A cluster randomised trial of the impact of a policy of daily testing for contacts of COVID-19 cases on attendance and COVID-19 transmission in English secondary schools and colleges

Background School-based COVID-19 contacts in England are asked to self-isolate at home. However, this has led to large numbers of missed school days. Therefore, we trialled daily testing of contacts as an alternative, to investigate if it would affect transmission in schools. Methods We performed an open-label cluster randomised controlled trial in students and staff from secondary schools and further education colleges in England (ISRCTN18100261). Schools were randomised to self-isolation of COVID-19 contacts for 10 days (control) or to voluntary daily lateral flow device (LFD) testing for school contacts with LFD-negative contacts remaining at school (intervention). Household contacts were excluded from participation. Co-primary outcomes in all students and staff were symptomatic COVID-19, adjusted for community case rates, to estimate within-school transmission (non-inferiority margin: <50% relative increase), and COVID-19-related school absence. Analyses were performed on an intention to treat (ITT) basis using quasi-Poisson regression, also estimating complier average causal effects (CACE). Secondary outcomes included participation rates, PCR results in contacts and performance characteristics of LFDs vs. PCR. Findings Of 99 control and 102 intervention schools, 76 and 86 actively participated (19-April-2021 to 27-June-2021); additional national data allowed most non-participating schools to be included in the co-primary outcomes. 2432/5763 (42.4%) intervention arm contacts participated. There were 657 symptomatic PCR-confirmed infections during 7,782,537 days-at-risk (59.1/100k/week) and 740 during 8,379,749 days-at-risk (61.8/100k/week) in the control and intervention arms respectively (ITT adjusted incidence rate ratio, aIRR=0.96 [95%CI 0.75-1.22;p=0.72]) (CACE-aIRR=0.86 [0.55-1.34]). There were 55,718 COVID-related absences during 3,092,515 person-school-days (1.8%) and 48,609 during 3,305,403 person-school-days (1.5%) in the control and intervention arms (ITT-aIRR=0.80 [95%CI 0.53-1.21;p=0.29]) (CACE-aIRR 0.61 [0.30-1.23]). 14/886(1.6%) control contacts providing an asymptomatic PCR sample tested positive compared to 44/2981(1.5%) intervention contacts (adjusted odds ratio, aOR=0.73 [95%CI 0.33-1.61;p=0.44]). Rates of symptomatic infection in contacts were 44/4665 (0.9%) and 79/5955 (1.3%), respectively (aOR=1.21 [0.82-1.79;p=0.34]). Interpretation Daily contact testing of school-based contacts was non-inferior to self-isolation for control of COVID-19 transmission. COVID-19 rates in school-based contacts in both intervention and control groups were <2%. Daily contact testing is a safe alternative to home isolation following school-based exposures.


41
Background 42 School-based COVID-19 contacts in England are asked to self-isolate at home. However, this 43 has led to large numbers of missed school days. Therefore, we trialled daily testing of 44 contacts as an alternative, to investigate if it would affect transmission in schools. 45 46 Methods 47 We performed an open-label cluster randomised controlled trial in students and staff from 48 secondary schools and further education colleges in England (ISRCTN18100261). Schools 49 were randomised to self-isolation of COVID-19 contacts for 10 days (control) or to voluntary 50 daily lateral flow device (LFD) testing for school contacts with LFD-negative contacts 51 remaining at school (intervention). Household contacts were excluded from participation. 52 53 Co-primary outcomes in all students and staff were symptomatic COVID-19, adjusted for 54 community case rates, to estimate within-school transmission (non-inferiority margin: <50% 55 relative increase), and COVID-19-related school absence. Analyses were performed on an 56 intention to treat (ITT) basis using quasi-Poisson regression, also estimating complier 57 average causal effects (CACE Daily contact testing of school-based contacts was non-inferior to self-isolation for control 78 of COVID-19 transmission. COVID-19 rates in school-based contacts in both intervention and 79 control groups were <2%. Daily contact testing is a safe alternative to home isolation 80 following school-based exposures. 81 Introduction 82 Since the start of the COVID-19 pandemic, there have been four different degrees of disease 83 control in schools, ranging from no controls at one extreme, to school closure at another 84 extreme. Between these poles, different degrees of control have been applied, including 85 isolation of suspected or confirmed cases, to isolation of close contacts of cases.
[1] 86 87 With widespread availability of point of care testing for SARS-CoV-2, daily contact testing 88 (DCT) has been modelled and piloted as an alternative to compulsory unsupervised isolation 89 of contacts.[2,3,4] Within the pilots contacts could continue to attend school provided a 90 daily SARS-CoV-2 test was negative. Daily testing performed with antigen lateral flow 91 devices (LFDs) has been shown to be feasible, [5] with rapid turnaround times and relatively 92 low cost and good detection of virus. [6,7] In addition to allowing students and staff to 93 remain at school, DCT might also make regular asymptomatic testing more popular or 94 improve reporting of contacts, as it removes the social penalty of a positive case triggering 95 isolation in contacts. [ A policy of routine self-isolation of contacts assumes this reduces the risk of onward 100 transmission in schools. In practice its impact is unknown; adherence to isolation is 101 incomplete,[10] and the number of isolation-days required to prevent an onward 102 transmission has not been calculated. Evidence is lacking that the benefit of the policy 103 outweighs the clear social [11,12] and educational [13,14,15] disadvantages. Recent 104 observational data from national English contact-tracing suggests that transmission 105 following a contact event in secondary schools is infrequent, and occurs in <3% of 106 educational contacts in teenagers. [16] 107 108 We undertook a cluster randomised controlled trial of DCT in students and staff at English 109 secondary schools and colleges. We aimed to determine if DCT increases school attendance 110 and to assess the impact of DCT on SARS-CoV-2 transmission. 111

112
Study design and participants 113 We conducted an open-label, cluster-randomised controlled trial to assess the effectiveness 114 of offering daily testing of contacts with cases of COVID-19. The study took place in 115 secondary schools and further education colleges in England. Schools and colleges 116 (hereafter collectively referred to as schools) were eligible to participate if willing to follow 117 the trial procedures and able to operate assisted testing on site. A representative of the 118 institution provided consent electronically. Participating schools were funded for a single 119 study worker located in the school. Participation in study procedures by student and staff 120 contacts was voluntary for individuals and those who agreed provided consent by written or 121 electronic completion of a consent form. Parents or guardians provided consent for 122 participants <16 years old and for those who were otherwise unable to give consent 123 themselves. The study protocol was reviewed and ethical approved granted by Public Health 124 England's Research Ethics and Governance Group (ref R&D 434). The study was done in 125 accordance with the Declaration of Helsinki and national legislation. The trial is registered as 126 ISRCTN18100261. A nested qualitative process study of acceptability and feasibility for 127 students, parents and staff will be reported separately. 128 129

130
Schools were randomly assigned 1:1 to either a policy of offering contacts daily testing over 131 7 days to allow continued school attendance (intervention arm) or to follow usual policy of 132 isolation of contacts for 10 days (control arm). Stratification was used to ensure schools 133 representative of those in England were balanced between study arms ( If a student or staff member had a positive LFD or PCR, close contacts ("contacts") were 145 identified by schools using national guidelines (see supplement). Those with close contact 146 with a case in the two days prior to symptom onset (or prior to positive test if 147 asymptomatic) were required to self-isolate for 10 days.
[18] 148 149 At schools in the intervention arm, close contacts were offered DCT as an alternative to self-150 isolation, provided the contact with was school-based (i.e. a staff member or student), the 151 contact did not have indicator symptoms of COVID-19 and they were able to attend for on-152 site testing at the school. Contacts were not eligible for DCT if they had a household 153 member who was isolating due to testing positive for COVID-19. Contacts who did not 154 consent to DCT were required to self-isolate for 10 days. 155 156 Participants who agreed to DCT swabbed their own anterior nose; swabs were tested by 157 school staff using a SARS-CoV-2 antigen LFD (Orient Gene).
[19] Participants who tested 158 negative were informed and were released from isolation that day to attend education, but 159 were asked to self-isolate after school and on non-testing days (weekends/holidays). Those 160 with 5 negative tests over ≥7 days were released from self-isolation, allowing for no testing 161 at weekends. Where a close contact tested positive, they were instructed to self-isolate 162 along with their household, their contacts were identified, and the process repeated for 163 these contacts. 164 165

Data collection
166 Schools provided a list of all students and staff, including personal identifiers and 167 demographics. For randomised schools that stopped active participation prior to providing 168 these details, a list of students was obtained from the UK Government Department for 169 Education (DfE Results of routine SARS-CoV-2 tests performed outside of the study in staff and students 185 were obtained from national public health data ("NHS Test and Trace"). Dedicated study 186 PCR testing was also undertaken in consenting contacts in both study arms on day 2 and day 187 7 of the testing/isolation period. In addition, study PCRs were obtained from all LFD/PCR 188 positive individuals for later analysis (see supplement).

190
Outcomes 191 The co-primary outcomes were (i) the number COVID-19-related absences from school 192 amongst those otherwise eligible to be in school and (ii) the extent of in-school  transmission. The latter was estimated from rates of symptomatic PCR-positive infections 194 recorded by NHS Test and Trace, after controlling for community case rates. Both these end 195 points could be assessed using study data for actively participating schools, but also using 196 national administrative data on student attendance and student and staff lists for non-197 participating randomised schools. Rates of symptomatic PCR-positive community tests were 198 compared as the incidence of these tests was not expected to be impacted by the study 199 intervention, whereas more intensive sampling of asymptomatic contacts in the 200 intervention arm may have detected more asymptomatic infection. 201 202 Secondary outcomes reported include DCT participation rates in the intervention arm, the 203 proportion of asymptomatic research PCR tests and symptomatic routine PCR tests in 204 contacts that were positive, and the performance characteristics of LFD vs. PCR testing in 205 participants in the intervention arm tested on the same day. 206 207

208
Rates of COVID-related absence were compared on an intention to treat (ITT) basis using 209 quasi-Poisson regression, adjusting for randomisation strata groups and participant type 210 (student/staff) and accounting for repeated measurements from the same school over time 211 (see supplement for details of this and following analyses). 212 213 We compared the incidence of symptomatic PCR-positive SARS-CoV-2 infection between 214 arms on an ITT basis using quasi-Poisson regression, adjusting for randomisation strata 215 groups, participant type and community SARS-CoV-2 case counts at the lower tier local 216 authority level (LTLA) in the prior week. 217 6 218 To account for incomplete participation in DCT, we present complier average causal effects 219 (CACE) estimates for both primary outcomes, which estimate the impact of the intervention 220 amongst those actively participating. 221 222 We report uptake of LFD testing for intervention arm participants, on a per day and per 223 participant basis. We used logistic regression to investigate factors associated with per 224 individual participation rates, including the randomisation stratification groups, participant 225 type, age, sex, and ethnicity. 226 227 The proportion of close contacts testing positive on an asymptomatic research PCR test or 228 symptomatic community PCR test was compared between study arms using logistic 229 regression. Given there were relatively few events, adjustment was made only for 230 randomisation strata groups and local case counts in the previous week. 231 232 We compared the performance of LFD to PCR testing in participants tested by both methods 233 on the same day, regarding PCR testing as the reference standard. 234 235 Sample size and power 236 The challenge with setting a non-inferiority margin for transmission events is that the 237 meaning of a non-inferiority margin is highly dependent on the control group event rate, 238 and it was not possible to determine the transmission event rate in the control group before 239 the start of the trial and it is subject to on-going change in any case. However, it was 240 considered at the time of writing the study protocol that an upper bound of the confidence 241 interval of a relative increase in transmission of up to 50% would be acceptable. Given the 242 uncertainties in the absolute rates of transmission events in each arm, we powered the trial 243 to detect a difference in school attendance (details in supplement).

245
Role of the funding source

246
The UK Government Department of Health and Social Care sponsored the trial and was 247 involved in study design and matching of NHS Test and Trace data with study records, data 248 curation and interim monitoring. Otherwise, the study sponsor had no role in data analysis 249 and interpretation or writing of the report. 250 251

252
201 schools were randomised (Table S1) and started participating in the study between 19-253 April-2021 and 10-May-2021 and continued until 27-June-2021; 76/99(77%) control and 254 86/102(84%) intervention schools actively participated in the study, returning student/staff 255 lists and attendance data ( Figure 1). The remaining 39 stopped active participation, between 256 randomisation and the study starting (of those providing reasons: 20 stated resource 257 constraints, 3 intervention schools cited concerns about the protocol, 2 control schools did 258 not wish to be in the control arm, 1 intervention school on local authority public health 259 advice). 260 261 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 25, 2021. was sent home to isolate, following either school or public health agency intervention 287 ( Figure 2A). These participation pauses occurred at 14 schools, 5 due to school capacity 288 issues, 6 due to school or public health agency concern about Delta variant, and 3 due to 289 public health concern about cases in the school as a result of transmission in the 290 community. No pause was instituted because of perceived excess transmission attributed to 291 the intervention. 292 293 Per day DCT participation was highest at the start of the study and lowest in the week prior 294 to the "half-term" holiday (31-May-2021 to 04-June-2021) when participation fell, 295 predominately due to school-wide participation pauses (Figure 2A,2B). 296 297 Using reporting of ≥3 LFD results or a positive LFD result to summarise participation per 298 contact rather than per day, 2432/5763(42.4%) contacts participated, with differing rates by 299 school ( Figure 2C). The median(IQR) participation across the 59 schools was 63%(40-79%). 300 Staff were more likely to participate than students (adjusted OR, aOR=2.67;95%CI 1.35-301 5.27;p=0.005). Participants identifying as Chinese ethnicity were more likely and those 302 identifying as "Other" ethnicity were less likely to participate compared with those 303 identifying as white. Amongst schools with ≤17% of students receiving free school meals, 304 participation rates were higher in schools with students aged 11-16 years compared to 11-305 18 years (Table 3). 306 307 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 25, 2021. rates in students were approximately linearly related to local case counts, plateauing as 367 community incidence rose ( Figure S2); estimates were similar with varying plausible lags 368 between community case counts and student and staff infections ( We also compared the proportion of contacts with a symptomatic PCR-positive test, which 394 included those initially testing positive while asymptomatic above who went on to have a 395 symptomatic test. This analysis is contingent on schools reporting contacts, with several 396 control arm schools with higher incidence not actively participating and reporting contacts 397 ( Figure S3). In the control arm 44/4665(0.9%) of contacts tested PCR-positive within 10 398 days, compared to 79/5955(1.3%) in the intervention arm. Adjusting for randomisation 399 strata groups and community case counts, there was no evidence that the proportion of 400 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 25, 2021.  Figure S4). 411

412
Daily LFD testing of school-based COVID-19 contacts was trialled as a voluntary alternative 413 to 10 days of self-isolation. Although DCT avoids students and staff missing school days 414 while isolating, at the conception of the trial there was uncertainty whether it would 415 substantially increase SARS-CoV-2 transmission, e.g. via infections missed by LFD testing.
[2] 416 The trial provides evidence this was not the case. 417 418 We investigated the incidence of symptomatic infection as an unbiased outcome measure 419 that could be ascertained across nearly all schools, as national public health policy was that 420 all symptomatic children, whether or not they had a LFD test, should obtain a PCR test for 421 SARS-CoV-2. As the intervention was not expected to impact the relative incidence of 422 asymptomatic versus symptomatic infection this measure should also indicate the impact on 423 all infections. Based on a non-inferiority margin of ensuring symptomatic infection did not 424 increase by >50%, we show allowing student and staff contacts to remain in school after a 425 negative lateral flow test was non-inferior to routine isolation. On an ITT basis, i.e. using 426 lateral flow testing at participation rates seen in the trial, using data for students from 427 197/201 schools and staff data from 161/201 schools, we can be 97.5% confident that any 428 increase in the rate of symptomatic infection did not exceed 22% more than seen in the 429 control arm. Were all those eligible to participate in daily lateral flow testing to do so, then, 430 based on a CACE model, we can be 97.5% confident that any increase does not exceed 34%. 431 In both analyses the point estimate favours a slight to modest reduction in incidence with 432 the intervention. 433 434 The range of absolute changes in symptomatic infection rates potentially seen with the 435 intervention, depends on prevailing incidence. At the average incidence in the control arm 436 during the study (0.06% students/week), the range of uncertainty in the impact of the 437 intervention is equivalent to 1.2 fewer to 0.9 more infections/1000-student-school/month, 438 or 3.6 fewer to 2.7 more at the highest weekly rate seen (0.18% students/week). 439 Throughout the study, cases in both arms remained well below the >1% level seen in 2020 440 when schools remained open.
[21] Staff had lower rates of infection than students. There 441 was no evidence of a difference in the effect of the intervention for students and staff. 442 443 In both control and intervention arms it was uncommon for school-based contacts to 444 become infected with no evidence of a difference in asymptomatic or symptomatic 445 infection: 1.6% and 1.5% of students and staff participating in research PCRs tested positive 446 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.
while asymptomatic, and 0.9% and 1.3% tested positive in symptomatic testing for the 447 control and intervention arms respectively. These figures are comparable to the estimates 448 for school age children from national contact-tracing data. [16] Therefore, given precautions 449 in place in schools during the trial (routine mask use was discontinued part way through the 450 trial on 17-May-2021, but other precautions were maintained), the overall risks to students 451 and staff following exposure to a contact at school are low. Indeed, whether the extent of 452 transmission is sufficient to make any contact testing necessary and cost-effective will 453 require careful discussion and may vary with changes in incidence, virus transmissibility or 454 the prevalence of vaccine evasive strains. Participation in research PCR testing in control 455 schools was lower than in the intervention schools, in part because participation in DCT 456 facilitated intervention arm PCR-testing. It is unclear whether this caused any bias in the 457 results for the research PCR tests, however we also found no difference in symptomatic 458 infection rates in contacts. 459 460 We did not clearly demonstrate superiority of the intervention in terms of avoiding student 461 and staff absences from school related to COVID. This possibly reflects that the trial was 462 relatively underpowered given the large extent of variation in absence rates over time and 463 between schools, requiring overdispersion to be accounted for in the regression models 464 fitted. Pooling the data on a per school basis, in an ITT analysis, our point estimate showed 465 a 20% decrease in COVID-related absences, but with a broad range of uncertainty (95%CI 466 0.62-1.03), similarly in the CACE analysis amongst those who participated the point estimate 467 was a 38% reduction, but with broader uncertainty (95%CI 0.29-1.33). 468 469 That reductions in COVID-related absences were not greater reflects firstly that not all those 470 eligible chose to participate, and secondly that not all absences were amenable to the 471 intervention, e.g. those who with household contacts were ineligible. However, despite the 472 lack of statistical evidence from the trial, in the absence of increased transmission, it is 473 reasonable to assume that a policy of allowing students and staff to remain in school, would 474 indeed lead to increased attendance, but this may be more limited than might be initially 475 anticipated. 476 477 Overall participation rates in LFD testing in intervention arm contacts were 42% of a per 478 person basis with marked variation between schools (range 0-100%). Although contacts at 479 government-funded schools with students 11-16 years old with a low percentage of free 480 school meals were most likely to participate, other school types were similar. Staff were 481 more likely to participate than students. A qualitative analysis of interviews with 482 participants to understand why some participated and others did not will be presented 483 separately. Additionally, at some stages, schools paused the intervention either because of 484 capacity limitation or because public health officials were concerned about the spread of 485 the Delta lineage or rising transmission in the community. No local public health teams 486 reported concern that transmission was observed to increase because of this study. 487 488 Previous estimates for the performance of antigen LFDs compared to PCR testing have 489 varied markedly.[6,22] Here we estimate the overall sensitivity of school-based LFD testing 490 in largely asymptomatic individuals as 53%, which falls within the range of previously 491 reported rates. It is worth noting the findings on transmission in this study are in the context 492 of this level of performance. Specificity was 99.93%. As LFD performance varies by viral 493 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021. The study has several limitations. Schools and colleges, despite provision of dedicated 499 resources, were not always able to participate due to competing pressures, and it is also 500 likely as a result that data capture was imperfect, e.g. it is possible that not all PCR-positive 501 cases were reported to schools, and not all contacts may have been documented for all 502 index cases. However, how the primary outcome measures are assessed is robust to this. 503 We used the incidence of symptomatically driven testing as a primary endpoint as this was 504 least likely to be affected by the two testing strategies; in fact, there was little difference in 505 the incidence of all community PCR tests between the study arms. Relying on linkage to Test 506 and Trace data is also a potential weakness, as it depended on imperfectly recorded 507 identifiers, however this would not be expected to differ between study arms. Furthermore, 508 using incidence data means we do not directly measure within school transmission, rather 509 we estimate it by controlling for the rate of community infections, as a proxy for the extent 510 of introductions into the school. The trial was conducted during periods of low to moderate 511 COVID-19 incidence. We therefore did not estimate the impact of DCT in high incidence 512 settings. In the last two weeks of the study, the community rate of infections rose making 513 the DCT protocol unwieldy for some schools, given the space and staff required to perform 514 testing. 515 516 Future work includes whole genome sequencing of positive samples from school members 517 and from the community, which may help analyse the transmission networks in schools, 518 including during periods of higher incidence in a manner successfully achieved for SARS-519 CoV-2[24,25] and a number of healthcare-associated pathogens. [26,27] This study includes 520 staff and students from secondary schools and colleges of further education but most of the 521 participants were students aged 11-18 years. Therefore, it is unclear the extent to which it 522 can be generalised to other settings, and other context-specific studies are required. 523 524 Overall, this study shows that in secondary school and college of further education students 525 and staff infection of following contact with a COVID-19 case at school occurs in less than 526 2%. There was no evidence that switching from isolation at home to daily contact testing, at 527 least in the settings of the schools studied, increased rates of symptomatic COVID in 528 students and staff. Daily contact testing is a safe alternative to home isolation following 529 school-based exposures and should be considered an alternative to routine isolation of 530 close contacts following school-based exposures. 531 532 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.  Table 3. Associations with participation in lateral flow testing in 5763 contacts in intervention arm schools where the 10 days following the 700 positive test in the index case included ≥1 school day. Participant age is omitted from the multivariable model due to collinearity with 701 participant type. Results from logistic regression, adjusting confidence intervals to account for repeated measurements from the same school. 702 1 n (%); Median (IQR); 2 OR = Odds Ratio, CI = Confidence Interval. Note week 7 is the school "half-term" holiday, when school-based lateral 703 flow testing was not undertaken. Note participation in the final week of the study appears lower than in Figure 2, as participation is 704 summarised as completion of ≥3 LFDs, and contacts in the final week may not have completed testing before the end of the study. 705  availability of daily school attendance data for students and staff aggregated at school level. 717 The latter depends on provision of student and staff lists to enable matching of identifiers 718 with NHS Test and Trace national community testing data. DfE, UK Government Department 719 for Education. School participation was defined based on submission of student/staff lists 720 and attendance data for at least part of the study. 721 722 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.   is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.

B
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021. ; https://doi.org/10.1101/2021.07.23.21260992 doi: medRxiv preprint i A cluster randomised trial of the impact of daily testing for contacts of COVID-19 cases on education and COVID-19 transmission in English secondary schools and colleges: Supplementary material

Supplementary methods
Randomisation Schools were randomly assigned 1:1 to either a policy of offering contacts daily testing over 7 days to allow continued school attendance (intervention arm) or to follow usual policy of isolation of contacts for 10 days (control arm). Randomisation was performed in blocks of 2 and stratified using nine strata to ensure a sample representative of schools and colleges in England. Stratification was performed according to school type, size, presence of a sixth form, presence of residential students and proportion of students eligible for free school meals (as a marker of social deprivation), the nine strata are listed in Table 1. Randomisation was performed by a trial team member in Stata (version 16).
10 schools participated in a non-randomised pilot of the study protocol in March 2021. During the main study they continued to follow the intervention procedures, but do not contribute to the analysis of randomised outcomes.

Procedures
Forms of close contact applicable to schools as defined in national guidelines were, face to face contact (within 1 metre for any length of time) or skin to skin contact or someone the case coughed on; or within 1 metre for ≥1 minute; or within 1-2 metres for >15 minutes. Any person who met the definition of being in close contact with a case in the two days prior to symptom onset (or prior to positive test if asymptomatic) was required to selfisolate for 10 days.
In the intervention group, daily contact testing was performed with a lateral flow device on arrival at school or college each morning. Day 1 of testing began the day after a case was identified. Where there was a delay to the start of testing, contacts could opt to start DCT within 3 days of a case being identified. Testing was done over 7 consecutive days, and a minimum of 5 test was required (allowing for no testing on weekends). Five negative tests, including one on or after the 7 th day of testing was required to complete DCT, at which point contacts were released from self-isolation. Contacts who opted to stop testing during the process reverted to self-isolation for 10 days. Contacts who tested positive during DCT were instructed to self-isolate for 10 days from the positive test.

Data collection
Data were collected using a web-based data capture system (Voyager, IQVIA).
Schools reported in aggregate the number of staff and students present on each school day, and numbers absent for COVID-19-related reasons and separately numbers absent for other reasons. Attendance data for individual participating students and staff members were not collected.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021. ; https://doi.org/10.1101/2021.07.23.21260992 doi: medRxiv preprint ii PCR testing Results of routine community tests performed outside of the study for SARS-CoV-2 in staff and students were obtained from national public health data ("NHS Test and Trace"). Matching of results to study participant identifiers was undertaken by the UK Government Department of Health and Social Care (DHSC). Results were matched based on an exact match of (surname, date of birth, home postcode) OR (first name, surname, date of birth, testing centre and school lower-tier local authority [LTLA]) OR (first name, surname, year of birth, home postcode). An iterative approach with manual review of school-reported and Test and Trace cases was used to define the matching rules. Test and Trace results recorded whether the individual was symptomatic or not prior to testing.
Routine community-based testing was undertaken by a network of accredited diagnostic laboratories, with high-throughput national "Lighthouse laboratories" undertaking testing with the ThermoFisher TaqPath assay undertaking the most tests.
Dedicated study PCR testing was also undertaken. All individuals who tested positive for SARS-CoV-2 by either LFD or PCR for SARS-CoV-2 infection who consented were asked to provide a swab of nose and throat for PCR testing. Additionally, all close contacts in either study arm who consented to participate were asked to provide a swab of nose and throat for PCR testing on day 2 and day 7 of their testing/isolation period. For contacts undergoing DCT the test was done on the nearest school day.
Swabs for PCR testing were sent by courier or mail to a central laboratory and forwarded for testing at an accredited clinical microbiology laboratory (Oxford University Hospitals NHS Foundation Trust). Samples were stored at -20°C for up to 2 weeks. RNA extraction was performed using the KingFisher (Thermo Fisher) automated extraction system. SARS-CoV-2 PCR was performed using the Thermo Fisher TaqPath COVID-19 kit. Detection of both N and orf1ab targets was required for a positive result, with the cycle threshold (Ct) for one target ≤32 and the other ≤33. Samples with no detected viral targets were considered negative and all other samples indeterminate.

Statistical analysis
The rate of COVID-19-related absences from school amongst those otherwise eligible to be in school (i.e. not absent for another reason) were compared between the study arms. Students and staff were considered at risk of a COVID-related absence, while not absent for other reasons, on school days following enrolment of the school into the study from 19-April-2021 onwards until 27-June-2021. Weekend days, national holidays, the school halfterm holiday (31-May-2021 to 04-June-2021), and individual school non-school days were excluded.
Total rates of COVID-19-related absence per school were compared on an intention to treat (ITT) basis, testing for superiority of the intervention, for all schools with available data irrespective of whether they participated after randomisation or not. Models were fitted using quasi-Poisson regression to account for overdispersion. Pre-specified adjustment was made for 6 study stratification groups (Government-funded, 11-16y, free school meals ≤17%; Government-funded, 11-18y, free school meals ≤17%; Government-funded, 11-16y, . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021. ; https://doi.org/10.1101/2021.07.23.21260992 doi: medRxiv preprint iii free school meals >17%; Government-funded, 11-18y, free school meals >17%; Independent schools; Other), combining several of the smaller original randomisation strata given small numbers in these strata, and for participant type (student or staff). Repeated daily measurements from the same school were accounted for using robust standard errors with clustering by school. We also present results combining data from each school during the study without robust standard errors.
We compared the incidence of symptomatic PCR-positive SARS-CoV-2 infection between arms using quasi-Poisson regression. Individuals were considered at risk of an infection on all calendar days (school days and non-school days) from the later of the date of the start of the study (19-April-2021) or enrolment of their school, up until the end of the last week of the study (27-June-2021). Weekly incidence data were used, adjusting for the 6 study stratification groups above, participant type, and community PCR-positive case rates in the local population in the prior week. Adjustment for community case rates was designed to allow the analysis to assess any excess in cases in the intervention arm over and above that expected from importation of community-acquired cases into the school. Sensitivity analyses examined the impact of using differing lag periods between community and school case counts of 1 and 4 weeks prior, and without adjustment for community case counts. Community case counts were obtained from nationally reported data, publicly available on the gov.uk website, at the LTLA level, using data from the LTLA within in which the school was situated. Repeated measurements from the same school were accounted for using robust standard errors with clustering by school. The relationship between community case rates in the prior week and the outcome was modelled using natural cubic splines to allow for non-linearity, up to 5 default-placed knots were allowed, choosing the final number of knots based on model fit according to the Bayesian Information Criterion. To avoid undue influence of outliers community case rates were truncated at the 2.5 th and 97.5 th centiles.
No interaction terms were included in either of the co-primary outcome models, however we tested for heterogeneity in the effect of the intervention on students and staff in separate models. We also present subgroup analyses in students and staff separately.
To account for incomplete participation in DCT, we present complier average causal effects (CACE) estimates for both primary outcomes, estimated using the randomisation arm as an instrumental variable and a two-stage regression approach. In this approach, we first fit two models: 1) the relationship between study arm and measured compliance, adjusting for the covariates above; 2) the relationship between measured compliance and the outcome, adjusting for covariates, but not study arm. These estimates are combined to estimate the impact of the intervention amongst those actively participating.
For the COVID related absence analysis compliance was calculated per school and participant type, as the sum over all study school days of individuals eligible for DCT returning a test result or already having completed follow up each day, divided by the sum of individuals eligible for DCT. For the symptomatic infection outcome, compliance was calculated per school, participant type and week, as other covariates varied by week. For schools in the control arm and those in the intervention arm not actively participating compliance was set to zero. For participating schools without any eligible contacts in a given week the median compliance per schools was used, and where no eligible contacts were . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021. ; https://doi.org/10.1101/2021.07.23.21260992 doi: medRxiv preprint iv identified during the study the median compliance per randomisation stratification group. Sensitivity analyses were performed using the 25 th and 75 th centiles for imputation instead of the median value.
For the symptomatic infection outcome, to account for repeated measurements by school, confidence intervals for CACE estimates were generated from 1000 bootstrap samples, using bias-corrected and accelerated bootstrap intervals, and sampling based on school clusters.
We report uptake of LFD testing for intervention arm participants, on a per day and per participant basis. For the per day analysis, we identified all school days between a contact being identified and day 10 following their first exposure to the index case. Participation was defined as either return of a test result or where testing had been completed, i.e. ≥5 test results were already available or a prior positive test had occurred. For the per participant analysis, we pre-defined participation as a school recording ≥3 negative or ≥1 positive LFD test result for the participant. We used logistic regression to investigate factors associated with per individual participation rates, including the randomisation stratification groups, participant type, age, sex, and ethnicity. We used variance adjustment as above to allow for clustering of results by school.
The proportion of close contacts testing positive on an asymptomatic research PCR test was compared between study arms using logistic regression, given there were relatively few events, adjustment was made only for randomisation strata groups and local case counts in the previous week (at the LTLA level as above). As individuals could be contacts on multiple occasions, including simultaneously with different index cases, we deduplicated our data to present one result per non-overlapping contact episode, defining each episode as the 10 days from the index case. We also use symptomatic community-based testing data from NHS Test and Trace to present the proportion of contact episodes associated with a symptomatic PCR positive result in the 10 days following the diagnosis of the index case. For both asymptomatic and symptomatic analyses we only consider contacts at risk prior to their first positive result in the study, as any subsequent result within the 70 days of the study could represent residual RNA from the first infection. We account for clustering of results by school as above.
We compared the performance of LFD to PCR testing in participants tested by both methods on the same day, regarding PCR testing as the reference standard. Additional data from a pilot phase of the study, involving 10 non-randomised intervention schools was included in this analysis only.

Sample size and power
We powered to trial to detect a difference in school attendance. We assumed of 100 similarly-sized schools randomised to each arm, ~50% would participate. In the control arm we assume 30% participation in national twice weekly LFD testing outside the trial, such . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021. ; https://doi.org/10.1101/2021.07.23.21260992 doi: medRxiv preprint v that index cases would be identified at a rate of 1 per school per month, with each associated with 50 contacts. Hence with an isolation period of 10 days, 510 isolation days per school per month would occur in the control arm. For the intervention arm, we assume the intervention would increase uptake of routine LFD testing two-fold to 60% with the barrier of potential isolation removed. Therefore, the expected rate of index case detection from routine testing doubles to 2 per month. We assume that 70% of contacts will participate in DCT, such that only 15 per index case self -solate, with an additional 2 per index case self-isolating following a positive LFD in DCT, but without further contacts outside of the existing contacts. This results in an expected 170 missed school days per index case or 360 per month. Based on these assumptions we estimated that 58 participating schools in each arm provides 80% power (two-sided alpha=0.05) to detect a difference in attendance between the study arms. However, the number of pupils varied substantially by school and therefore the original analysis based on the sample size calculation (which assumed approximately equal school sizes) was not appropriate. Further, there was substantial evidence of over-dispersion which we also had to account for in the analysis.

Analysis Group
Bernadette Young, David Eyre, Tim Peto, (thanks to Sarah Walker for statistical advice) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.  Table S1. Participating schools and randomisation strata.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.  Table S4. Co-primary outcome, sensitivity analysis: rate of COVID-related absence in students and staff and compliance imputation strategy. Results of quasipoisson regression models using data accounting randomisation strata group, participant type and for clustering by school using variance adjustment are shown. IRR, Incidence Rate Ratio, CI = Confidence Interval, CACE, complier average causal effect.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.  Table S7. Secondary outcome: rate of all-cause absence in students and staff. Results of a quasipoisson regression model using data accounting for clustering by school using variance adjustment. 1 IRR = Incidence Rate Ratio, CI = Confidence Interval. ITT, intention to treat; CACE, complier average causal effect. Overall, all-cause absences were considerably higher than COVID-related absences, 19.7% in students and 8.3% in staff, in part because students in two school years were granted study leave during weeks 7-10 of the study, and only a minority of several large further education college students were expected to attend each day.  Table S8. School reported index cases and national community-based testing results reconciliation. Index cases were reported to schools by students and staff and recorded by schools in study records. Details of students and staff at schools allowed matching to national testing data (NHS Test and Trace).
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.  Table S9. Co-primary outcome: incidence of symptomatic PCR positive infection in students and staff. Results of a quasipoisson regression model accounting for clustering by school using variance adjustment. In the adjusted analysis, adjustment is also made for community case counts in the prior week using a 4 knot spline (default placed knots, with number up to five chosen on the basis of BIC in a Poisson regression model) (see Figure S2). . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.  Table S11. Co-primary outcome, sensitivity analysis: incidence of symptomatic PCR positive infection in students and staff and compliance imputation strategy. Results are shown of quasipoisson regression models using data adjusting randomisation strata group, participant type, and community case rates in the prior week, with allowance for clustering by school using variance adjustment. IRR, Incidence Rate Ratio, CI = Confidence Interval, CACE, complier average causal effect.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 25, 2021.  Table S12. Secondary outcome: incidence of any PCR positive infection in students and staff. Results of a quasipoisson regression model accounting for clustering by school using variance adjustment. In the adjusted analysis, adjustment is also made for community case counts in the prior week using a 4 knot spline (default placed knots, with number up to five chosen on the basis of BIC in a Poisson regression model) (see Figure S2).  Table S13. Co-primary outcome, subgroup: incidence of symptomatic PCR positive infection in students. Results of a quasipoisson regression model accounting for clustering by school using variance adjustment. In the adjusted analysis, adjustment is also made for community case counts in the prior week using a 4 knot spline (default placed knots, with number up to five chosen on the basis of BIC in a Poisson regression model) (see Figure S2).  Table S14. Co-primary outcome, subgroup: incidence of symptomatic PCR positive infection in staff. Results of a quasipoisson regression model accounting for clustering by school using variance adjustment. In the adjusted analysis, adjustment is also made for community case counts in the prior week using a 4 knot spline (default placed knots, with number up to five chosen on the basis of BIC in a Poisson regression model) (see Figure S2).  Table S15. Secondary outcome, subgroup: incidence of any PCR positive infection in students. Results of a quasipoisson regression model accounting for clustering by school using variance adjustment. In the adjusted analysis, adjustment is also made for community case counts in the prior week using a 4 knot spline (default placed knots, with number up to five chosen on the basis of BIC in a Poisson regression model) (see Figure S2).  Table S16. Secondary outcome, subgroup: incidence of any PCR positive infection in staff. Results of a quasipoisson regression model accounting for clustering by school using variance adjustment. In the adjusted analysis, adjustment is also made for community case counts in the prior week using a 4 knot spline (default placed knots, with number up to five chosen on the basis of BIC in a Poisson regression model) (see Figure S2). Table S19. Secondary outcome: performance of lateral flow device (LFD) testing in close contacts compared with paired polymerase chain (PCR) testing. Sensitivity, specificity, positive predictive and negative predictive values given, with 95% confidence intervals calculated by exact binomial method.
xxxi Supplementary figures Figure S1. Student (panel A) and staff (panel B) attendance data completeness by study day. Individuals were considered at risk of a COVID-related absence on school days following enrolment of the school into the study from 19-April-2021 onwards up to 25-June-2021. National holidays, the school "half-term" holiday (31-May-2021 to 04-June-2021), and individual school non-school days were excluded. The total height of the bar represents the number of randomised schools entered into the study on that day excluding any schools with a non-school day. Although 4 schools continued throughout the half-term holiday, this period was removed from the analysis for all schools. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 25, 2021. ; https://doi.org/10.1101/2021.07.23.21260992 doi: medRxiv preprint xxxii Figure S2. Relationship between community case rates and weekly incidence of PCRconfirmed infections in students. Model, with a 4 knot spline (with default positioned knots) adjusted for strata group and study arm, shown for Government-funded, 11-18y, free school meals ≤17% schools in the control arm. Weekly community case rate per 100k population Weekly student infections per 100k students . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 25, 2021. ; https://doi.org/10.1101/2021.07.23.21260992 doi: medRxiv preprint xxxiii Figure S3. Incidence of symptomatic PCR-confirmed infection by study arm and school. Schools actively participating in the study and therefore potentially reporting contacts are shown in blue. Schools not actively participating, for which, student lists where obtained from the Department for Education (DfE) are shown in orange. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 25, 2021. ; https://doi.org/10.1101/2021.07.23.21260992 doi: medRxiv preprint