Modeling for COVID-19 college reopening decisions: Cornell,a case study

We consider epidemiological modeling for the design of COVID-19 interventions in university populations, which have seen significant outbreaks during the pandemic. A central challenge is sensitivity of predictions to input parameters coupled with uncertainty about these parameters. Nearly 2 y into the pandemic, parameter uncertainty remains because of changes in vaccination efficacy, viral variants, and mask mandates, and because universities’ unique characteristics hinder translation from the general population: a high fraction of young people, who have higher rates of asymptomatic infection and social contact, as well as an enhanced ability to implement behavioral and testing interventions. We describe an epidemiological model that formed the basis for Cornell University’s decision to reopen for in-person instruction in fall 2020 and supported the design of an asymptomatic screening program instituted concurrently to prevent viral spread. We demonstrate how the structure of these decisions allowed risk to be minimized despite parameter uncertainty leading to an inability to make accurate point estimates and how this generalizes to other university settings. We find that once-per-week asymptomatic screening of vaccinated undergraduate students provides substantial value against the Delta variant, even if all students are vaccinated, and that more targeted testing of the most social vaccinated students provides further value.

When is it safe to offer in-person university instruction during the COVID-19 pandemic? What interventions, if any, provide the level of safety required? Colleges and universities across the globe faced this question in summer 2020 as they considered whether to offer in-person instruction. They continue to face this question today as they contemplate partially vaccinated student populations, waning immunity, booster shots, and the potential for new variants to emerge.These questions are significant because outbreaks in university student populations have occurred regularly (1) and may harm the health of students and more-vulnerable employees and community members that interact with them (2). Even when vaccination protects the bulk of the population against the most severe health outcomes of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, widespread breakthrough infections would threaten the health of unvaccinated and immunocompromised individuals in their midst. At the same time, social distancing, masking, asymptomatic screening, the migration of in-person instruction to a virtual format, vaccine mandates, and other interventions that can be brought to bear against university outbreaks all incur social and financial costs (3, 4). Better understanding the protection offered by these interventions would support providing safety while minimizing these costs.These questions remain difficult to answer because vaccination levels, SARS-CoV-2 variants, and other conditions continue to change and because experiences at the city, state, and national level do not easily generalize to university populations. Indeed, university populations are younger than the general population and thus have increased rates of contact (5) that may elevate virus transmission (2, 6). In addition, universities can implement interventions that would be substantially more difficult for the general population, such as mandatory vaccination and mandatory asymptomatic screening (7, 8).Universities have responded to this central question in dramatically different ways. In the 2020–2021 academic year, many schools went fully online, while many others opened for in-person instruction with a modest set of interventions centered around symptomatic testing, contact tracing, and social distancing (9). Moreover, those schools that opened for in-person instruction pursued dramatically different testing strategies (10). Some tested only symptomatic students, others tested all students once on arrival, and others tested all students at least once per week. In the fall 2021 semester, schools differ in whether they mandate vaccines, their testing strategies, and masking policies (11).This diversity in approach reflects, in part, a diversity of circumstance, such as proximity to, and interaction with, population centers, prevalence in those population centers, availability of housing to quarantine students, and the desires of the surrounding community (12). However, it also reflects substantial continued uncertainty about how policy translates into outcomes. Such uncertainty and diversity in approach among universities reflects the larger response to the pandemic, in which US states and national governments adopted dramatically different responses to the pandemic despite apparently similar circumstances.Simulation-based epidemic models would seem to offer the power to resolve this uncertainty in support of high-quality decisions. They allow prediction, customized to the circumstances of a university, city, state, or nation. By varying the interventions in silico and observing predicted outcomes, one can hope to choose the best course of action. Unfortunately, epidemic models only approximate reality (13). Ever-present uncertainty in model input parameters coupled with the potential for exponential growth significantly limit accuracy. Small differences in behavioral and biological parameters can cause huge differences in predicted case counts. As a consequence, epidemic models have been maligned for producing inaccurate point estimates (13, 14).This article demonstrates that simulation models can support effective selection of COVID-19 interventions even when they are unable to provide accurate point estimates of epidemic outcomes. We demonstrate this through a case study of how simulation models supported the design of COVID-19 interventions that were subsequently implemented at Cornell University. We also present a modeling framework that can support decisions at other universities. (The use of epidemic models in the presence of significant parameter uncertainty is also discussed in, for example, ref. 15. In such settings, clear communication of uncertainties is key; see, for example, ref. 16).)In close communication with Cornell University’s administration, we conducted a simulation-based analysis in summer 2020 using a compartmental Susceptible, Exposed, Infectious, and Recovered model with multiple subpopulations; see refs. 17–19 for closely related models. Our work was the basis for the decision to reopen Cornell’s Ithaca campus for residential instruction in fall 2020 (20) and was used to design an asymptomatic screening program that was and remains a critical part of Cornell’s strategy.Based on these modeling recommendations, all students were invited to return to Cornell’s Ithaca campus for residential instruction during the 2020–2021 school year under an asymptomatic screening program, and 75% of students returned (21). The surveillance program used pooled PCR testing with the testing frequencies obtained through our modeling. The surveillance program used less-sensitive but more-comfortable anterior nares (AN) sampling over nasopharyngeal (NP) sampling, because modeling suggested that the benefits of comfort to test compliance outweighed a potential loss in sensitivity. Asymptomatic surveillance was enabled at Cornell through a major effort to support large-scale sample collection and develop a new COVID-19 testing laboratory based on diagnostic expertise in Cornell’s College of Veterinary Medicine, and through a unique partnership with a local health care provider. Based on recommendations from our simulation modeling approach, this strategy was updated for the spring 2021 semester to test varsity athletes and students in Greek-life organizations more frequently (contact tracing data showed them to have more social contact than other individuals) and again in fall 2021 to adjust for the Delta variant, changes in social distancing policies, and the protection offered by vaccination. Over the course of the 2020–2021 academic year, there were fewer than 1,044 infections among students and employees, fewer than many schools with similar student populations offering only virtual instruction (1, 22).Our modeling approach hinges on delineating those simulation model input parameters yielding epidemics that can be successfully controlled versus those that cannot. If the set of plausible input parameters are contained within the set of safe parameters, then we can be highly confident, although never certain, that the epidemic can be controlled. At Cornell in summer 2020, we demonstrated this to the university administration for a suite of interventions available with in-person instruction: frequent asymptomatic screening, testing students on arrival to campus, contact tracing, social distancing on campus, limits on student and employee travel, masking requirements, and a behavioral compact curtailing student social gatherings. It was also possible that we would have found that plausible ranges of the input parameters overlapped the portion of parameter space where epidemics would grow out of control, in which case we would not have been able to recommend reopening.We found that access to regular asymptomatic screening (7, 23), with an ability to increase testing frequency if needed, was critical. Indeed, those few universities employing a similar asymptomatic screening approach succeeded, by and large, in controlling campus outbreaks (24–27). See also refs. 28–33 for explorations of the interaction of pooled testing and asymptomatic surveillance for controlling epidemics.We also found it was critical to analyze epidemic growth if in-person instruction were not offered, to quantify the relative merits of the alternative to in-person instruction. Survey results (20, 34) suggested that a significant number of students would return to the Ithaca area even if in-person instruction were not offered. Without the benefits of the legal framework offered by in-person instruction, frequent asymptomatic screening would have been difficult to mandate for this population. Moreover, our analysis suggested that many of those parameter settings in which asymptomatic screening would not ensure safe in-person instruction would also be ones in which a significant outbreak would occur in the local student population under virtual instruction. This resulted in the decision to reopen Cornell’s Ithaca campus with a fully residential semester in fall 2020 (20).We additionally measure key parameters of a university population needed for understanding the dynamics of epidemic spread, including university subpopulations’ intergroup and intragroup rates of viral transmission and how it has changed over time with vaccination, the Delta variant, and relaxation in social distancing. We find that a small group of students has significantly more intergroup viral transmission than other groups and plays an important role in determining the risk of an outbreak. We find that targeting interventions to this group provides substantial protection against outbreaks. Unlike students, we find that employees have very little transmission at work and are well separated from students, with extremely little transmission across the two groups. This has implications for understanding the risk to older and more vulnerable individuals from student infections.When considering a range of interventions against the Delta variant, we find that achieving high levels of vaccination provides significant protection, but that, even in a 100% vaccinated student population, there is significant potential for breakthrough outbreaks in the absence of asymptomatic screening and social distancing. This is consistent with findings from other modeling studies (19). While once per week asymptomatic screening of vaccinated students might be sufficient in many situations, we find that testing vaccinated student groups with high rates of social contact twice per week substantially reduces risk even when the entire population is vaccinated. We also find that moving from 75% vaccination to full vaccination provides substantial additional protection.To summarize, the key contributions of this paper are 1) providing a simulation framework for supporting the design of COVID-19 interventions despite parameter uncertainty; 2) demonstrating this framework through its implementation at Cornell University; 3) measuring key parameters of the dynamics of the spread of SARS-CoV-2 in university populations and the effectiveness of interventions; and 4) providing a framework for making decisions moving forward, including the design of asymptomatic screening strategies in the presence of partial vaccination and the Delta variant.Our work adds to the broader literature using epidemic modeling in the context of universities. See, for example, ref. 35 for a perspective on the challenges of reopening as informed by a variety of epidemic models, refs. 36 and 37 for the use of agent-based modeling to evaluate mitigation strategies to enable safe in-person instruction, ref. 38 for probabilistic modeling of strategies to suppress virus spread in dorms and classrooms, and ref. 39 for a study of interventions for generic small residential campuses.     

The treatment of advanced cardiac allograft rejection

Severe cardiac allograft rejection remains a serious problem despite the advances of cyclosporine-based immunosuppression. This study analyzes our experience with 202 recipients of cardiac allografts who were treated primarily with cyclosporine and prednisone. Failure of such therapy in 86 patients (43%) resulted in 105 episodes of advanced cardiac allograft rejection as diagnosed by endomyocardial biopsy. Of 101 rejection episodes that were initially treated with intravenous pulse therapy, 48 (48%) were successfully resolved, yet 60% of these successes were associated with major infections. Patients in whom steroid therapy failed or was contra-indicated received intravenous antithymocyte globulin (ATG) or intravenous monoclonal antibody (OKT3). ATG and OKT3 successfully reversed severe rejection in 26 (81%) of 32 and in 13 (93%) of 14 episodes, respectively. Infectious complication rates were 54% and 21%, respectively. Because the majority (87%) of these rejection episodes occurred within the first 30 days after treatment, many of them may have resulted from inadequate immunosuppressive induction therapy. Based on our results, we believe that advanced cardiac allograft rejection may be managed best by individualizing immunosuppressive therapy, thus enhancing prevention, and by adding OKT3 to the regimen when rejection occurs.