the hands of students in a virtual course for the fall: ME233: Data-Driven Modeling of COVID-19.
In the course, students will build upon and customize Kuhl’s existing code to explore various policy strategies for bringing students back to campus. Students will work in teams to solve various aspects of the COVID conundrum, building through the quarter toward a final project in which the students make risk assessment recommendations for a safe, phased reopening of the Stanford campus.
Kuhl has drawn global attention in recent months applying the model — once used to predict electrical signaling in the brain — to understand human travel patterns and predict how various disease control policies might affect the spread of COVID-19. The model was recently used as the scientific basis in court proceedings in Newfoundland, Canada, exploring various reopening strategies for the province.
Stanford Engineering talked to Kuhl and a postdoctoral scholar in her lab, Kevin Linka, the key developer of the model, to discuss what led them to turn their research into a course and what their models say about the pace of reopening campuses across the country.
Kuhl: Over the last six months or so, we’ve published several papers, and we received an email from the office of the Attorney General of Newfoundland. Newfoundland had closed itself to all external traffic in March and, as a result, had zero new cases of COVID since April. Closing Newfoundland was easy; the island is only connected by two ferry routes and one airport to the mainland. But then some private citizens started to challenge the policy because they wanted to travel to Newfoundland and the government had to defend itself in court. The representing attorney of the government contacted us and said, “We’ve seen your papers and your work. Can you help?”
Linka: They were interested in forecasting different reopening scenarios. So one scenario that they had in mind were these so-called travel bubbles, where they just open up travel to a specific region — either to only the three Atlantic provinces of Canada, or to all of Canada, or to the entirety of North America. They wanted to see what enforcing a 14-day quarantine for all incoming travelers would look like and what would happen if certain people don’t follow the rules.
Kuhl: The model has confirmed that even a very small quarantine violation can have huge impact. In an ideal world, students will come to campus and get tested, then isolate and get tested again before being cleared for class. But as we’ve seen just recently in many examples where travel is involved — not just in campus reopenings — people are people and they don’t always follow these strict rules. Our model is able to account for some form of quarantine violation.
Linka: I think the insight I take is that reopening has to be very careful. In Newfoundland, we found that even if you quarantine 95% of the population, you would still see 500 new cases, that’s 0.1% of the population, within less than three months. It just multiplies so quickly, even with two or three uncontrolled cases. Ninety-five percent quarantine seems pretty good, but that would mean that still one in 20 students could spread the disease. And this one person doesn’t just infect one other person on average, but several others. It’s a multiplier effect. Without control, the numbers could rise dramatically, even within a single quarter.
Kuhl: We hear a lot of frustration from people, from students, parents, and faculty, who want the students back on campus. The in-person Stanford experience is so important, for undergraduates especially, and they are eager to return. And, if you think about it, this sort of travel to Newfoundland is actually very similar to reopening a campus, where students come from far away to one single central location. Our campus is currently healthy and more or less closed to external traveling. There is a thoughtful plan to regularly test all students when they return. But there’s a human factor to it, too. Our students are young, they want to get together with other students, and they don’t always understand why they have to quarantine, especially when they don’t feel sick. So, that’s exactly what our model can explore: How does the disease spread if some percentage of students doesn’t quarantine? What happens if we invite only various subgroups — freshmen and sophomores, or juniors and seniors — to return.
Our model is really optimal for this kind of study and we thought it would make an interesting course. So, we were thinking this would be something where we could teach about modeling and, at the same time, the students actually solve their own problem — a solution for students from students. It’s rare that a modeling class gets to work on subject matter that hits so close to home. It’s a timely problem where our students actually impact on their own future.
Linka: Our model will use disease dynamics, in real time, and personalized travel data for every individual Stanford student. Current models don’t have this. That’s something we as engineers bring to the problem. I’m an engineer, not an infectious disease expert. This class and this model will combine classic epidemiology with the latest in machine learning, computer modeling, concepts of the effective reproduction number and herd immunity, network modeling, outbreak dynamics and outbreak control and uncertainty quantification. It’s all in there. If, by the end of the fall quarter, our students can integrate all these things into their final project, we might end up with a crowdsourced solution that can make robust informed recommendations about phased campus reopening in the winter. Wouldn’t that be cool?