Who thought it would be a good idea to move thousands of teenagers and young adults across the country to college campuses, where, unencumbered by parental supervision, many college kids did what college kids do?
Actually, Nigel Goldenfeld and Sergei Maslov, two University of Illinois at Urbana-Champaign physics researchers, thought they had it figured out. They created a predictive model for the campus, which showed that with a robust, twice-a-week testing program for students, faculty and staff who are regularly on campus, a mask mandate and an app for contact tracing, COVID-19 cases could be kept below 500 people for the whole semester on a campus that includes about 46,000 people. They even accounted for close interactions among college students.
But that model failed to take into account that kids who test positive for the virus, whether sick or asymptomatic, might continue to party. From Aug. 16, when campus reopened, to Sept. 13, more than 1,800 new cases of COVID-19 were detected, according to the university’s COVID-19 dashboard. One thousand cases occurred in the first two weeks of the fall semester.
“What is not in the models is that students will actually fail to isolate,” said Goldenfeld during a Sept. 2 press briefing, “that they would go to a party even if they knew they were COVID-positive or that they would host a party while they were COVID-positive … We didn’t include that behavior in the model.”
Many other colleges across the country also thought through how to bring students back to campus. Several schools looked at computer models to see how COVID-19 would affect students and staff. But, as with the plan developed at Illinois, these models were sometimes based on a set of assumptions that ended up being wrong. In other cases, models that showed what could happen without mitigation strategies were ignored by university administrators, who went forward with plans to bring students back.
Either way, the great student migration has resulted in COVID-19 outbreaks on college campuses nationwide. The University of Central Florida: 286 cases since the week ending Aug. 8. Texas Christian University: 600 cases in the month of August and 212 in September so far. The University of Iowa: 1,732 cases from Aug. 18 to Sept. 9. The University of South Carolina: 2,074 cases since Aug. 1. Making matters worse, some afflicted schools are setting off a second student migration by sending their students back home.
The administration at the University of Illinois at Urbana-Champaign asked students to lock down for two weeks on Sept. 2. And Goldenfeld said during a Sept. 2 news conference that it was too early for him to make a new prediction whether COVID-19 cases could be kept under control for the semester.
He said he and Maslov would be adjusting their model, but were waiting to see how students would respond to the lockdown. Cases of COVID-19 on campus have declined since the lockdown was implemented.
The administration of the University of Illinois at Urbana-Champaign has collaborated directly with Goldenfeld and Maslov, and has been transparent about the model it is are using to base its decisions. Other universities haven’t been as upfront.
After hearing that Penn State planned to open again for the fall, a concerned faculty group, Coalition for a Just University, created a model predicting what COVID-19 spread would look like at the University Park campus in State College, Pennsylvania. The Coalition’s modeling group, composed of engineering and science faculty, chose to remain anonymous, fearing retribution from the university. Its predictive model showed that more than 1,800 students could become sick and two could die from COVID-19 during the semester if only 1% of students were tested each day, which is Penn State’s plan. Since Aug. 28, 650 students at the University Park campus (which is attended by some 47,000 students total) have tested positive for COVID-19.
The team sent the model to university administrators but received no response. A Penn State spokesperson told the Centre Daily Times, a local newspaper, that the methodology of the model was “flawed” and the group that released it had “advocated against any reopening of campuses.” The Coalition has asked that faculty be given an option to conduct classes remotely and for more transparency in the data behind the decision to reopen campus.
The Penn State spokesperson later said that the university had developed its own predictive model but declined to share its results with the paper. Penn State did not respond to a request for comment.
Penn State isn’t alone in its lack of transparency. Edwin Michael, a professor of epidemiology who recently left the University of Notre Dame to work at the University of South Florida, said he created a simulation in April to show how COVID-19 could spread on Notre Dame’s campus in South Bend, Indiana. He said he shared it with the university officials but never heard back.
The model showed that on a campus of 20,000 people, if 25 students returned to campus with COVID-19 and there were no mitigation strategies, up to 7,500 students could soon be infected. Roughly 470 would need hospitalization and 365 would need treatment in the intensive care unit.
It was a dire prediction with a purpose. He said it was created “simply to highlight that an outbreak is inevitable if students were to return infected.”
Dennis Brown, a spokesperson for Notre Dame, said that Michael’s predictive model was forwarded to members of the planning committee in May “and subsequently taken into consideration.”
“However, because it made certain assumptions that did not align with the plans being made at Notre Dame, we did not find it relevant to our situation and decided to use other predictive models,” Brown wrote in an email.
Brown declined to give more information on what predictive models Notre Dame did use. Notre Dame has implemented mitigation strategies, such as requiring mask wearing on campus at all times and limiting gatherings to ten people, but on Aug. 18, imposed two weeks of remote classes for all students after a spike of cases on campus the first week back. The university has documented 642 cases among students since Aug. 3. It started phasing in-person classes back in on Sept. 2.
Professors elsewhere have, like Michael, developed models not necessarily to make accurate predictions, but to make a point that without some kind of mitigation strategy there would inevitably be a COVID-19 outbreak on campus—and that part has held true.
On Aug. 15, five days before the University of Georgia started classes for the fall semester, John Drake, director of the Center for the Ecology of Infectious Disease there, predicted that from 210 to 1,618 students could bring COVID-19 back with them to campus. He also predicted that without any type of risk mitigation, reopening campus could result in more than 30,000 infections among the campus population—about 60% of all students and staff.
“Campuses should anticipate explosive localized outbreaks,” Drake wrote when making his model public. (Like most of the university COVID-19 models mentioned here, his was not peer-reviewed or published in a journal.)
There’s no way to know whether Drake’s prediction was right, since the University of Georgia didn’t conduct entry testing for students who returned. Instead, the university is conducting voluntary randomized testing of asymptomatic individuals on campus and asking anyone who has symptoms to get tested.
On Sept. 9, the university reported more than 1,400 cases of COVID-19 among students in a week. University of Georgia officials did not respond to questions about whether they used Drake’s model or others when opting to reopen.
About 70 miles away, Joshua Weitz, a professor who studies viral dynamics at the Georgia Institute of Technology in Atlanta, created his own predictive model, this one with a more dire message: Without any mitigation strategies, 50% of people on Georgia Tech’s campus of about 31,500 would be infected with COVID-19 and 75 would die. The majority of those deaths would be among older faculty and staffers.
He hoped laying out this extreme scenario would show why the school needed to test everyone on campus once a week. Although Georgia Tech has enough tests available and encourages students to be tested once a week, it is not mandatory. Georgia Tech confirmed that Weitz’s model had been taken into consideration when it planned its COVID-19 response. Georgia Tech reported 571 cases of COVID-19 for the month of August.
While some professors created models without mitigation strategies as a cautionary tale to show university administrators what would happen without interventions, others were developed to help campuses adopt a framework to reduce infections once students arrived. Though, the limitations of these models run the gamut, their message seems to be the need for constant agility in enforcement policies and awareness about COVID-19’s local spread.
After all, models can’t change one underlying risk that continues regardless of testing plans and other public health strategies: in the end, some college students are still going to be college students, says Michael Osterholm, director of the Center for Infectious Disease Research and Policy at the University of Minnesota. (The University of Minnesota delayed the moving of students into university housing by two weeks and started classes online on Sept. 8. The university has had 87 students test positive for COVID-19 through Sept. 10, though students are just this week beginning to move back in to residence halls).
“You don’t need a model to understand that bringing together all the young adult population in college campuses around the country is putting a lit match in a gas can. You don’t need a model to know what’s going to happen next,” Osterholm says.