key: cord-0983271-bn44i96z
authors: Sparks, S. R.; Aspinall, W. P.; Brooks-Pollock, E.; Danon, L.; Cooke, R.; Barclay, J.; Scarrow, J.
title: A novel approach for evaluating contact patterns and risk mitigation strategies for COVID-19 in English Primary Schools with application of Structured Expert Judgement
date: 2020-08-14
journal: nan
DOI: 10.1101/2020.08.13.20170068
sha: 75fce0354998ed16f9a2ff5ea54addd00622c737
doc_id: 983271
cord_uid: bn44i96z

Background Contact patterns are the drivers of close-contacts infections, such as COVID-19. In an effort to control COVID-19 transmission in the UK, schools were closed on 23 March 2020. With social distancing in place, Primary Schools were partially re-opened on 1 June 2020, with plans to fully re-open in September 2020. The impact of social distancing and risk mitigation measures on childrens contact patterns is not known. Methods We conducted a structured expert elicitation of a sample of Primary Headteachers to quantify contact patterns within schools in pre-COVID-19 times and how these patterns were expected to change upon re-opening. Point estimates with uncertainty were determined by a formal performance-based algorithm. Additionally, we surveyed school Headteachers about risk mitigation strategies and their anticipated effectiveness. Results Expert elicitation provides estimates of contact patterns that are consistent with contact surveys. We report mean number of contacts per day for four cohorts within schools along with a range at 90% confidence for the variations of contacts among individuals. Prior to lockdown, we estimate that, mean numbers per day, younger children (Reception and Year 1) made 15 contacts [range 8..35] within school, older children (Year 6) 18 contacts [range 5..55], teaching staff 25 contacts [range 4..55) and non-classroom staff 11 contacts [range 2..27]. Compared to pre-COVID times, after schools re-opened the mean number of contacts were reduced by about 53% for young children, about 62% for older children, about 60% for classroom staff and about 64% for other staff. Contacts between teaching and non-teaching staff reduced by 80%, which is consistent with other independent estimates. The distributions of contacts per person are asymmetric indicating a heavy tail of individuals with high contact numbers. Conclusions We interpret the reduction in childrens contacts as a consequence of efforts to reduce mixing with interventions such as forming groups of children (bubbles) who are organized to learn together to limit contacts. Distributions of contacts for children and adults can be used to inform COVID-19 transmission modelling. Our findings suggest that while official DfE guidelines form the basis for risk mitigation in schools, individual schools have adopted their own bespoke strategies, often going beyond the guidelines.

Background 2

Contact patterns are the drivers of close-contacts infections, such as COVID-19. In an 3 effort to control COVID-19 transmission in the UK, schools were closed on 23 March 4 2020. With social distancing in place, Primary Schools were partially re-opened on 1 5

June 2020, with plans to fully re-open in September 2020. The impact of social 6

distancing and risk mitigation measures on children's contact patterns is not known. 7

We conducted a structured expert elicitation of a sample of Primary Headteachers to 9

quantify contact patterns within schools in pre-COVID-19 times and how these patterns 10 were expected to change upon re-opening. Point estimates with uncertainty were 11 determined by a formal performance-based algorithm. Additionally, we surveyed school 12

Headteachers about risk mitigation strategies and their anticipated effectiveness. 13

Expert elicitation provides estimates of contact patterns that are consistent with 15 contact surveys. We report mean number of contacts per day for four cohorts within 16 schools along with a range at 90% confidence for the variations of contacts among 17

individuals. Prior to lockdown, we estimate that, mean numbers per day, younger 18 children ( other staff. Contacts between teaching and non-teaching staff reduced by 80%, which 24

is consistent with other independent estimates. The distributions of contacts per 25

person are asymmetric indicating a heavy tail of individuals with high contact numbers.

Conclusions 28

We interpret the reduction in children's contacts as a consequence of efforts to reduce 29 mixing with interventions such as forming groups of children (bubbles) who are 30 organized to learn together to limit contacts. Distributions of contacts for children and 31

adults can be used to inform COVID-19 transmission modelling. Our findings suggest 32 that while official DfE guidelines form the basis for risk mitigation in schools, individual 33 schools have adopted their own bespoke strategies, often going beyond the guidelines. of Primary school children in England before the summer holidays. There is now an 45 expectation that schools will fully re-open in September. 46

The partial re-opening of Primary schools has been widely debated with concerns from 47 some parents, teaching unions and teacher associations about the safety of the children 48

and school staff. There was also concern about the effect on infection rate in the wider 49 community, for example by triggering a second wave. Some schools re-opened on 1 st 50

June, some have delayed their re-start while, in other cases, schools have not re-51 opened under advice of local education authorities. Data from DfE indicates on 15 th July 52 that about 88% of state-funded Primary Schools have re-opened to some extent. The 53 community response has been variable and between 1 st and 15 th June 2020 54

approximately one third of eligible children returned. The numbers have increased 55 somewhat and were 41% (year 1) and 49% (year 6) on 2 nd July. Between 18 th May and 56 31 st July 2020 there have been 247 COVID-19 related incidents in schools of which 116 57

were tested to be positive test (PHE 2020) . Primary Schools as experts. SEJ is a well-established approach to quantifying 65 parameters and their attendant uncertainties where there are no data or the data are 66 sparse or of poor quality or are highly empirical in character or have large associated 67 uncertainties (Cooke 1991; Colson & Cooke 2017) . This is the case for many 68 epidemiological parameters, including contact patterns which are used in combination 69

with the probability of transmission and the infectious period to describe the 70 reproduction number in a population. SEJ has been widely applied to risk assessment 71 and uncertainty analysis in many areas of science, engineering, the environment, 72

business and public health (Colson & Cooke 2017 were made as a consequence of these reviews. Normally experts are convened at least 111 in one or more face-to-face meetings in an SEJ. A meeting usually covers: introduction 112 to the methodology; calibration of the experts using seed questions; presentation and 113 discussion of the questions; and a time for the experts to answer the questions. A 114 meeting might last 1 or 2 days and with plenty of time for discussion on the evidence 115 that informs responses to the questions. After the data have been processed there then 116 an opportunity to discuss the results and there may be opportunities to re-elicit some 117

questions. 118

During national lockdown and with experts distributed in Primary Schools across 119

England the normal procedure was not possible. A briefing session was held by zoom 120

and most of the experts participated. The briefing session was recorded enabling those 121 who could not attend the chance to hear the proceedings. For both the seed and 122 elicitation questions Aspinall and Sparks were available to respond to queries and 123 clarifications by email. To compensate for the lack of a meeting six experts were chosen 124

for structured interviews (Longhurst, 2015) . The question protocol ensured participants 125

described the thinking behind their quantitative answers but also allowed a free 126 exploration of topics the headteachers perceived of relevance to adult and child 127

contacts The experts 187

The elicitation produced 26 complete responses from experts who had undergone 188

Classical Model (CM) calibration process. Thus, each had a personal statistical accuracy 189 score and an information score (Table 1) . These scores are the basis for ascribing a 190

relative performance weight to each participant for pooling judgements in the context 191 of enumerating uncertainty assessments for specific target/query items. Classical Model 192 non-optimised ltem weights combination solutions (i.e. all teachers given some weight 193

based on calibration scores) are listed in Table 1 1 . These weights are not the same as 194

ascribing equal weights to all participants. When combining judgements collectively, 195 each person's uncertainty distribution is weighted according as their own performance 196 score, so each contributes with some real positive weight to the overall outcome. 197

The calibration "p-value" score is calculated on the basis of a Chi-squared test on the 198 deviations of the expert's judgements over the seed questions, relative to the known 199 values used for calibration. Using Shannon's relative information statistic, the Chi-200 squared test takes the frequencies with which the expert's assessed calibration variable 201 values fall within various ranges, and compares these with the counts of actual (known) 202

item values in the same ranges; this produces relative probabilities of match per item, 203 which can be summed over all calibration items to form a measure of the expert's 204 statistical accuracy. When combined with the expert's information metric (see Cooke, 205 1991) in a product, this p-value provides the 'statistical accuracy' part of the expert's 206 performance score. In effect, a calibration p-value can be thought equivalent to the 207 probability that an expert's performance would be regarded as statistically accurate by 208 chance but, in the Classical Model, it is not used in the sense of an hypothesis test. 209

Rather, it is the formal mathematical basis for enumerating the metric for the statistical 210

accuracy of the expert's assessments and is used, with a companion information metric, 211 for scoring performance in assessing uncertainties. A high p-value indicates a close 212 correspondence between the expert's assessment values and the known values (a 213 perfect score would be p = 1); a very low p-value signals major deviations exist between 214 the assessed and actual values. The calibration metric is a 'fast' function and typically 215 changes markedly between experts in an elicitation . 216

To provide some context for the teachers' panel revealed performance scores, the 217 calibration and weighting profile of the group is very similar to that of other 218 professional expert elicitations; for example, in Figure 1 a profile is shown for a medical 219 panel with the same number of participants (n.b. some points overlap). While the 220 overall range of the teachers' statistical accuracy p-values -from highest to lowest -is 221 not atypical, in this case there are fewer individuals with very low p-values (i.e. p < 10 -6 ) 222 than in the medical panel; the latter represents the more usual case. With the 223 exception of one teacher (a very low p-value outlier not shown here), the majority of 224 the teachers' p-values are more clustered at higher p-values than those obtained for 225 the medical panel. On the other hand, the relative information scores of the teachers 226 are generally lower than those of the medical experts; this may reflect inherent 227 differences in the natures of the precision of the data types each group were 228

considering. 229

Our conclusion is that the teachers' elicitation does not exhibit any substantive 230

shortcomings compared to other cases, notwithstanding it was conducted with minimal 231

briefing and without the benefit of a plenary workshop to help focus judgements. In 232 short, the Head teachers' performances proved to be as strong, collectively, as those of 233 many other groups of experts (Colson and Cooke 2017, for other case profiles). 234

The combination Decision Maker (DM) --the combination of all experts --is statistically 235 more accurate than any individual expert. The range of relative information metric 236 scores is also typical; this is a 'slow' function that does not vary greatly from one expert 237 to another. Higher values indicate experts who provided tighter (more informative) 238 uncertainty ranges. However, there is here, and usually, an inverse relationship 239 between informativeness and statistical accuracy: experts who are too narrow or too 240 precise with their uncertainty judgements tend to 'miss the target' too frequently and 241

their performance scores are penalised as a consequence. On the other hand, the goal 242 is to identify those experts who are both informative and statistically accurate. This is 243

usually a minority of a group, and such persons are not identifiable a priori on common 244 grounds such as professional standing; the best uncertainty assessors are discovered 245 only after calibration. The DM has a low information score compared to individual 246 experts, and this is the price paid for the DM's superior statistical accuracy. As noted 247 above, it is well-established that individual experts tend to under-estimate true variable 248

uncertainties (Lin & Bier 2008) , and pooling via the DM redresses this trait. The 249

combination DM out-performs the best single expert and provides a set of target item 250

solutions that is superior to any single expert. 251

Individual expert's relative performance scores (column 5 in Table 1) are computed  252 from the products of their p-value and information scores per item, and are un-253

normalised. When the DM is included as a synthetic expert, performance scores are 254

adjusted and normalised to sum to unity across the group (with DM included). These 255 are the relative weights (column 6 in Table 1 ) that are used in the Classical Model for 256 combining judgements on target items. 257

Two experts (E09 & E10) jointly achieve the best statistical accuracy scores (0.182); 258 their judgment influences are differentiated in the analysis by their different 259

information scores -one is more informative than the other and is consequently 260 rewarded with a marginally higher overall weight than the other (see two rightmost 261 black points on Figure 1 ). 262

Prior to the main elicitation and before 1 st June we asked the experts to forecast the 263 proportion of returning pupils and teachers on two dates (1 st and 15 th June). When 264 completing these forecasts some teachers indicated that their estimates and ranges 265

were based on surveys of parents before 1 June, conducted for planning purposes. As 266 such, these percentages did not represent personal judgements, sensu 267 stricto. However, absent comments from other teachers, it was not possible to know 268 how general pre-return surveys were, or how many respondents had provided 269

percentages data based on similar surveys. Thus, for the purposes and goal of our 270 elicitation exercise, we chose to regard all the inputs on the percentages of pupils 271 returning as representing objective, informed judgements and treated them uniformly 272 when processing the entire group's responses. 273

The forecasts were received before or by 27 th May 2020. There was a wide range of 274 responses ( Figure 2 and Table 2 ), indicating a remarkable diverse set of circumstances 275

in individual schools and community enthusiasm or lack of enthusiasm for a return to 276

school. This was borne out by the expert interviews where the actual return had been 277

highly variable across just six schools driven both by community perception and 278

school's mitigation measures. However, when averaged over all of the teachers, the 279 median forecasts are very close (indeed, for pupil's attendance, identical) to the 280 national attendance on 1 st June ( Table 2 ). The national attendance on 15 th June was 281 similar to 1 st June but had increased by 2 nd July to levels similar to those anticipated by 282 the teachers for 15 th June. These results demonstrate the ability of the teachers, as 283 experts, to make good forecasts and strengthen the belief that the study schools are a 284 dependable representative sample of primary schools in England. 285

The six structured interviews strengthened the assessment of the teachers as 286 thoughtful and careful in thinking through the responses to the questions even thought 287 they were regarded as challenging. All thought about variations between different 288 children, differences in the school day, differences between class time and breaks, and 289

in roles and interactions of staff. Two of the interviewees had consulted with other staff 290

to help them think about contacts. 291 292

Here we present and interpret the quantitative responses from the elicitation. Table 3  294 itemises the questions for which quantitative answers were requested. The questions 295

largely focus on contacts between persons in schools as epidemiological models use 296 contact data as a basis for modelling transmission of infection (Danon et al. 2013 ).

Essentially the greater the number of contacts and the longer the duration of those 298

contacts the greater the chance of infection transmission. 299 Table 4 list the results, noting that triplets of quantiles for the elicitation of a single 300 central are variance spreads on the single values, rather than the usual elicited 301 uncertainty ranges per expert. In several questions the experts are asked a pair of 302 questions to estimate contact numbers for an individual (Q2a, Q3a, Q5a,Q6a) and then 303

to requested to estimate the range of contacts considering the variations between 304 individuals (Q2b, Q3b, Q5b, Q6b). Figure 3 shows examples of responses to some 305

questions to illustrate variation in responses among the experts. Figure 4 shows the 306 range graphs for the DM contact distributions, while Figure 5 shows a typical 307 distribution for an individual child. The amalgamated DM distributions are 308 characteristically heavy tailed. Figure 6 plots the 5 th , mean and 95 th values to compare 309 normal (pre-COVID) and new normal (COVID) times. 310

The interviews conducted suggested the teachers were largely thinking about how 311 contacts might vary from day to day, with this being much harder to estimate during 312 normal times due to the diversity of activities pursued by adults and children. For the 313 second set of questions the variation between experts was also dependent on the 314 make-up of the school; for example some schools had special-learning units which 315 meant some children in the cohort spent only part of the day engaged in mainstream 316

learning. 317

Before presenting the responses to specific questions we consider sources of 318 uncertainty in the results with a focus on the contact data. Some questions concern the 319 contacts of individuals while others concern variations among groups of individuals. 320

Each teacher has been asked effectively to make measurements of contacts. The 321

definition of a contact (conversation at 1 metre for five minutes or more) is challenging 322

and quite large uncertainty is implicit in making the measurement. The large range for 323 most experts in the graphs for individuals (Figure 3 ) represents the accuracy of the 324 measurement. In the interviews teachers also placed different emphasis on thinking 325 this through; in some instances 5 minutes of contact was considered a long time for 326 some children during the course of play-led learning. 327

In comparing schools and combining experts into a DM two additional sources of 328 uncertainty arise. First there are likely to be real differences between schools because 329 of variations in bubble sizes and school characteristics. In all of the interviews the 330 teachers attested to the strict way in which the bubbles were observed, and thus 331 influenced their thinking around contacts. Second there is a calibration issue with each 332 expert working out how to make the measurement. These uncertainties are manifest in 333 the wide variation of elicited values observed in Figure 3 . In normal circumstances face-334

to-face discussions between experts might have been able to reduce the calibration 335 effect. In creating the DM we are combining measurement accuracy at schools, real 336 variations in contacts between schools and significantly different calibrations. 337

The DM distribution aggregates all these uncertainties and results in markedly skewed 338 distributions (Figure 4a ). Here we take the ratio of the right to the left tail, commonly 339 termed eccentricity, as an approximate measure of skewness. are organized to learn together to limit contacts with many children. The DFE guidelines 364

issued during the study period were for schools to form bubbles of up to 15 children. 365

The responses indicated bubbles between 6 and 15 with a mean of 11 (Cohort 1) and 13 366

(Cohort 2). The response to the question on spacing between bubbles indicated that 367 each bubble was in a separate room and so these results are not reported as they are 368

deemed not relevant to the risk of transmission between children in different bubbles 369 during class time. Teacher interviews substantiated this view, with teacher-teacher 370 contacts between bubbles being substantially restricted. The interviewees cited this 371

change of behaviour as one of the most significant in terms of reduction in contacts for 372

teachers. 373

We first consider results for contacts of children in normal pre-COVID times (Q2, Q5). of high contact individuals. These studies do not report 5 th and 95 th percentiles on the 399 contact distributions. The teacher elicited data for pre-COVID times (Q2c and Q5c) are 400

qualitatively consistent with previous studies. 401

The contacts between children in new normal time under risk mitigation regimes are 402

substantially reduced ( Figure 6 ): comparing mean contact numbers are reduced by 53% 403

for Cohort 1 (Q3a) and 62% for Cohort 2 (Q6a). The latter reduction is slightly less than 404 the reduction of 74% for adults found by Jarvis et al. (2020) . The somewhat greater 405 reduction in contacts in Cohort 2 compared to Cohort 1 in new normal times, however, 406

indicates that older children are slightly easier to manage with respect to social 407

distancing. This interpretation was supported by interviewees. Some thought that older 408

children were more capable of more sustained (> 5m) social distancing (talking face-to- The formation of bubbles, however, is unlikely to be the only factor in the decrease in 420

contacts. Children will naturally form smaller groups in schools due to either 421 juxtaposition in class or formation of friendship groups that are much smaller than a 422 classroom (Conlan et al 2011). Thus most contacts will be restricted to a fraction of a 423 class in normal times, while in bubbles the size of friendship groups will become more 424

comparable. The interviews indicated, however, that reception classes in particular 425

were more free flow with gregarious children interacting with sibling groups in other 426

years as well as their peer group within bubbles. These observations help explain the 427 significantly smaller reduction in contacts in Cohort 1. 428

We asked questions about differences between different cohorts (Q4 and Q7) In new normal times mean daily contacts for Cohort 3 (Q9c) decrease from 25 to 10, a 459 60% reduction ( Figure 6 ). The reduction in contacts compared to normal times is a 460 factor of 2.5, compared to between 2.1 (Cohort 1) and 1.5 (Cohort 2) for children. Note 461 that the mean of 10 is similar to the bubble size of 11-13 (Q1b). The contacts between 462 adults between normal (Q13) and new normal (Q15) reduces by 80% ( Figure 6 ). This 463

result is comparable to a 74% reduction in adults found by with Jarvis et al (2020).

These results indicate that teachers are social distancing to a greater extent than 465 children, limiting close contacts as far as is feasible. Answers to Q9b indicate variations 466

in the roles of different teachers and effects of tasks like supervising breaks and meal 467

times. There is still a heavy tail to the responses but this is not as great as for other 468

responses. In the interviews the teachers perceived that those facing older children 469

were strongly socially-distanced but those facing younger children were less able to 470 strictly observe this. 471

For Cohort 4 the changes between normal (Q10) and new normal times is similar to 472

Cohort 3, but in general contacts are fewer (by ~ 30%) and reflect different roles of 473 ancillary staff some of which involved much less interactions with children (e.g. 474

administrative staff). Responses to Q11 indicate a 64% reduction of contacts in new 475 normal time (Figure 6 ), reflecting the deliberate policy of limiting ancillary staff contacts 476 with children and the efforts of these staff to observe social distancing. Adult to adult 477 contacts are reduced by 80% between normal and new normal time (Figure 6 ). At 478

interview Cohort 4 were uniformly perceived as adults with the greatest change to their 479 day-to-day contacts in the working day, with some able to be entirely socially distant. 480

This is similar to the overall reduction in contacts of 20% in the general community We asked about adherence and the response indicates that the experts considered 491 adherence to be high (Q16). This view is supported by the analysis of the qualitative 492 parts of the questionnaire related to risk mitigation measure. From the interviews the 493 string was driven often by thinking about consequences, for children, their families but 494 also their colleagues. Questions related to understanding risk reduction (Q20, Q21, 495 Q22) are discussed in the next section. 496 497

We asked the Head Teachers about risk mitigation measures that they have put in 499

place. Questions are listed in Table 4 . Reference is made for some topics to literature on survey question answers structured interviews were conducted to help in the 505 interpretation of the response. 506

Teachers have attempted to follow the Government guidelines, with respect to school 507

reopening. Links to guidelines from DfE can be found as footnotes. These guidelines 508

have been updated regularly so it is not easy to categorically state which guidelines 509

were followed when teachers completed their elicitations. It is apparent that in 510

instances where the guidelines are not defined, such as 'minimising contact and mixing 511 by altering, as much as possible…' teachers have typically been very cautious, putting in 512 measures that go beyond the guidelines. Equally, there are no explicit guidelines related 513

to the need to 'wash hands frequently' and consequently there is considerable variation 514 in interpretation. One interviewed teacher made the point that the younger children 515

are in the process of learning, and that learning needs to incorporate the making of 516 mistakes. Many schools had made a good job of initiating procedures for handwashing 517 that were age appropriate and encouraged sticking within the guidelines (e.g. no 518

singing). Teachers have interpreted the guidelines to suit their own settings and then 519 gone one step further to ensure the safety of their charges. Further comparisons to the 520 guidelines are included in each section below. The appendix reports all responses and 521 provides a record of good practice and innovation by the schools which will shared 522 between the schools. 523

Answers to Q1a concerning the strategy to reduce close contacts of children were 524 broadly similar for cohort 1 and 2 and have been combined for this summary. Four 525

responders did not answer this question. The Government guidelines recommend 526 smaller group sizes, with consistent students and teachers, kept 2 m apart where 527 possible with staggered break times. Outdoor activities are recommended where 528 possible and students are to stay in the same desks and rooms as much as possible. 76% 529 of responders noted social distancing measures that were being put in place with visual 530

indicators for children to follow. Over 90% of responders followed the guidelines and 531

indicated the use of classroom bubbles of varying sizes below the recommended 532 number of 15. The answer to Q1b (Table 7) indicated that many bubbles were about 10 533 children. Some have indicated that children will only be in school part time to allow 534 teachers to accommodate the reduced class sizes. 535

Other measures included the removal of furniture to allow for more space, rotation of 536 toys and/or removal of certain play items and lunches taking place in the classroom 537 (either packed lunches or 'take-away' style cartons to remove the need for cutlery). In 538 line with the guidelines, a third of the responses indicated that learning would be 539 moved outdoors as much as possible and a similar number noted the need for 540

individual desks and resources. Around a third of teachers have extrapolated the 541 guidelines and ensured that student bubbles are allocated their own areas of the 542 playground, their own toilets or their own lunchtime spaces and over 50% refer to the 543 suggested staggered break times, start times etc. This was also borne out at interview, 544 almost all interviewees used the word 'strict' to describe their bubble, and indicated 545 that adults avoided connecting between bubbles. 546 Unsurprisingly all responders in Q18 concerning cleaning have referred to changes to 547 the usual cleaning regimes in their school with over 70% making specific reference to 548 ongoing cleaning throughout the day, highlighting high touch areas such as computers 549

(3 responses) and toilets (13 responses); this is broadly in line with Government 550 recommendations 2 . In the main, deep cleaning is taking place before and/or after 551 school and reference is made to deep cleaning taking place after classroom bubbles 552 have used specific areas; Government guidelines recommend thorough cleaning at the 553 end of the day 3 ; 3 responders made specific reference to hard plastic toys being 554 sterilised in Milton after use following DfE guidelines that recommend that cleaning of 555 toys takes place between use 4 . Some teachers referred to the removal of soft toys and 556

furnishings in their answer to question 1a). Three responders have referred to their 557 employment of additional cleaning staff and three have indicated that staff and children 558

will, themselves, engage in cleaning processes. Several responders noted that classroom 559 doors will be left open, presumably to avoid the need to touch door handles as 560 recommended in the Government guidelines 5 . Some of the responses in the appendix 561

indicate that some schools have gone beyond the guidelines, such as: bubbles having 562 their own toilet cubicle and sink; cleaning staff using PPE which are double bagged and 563 stored for 72 hours before putting in bins; disinfecting laptops after every use. Allergies 564

to disinfectant in some of children hampered cleaning of some areas. 565 SARS-CoV-2 can persist on surfaces for hours to days (Eslami and Jalili 2020) and 566 disinfecting surfaces is recommended by major health authorities (WHO 2020). 567

Exposure via contaminated environmental surfaces appears to be a secondary vector of 568 transmission; transmission via airborne droplets is more important. There is very little 569 direct research on SARS-CoV-2 =so studies rely on tests performed on similar viruses, or 570

other viruses that are hard to kill. Surface disinfection to remove viruses include 571

Quaternary ammonium, ethyl alcohol, hydrogen peroxide and sodium hypochlorite 572 (Henwood 2020 accordingly. 39% had put in place one-way systems to reduce contact between 586 parent/children arriving and those leaving (at drop off and collection times). 52% noted 587 that parents were either not allowed on the school site or that only one parent could 588

drop the child off, with teachers meeting children at the school gates. 48% referred to 589 social distancing measures being put in place for parents waiting and/or for children in 590 the playground. Approximately 20% referred to children not being allowed in the 591 playground at the start of the school day and needing to go straight to class and a 592 further 20% referred to classes and bubbles having separate entrances and exits. 593

Accurate quantification of the risk reduction gain from such measures is difficult. A 594 scoping calculation, however, can give an indicative estimate. If the average parent has 595 a daily contact hours of 30 (Danon et al. 2013 ) then the mitigation measures described 596 might reduce the contact hours by 1 or 2 hours compared to normal mixing associated 597

with delivering children to and from school, so we estimate an overall effect of a few 598 percent (~3-6%), which can be compared to the 80% reduction associated with general 599 lockdown (Brooks Pollock et al. 2020). The contribution to risk reduction within schools 600 is likely to be very small, but larger and tangible in the wider community. 601

Responses to the question on weather (Q20) indicate that contacts were not changed 602 between indoors and outdoors. It is now widely thought that outdoors is much less 603 risky than indoors but this was not the question. Interviews indicated that the physical 604 layout of the school informed their responses, and some commented the influence of 605 very good weather during the course of the study. 606

In relation to handwashing (Q21) Government guidelines recommend that adults and 607 children should frequently wash their hands with soap and water for 20 seconds and 608 dry thoroughly 7 . Responses (Table 7 ) gave a range from 3 to 13, with over half opting 609 for a range between 3 and 10 handwashes. One teacher referred to the number of 610

handwashes being dependent on the number of visits to the toilet. Those teachers who 611 gave a specific lower number with a wider range have implemented a policy of 612

handwashing at certain times with variance added for toilet visits (one responder has 613 indicated scheduled handwash times). All responses are shown in the elicitation results. 614

There is limited evidence on the efficacy of handwashing in reducing transmission. Soap 615

is very effective at neutralising the virus (Eslami and Jalili, 2020). The chances of an 616

infectious individual passing the virus by touching a person or contaminating a surface 617 are reduced by regular hand washing. The weight of evidence is that handwashing 618 reduces transmission but may become counterproductive, less effective and even 619

harmful if the handwashing becomes excessive. Risk reduction is estimated as 6% to 620 44% for respiratory diseases (Rabie and Curtis, 2006) . Beale et al (2020) studied risk 621 reduction of handwashing for flu and common coronavirus infections in the UK. They 622 estimate significant reduction for between 6 and 10 handwashes per day (5 th per = 58%; 623 50 th per = 36%; 95 th per = 1%), results consistent with those of Rabie and Curtis, 2006) .

A reservation for these studies is that they involve extrapolation to transmission of 625 SARS-CoV-2. This distribution is included in the risk modelling. The range in the elicited 626 answers indicate that most schools have adopted an optimal regime of hand washing 627 too reduce risk and have avoided excessive handwashing that are considered to be 628 potentially harmful. 629

The schools follow Government guidelines for those displaying COVID19 symptoms 630 (Q22), but some schools have gone beyond these recommendations. In general anyone 631

showing symptoms is self-isolated with some mentioning designated areas set aside for 632 this purpose. A few will use of PPE, which is recommended if the adult cannot keep 2 m 633 apart. The symptomatic persons are then sent home immediately and asked to get 634 tested. School areas are then cleaned following the guidelines 8 to use PPE. Two 635 responders plan to notify families of children in a bubble. One school proposed that 636 everyone in that bubble would be recommended to get a test and, if it comes back 637 positive, the whole bubble is required to self-isolate for 14 days. The individual stays at 638 home for 7 days or until a negative test is achieved. One responder has said that if a 639 positive test is confirmed the whole school will close. For case where infection is 640

suspected outside school hours, those that answered required the school to be 641 informed, a test will be undertaken, and the school notified of the result. A small 642 number mentioned that the person showing symptoms would not be able to attend 643 until a negative test was achieved. One responder stated that a positive test would 644 result in self-isolation of the whole bubble for 14 days. Only a few responded to the 645 policy for dealing with anyone with symptoms identified outside of school. Most 646

interviewees had experienced at least one instance of a suspected case that required 647

testing. While the protocols all required that the child stayed away from school, the 648 responses varied in terms of the notification or isolation of the child's bubble until test 649 result was received. The DfE guidance states that the bubbles only close if a positive 650

test is returned but one school will require immediate self-isolation until results come 651

back. 652

The survey indicates that all schools had strong measures in place following and going 653

beyond government guidelines to minimise risk. 654

Most responders to Q23 stating that the child cannot attend school and must self-655

isolate -Government guidelines are limited to just this piece of advice, with respect to 656 schools 9 . A small number will inform other families within the same bubble. One school 657 stated they had no policy for such events and 2 schools indicated that the child of said 658

adult could still come to school. One school submitted this guidance from a local health 659 protection team. 660

Other comments (Q24) are unique to individual schools and are all included in the 661 appendix below. Some noted issues of EAL and SEND students 10 and risk factors 662

involved with school transport. 663

The structured interviews augmented these findings from the risk mitigation survey, but 664

highlighted some additional issues. Interviewees mention the difficulties and stress for 665 staff in maintaining social distancing and risk mitigation measures. The tension between 666 social distancing and key educational objectives of learning and developing social skills 667 was highlighted. There was comment that some measures would be impossible or much 668 more difficult with a full return of school. 669 670

The circumstances in English primary schools in June and July 2020 are unprecedented 672 and unlikely to be repeated. The partial re-opening of schools was undertaken under 673 strict guidelines of social distancing and a range of risk mitigation methods to reduce 674 the transmission of COVID-19. These unique circumstances provide an opportunity to 675 evaluate the efficacy of different risk reduction strategies and to add to the data on 676

contact patterns for young children and staff in the school environment. 677

A striking aspect of our results is that the 34 participating schools proved to be 678

representative of schools nationwide. Thus as in opinion polls an accurate picture of 679

what is happening in nationally in schools can be gleaned from a modest sample size. 680

We were fortunate here that volunteerism led to a group of schools covering a wide 681 range of sizes and communities. Even though the schools individually predicted and 682 then reported a very wide range of returns for pupil and teacher attendance on 1 st June 683 the average for the 36 schools was very close to the national picture. The prescience of 684 experienced school leaders could be used to anticipate what will happen in September 685 when a full return to school has been mandated by Government. 686

Breaking social networks is a key non-pharmaceutical intervention for countering the 687 spread of infectious disease. Our study indicates that contacts within schools were 688 reduced in the range 45% to 80% ( Figure 5 ). These strikingly successful outcomes 689

highlight the tremendous work of school staff and the role of greatly reduced class sizes 690 through creation of bubbles of children which are much less than normal class sizes. 691

Although the marked reduction in contacts can be partly explained by much smaller achieved similar reductions in contacts. The data also confirm that older children 711

(Cohort 2) are easier to manage than younger children (Cohort 1) with more than a 712 two-fold greater reduction of contacts in the older children compared to the younger 713 ones. 714

The elicited data can be compared quantitively with other kinds of contact survey data 715

because of the variation in methodologies. The results indicate the same kind of 716 heterogeneity as documented in Danon et al. (2013) in which the data are a mixture of 717

individual contacts involving significant conversations at a distance of 3 m or less or 718 touching plus contacts related to groups where each person in a group counts as a 719

contact. Results were expressed as contacts per day and total contact hours, noting that 720 the latter parameter can exceed 24 hours because contacts within groups occur 721

simultaneously. In a school setting, classroom adults (Cohort 3) have a median of 26 722 contacts (Q8a), of which about 2/3's are with children (Q12) and 1/3 with adults (Q13). 723

The distance specified in the question is 1 m with social distancing at the time of 724 elicitation being 2 m. Contact hours are of the order of 2 hours. 725

In the context of a typical class of 30 the group definition of contacts in Danon et al. 726 (2013) seems less relevant. There must be an additional risk factor related to being in 727 an enclosed space with widespread aerosol circulation, but this is not quantifiable with 728 the elicitation data. However, reducing class sizes from 30 to roughly 10 reduces the 729 random risk of an infectious person being in the class-room by about 1/3. 730

All of the risk mitigation strategies contribute to reduction of risk. Specific risk 731 management strategies reflect individual circumstances in schools. Many relate to steps 732

which are likely to reduce contacts and disrupt break contact networks. Each of the risk 733 mitigation measures will contribute to risk reduction, but this is hard to quantify. 734

It seems unlikely that the significant reduction of risk, implied by these results, can be 735 maintained with a full return to school without greatly expanding the accommodation 736

to maintain reduced class sizes, as suggested by the factor of 2 to 3 reduction in 737

contacts between children. Adult staff can continue to observe strict social distancing 738 behaviours and can continue to organise the classroom and break times which reduces 739

contacts. 740

We are particularly appreciative of the Primary School leaders who volunteered to join 742 the expert panel. Their enthusiasm, support and expert knowledge was paramount. The 743

University of East Anglia fact finding team of Jade Eyles, James Christie, Nicola Taylor 

Anticipated impacts of Brexit scenarios on UK food 750 prices and implications for policies on poverty and health: a structured expert 751 judgement approach

Hand Hygiene Practices and the Risk of Human Coronavirus Infections in a UK 755 Community Cohort

Attribution of Illnesses Transmitted Commonly by Food and 758 Water in the United States to Comprehensive Transmission Pathways using Structured 759 Expert Judgment

Population Attributable Fraction (PAF) of cases due to gatherings and groups with 763 relevance to COVID-19 mitigation strategies

Adherence to physical distancing 766 dominates the impact re-opening schools on SARS-CoV-2

Experts in Uncertainty

Cross validation for the Classical Model of structured 771 expert judgment

Gog JR 774 Measuring social networks in British primary schools through scientific engagement

Social encounter networks: 778 characterizing Great Britain

Mixing patterns between 781 age groups in social networks

Contact Patterns Explain the Spread of H1N1v Influenza

The role of environmental factors to transmission of SARS-CoV-786 2 (COVID-19)

World Health Organization Estimates of the Relative Contributions of Food to 791 the Burden of Disease Due to Selected Foodborne Hazards: A Structured Expert 792 Elicitation

Coronavirus disinfection in histopathology

Long-term 796 exposure to respirable volcanic ash on Montserrat: a time series simulation

CMMID COVID-19 working group

Quantifying the impact of physical distance measures on 800 the transmission of COVID-19 in the UK

Contacts in 802 context: large-scale setting-specific social mixing matrices from the BBC Pandemic project

A study of expert overconfidence. Reliability Engineering 805 & System Safety

Semi-structured interviews and focus groups

Social contacts and mixing patterns relevant to the spread of infectious diseases

Weekly Coronavirus Disease 2019 (COVID-19) Surveillance Report Summary 815 of COVID-19 surveillance systems Weeks 28

Handwashing and risk of respiratory infections: a quantitative 817 systematic review

Expert 819 elicitation on the uncertainties associated with chronic wasting disease

Expert judgement and re-elicitation for prion disease risk 824 uncertainties

WHO (2020) Infection prevention and control during health care when novel 827 coronavirus (nCoV) infection is suspected. World Health Organization, Geneva. Interim 828 guidance

Changes in contact patterns shape the dynamics of 832 the COVID-19 outbreak in China Science