Feasibility and efficacy of mass testing for SARS-CoV-2 in a UK university using swab pooling and PCR

Transmission of SARS-CoV-2 without symptoms is well described, and may be mitigated by mass testing. Nonetheless, the optimal implementation and quantitative real-world impact of this approach remain unclear. During a period of rising SARS-CoV-2 prevalence, students at the University of Cambridge were enrolled in a voluntary programme of weekly PCR-based asymptomatic screening. Swab pooling by household reduced the total testing capacity required by five-fold, without affecting laboratory workflows or compromising test sensitivity. Participation remained >75% throughout the study period. 299/671 (45%) of students diagnosed with SARS-CoV-2 were either identified or pre-emptively quarantined because of the screening programme. After a negative screening test, the risk of developing COVID-19 over the following 7 days was decreased by 51%. Modelling transmission using parameters from our study suggests a reduction in R0 of up to 31% attributable to weekly screening. We therefore demonstrate the feasibility and efficacy of regular, voluntary mass testing for COVID-19.

population also has the potential to reduce viral spread to university staff and members of 92 the local community, who may be at higher risk of developing severe disease 27,28 . Lessons 93 learnt from mass testing in universities should be broadly applicable to asymptomatic 94 screening for COVID-19 in other settings. 95 Variable levels of information on asymptomatic screening programmes at European and 96 North American universities are available from institutional websites, but outcomes of these 97 In this study, we describe the implementation and validation of swab pooling and PCR 115 testing in a UK university, and demonstrate how this approach can be used for weekly 116 asymptomatic screening of student households. At the same time, we present a detailed 117 epidemiological description of the characteristics and spread of COVID-19 within the 118 university, identifying risk factors for infection and linking secondary attack rates amongst 119 household contacts with the characteristics of index cases. By combining these data, we 120 provide evidence for the efficacy of asymptomatic screening which goes beyond simple 121 numerical case ascertainment. We demonstrate how regular screening is able to identify 122 many students with presymptomatic as well as asymptomatic SARS-CoV-2 infection, 123 quantify the reduction in risk of COVID-19 associated with a negative screening test, and 124 use model-based analysis parameterised from our study to quantify the ability of mass 125 testing to reduce transmission. 126

127
Implementation of asymptomatic screening programme and characteristics of study 128 participants 129 The prevalence of COVID-19 amongst older teenagers and young adults in the UK rose 130 steeply in autumn 2020, reaching approximately 1.5% by October 1, 2020 49 . Against this 131 backdrop, 12,781 students living in University of Cambridge accommodation participated in a 132 weekly asymptomatic COVID-19 screening programme during the autumn term, spanning 9 133 weeks between October 5, 2020 and December 6, 2020 (Fig. 1). In addition to 134 asymptomatic screening, the University of Cambridge also provided a dedicated, PCR-135 based testing service for all students and staff with cardinal symptoms of COVID-19 (fever, 136 cough, and/or anosmia/ageusia). Amongst study participants, 1,031 undertook at least 1 of 137 these symptomatic tests during the study period (Fig. 1). 138 A detailed description of the asymptomatic screening programme is provided in the Methods, 139 and further information is available from the University of Cambridge website 140 (https://www.cam.ac.uk/coronavirus/stay-safe-cambridge-uni/asymptomatic-covid-19-141 screening-programme). In brief, students were screened using swab pooling and two-step 142 confirmatory PCR testing (Fig. 2). Combined nose and throat swabs were obtained by self-143 administration in the students' own accommodation. Swabs from up to 10 students were 144 then immediately pooled in the same tube of viral transport medium. In general, testing pools 145 corresponded with student households. In the event of a positive pooled screening test, 146 students in the pool were instructed to self-isolate, and invited for same-day individual PCR 147 confirmatory testing. Students with positive individual confirmatory tests were treated in the 148 same way as students with positive results from symptomatic testing, including self-isolation, 149 household quarantine and contact tracing. If all individual confirmatory tests were negative, 150 students were released from self-isolation, typically after 1-2 days (depending on whether 1 151 or 2 rounds of individual tests were required) (Fig. 2). 152 University accommodation was divided into 3,094 households (mean 5.0 household 153 members, range 1-20) (Fig. 3a). For pooled sample collection, students were organised into 154 2,275 testing pools (mean 6.8 students, range 1-10) (Fig. 3b). To minimise the number of 155 screening tests required, 1,476 (47.7%) smaller households were combined in merged 156 testing pools. To ensure a maximum of ten swabs per sample tube, 151 (4.9%) larger 157 households were spilt between multiple testing pools. Overall, 65% of testing pools 158 corresponded exactly with households, 21% of pools included more than one household, 159 10% of pools included a part of one household, and 4% of pools included students from 160 more than one household and part of one household. Participation was higher by household 161 than by individual student, increasing from 91.3% (week 1) to 92.9% (week 9) of all 162 households with eligible students (Fig. 3c). 163 Compared with the overall population of eligible students, the characteristics of participating 164 students were similar (Supplementary Table 1). Undergraduates (particularly first year 165 students and those studying science and technology courses) were somewhat better 166 represented, with small but significant differences in sex, age, ethnicity and UK/international 167 residency. The proportion of students consenting to participate increased over the course of 168 the term, from 75.2% (week 1) to 81.9% (week 9) of all eligible students ( Fig. 3d and  169 Supplementary Table 2). Spikes in recruitment were observed after email communications 170 to students providing information about the programme (Extended Data Fig. 1). 171 In theory, the total testing capacity required each week (including both pooled screening and 172 individual confirmatory tests) was expected to vary according to testing pool size and the 173 prevalence of infection amongst screened students (Fig. 3e). In practice, the programme 174 was scaled up in stages over the course of the university term, by increasing the number of 175 students invited to participate in pooled sample collection each week (weeks 1-2, 2 students 176 from each testing pool; weeks 3-7, half the students from each testing pool, alternating week 177 by week; weeks 8-9, all the students from each testing pool, every week) (Fig. 3c,d). At full 178 scale (weeks 8-9), utilisation of testing capacity was almost 5x more efficient than screening 179 with individual tests alone ( Fig. 3e and  positive pooled screening tests, the proportion of operational false positives was lowest in 195 weeks 2 (1/28, 3.6%), 3 (2/38, 5.3%) and 6 (8/67, 11.9%) when the prevalence of infection 196 amongst screened students was highest. Conversely, the fraction characterised as false 197 positives was highest in weeks 8-9 (14/17, 82.4%), when the prevalence of infection 198 amongst screened students was lowest (Supplementary Table 3). 199 Because students contributing swabs to negative pooled screening tests did not undergo 200 paired individual tests, it was not possible to calculate an operational false negative rate for 201 pooled screening tests. Nonetheless, we did not observe any loss in analytical sensitivity for 202 pooled samples containing ≥5 swabs tested using spike-in standards (Extended data Fig.  203 2), and the limit of detection of the assay ≤125 SARS-CoV-2 digital copies/mL viral transport 204 media remained well within the UK Medicines and Healthcare Products Regulatory Agency 205 (MHRA) performance specification for laboratory-based SARS-CoV-2 PCR tests 206 (Supplementary Table 4) 50 . There was a good correlation between CT values from pooled 207 screening tests and paired individual confirmatory tests (Fig. 4b), and no differences in the 208 distributions of CT values were observed between paired tests stratified by number of swabs 209 (Fig 4c). Taken together, and in agreement with previous studies 46-48 , these data strongly 210 argue that swab pooling does not result in any significant loss of PCR test sensitivity. 211 212

Characteristics of students with SARS-CoV-2 infection 213
Over the entire term, 5.2% (671/12,781) of participating students were diagnosed with 214 Table 3). After an initial 215 increase in week 2, the incidence of infection fell gradually during weeks 3-5 (Fig. 5a). A 216 second increase in week 6 followed Halloween and the announcement of the second UK 217 national lockdown, with a dramatic decline in cases over weeks 7-9 with lockdown measures 218 in place. Compared with the overall population of participating students, students diagnosed 219 with SARS-CoV-2 were more likely to be male, of white ethnicity and resident in the UK out 220 of term time, according to both single variable and multivariable models (Supplementary 221  Fig. 3a,b). The distribution of time intervals between positive 233 pooled screening tests and symptom onset for these students implies a variable 234 presymptomatic infectious period, with a mean of 3.6 days (Extended Data Fig. 3c). This is 235 consistent with estimates from other populations, but slightly longer than most (typically 1-4 236 days) 52 . This may reflect methodological differences, or the particular demographic 237 characteristics of our study population. 238

SARS-CoV-2 infection across all testing routes (Supplementary
Of the respondents to the survey, 29/140 (20.7%) remained entirely asymptomatic. An 239 additional 43/140 (30.7%) students detected by the asymptomatic screening programme 240 reported minor symptoms, excluding fever, cough and anosmia/ageusia. Since these 241 students did not meet criteria for self-isolation or symptomatic testing, they were classified 242 pragmatically with asymptomatic students. typically on or just before symptom onset, and coinciding with symptomatic testing or 256 identification of presymptomatic students 53,54 . Conversely, within the limit of the screening 257 interval, asymptomatic students may be sampled at any point in their infection. 258 Next, we assessed the number of secondary infections amongst household contacts of 259 students with SARS-CoV-2 infection (index cases). Compared with index cases identified by 260 symptomatic testing, the secondary attack rate for index cases identified by asymptomatic 261 screening was lower (6.9% vs 12.0% of household contacts, P=0.0003) (Extended Data 262 Fig. 5a). Similar to CT values, we therefore stratified secondary attack rates for index cases 263 sampled by our telephone survey, according to the presence (presymptomatic students) or 264 absence (asymptomatic students) of cardinal symptoms of COVID-19 at some point in their 265 infection. Remarkably, the secondary attack rate was much higher for presymptomatic 266 students (15.2%), similar to index cases identified by symptomatic testing (Extended Data 267 Fig. 5b). Conversely, the secondary attack rate for asymptomatic students was much lower 268 (2.3%), consistent with the differences observed in CT values, and suggesting a lower 269 infectiousness (transmission potential) of students who never develop symptoms (relative 270 risk 0.17 compared with symptomatic students, 95% C.I. 0.08-0.38). This is consistent with 271 estimates from other populations 8 . Whilst some infections amongst household contacts may 272 have been acquired outside the home, that would tend to bias towards the null (smaller 273 relative risk reduction for asymptomatic students). 274 Infectious virus is more readily recovered in the laboratory from patient samples with low CT 275 values (higher viral loads), and individuals with low CT values are therefore presumed to be 276 more infectious 55 . To test this hypothesis, we examined the relationship between the 277 secondary attack rate amongst household contacts and the CT value of the index case. As 278 predicted, there was an inverse correlation between the CT value and the secondary attack 279 rate (Extended Data Fig. 5c). Nonetheless, the magnitude of the effect was relatively 280 modest, ranging from 10.3% (CT<20, highest viral loads) to 5.5% (CT>30, lowest viral loads) 281 (Extended Data Fig. 5d). 282 283 Efficacy of asymptomatic screening 284 Combining our data on case ascertainment with the results of our telephone survey (and 285 assuming that sampling was random), we estimate that 20% (95% C.I. 16.4-22.9%) of all 286 students with confirmed COVID-19 were asymptomatic (no cardinal symptoms) (Fig. 5b). Of 287 the remaining students with confirmed COVID-19 who developed symptomatic disease, 23% 288 (95% C.I. 19.0-27.1%), were detected by the asymptomatic screening programme. A further 289 8% of these students were detected by symptomatic testing, but were already quarantined at 290 the time of symptom onset because of detection of a household contact by the asymptomatic 291 screening programme. Taken together, the asymptomatic screening programme may 292 therefore have contributed to the interruption of onward transmission from 45% of all 293 students with confirmed SARS-CoV-2 infection. 294 To measure directly the efficacy of asymptomatic screening for the detection of students with 295 presymptomatic infection using an orthogonal approach, we focussed on weeks 3-7 of the 296 programme. During this period, half the students from each testing pool were screened each 297 week, with a screening interval of 14 days for individual participating students. The risk of a 298 positive symptomatic test was smallest immediately following a negative screening test (Fig.  299 6a), and the likelihood of a positive symptomatic test 1-7 days after a negative screening test 300 was reduced by approximately 51% (52 vs 106 students) compared with days 8-14. This 301 reduction corresponds to the successful detection of presymptomatic cases. It further 302 suggests that weekly asymptomatic screening will detect approximately half of all students 303 who ultimately develop cardinal symptoms of COVID-19 whilst still presymptomatic, in 304 agreement with the presymptomatic infectious period observed in our study (mean 3.6 days). 305 Finally, to quantify the potential of different screening strategies to reduce SARS-CoV-2 306 transmission, we parameterised a discrete-timestep forward-simulation SEAIR network 307 model using key real-world metadata derived from our study (Supplementary Table 8). 308 These metadata included: distribution of household sizes (Fig. 3a); within and between 309 household secondary attack rates (Extended Data Fig. 5); proportion of asymptomatic 310 infections (Fig. 5b); and duration of presymptomatic infectious period (Extended Data Fig.  311 3c). In all simulations, we assumed that symptomatic individuals self-isolate with their 312 households from the day of symptom onset. We projected new cases over time (Fig. 6b,c), 313 and compared the effect of each screening strategies on the basic reproduction number (R0) 314 and the number of students quarantined (Fig. 6c,d). 315 Weekly screening of all students (as implemented during weeks 8-9 of the programme) 316 resulted in a 31% reduction in R0 from a median of 1.78 (95% CIs 1.37-2.23) to a median of 317 1.22 (95% CIs 0.82-1.62), with screening of half the students from each testing pool each 318 week (as implemented during weeks 3-7 of the programme) intermediate in effect (median 319 R0 1.45, 95% CIs -1.02-1.88). Similar effects on transmission were observed in all 320 sensitivity analyses (Extended Data Fig. 6), and the number of students quarantined was 321 reduced, rather than increased, by more frequent screening (Fig. 6c,d). These simulations 322 therefore support the efficacy of the asymptomatic screening programme for the control 323 SARS-CoV-2 transmission in the university. 324

325
To our knowledge, this is the most comprehensive epidemiological evaluation to date of 326 regular, large-scale asymptomatic screening for COVID-19 in a university or any other 327 defined residential setting. The duration of SARS-CoV-2 RNA shedding from the upper 328 respiratory tract is typically 2-3 weeks 56,57 . Even allowing for the possibility of false negative 329 PCR tests 58,59 , the combination of regular asymptomatic screening with readily accessible 330 symptomatic testing is therefore likely to have detected almost all cases of SARS-CoV-2 331 infection in our study population, which could then be mapped to the known household 332 structure. 333 Amongst all students with confirmed SARS-CoV-2 infection, the proportion of asymptomatic 334 students (20%) was comparable with overall estimates from other studies 7,8 , but lower than 335 predicted given the age range of study participants. This may reflect the retrospective 336 ascertainment of symptoms, including subjective loss or change to the sense of smell or 337 taste (which may be more frequent in young adults, and mild disease) 60,61 . Conversely, the 338 proportion of infected individuals who ultimately developed symptoms identified by 339 asymptomatic screening (25%) was greater than predicted, given that weekly screening of 340 all participating students was only achieved by week 8. Nonetheless, this proportion is in 341 keeping with the observed reduction in likelihood of symptomatic disease in the week after a 342 negative screening test (51%), as well as the observed presymptomatic infectious period 343 (3.6 days). 344 Consistent with the known peak in SARS-CoV-2 viral load early in infection (on or just before 345 symptom onset), CT values in presymptomatic students were low, and comparable to CT 346 values in symptomatic students 53 . In agreement with this, the observed secondary attack 347 rate amongst household contacts was significantly greater for presymptomatic and 348 symptomatic index cases, than index cases who never developed symptoms. Taken 349 together, these observations support our classification of presymptomatic and asymptomatic 350 students, and imply that the infectiousness of students who never develop symptoms is 351 significantly reduced. They are consistent with data from another recent study 62 , and suggest 352 that screening programmes should be designed to target individuals with presymptomatic 353 infection, for example by maximising testing frequency. infection, even when they are no longer infectious 65 . Whilst this is a valid concern for one-off 399 mass testing, it is not relevant for programmes based on regular, frequent screening. 400 Conversely, the high sensitivity of PCR tests ensures that the risk of false negatives is 401 minimised. This may be particularly important for the detection of breakthrough infections 402 following vaccination, when viral loads may be reduced 66  Notwithstanding the rapid turnaround of LFTs, and asymmetric profile of SARS-CoV-2 viral 419 shedding, it is therefore likely that adherence to twice weekly mass testing using home LFTs 420 will be required to match or even exceed the benefits of weekly PCR-based screening 54 . 421 The development of SARS-CoV-2 variants with mutations enhancing transmissibility and 422 enabling at least partial escape from vaccine-induced and natural immunity 72-74 , together 423 with the limited availability of vaccination on a global scale 75 , suggest that non-424 pharmaceutical interventions to control the transmission of SARS-CoV-2 will be required for 425 some time yet. Our study was conducted in the UK before the widespread circulation of 426 (Michaelmas) term 2020, screening took place over 9 weeks from October 5, 2020 to 450 December 6, 2020. Students were screened using swab pooling and two-step confirmatory 451 PCR testing (Fig. 1). Additional tests were provided by the university for all students and 452 staff with symptoms of possible COVID-19. 453 The University of Cambridge comprises 31 colleges. All undergraduate and postgraduate 454 students resident in university (college) accommodation were eligible for participation in the 455 programme, along with resident students of the Cambridge Theological Federation (weeks 456 1-9). To accord with UK government guidance on student testing prior to the Christmas 457 vacation 2020, eligibility was extended to include students resident in private 458 accommodation on November 30, 2020 (week 9). Results from these students will be 459 described in a separate report. 460 Eligible students were invited to participate by email. The programme was initially 461 announced to students on September 9, 2020 by email communication. Formal email 462 invitations were sent to all eligible students on September 26, 2020 and weekly thereafter. 463 Each email contained links to the programme website, which included the consent form, 464 student information sheet and privacy notice. Additional communications were sent via email 465 from individual colleges and promoted on social media by student representatives. 466 The screening programme was introduced in phases, to ensure that all aspects of the 467 system operated effectively and had capacity to manage greater numbers. Initially, two 468 students from each testing pool were randomly selected to contribute swabs for pooled 469 sample collection each week (weeks 1-2). Then, half of the students in each testing pool 470 were asked to contribute swabs each week (weeks 3-7, ensuring every student was tested 471 every two weeks). Then, all students in each testing pool were asked to contribute swabs 472 each week (weeks 8-9, ensuring every student was tested weekly). 473 At any point in the term, students were excluded from pooled sample collection if: 1) they 474 had tested positive for SARS-CoV-2 within the preceding 8 weeks; 2) they had symptoms of 475 possible COVID-19 or were awaiting an individual symptomatic test result; 3) they were self-476 isolating for any other reason, for example due to contact with a confirmed or suspected 477 COVID-19 case or international travel from a high prevalence area; or 4) for weeks 1-7 only, 478 they had not been selected to participate (see above). In addition to asymptomatic screening, the University of Cambridge (in partnership with 537 CUHNFT) also provided a dedicated testing service for all students and staff with cardinal 538 symptoms of COVID-19 (fever, cough, and/or anosmia/ageusia). Combined nose and throat 539 swabs were obtained by self-administration at a university testing facility in the city centre, 540 Mon-Fri during the study period. Alternative testing via NHS pathways was also available to 541 all students. individual sample). Where multiple index cases were identified in the same household on the 602 same day, they were either reported as a separate "mixed" category or the household was 603 excluded from analysis. Secondary attack rates were compared using Fisher's exact test. 604 Confidence intervals for the relative risk estimates were calculated using the Koopman 605 asymptotic score method. To assess the relationship between the CT value of each index 606 case and the probability of household contacts testing positive within 14 days, fitted 607 probabilities were obtained from a logistic regression model. 608 To calculate the presymptomatic infectious period in our study population, a positive PCR 609 test was taken as proxy for infectiousness. Data were analysed from 68 students identified 610 by asymptomatic screening, who reported cardinal symptoms of COVID-19 at some time 611 during their infection (presymptomatic students). Provided the screening interval is longer 612 than the presymptomatic infectious period, presymptomatic students will be sampled at 613 random during this period, and (on average) halfway through. We therefore calculated the 614 mean presymptomatic infectious period as twice the mean observed time interval between a 615 positive pooled screening test and symptom onset. 616 To estimate the number of students with presymptomatic SARS-CoV-2 infection detected by 617 asymptomatic screening, we reasoned that successful identification of these cases should 618 correspondingly reduce the number of students identified by symptomatic testing in the 619 period after a negative screening test, in accordance with the duration of the presymptomatic 620 infectious period. We therefore analysed data from 158 students with SARS-CoV-2 infection 621 identified by symptomatic testing during weeks 3-7 of the programme. Over this period, half 622 the students from each testing pool were screened each week, corresponding to a screening 623 interval of 14 days for individual participating students. For each student, we assessed the 624 time interval between their positive symptomatic test and their most recent negative 625 screening test. We then compared the number of students testing positive 1-7 days or 8-14 626 days after a negative screening test. We selected these periods because: 1) we observed a 627 mean presymptomatic infectious period ≤7 days in most students (mean 3.6 days); 2) we 628 Health and Social Care. AstraZeneca and Charles River Laboratories reviewed the data 714 from the study and the final manuscript before submission, but the academic authors 715 retained editorial control. Other funders had no role in the study design, data collection, data 716 analysis, data interpretation, or writing of the report. All authors had full access to all the data 717 in the study and had final responsibility for the decision to submit for publication. 718 719 Data availability 720 Weekly reports are available from the programme website 721 (https://www.cam.ac.uk/coronavirus/stay-safe-cambridge-uni/asymptomatic-covid-19-722 screening-programme), together with data from the University of Cambridge symptomatic 723 testing programme (https://www.cam.ac.uk/coronavirus/stay-safe-cambridge-uni/data-from-724 covid-19-testing-service). Reasonable requests for additional de-identified data will be 725 facilitated by the authors, provided they are in accordance with the terms of consent and 726 ethical approval, and relevant legal and regulatory requirements (such as, relating to data 727 protection and privacy).   households where multiple students were identified on the same index day. d, As for (a), but 896 data are stratified according to CT value of the index case. Where multiple students from the 897 same household were identified on the same index day, data are shown separately (multiple 898 index cases). 899 900 Extended Data Fig. 6