How Cornell University Contained the Spread of Covid-19
https://news.cornell.edu/stories/2020/11/data-testing-helped-cornell-curb-covid-19
As described by Melanie Lefkowitz in the Cornell Chronicle, Cornell University successfully contained the spread of Covid-19 on its campus. In this blog, I will provide background on Covid-19 and college campuses, summarize Cornell’s success containing the virus, and then add my own intuition as to why Cornell was able to succeed. I will provide a more in depth analysis of the containment using the basic reproductive number (R) and the SIR (Susceptible, Infectious, Recovered) model. As a disclaimer, I am not a medical professional or an expert on infectious diseases. I will be using the perspective I have as an undergraduate student at Cornell along with my intuition to make this analysis.
The Covid-19 pandemic has been spreading rampantly in the United States since August. Many of these cases have been attributed to the start of the university school year. Students all over the country returned to college campuses and carried the virus with them, spreading it throughout heavily connected networks of college students and staff. The fact that people on college campuses live so closely together combined with the seemingly more reckless behavior of young people makes it easy for the virus to spread. As a result, many of the schools that “opened up” (allowed students back on campus) had to send students home after failing to contain the outbreak.
However, Cornell University has surprisingly contained the spread of Covid for the duration of its in person semester, which lasted from August to November. Overall, there were less than 200 cumulative Covid-19 cases on campus, even lower than the projected amount. Lefkowitz’s article explained how Cornell managed to accomplish this feat. The key was the twice a week testing and contact tracing protocol for all undergraduate students. Cornell successfully processed 7,000 tests per day, isolated students who were positive, and isolated students who may have been exposed. Dr. Gary Koretzky, a professor in the Department of Medicine at Weill Cornell Medicine, attributed the success to the responsibility of the students as well, saying “They care … They’re part of the solution.”
While it may seem obvious that testing and contact tracing reduce the spread of a virus, there are metrics and models that we can use to better understand why it works. The basic reproductive number of a virus, represented by R, is a metric that allows us to predict whether a virus will continue spreading or die out. R is calculated by multiplying P and K. P is the probability of the virus spreading between two people who were in contact. K represents the number of different people each typical person comes into contact with. In general, if R is less than 1, then the virus will die out in a finite number of steps. If R is greater than 1, then the virus may persist forever.
The SIR model helps us visualize how this works. By representing people as nodes and connecting nodes based on which people came into contact with each other, we can create graphs that model the spread of a virus. If the chance of the virus spreading from an I (infected) node to an S (susceptible) node is simulated with P and the number of links a typical node has is K, we can see how higher values of P and K make it more likely for the virus to spread throughout the entire graph.
Clearly, Cornell University was able to keep the R of Covid-19 low. However, I believe the levels of P and K were not dramatically reduced compared to previous semesters. From my personal experience, it appeared on the “outside” that we effectively reduced P by wearing masks and washing hands. Masks are worn in the collegetown streets and on campus. However, I believe students did not practice these sanitary precautions in places where it was not enforced. I think I can safely assume a fair number of students did not wear masks when visiting each other in their residences. It is even more important to wear masks inside versus outside in terms of lowering P.
In addition, I think many students also likely did not lower their own values of K, or how many people they came in contact with. While I admit that I did not observe any super large gatherings such as parties take place, observing people’s behavior in collegetown and on social media tells me that most people have been hanging out with their friends as they normally would. For example, anyone on a Friday night walking by the glass windows of Koko, the Korean restaurant on College Ave, would notice how busy it is. The restaurant is filled with students and has reached maximum capacity. All the students are eating in one room, they are on tables that are not very far apart from each other, and because they are eating, nobody is wearing a mask. The students are clearly not very concerned about high amounts of people in one place, so it seems likely they wouldn’t be concerned about hanging out in groups in their homes.
Based on my observations, it seems like Cornell has not effectively reduced R. Yet the rate of infections continues to stay low. To understand why this is, I am going to combine the concept of R with the SIR model. At Cornell, I am going to assume that the SIR model of students is made up of a bunch of clusters. The people that a typical student interacts with also likely interact with each other because of strong triadic closure, belonging to the same organization such as a fraternity, or living together. Overtime, students eventually form friend groups that mostly hang out with each other. These friend groups are clusters on our SIR model graph.
These clusters, combined with Cornell’s rapid testing program, are key to understanding how Covid-19 did not spread. When Cornell finds out that a person tests positive, that person is quarantined along with everyone they have come into contact with. This most likely will consist of an entire cluster becoming isolated and effectively removed from our SIR model. In other words, if we were to give each cluster an R value, the clusters with a positive person in them have an R value of 0 because their value for K is 0. This effectively removes all infected nodes from our SIR model before the virus gets a chance to spread. Now, obviously before a cluster gets quarantined, there is a chance that the virus spreads to another cluster. However, I observed the Cornell students have not been having large gatherings like parties, which is why I think infection across different clusters is unlikely.
For all the clusters that are not quarantined and seem to be going about their life as normal, the students in them are protected by knowledge. In other words, because each student is tested so frequently, they each have highly accurate knowledge of whether or not they are infectious. This is what allows them to be comfortable even without making great effort to reduce P and K. This knowledge is not reflected in our SIR model or our Basic Reproductive Number. My intuition tells me that predicting the spread of a virus with just P and K is not accurate enough. We need to factor in 1) If the SIR model consists of clusters and 2) whether or not people have accurate knowledge of their infection status (negative or positive).