President Martha E. Pollack announced on Tuesday the University’s intentions to reactivate campus for the fall semester, which would begin on Sept. 2. Despite criticisms that this decision could put the health and safety of the Cornell community at risk, Pollack and Provost Michael Kotlikoff cited modeling that indicated it would be safer to reactivate campus, and subject students, faculty and staff to extensive testing than to commit to a virtual semester.
The modeling was done by Prof. Peter Frazier, operations research and information engineering, who has a background in viral pool testing and was a member of the teaching reactivation committee. Frazier spoke with The Sun about limitations in modeling and Cornell’s plans for the upcoming semester.
This Q&A has been lightly edited for clarity and grammar.
The Sun: So I thought a good place to start was what the impetus for this research was, when did you start doing this analysis? When did the administration ask you to do this? Or is this something that you just started?
Frazier: I’ve been interested for a long, long time in something that’s called pooled testing or group testing. And that is not specific to COVID-19, it’s a way of scaling up your ability to test to be able to cover more people. When the COVID-19 outbreak happened, testing wasn’t available. And I thought, “Well, why don’t people do pool testing?” That seems like a really good tool that we could really use here. Nobody was really talking about it. So I started, I wrote a little white paper that was about how you could use pool testing while letting people exit social distancing restrictions. This was when we had really strong social distancing restrictions in place in New York State and wrote about how you could use pool testing in order to exit from those kinds of restrictions safely.
And then it turns out that our provost, Mike Kotlikoff, was also interested in pool testing. And one of the people that I was talking to was like, “Oh, you know, the provost is interested in that.” Then, somehow, I got on an email chain with him and he saw the white paper. We started talking about design and testing protocols and then he asked me to serve on a committee that was looking at teaching reactivation options. Originally, I think that the goal there was just, “Hey, we’re thinking about using pool testing. Here’s a guy that knows about it. Why don’t we have him be on the committee?”
On that committee, we were facing all kinds of other questions: Should we reopen? If we reopen, how do we do it? I thought, well, we should probably expand the modeling efforts, because we had already been doing some modeling efforts in order to try to understand the impact of pooled testing. So five Ph.D. students are working on this with two other faculty members. We started doing this modeling effort and tried to support the teaching reactivation committee and help them think through questions that they had and try to help them think through options.
The Sun: What is the strategy for testing? It seems like there’s not a lot of tests available in a lot of places. So to me, it seems pretty ambitious to test everyone in the Cornell community every five days, especially when I presume there will be similar testing needs for the Ithaca community and for the Ithaca College community.
Frazier: Number one, those who live in Ithaca and are associated with Cornell are actually extraordinarily lucky in that there is a very large ability to do [polymerase chain reaction] tests, the chemical reaction used for doing testing for the presence of the virus. There’s a large capability to do that in the Vet School. That exists in order to do testing for dairy cows and other animals in order to control outbreaks in that population. But it can be repurposed. There’s a whole bunch of complicated regulatory stuff that we could talk about, but that can be repurposed to provide testing for our community. And that’s a thing that most other universities don’t have access to.
You asked me [earlier] about talking through analyses of other universities reopening plans, and actually I was just on a call with, I think, three or four other people from other universities who are engaged in similar modeling efforts.
One of the things that really separates what we’re proposing to do from what other universities are proposing is our ability to test everyone regularly and at such a high cadence. Part of the reason we can do it is because of the Animal Health Diagnostics Center. Another reason that we can do it is because we’re building capabilities in pool testing, based on the work of people in the Vet School who have expertise in the experimental side of that and based on early work, that we did thanks to our provost. That’s a second reason that we’re able to test so many people so often.
As to how that testing capability is being deployed in Ithaca and the extent to which it is being applied only to Cornell staff, faculty, students versus other people who live in the area is an important question, and I believe that discussions are ongoing.
The Sun: Are we making those changes in the Vet School to transition our capacity from cows and other animals to humans? Is that something that’s happening right now?
Frazier: Yeah, it is. The biggest thing is the legal change you need to get. The [Food and Drug Administration] made that a lot easier because they said last week or the week before, that the use of pool testing for surveillance can fall under an emergency use authorization — which is much easier to get opposed to what you would need for what’s called diagnostic testing. The way that it’ll work is we’ll do pool testing, and I should also say that we’re planning to partner with Cayuga Medical Center.
So there’s sort of two avenues that we’re going down and [Cayuga Medical Center] is also building pool testing. They did a test of people in ILR and CALS, and maybe Human Ecology, I’m not sure. CMC tested roughly 800 people, where they did pool testing in pools of size five. So that’s also a capability being done. The Vet School is working hard to set that up and get all their assays in place and get the proper legal requirements. It is really a gargantuan task. So we do need to work hard every day over the summer to make sure that that happens.
The Sun: Backtracking to pool testing — do you just take samples from, in this case, five people and run the PCR together and if one comes back positive, then you test those five people again?
Frazier: Yeah, there are a couple different ways to do pool testing. What you just described is one of the primary ways that people do it. We’re pursuing two different avenues, and one of the avenues that [we’re] pursuing is what you just described.
The other avenue is a little bit more complicated, but it’s a little bit more efficient. The second way is that you imagine taking samples from 25 people, and then imagine putting those samples into a square. So it’s a five-by-five square, and then imagine forming a pool from every row in the square and a pool from every column. Then you test those 10 pools. And if a person comes up positive in both the row and the column, then you do follow-up testing on that individual person. Five-by-five doesn’t really necessarily make sense to do. But, we’re looking at 24-by-24 protocols that are like that. It’s called a square array protocol. So then you can test 24 squared people with 48 tests, and it has some slightly better error properties.
The Sun: Right, I was about to say, I assume you get less false negatives that way.
Frazier: False negatives are interesting and important. This needs to be confirmed, but we think that one of the primary reasons that you get a false negative is because you have a low viral load in the person that you’re testing. So that ends up causing correlations and errors and makes pooled testing perform better than you would think, based on a more naive sort of analysis.
The Sun: Back to the study itself, could you explain to me in layman’s terms, what some of the numbers of this analysis showed in terms of how many people would get sick? How many people may need to be quarantined, hospitalizations?
Frazier: So I can walk you through the numbers from what we call the nominal scenario, which is in the main report.
So in the nominal scenario that we study, where everybody comes back, and we assume that the the rate of transmission among individuals is comparable to what’s called an R0 (basic reproductive number) of 2.5, R0 measures how many other people, each infectious person infects over the course of their over the course of their infection. 2.5 is a number which represents the general population, so different from a university population. In a world without strong social distancing measures — so basically before we all went into lockdown — that number comes from the [Centers for Disease Control and Prevention]. Assuming that and assuming testing everyone once every five days, this model says that roughly 1,200 people in the Cornell community would be infected and it says that roughly 10 to 15 people would be hospitalized.
Something that people have commented on is that we don’t actually estimate deaths in the report which is true, because we felt that it was hard to estimate them and that one could simply base decisions based on hospitalizations. It is possible that someone would die or, two people could die, or even more. But basically, 10 to 20 hospitalizations and roughly 1,200 infections. Now, we have a lot of uncertainty. So it’s not like I’m saying that is absolutely what’s going to happen. It’s just that those numbers give you a flavor for a plausible scenario for what might happen under kind of a plausible set of parameters.
The Sun: Did you talk to any kind of biologist epidemiologist? How viable is it really to have, let’s say, a clean classroom or a dorm room, even if you have mass gatherings and things like that?
Frazier: We talked to quite a few people with a biomedical background. Gary Koretzky, who is a biologist, is the head of the subcommittee so he was closely involved. And, [Kotlikoff] looked over a number of early versions of the report. We got formal reviews from two different groups within the Vet School, Renata Ivanek and Yrjo Grohn.
We had the report also reviewed by David Hamer, who’s a professor of global health at Boston University. A number of [other] people and Isaac Weisfuse, who was in the New York City Public Health Department for a number of years and had a leadership role there as an epidemiologist, [were involved]. So a number of people on the [teaching reactivation committee] have expertise in different aspects of this. I should also mention that we got a lot of help from the Cornell library, who did a little literature review for us on the enormous number of different aspects of the disease.
With mechanisms for transmission for COVID basically, droplet transmission is the one that is probably the most commonly discussed. Let’s say that two people are standing next to each other, talking. One person coughs, that person is infectious, then the droplet — these kinds of spheres of fluid — come out of their respiratory tract, and they may land on the other individual. They might land on the eyeball or they might land in the mouth. And then the virus in those particles might then directly infect cells in the body, or they might land on the hand and then that person might rub their nose, some mucous membrane — that’s the primary method and that’s why the CDC talks about close contact as being a primary risk factor. Close contact, at least according to the CDC, means that you’re six feet or less, 15 minutes or more.
You can definitely have transmission in ways that don’t constitute close contact. So it could just be that, very briefly, we stand next to each other on the TCAT and the person coughs on me and then I get infected — we were only together for 45 seconds.
There’s also surface transmission. Can you clean a classroom? I think the answer is that you can clean a classroom to some extent and you can’t clean it perfectly. In the modeling, we do not we do not model every detail, because there’s so much uncertainty about each detail. What we do instead, is we take this R0 number which is estimated from the general population. In the general population, even though there’s not a lot of people entering lecture halls, there are a lot of people that are, sitting; one person walks onto the subway, rubs their nose, puts their hands on the subway seat and another person walks in, sits on the same subway seat; or, people are walking in and out of a busy building, and you put your hand on the door, and then the next person comes along and puts their hand on the door and you get transmission that way.
It’s not like surface transmission only happens in universities. We believe that that kind of aspect of transmission is reflected in the 2.5 number. An important thing that we’ve done is to try to understand how much uncertainty we have about that and if it turns out that it’s larger than we think, what can we do? So if we found out that there was a lot of transmission happening in classrooms, which we will find out from testing, we would then say, “Hey, what can we do in order to reduce transmission in classrooms?” One of those things would probably be to clean more aggressively, or if we find out there’s a lot of transmission through shared bathrooms, then we would change the way those bathrooms are laid out: maybe do more frequent cleaning, maybe set up partitions.
The Sun: That’s really interesting. I hadn’t heard that talked about a lot of other college reactivation plans. I think something that concerns a lot of people, especially as a student, myself, is the aspect of student behavior and enforcement. So I think one of the bigger unknowns is how do you account for, you know, if someone throws a party and someone’s infectious, you’d almost think that everyone gets sick? So how did you model for the potential of social interactions outside of the classroom? I know you mentioned to Inside Higher Ed that there would be some enforcement policy. What kind of enforcement policies do you think are viable at Cornell?
Frazier: In terms of parties, we looked at literature on social interactions that broke out those interactions by age group. So that was one way that we tried to understand the effect of parties specifically. I do agree that parties do pose a risk, especially if people on campus have large parties.
In terms of their effect on the modeling, our uncertainty about our ability to control large social gatherings is reflected in uncertainty that we have about this R0 value and the number of contacts per day students will have. And we find that if we don’t do anything, and if we’re wrong about the amount of content, the amount of transmission that’s happening on campus, what ends up happening is that you kind of get a gradual increase in infections and hospitalizations as contacts per day rise until it’s about double what we modeled — if parties are so out of control that people are having, on average, twice as many contacts per day as they would in our model.
If you do more frequent testing, you can really handle a very large number of contacts per day. So the key thing for us will be, number one, to make sure that we can do regular testing; number two, to make sure that, like if we need to, we can increase the frequency. Number three is to make sure that we’re looking at the results from the testing in an intelligent way so we can figure out quickly if it turns out that people are having parties more frequently than we thought, then we can do something about it.
Also Ithaca is just not that big of a place. It’s probably hard to find a party with 15 people that are being really quiet and listening to, you know, smooth jazz softly. But if it’s a really raging party, and it’s in Collegetown, I need 15 minutes to drive around Collegetown and realize, “Oh, there’s a giant party.” It’s not really that hard to figure out. It’s not like people live in caves or something.
The Sun: What was the number of contacts that you modeled for?
Frazier: In the model’s nominal setting, we modeled 8.3 contacts per day — and those are close contacts.
The Sun: Does that mean you were close to the contact or interacted with them?
Frazier: The probability transmission we used there is consistent with a CDC close contact — which would be less than 6 feet and last 15 minutes. So it’s more than just like, if I saw you in the hallway, and I said “Hey, how’s it going?” That wouldn’t be a close contact.
The Sun: One thing I noticed is that the paper was dated June 15. Since then there has been a huge rise in cases in a lot of places — Florida, Texas — that we have students coming from. Do you think that changes anything in the model?
Frazier: I think it’s consistent with what we saw. There are two things where I think that [spike] informs. One thing that it informs is what we think transmission rates are among students — just among people in general — when we lift social distancing restrictions. Cases have escalated in places that have lifted social distancing restrictions in an indiscriminate way. I think it’s consistent with an R0 that’s bigger than one, maybe at least 1.5. And I think we kind of knew that.
However, yes, it’s totally relevant for the prevalence [of COVID-19] among the students who are returning. We model a prevalence, among the three models, that has a lot of students already being here. But, of the students that we model as returning to campus from elsewhere, we assume a prevalence of 5 percent.
In light of these rising prevalence values in other states, I think it would make sense to get ready for a larger amount of quarantine as students return. What we’re planning on doing is a gateway testing program, where we would test students definitely when they arrive and, I believe in some cases, we are planning to test them before they leave. So when you do that, you end up catching most of the positive cases.
So it’s bad, right? If you’re COVID positive and you come to Cornell, and you think you’re fine, you feel fine, but it turns out you’re asymptomatic and then you come out positive, then you do need to quarantine for two weeks and that’s not super fun. So more people that have to quarantine for two weeks is something I would love to not have to do. That’s an impact [of the rising cases nationally]. It also has an impact if positive cases are found here in Ithaca, then we need to quarantine them in Ithaca and that consumes local quarantine capacity, which is an issue.
But, in terms of the safety of other people that are in Ithaca, as long as the testing is executed as planned, we’ll catch — not everybody — who’s positive, but we’ll catch everyone except for those that got infected during transit. If you get infected only yesterday, I’m not going to be able to see that on a PCR test. And then also, there’ll be some false negatives. But if you have 5 percent prevalence, then the fraction of people that you’d miss when you do tests is pretty small.
The Sun: What do you think the odds are that testing isn’t enough to prevent an outbreak? Is there any scenario where an outbreak just blows up so quickly that the five-day testing range isn’t fast enough to catch it in time?
Frazier: So Martha Pollack sent an email and said that there’s no such thing as no risk. Yeah, I think there’s a couple of things that I worry about. One is that aerosol transmission turns out to be more important than we thought. So, if that’s true, that’s a problem for the entire world and things are just really bad.
Another thing I worry about is that we won’t be able to achieve our goal of setting up testing, that there’ll be some logistical problems. I don’t know what would go wrong exactly, but there’s a lot of details to get in place. Maybe, I think a thing that would be plausible is that, let’s say that we set up testing, but we don’t test the testing. And then we will have the testing, but if we don’t test the testing, what would happen is that the first time you try something it always takes a little bit before you fix all of the gotchas. So you could imagine, maybe we realized that there are bottlenecks in bringing the samples into the building where we do the testing. And then we get long delays. Or maybe we forget that we need to have redundancy in the machines; we own the one machine and the machine breaks. But those are all things that we’re thinking through so it doesn’t happen. But you never know, there could be something that we’re not thinking of that could break.
And non-compliance is another issue. I have been doing less of this, but other people have been looking really hard in order to make sure that you have a good plan for ensuring compliance while also creating a good experience for everybody. But there’s a lot of unknowns there. So that’s also something to worry about.
The Sun: Is there a breaking point where you send students home, or you just realize that the test, quarantine, and isolate strategy just isn’t enough and you need to call it off?
Frazier: We had this committee and we released a report and then the President made a decision. But things aren’t over and there’s a lot more planning we need to do. Part of the planning that we need to do is to figure out, you know, exactly what does that look like? What would the trigger be? How would we executive that kind of scenario?
But, in many settings, it’s probably safer to have students stay in place rather than go home. I think the one scenario where I think it really would make sense to send everyone home would be if we start seeing outbreaks in dorms and we don’t have the quarantine capacity. But if we start to see that happen, we would put in pretty strong social distancing measures and we would increase the capacity of testing. And then, over a two week period, get control over things.