By late Spring 2020, large-scale viral testing for SARS-CoV-2 followed by case isolation and contract tracing was identified as a potentially effective mitigation mechanism, particularly given the absence of vaccines and other pharmaceutical interventions. In mid-July 2020, the Georgia Institute of Technology, a community of around 27,000 students and 15,000 staff and faculty, committed to a comprehensive viral surveillance program aiming to mitigate the spread of SARS-CoV-2 among the on-campus student population, as well as staff and faculty. The program included the capacity to test over 1,500 individuals per day initially, rising to a maximum of 2,850 per day. With an estimated 7,370 students in residence (approximately 1,170 in 38 fraternity or sorority houses and 6,200 in 48 residence halls), and up to 5,000 staff and non-resident students visiting campus daily, testing capacity meant that each individual could potentially be tested biweekly at inception and weekly by mid-Fall semester. The import of approximately 50 to 100 cases upon start of the Fall semester was anticipated in light of estimated prevalence amidst a late summer surge in the Southeast.1 Identifying and restricting the spread of cases in a high-density live–learn environment formed the basis for the design and scope of the surveillance program. Positive individuals would be isolated for 10 days per CDC guidelines,2 allowing a reduced level of in-person instruction and research to continue.
We preassessed the scope of the program by evaluating the synergistic effectiveness of nonpharmaceutical interventions and testing.3,4 Nonpharmaceutical interventions reduce baseline expectations for R0 (anticipated to be approximately 2.5 for SARS-CoV-2 https://www.cdc.gov/coronavirus/2019-ncov/hcp/planning-scenarios.html#table-1) such that the benefits of testing are synergistic with reduced spread. Since we estimated the average time between infection and recovery (including both latent and infectious period) to be approximately 7 to 10 days,5 we expected rapid testing and isolation that occurs at a similar timescale to constitute a form of mitigation.6
Identification and isolation of individuals via surveillance testing has the potential to reduce the number of transmission events to susceptible individuals. Contemporaneous models had suggested that comprehensive testing with a cadence of 2 days would be required to keep infection levels at a fraction of one percent (reported in https://covid19.illinois.edu/more-information/covid-19-briefing-series/). However, University System of Georgia’s system-wide policies prevented mandatory testing and Georgia Tech did not have the resources to conduct thrice-weekly testing of the entire campus community. Nevertheless, we conducted scenario-based epidemic model analyses, consistent with other reports,7,8 which indicated that achieving testing at approximately 7-day to 10-day intervals combined with mask wearing, online courses, and distancing (i.e., nonpharmaceutical interventions), could restrict the attack rate in the campus community to ~1000 cases as opposed to >5000 in the absence of testing. The Susceptible-Exposed-Infectious-Removed modeling considers the spread of SARS-CoV-2 among 15,000 individuals in a focal community with large-scale testing-based interventions (2 day exposure period, 7 day infectious period, with transmission rate varied to match a range of Reff values; full model simulation available on GitHub). Figure 1 shows a scenario for how expected cumulative caseloads over a semester are expected to vary as a function of Reff and frequency of testing. We contrasted scenarios with or without entry testing, and assuming either a 90% (see Figure 1) or a 75% sensitive test (see GitHub), finding that testing speed was more critical than sensitivity differences at this scale in providing population-wide benefits.
;)
FIGURE 1.: Epidemic modeling of the predicted cumulative number of cases in one 14-week semester expected on campus. A simplified SEIR (susceptible-exposed-infected-removed) model assumed a total community of 15,000 individuals, a 2-day latent period on average between infection and positivity, and a 7-day infectious period. Transmission rates were varied to yield the basic reproduction number on the x axis. A, Modeling without testing as students return to campus. B, With testing, which reduces the initial positivity. Contour points show the relationship between average test frequency per individual and baseline transmission ratio. We shared these data with campus leadership in arguing for implementation of the program and shared with the community as part of communication efforts in August 2020.
We presented these results to Institute leadership prior to re-entry and then to the Institute community during Fall 2020 entry.9 Although the program was not fully operational on re-entry, we predicted that even 7-day testing with Reff in the range of 1.5 to 2 could halve the case rate and potentially limit it to <10% of the community; benefits continue to accrue with even more effective nonpharmaceutical interventions. The rationale for testing effectiveness as a form of mitigation is that the bulk of infectious individuals would be ‘removed’ due to two mechanisms: (1) natural recovery and loss of infectiousness at a rate γ; (2) identification and isolation of infectious individuals via testing at a rate ω with sensitivity se. As a result, the reduction in infectiousness is expected to be the ratio of the removal rate to the sum of rates: Hence, effective testing on the order of one week has a chance to compete with the recovery period in reducing the effective reproduction number (and overall transmission). The synergistic benefit leverages the fact that epidemic dynamics are nonlinear, implying that masking, distancing, and online instruction were also essential components of an intervention approach.10 Given uncertainty of parameter estimates and heterogeneity in spread, we viewed the purpose of epidemic models to guide principles of the testing strategy rather than claim precise prospective quantification of the location and extent of cases (which were almost certainly going to be influenced by behavior and stochastic effects). Nonetheless, we show here that the outcomes in terms of mitigation of COVID-19 incidence were fully consistent with this modeling.
METHODS
For safety and simplicity, we developed a saliva-based test, and used it with informed consent of all participants prepared in consultation with the institute IRB, also including external legal review. In the absence of a testing mandate, participation was voluntary and conducted under the auspices of quality improvement in the interest of campus safety amidst a public health emergency. We established a half-dozen testing locations at convenient locations near residences and at campus hubs, and encouraged participation by messaging, word-of-mouth, and incentivization with frozen treats and/or fresh cookies depending on the time of year. Additional testing locations were added via pop-up opportunities on weekends, at special events, and in response to anticipated interest surges (e.g., before the end of the Fall semester). Individuals spit into a plastic cup, transferred approximately 0.4 mL via a dropper into a 1.5mL Sarstedt screw-top tube (up to a premarked indicator line), preloaded with 0.4mL viral deactivation buffer, which also preserves viral RNA at room temperature for subsequent analysis. The testing procedure was supervised but self-administered and took approximately 2 minutes to complete. In contrast, a nasal swab test was expected to be less acceptable to most students, would require storage in live viral transport medium, and involve more labor-intensive collection and laboratory processing. Subsequent studies have confirmed superior analytical test features of saliva relative to nasal or oropharyngeal tests consistent with our modeling expectations, e.g., meta-analysis of 46 studies indicates a pooled sensitivity of 88% (95% credible interval = 81%, 93%) for saliva and 82% (95% CI = 73%, 90%) for nasal swabs, although nasopharyngeal swabs remain the gold standard.11 Two members of the team (M. Farrell and M. Shannon) had been engaged with the Georgia State testing taskforce for several months and were able to conduct a field study south of Atlanta in July, in conjunction with partners at other USG universities. The saliva-based test prototype had >95% sensitivity and specificity that met the requirements for an Emergency Use Authorization submission to the FDA. Approval to establish a new CLIA-certified diagnostic laboratory was also obtained from the Georgia DPH the week before student re-entry began on August 1. Individual samples were retained for follow-up CLIA diagnostic testing of presumptive positives identified in initial surveillance; this 2-step strategy reduced costs and increased testing capacity.
We designed a novel double-pooling scheme to evaluate 15 individuals in 6 wells of a 96-well plate where each sample was present in 2 of the wells in a unique combination that, given prevalence around 1%, usually allowed us to infer which sample was positive. This approach reduces false positives and controls the number of follow-up diagnostic tests, also reducing costs—consistent with subsequent reports.12 Leaving 6 wells for controls, we surveyed 15 × 15 = 225 individuals per plate, using Integra robots to generate pools, Kingfisher robots to isolate RNA, and QuantStudio quantitative RT-PCR testing. The PCR test was the CDC-approved IDEXX kit supplied by OptiMed, with control human RP and combined N1 and N2 SARS-CoV-2 primers. Individuals in negative pools were informed by email, usually within 24 to 36 hours, that “no further action is required” without giving an actual clinical result, while presumptive positives, ambiguous samples, and some negatives were sent a provisional email asking them to self-isolate while awaiting the result of a CLIA-approved diagnostic test. Thus, although each sample was tested multiply, only the diagnostic test counted as a case positive and each reported case was a single evaluation. Mean time to final test report in the event that a sample was referred for CLIA diagnostic testing was 48 hours (standard deviation 14 hours); all positives were referred to Stamps Health Services (Georgia Tech’s health care center), reported to the Georgia DPH, and contact tracing was initiated. Students were given the option to return home or preferably to isolate for 10 days in a local hotel contracted for this purpose. All staff and faculty positives were reported to the Georgia DPH as required, contract tracing was initiated, and individuals were advised to consult a physician and isolate at home for 10 days. Also per CDC guidelines, close contacts defined by at least 15 minutes within close proximity were also required to isolate, but some noncompliance is likely since there was no enforcement of the movements of students who elected to quarantine off-campus. Institute policy is outlined here: https://health.gatech.edu/coronavirus/isolation-quarantine.
In addition to the asymptomatic surveillance program, 3 other sources of tests were available for symptomatic individuals. First, Stamps Health Services provided Rapid LAMP tests for up to 20 students a day (with observed positivity between 10% and 20% across the semester). Second, commercial Vault tests were available for some staff and on weekends. Third, some members of the Georgia Tech community preferred to be tested off campus with a variety of commercially available options. We do not have any data on compliance with reporting recommendations for individuals who tested off-campus, and consequently cannot ascertain the true off-campus positivity rate. Nor do we know what the positivity rate was for positive tests off campus, since negatives were not reported. However, the majority of positives (62%) were reported from the asymptomatic surveillance testing program, and all reported positives were from CLIA-approved clinical diagnostic tests. A Dashboard available at https://health.gatech.edu/coronavirus/health-alerts visualizes daily case numbers by type of test or campus community, surveillance positivity, and tests per day.
RESULTS
A total of 1,825 positive tests were reported combining all testing modes in Fall semester of 2020, from August 9 through December 19. Figure 2A shows the number of cases per day along with a 7-day sliding average (including both surveillance positives and total positives) in the Fall term. In Fall, these 1,825 positive tests corresponded to 1,508 individual cases since 236 participants tested positive twice, 31 participants thrice, and 6 participants 4 or more times. Most repeat positive tests were within a day or 2 of one another (either because the individual visited both Stamps and the surveillance program if they suspected infection, or they elected off-campus isolation and nevertheless returned for a second test, which could be interpreted as confirmation-seeking behavior). However, multiple cases of positivity 3 or more months apart were observed. There is no indication that these are indicative of disease recurrence and infectiousness, nor is there any information on whether they represent repeat infections. The vast majority of positive cases were students (1,351, 90% of cases, 9.7% cumulative positive) with just 157 Staff/Faculty infections of the 4,078 tested (3.8% cumulative positive), whereas approximately 15% of the tests were of Staff/Faculty. The asymptomatic surveillance program recorded 940 independent cases from 112,500 tests, an average positivity of 0.84%, and the cumulative estimated proportion of infected individuals was 8.4% based on approximately 18,029 registered participants. Note that in the Spring semester of 2021 (January 5 through May 8), there were 616 surveillance positives on 142,241 tests for an average positivity of 0.43%. Including another 227 cases through the other sources, the cumulative proportion of documented, infected individuals rose to 13%.
FIGURE 2.: Summary distribution of SARS-COV-2 cases and positivity on campus. A, Total cases per day in the Fall term, 9 August 2020-19 December 2020, from all test sources, including 7-day sliding window, generated from GT dashboard
https://health.gatech.edu/coronavirus/health-alerts, which allows interactive parsing of data by groups (residence, staff/student) and visualization of testing rates and positivity. B, Surveillance positivity rate per day from surveillance program with 7-day sliding window (red line), including 80% Bayesian credible intervals. C, For completeness, positivity is also included for Spring 2021.
A large fraction of the campus community did not engage. At least one half of the students did not visit campus at all, working virtually and taking online instruction only (enrolment in Fall 2020 was approximately 27,000) and an even greater proportion of the over 14,000 staff and 1,000 faculty never tested and likely rarely visited campus if at all. Among students, the weekly testing rate was 59% (80% CI = 41%, 76%) for all undergraduate residences from weeks 3 to 14, being markedly higher (74%) during a period of very active messaging in response to the high entry positivity, dipping to 50% and then rising again toward the end of the semester. There was variation in participation rates among residences (ANOVA F39,440, P < 10−32; range 41% to 85%) and this was not correlated with the size of the dormitory. Notably, up to 20% of students did not participate. Supplementary eFigure S1A; https://links.lww.com/EDE/B881 shows the daily participation rate, which stabilized between 8000 and 9500 tests per week, 91% of which represent a single individual testing at most once in a given week.
Three notable aspects of the results provide implications of public health relevance. The first is the ability of large-scale testing to rapidly identify and control case surges (Figure 2A), as revealed twice in Fall 2020 (Figure 2B) and subsequently early in the Spring 2021 semester (Figure 2C) prior to the availability of vaccines. After a low initial re-entry positivity fraction under 0.5% in Fall 2020, a severe outbreak occurred in week 3. Seven symptomatic cases were identified in one fraternity (hereafter referred to as “Greek”) house, which was followed by transmission through the Greek community and into nearby Residence Halls. An intensive campaign was initiated to test almost 90% of the on-campus residents over the next 2 weeks resulting in a peak asymptomatic positivity rate of 4.1%. Due, in part, to rapid case identification and isolation, the asymptomatic positivity rate steadily declined, and by mid-September positivity was consistently below 0.5% and on some days no positives were detected in > 1500 tests (Figure 2B). Case rates increased slightly in October, concordant with increasing community levels of transmission in Georgia (https://dph.georgia.gov/covid-19-daily-status-report; eFigure S2; https://links.lww.com/EDE/B881), with the asymptomatic positivity rate approaching 2% on a single day in late October. This second wave was more diffuse than the first and included a higher proportion of off-campus participants (likely due to community transmission unrelated to on-campus spread, noting that GT is located in Midtown Atlanta). Targeted surveillance testing of the most-affected dormitories was carried out in early November, though eFigure S1B; https://links.lww.com/EDE/B881 does not reveal any clear relationship between spikes of positivity and frequency of testing in a typical focal dorm, likely because testing rates were already high by that point. Positivity declined to < 0.5% in the final days of the Fall semester, again illustrating how intensive testing can be part of an integrative strategy to reduce infections. A similar pattern of an initial peak followed by sustained low positivity in the vicinity of 0.2% to 0.4% was seen in re-entry for the Spring 2021 semester (see Figure 2C).
Our second observation is that there is heterogeneity with respect to COVID-19 risk. This observation is consistent with growing recognition of the importance of over-dispersion in transmission.13,14 The heterogeneity of risk is reflected in cumulative incidence data through September 25th summarized in Figure 3A and 3B. Figure 3A illustrates that 75% of the positive tests were attributed to 25% of the residence halls or Greek houses, and reciprocally that 20% to 33% of these living residences had almost no cases. Testing rates in residence halls was approximately 55% weekly after mid-semester. Testing rates in Greek houses was approximately 75% weekly after mid-September, though it is notable that a subset of Greek houses had nearly 100% weekly participation whereas a few had <20% participation. The majority of student cases were restricted to the east side of campus in residences located close to the Greek community. Figure 3B contrasts the number of tests per individual through 6 November 2020, for all individuals who had at least one test, showing clear bimodality in the general community (top), with a left-shift toward lower testing rates for the individuals who at some point had a positive test (bottom). Note that both distributions are subject to joint effects of sampling biases, confirmation bias (given that individuals with positive cases are recommended not to test again for a 3-month period), as well as intrinsic variation in individual risk and tendency to test. Nonetheless, we interpret both datasets as indicative of different behavioral approaches to the pandemic, with variation in the level of engagement with the testing program and practices that minimized individual risk. Although 44% of the individuals in Figure 3B tested at least biweekly (7× or more), some groups of individuals refused to test at all or tested off-campus.
FIGURE 3.: Heterogeneity in infection and testing. A, Positive cases were heterogeneously distributed in both the residence halls and Greek houses by September 25, when almost 4% and 18% of individuals had been infected, but most of the cases were due to a few residences. In each pair of columns, the proportion of occupants or positives is represented by a different color for each residence. Note that 1/5 of the residence halls and 1/3 of the Greek houses had 0 cases. B, Histogram distributions of the number of individuals testing (top) or diagnosed positive (bottom) who took from 1 to 32 tests across most of the Fall semester, 9 August 2020 through 6 November 2020.
A third observation was that shared double rooms in which 2 individuals sleep had elevated positivity. During the first peak, when one individual of a double tested positive, the second also tested positive either through surveillance or other sources over 30% of the time (38 of 125), 40% on the same day, 30% within 2 days, with just 3 cases > 5 days apart. Almost one half of the first wave cases were in double rooms, i.e., rooms with 2 individuals in the same sleeping space. There were several instances of a mini-cluster arising in an adjacent block of rooms. Subsequently, all students were offered the option of moving to their own single room, but many chose to remain with doubles, citing the preference to have more social interaction, and mental health concerns. The finding of elevated infection rates in shared living spaces is consistent with findings of elevated household transmission.15Figure 4 shows the proportion of positives observed in clusters of >1 individual sharing a room or suite of 2 or 4 bedrooms around a shared living area, for the 20 residence halls in which students shared accommodations at the start of the Fall semester. Although there was no overall relationship between cumulative positivity (orange bars) and proportion of cases in clusters (blue bars), 6 of 7 of the dorms with at least 10% cases across the semester had at least 30%—and in some cases 60%—of cases in clusters. Notably, 44 of the 50 clusters total clusters over the semester occurred prior to mid-September when the university implemented the policy to encourage students to move to their own individual rooms.
FIGURE 4.: Proportion of cases in clusters, Fall 2020. The bar graph shows the cumulative positivity paired with the proportion of cases observed in clusters within the same room or suite, both as percentages, for 20 of the dormitories.
DISCUSSION
Our results support the use of large-scale surveillance as a form of mitigation on college campuses, but contrast in important ways with other reports. Duke University for example reported just 84 total positives from a total of 68,913 tests performed on 10,265 students, only one-third of which were detected directly from pooled surveillance, the remainder being discovered through contact tracing or as symptomatic cases.16 Duke’s weekly positivity rate averaged just 0.08, though it should be noted that they report that just 18.4% of their positive pools led to a confirmed case, whereas Georgia Tech’s conversion rate for double-positive wells was very close to 100%—implying that gaps in test sensitivity and/or use of retrospective serological surveys should be considered when comparing implementation results. A survey of 9 Boston-area colleges17 where testing of all resident students was performed at least weekly found a cumulative documented incidence of 1.6% (range = 0.3% to 3.0%), also at least 5-fold lower than the cumulative documented Georgia Tech incidence. Note that documented new cases per capita in August were at least 3× higher in Atlanta than in Boston, though documented new case rates were more similar in September and October. The Boston study found no relationship between dormitory occupancy and positivity and no evidence for transmission within on-campus student housing,17 in direct contradiction to our findings. The differences may be a result of dorm density, room configuration, ventilation, and/or other factors.
The closest contextual comparison is provided by the University of Georgia (UGA), which reported (https://www.uhs.uga.edu/healthtopics/covid-19-health-and-exposure-updates) that the positivity rate exceeded 9% during their third week of testing and remained over 1% until mid-November, when another post-Thanksgiving increase to in excess of 4% occurred. In total, UGA reported 819 surveillance positives out of 33,435 tests for an average positivity rate of 2.5%. Given presumed similarity in nonpharmaceutical interventions, this contrast in outcomes provides some evidence for direct reduction of transmission via large-scale testing with relatively short turnaround times (consistent with related analyses in 6,18). GT had a 3-fold higher level of testing (at least 5-fold relative to community size) and 3-fold lower positivity fraction relative to that of UGA. Relatedly, comparing our rates of positivity (Figure 2B) with estimates for Fulton County, which includes Midtown Atlanta and Georgia Tech, in eFigure S2; https://links.lww.com/EDE/B881, we estimate that we achieved 3- to 5-fold reduction in incident cases, depending on assumptions of the Fulton County modeling used to infer true positive rates from reported cases given that the City of Atlanta did not pursue large-scale surveillance.
Notably, dashboard results from the University of Illinois Urbana-Champaign (https://splunk-public.machinedata.illinois.edu/en-US/app/uofi_shield_public_APP/home) reveal > 4500 positive cases out of > 1,000,000 saliva-based surveillance tests in the Fall semester, for an aggregate 0.44% positivity; that is approximately one-half of GT’s positive % with approximately 9× as much surveillance testing. These contrasting examples provide context for efforts to assess costs and benefits in the implementation of large-scale viral surveillance testing.
In summary, we estimate that large-scale testing along with nonpharmaceutical interventions and sustained communication were critical to GT’s ability to reduce transmission even amidst elevated community spread. Operationally, we note that testing of high-risk campus residences, including data-driven testing targeted at the level of individual floors and even clusters of adjacent rooms, is consistent with the identification and isolation of clusters; additional improvements in the speed and targeting of adaptive testing could be critical to increased efficacy. Implementation of large-scale testing programs that include increases of testing and targeted responses to clustered outbreaks represent an ongoing opportunity to mitigate spread amidst college, K-12, and business re-openings.
Acknowledgments
We are extremely grateful to the students and staff who embraced the program, and to dozens of colleagues in housing, communications, legal, and the PCR lab, who volunteered their time and, in many cases, redirected their effort to implement this program. Particular thanks go to Frank Neville, Aisha Oliver-Staley, Emily Ryan, True Merrill, and German Khunteev for their critical roles in establishing the program.
References
1. Chande A, Lee S, Harris M, et al. Real-time, interactive website for US-county-level COVID-19 event risk assessment. Nat Hum Behav. 2020;4:1313–1319.
2. Paltiel AD, Zheng A, Walensky RP. Assessment of SARS-CoV-2 screening strategies to permit the safe reopening of college campuses in the United States. JAMA Netw Open. 2020;3:e2016818.
3. Matrajt L, Leung T. Evaluating the effectiveness of social distancing interventions to delay or Flatten the epidemic curve of coronavirus disease. Emerg Infect Dis. 2020;26:1740–1748.
4. Watkins AE, Fenichel EP, Weinberger DM, et al. Pooling
saliva to increase SARS-CoV-2 testing capacity. MedRxiv 2020;
https://doi.org/10.1101/2020.09.02.20183830
5. Wiersinga WJ, Rhodes A, Cheng AC, Peacock SJ, Prescott HC. Pathophysiology, transmission, diagnosis, and treatment of coronavirus disease 2019 (COVID-19): a review. JAMA. 2020;324:782–793.
6. Brooks-Pollock E, Christensen H, Trickey A, et al. High COVID-19 transmission potential associated with re-opening universities can be mitigated with layered interventions. Nat Commun. 2021;12:5017.
7. Panovska-Griffiths J, Kerr CC, Stuart RM, et al. Determining the optimal strategy for reopening schools, the impact of test and trace interventions, and the risk of occurrence of a second COVID-19 epidemic wave in the UK: a modelling study. Lancet Child Adolesc Health. 2020;4:817–827.
8. Lopman B, Liu CY, Le Guillou A, et al. A modeling study to inform screening and testing interventions for the control of SARS-CoV-2 on university campuses. Sci Rep. 2021;11:5900.
9. Weitz JS, Gibson G. Science and projections for a return to campus. figshare 2020.
https://doi.org/10.6084/m9.figshare.12761357.v1
10. Grassly NC, Pons-Salort M, Parker EPK, White PJ, Ferguson NM; Imperial College COVID-19 Response Team. Comparison of molecular testing strategies for COVID-19 control: a mathematical modelling study. Lancet Infect Dis. 2020;20:1381–1389.
11. Lee RA, Herigon JC, Benedetti A, Pollock NR, Denkinger CM. Performance of
saliva, oropharyngeal swabs, and nasal swabs for SARS-CoV-2 molecular detection: a systematic review and meta-analysis. J Clin Microbiol. 2021;59:e02881–e02820.
12. Mutesa L, Ndishimye P, Butera Y, et al. A pooled testing strategy for identifying SARS-CoV-2 at low
prevalence. Nature. 2021;589:276–280.
13. Wong F, Collins JJ. Evidence that coronavirus superspreading is fat-tailed. Proc Natl Acad Sci U S A. 2020;117:29416–29418.
14. Hébert-Dufresne L, Althouse BM, Scarpino SV, Allard A. Beyond R0: heterogeneity in secondary infections and probabilistic epidemic forecasting. J R Soc Interface. 2020;17:20200393.
15. Madewell ZJ, Yang Y, Longini IM Jr, Halloran ME, Dean NE. Household transmission of SARS-CoV-2: a systematic review and meta-analysis. JAMA Netw Open. 2020;3:e2031756.
16. Denny TN, Andrews L, Bonsignori M, et al. Implementation of a pooled surveillance testing program for asymptomatic SARS-CoV-2 infections on a college campus - Duke University, Durham, North Carolina, August 2-October 11, 2020. MMWR Morb Mortal Wkly Rep. 2020;69:1743–1747.
17. Stubbs CW, Springer M, Thomas TS. The impacts of testing cadence, mode of instruction, and student density on Fall 2020 COVID-19 rates on campus. medRxiv 2020;
https://doi.org/10.1101/2020.12.08.20244574
18. Larremore DB, Wilder B, Lester E, et al. Test sensitivity is secondary to frequency and turnaround time for COVID-19 screening. Sci Adv. 2021;7:eabd5393.
