The modeling team would like to thank the Cornell community for the sincerity and energy put into providing comments on our work. The webpage provides links to our reports and updates.
Modeling Update for Gateway Testing (August 5)
We provide this updated modeling analysis of gateway testing prompted by three recent events: COVID-19 prevalence has risen in parts of the U.S. since we wrote our report in June; a potential lack of test access for some Cornell students in their home location; and the recently instituted requirement that people coming to NY State from some high-prevalence are as must self-quarantine upon arrival regardless of test results.
Addendum (July 17)
This provides an in-depth analysis of some of the questions posed in response to the June 15 Report. In particular, the addendum studies (a) alternative methodologies for estimating contacts / day and transmission rate, (b) the effectiveness of increased test frequency in mitigating the effect of higher-than-modeled contacts / day, (c) the effect of non-compliance with testing, and (d) the effect of offering testing to virtual instruction students.
Original Report (June 15)
Comments on the June 15 report (and responses) are given below. Comments are indexed for easy referral.
Issue (Last Updated July 3) |
Comments |
| Contacts-Per-Day | F28, F19, F16, F14, F10, F8, F7, F6, F1 |
| Higher-than-anticipated transmission rate | F32, , F19, F16, F14, F10, F9, F8, F7, F6, F1 |
| Effectiveness of Testing | F9 |
| Testing in the virtual instruction setting + testing compliance in the residential setting | F29, F27, F26, F25#1, F17, F11 |
| Off-campus students are tested in the residential scenario | F31, F27A |
| Number of Students Returning in the Virtual Instruction Scenario | |
| Modeling fatalities | F29, F13, F12 |
| Racial and Ethnic Disparities | F25#2, F24, F23 |
| Pressure from university leadership | F18 |
| Effect of raising transmission rate equally in virtual and residential instruction settings | F21 |
| Capacity in local hospitals | F22 |
| Framing of uncertainty | F30, F9, F5, F2 |
| Crediting of experts in disease modeling | F9, F1 |
| Impact on Tompkins County | F20 |
| Impact of students arriving early because we would start Sep 2 | F15 |
| Number of cases missed in gateway testing | F32 |
Note On Data
Data is central to the scientific approach that is being taken. Two surveys taken early in the summer had a big impact on the university’s approach to F20.The Student survey gets at residential vs online instruction and the likelihood of coming to campus/Ithaca for the fall semester. The Faculty survey that gets at in-person vs online instruction and risk assessment of being on campus.
If the work on this is ongoing, as suggested by a Github repository that seems associated with the model [https://github.com/peter-i-frazier/group-testing], will the modeling team continue to provide updates to the university community during the fall semester?
The most recent Python notebook [https://github.com/peter-i-frazier/group-testing/blob/master/notebooks/NYS_lockdown_threshold_analysis_Aug_29.ipynb] looks like it presents simulations that are related to the the new NYS guidelines for “locking down” universities. Why are Vet and/or Law school students included or excluded from some simulations? Is the modeling still considering the overall impact on the community or now just focused on anyone who would be part of the ‘threshold count’?
I encourage the team doing this work to be transparent and continue communication. If the models are being changed or the outputs of the models are being used in ways that are different than those communicated to the public (ie. the Cornell community) so far, it seems important to let us know both here and at the Cornell COVID page [https://covid.cornell.edu/testing/modeling/].
Figure 3 of the addendum suggests that there is not delay between the testing and quarantining the infected individuals (otherwise there is no other way to explain how an epidemic can be controlled with R_0 is quite big as in the case of 40 contacts per day). This is not a very realistic assumption, at best one can hope for 12h delay from taking the test to quarantining — actually 24h or even 48h is more feasible given the current delays in testing in the US.
If people are tested every 5 days adding such a delay will not make a huge difference, but if one goes to testing every other day such delay is essential. It seems that accounting for such a delay will turn testing frequency from 50% to about 33%. Thus even in the pessimistic scenario in order to control the epidemic one need to test every other day (or even every day) which likely is outside the capabilities of the Vet school and Tompkins county health services.
One very important point not addressed in the addendum is how to detect if the spread factor is higher. I.e. the university starts with testing every 5 days and the number of contacts is say in the pessimistic scenario of about 11 per day, how log is the window where switching to testing every other day will still control the spread? Clearly if one waits a month it will be too late, however one needs to waits a few weeks to see that the number of infections is higher that the predicted by the model and switch to more frequent testing.
In short, the addendum does not address many of the short comings of the model, and it is likely that the model prediction are overly optimistic.
Reading the report from June, as well as the addendum from July, I have a few concerns about the model.
First, looking at Figure 3 – you’re plotting two scenarios. 1) Full re-open with active surveillance and 2) No re-open and no surveillance. Your conclusion is that scenario 1 will be better, but you’re changing not one but two variables across the two scenarios. Could you share with us 2 other scenarios – re-opening campus without surveillance and no re-open with surveillance? This would really help differentiate between the effects of re-opening and adding testing.
Second, I’m very concerned about the assumption that there are homogeneous contacts between individuals. As an alum, my cursory familiarity with housing options for students suggests to me that there could be wild differences in the exposure risk between students who live in single apartments off-campus and students who live in fraternity houses. My concern here isn’t that the assumption is unrealistic – my concern is that it is well-understood among contact network epidemiologists that contact heterogeneity is a strong driver of outbreak size. Having read your explanations in your report for why you’ve chosen 8.3 contacts per individual, I do not see how you can justify not accounting for contact heterogeneity. I consider this important enough that providing actionable advice to Cornell and Ithaca based on a model that doesn’t account for this would be irresponsible.
It does sound like you’re planning on incorporating heterogeneity into your model for the future – I’d be very excited to see how this changes things.
To what extent if any, has this model been revised to reflect the alarming and growing surge in cases in large sections of the country?
I completely agree with this comment. I cannot find the survey results and data to support the Frazier report.
After reading the Frazier report, we need to see the survey data this report’s simulations were based on. The LOWER bound (“optimistic”) scenario the Frazier modeling uses 8000/15000. In an article to the WSJ, Pollack stated that this survey found that “as many as 50%” of undergrads reported in the survey they would return to campus regardless of the more of instruction.
So what is it? As many or as little as 50%?
We need some academic transparency and a peer review process for these findings. This is a university after all. I urge all faculty, especially those is quantitative social science fields to review this study and Pollock’s conclusions.
Where can we see the survey used to gather data on how many Cornell students would return to Ithaca regardless of the mode of instruction? Additionally, how many students was this sent to and how many responded? During what time period was it collected?
Whatever was gathered is only a snapshot from a particular moment in time before recent surges, and without complete information regarding what campus life would really look like in a COVID world, and without parental input. Given all this, how useful and truthful is this data?
What is Cornell doing now to know who has returned, if they have been tested, to notify those in states currently under quarantine upon arrival orders from NYS?
Where will temporary hospital wards be set up given the possibility of exceeding what our small medical community has to offer?
Why not keep COVID infected students in Statler rather than spreading them out into the Ithaca community?
What are the ethical obligations that our higher ed institutions have to protect the welfare of the permanent local residents?
This report fails to address what I think is the most important issue – the risk of ruin. There is likely a chance that there will be a widespread, uncontrollable infection on the Cornell campus, this would probably lead to cancelling of classes and research and put the university in a worse situation than it would be if it were not reopened in the fall. The potentiality of ruin would be represented as the tail of the distribution of possible outcomes and the important questions to ask become: how probable are these tail events? How sensitive is the probability to our model parameters? Would reopening be wise given the risk of ruin? Until these questions can be answered it seems foolish to commit to reopening the university.
Thank you for all your efforts in the full report. I wonder how the rapid increase in the past two weeks would change the result of your model. In the executive summary, you mentioned that “Toward the goal of quantifying uncertainty, we are continuing efforts to estimate parameters, provide ranges of plausible parameter values against which we should plan, and investigate the impact of modeling assumptions.” Could you provide some insights on how the parameters are changed based on the past two weeks’ data? And more importantly, has such an extreme scenario been tested in the Jun 15 report? If not, I am worried about the fidelity of the report.
Does the modeling team understand how draconian it is to threaten to remove students’ access to their own email to encourage compliance with testing? (This is proposed in section 4 of the Replies to Comments.) Does the modeling team understand the *numerous* equity problems that such a threat creates?
This preliminary study implicitly frames the institution and its prerogatives as unquestionable. This is a serious error.
The study cites Tompkins County as “a significant source of infection”. Tompkins County has had one new case in the last three weeks. We sacrificed for months to bring infection to near-zero levels. WE aren’t the source of infection. CORNELL would be the agent of infection to its surrounding community overwhelmingly!
This is a serious error because it is being used as cover for the university to assume prerogative for itself. It is OUR prerogative whether to allow you to conduct your business at the expense of our health! Are you offering to pay the medical expenses for the infections you will be causing? Funeral expenses?
And you cannot cover this mistake with talk about human endeavor and risk, pap about depression and suicide. You are a wealthy university that squanders its resources on 2-and-20 deals and a top-heavy management style. If you cannot support both your employees and the surrounding community until you truly get this sorted out, then the manner in which you conduct your core mission must be brought under question and re-examined at a fundamental level.
Having not read the report yet but listening to the town hall, I have a question. Since a significant portion of the student body lives off-Campus including seniors, graduate students and professional students how are you going to control contacts and testing. This is basically the open with virtual teaching but the students coming back. The term residential is misleading. Does it mean living in a dorm or university housing or being in Ithaca?
Most if not all of the proposed academic calendars involve ending in-person/residential instruction in advance of the Thanksgiving break, with the rest of the Fall semester to be completed virtually. In the “no reopen” modeling scenario, however, it is assumed that several thousand students will be physically in Ithaca even if there is no in-person instruction. So it is perhaps not unreasonable to assume that some fraction of students might leave Ithaca for Thanksgiving, only to return after Thanksgiving to complete the rest of the semester virtually. Has that possibility been included in any of the modeling efforts?
To clarify the comment above (F33), the question is whether — in the residential/reopen scenario — any thought has been given to off-campus students who leave Ithaca at Thanksgiving and then return for some period of time before the winter break. (Presumably, in the no-reopen scenario, off-campus students might be coming and going throughout the semester, which might or might not be reflected in parameter choices that have been made.)
I find the last paragraph in point 2 of the response very confusing. In the paper the authors of the model clearly state that during the first two days after exposure the disease is not detectable by the PRC test (page 17). Therefore one can not use frequent testing to leverage this period to slow the spread, as suggested by the response.
A back of the envelope computation based in the current number of new cases, shows that this two 2 undetectable period will lead to about 10 case of infected students which will no be detected by the rigorous testing during the move-in days. This number grows to 15-20 if one includes the the test has false negatives.
I do agree with one of the conclusions of the model, that a significant student population outside campus is likely to have a very significant infection rate absent any testing.
The other main conclusion is a bit more problematic: opening the campus with significant surveillance testing might also lead with significant infection. The chosen parameters avoid this scenario, but increasing the infection rate/ number of contacts (to a number which is plausible) leads to significant infection. The author stated multiple times that the testing can detect this in very early stages, but he did not provide any evidence supporting this claim.
The chosen model gives about 10 detected infections per day, but I am afraid that an epidemic can start with something like 15 cases per day for the first two weeks, and an explosion afterwards which ramps to 50 cases over a week or so. However at that point the epidemic can be stopped.
I understand that the authors had put a lot of effort in this model, but it is clear that it has significant shortcomings and the response does not seem to address most of them.
The no reopen scenario assumes that there will be are large number of students living off campus that will be sufficiently beyond the control of the university so as to preclude surveillance testing. But in the reopen scenario the off campus students are also not included in the surveillance testing. Why is this a major problem in the no reopen scenario, but not in the reopen case? It could still be the same number of off campus students either way, possibly even more in the reopening scenario.
Huge thanks for Frazier and his team for their detailed and ongoing effort to address concerns raised here and in other fora.
I’m not sure, but I suspect some of the comments here about the “executive summary” being misleading are referring not to the executive summary in Frazier’s report itself, which he (reasonably) defends, but to the C-TRO’s executive summary. The latter states, “Paradoxically, the model predicts that not opening the campus for residential instruction could result in a greater number of infected individuals…. Hence, it appears that in addition to the educational advantages that would arise, opening the campus for residential instruction may be in the best interest of the health of the Cornell community, Ithaca, and surrounding communities.” In that text, all the complexity and uncertainty in this prediction is left out. This simplified and perhaps overly rosy view of the Frazier model is likely to be far more widely read and internalized by the Cornell and Ithaca community than the Frazier report itself.