Background
Since the beginning of the COVID-19 pandemic, school closures have been implemented in many countries as part of a broader response to suppress transmission [
1]. It is well established that children are at low risk of hospitalisation and death as a direct result of infection [
2,
3]. Despite this lower risk, there is concern that allowing transmission amongst younger age groups increases the risk of infection in adults, who are at substantially higher risk. The role of schools in the transmission is therefore an important question. On 4 January 2021, a third national lockdown was announced in England to curb the transmission of SARS-CoV-2 [
4]. This included the closure of schools, a measure the UK government reversed on 8 March.
At the time of the decision to reopen, the direct and indirect impact of school closure and reopening was still unclear. To date, there is mixed evidence around the role of schools in community transmission. Existing studies of transmission within schools have wide ranging results [
5‐
7]. Other work demonstrates an increased prevalence amongst school-aged children in the months after schools returned in September 2020 [
8,
9] and a higher risk of infections entering households through children than adults. However, studies have failed to find evidence that schools drive transmission in the community [
10,
11]. A particular challenge for many observational analyses, based on reported cases, is bias resulting from the age dependence in case ascertainment due to varying rates of asymptomatic infection [
12]. This challenge is further complicated by changes in epidemiology due to the emergence of new variants [
13].
The potential change in transmission of SARS-CoV-2 upon reopening schools predominantly depends on a combination of two factors: firstly, the age-specific risk of transmission upon contact, and secondly, the likely increased rate of contact between members of the population due to school reopening. Multiple studies aimed at understanding the relative transmission risk associated with children indicate lower susceptibility [
14‐
16] and some indicate lower infectiousness [
14]. However, evidence of lower transmission risk amongst children alone is insufficient to quantify the impact of reopening schools. There is a need to combine the estimates of reduced susceptibility and infectiousness with age-specific contact patterns in these age group social contacts amongst school-aged children.
There is abundant evidence that children’s contacts increase when schools are open, presenting opportunities for increased infectious disease transmission which is well documented in other pathogens such as influenza [
17]. Nonetheless, it is important to capture how these contacts vary under the specific conditions presented during the current pandemic response, where social distancing and other mitigations are in effect within schools.
Although schools were closed during the third national lockdown, during the second national lockdown (November 2020), restrictions were similar for the majority of the population, but in-person lessons continued in schools. We used data collected as part of the CoMix social contact survey [
18] to compare contacts made during these two lockdown periods. We combined these data with estimates of age-stratified susceptibility and infectiousness [
14‐
16] to evaluate the potential impact of reopening schools on the reproduction number in England [
14‐
16].
Discussion
When the UK government made plans to reopen schools on 8 March 2021, the potential impact on the transmission of SARS-CoV-2 was uncertain. Although there have been many attempts to quantify the relative susceptibility and infectiousness of children and adults, these estimates need to be assessed alongside rates of contact to give an indication of the overall risk of transmission in any given setting. We combined social contact data from a large-scale survey in England during two periods of national lockdown, one with schools open and the other with schools closed, with estimates of relative susceptibility of children and adults. We used these data to project the potential impact of reopening schools on reproduction number when schools reopened in March 2021.
Whereas adults’ contacts were generally similar between the two periods of lockdown, there was markedly higher contact between children during the second lockdown (November 2020), when schools were open than the third lockdown (January to March 2021) and when schools were closed. We observed the change in contacts at school but also in other contacts outside of the home. Increased contact outside of school and home settings includes contacts in childcare outside of school hours, which would be expected to rise; however, it could also indicate reduced overall adherence to restrictions amongst children when attending schools physically.
The differences in contacts suggested that reopening all schools would likely increase
R above 1.0, from an assumed current value of 0.8, if no additional measures (not imposed in the second lockdown) were effective. Reopening primary or secondary is likely to increase
R above 1.0. This would, in turn, be expected to stop or reverse the fall in cases that had been observed since January 2021 [
25]. The risk of cases increasing following the reopening of schools depends greatly on the assumed value of R before schools are reopened. Although cases of the alpha variant (B.1.1.7) appeared to be increasing whilst national lockdown was still in place in November [
10,
13], the latest national serology surveys suggest that immunity levels have substantially increased across the UK [
25], resultant from both infections and the national COVID-19 vaccination programme. These changes in overall immunity should be reflected in real-time estimates of
R, but
R estimates are lagged due to delays in reporting [
26]. Although the results vary depending on the estimate of relative susceptibility and infectiousness in children, the qualitative interpretation remains consistent between them. We highlight that we included estimates based on equal infectiousness and susceptibility in all age groups for completeness, but stress that assuming that children are equally infectious and susceptible as adults is not compatible with results from previous studies or our own estimates (Fig.
3).
We found that whilst the increases in R due to school reopening would be reduced by additional vaccine-derived immunity, it is unlikely that they would be reversed. We emphasise that these results are indicative only: firstly, because treatment of vaccine-derived immunity is simplistic and based on broad scenarios of vaccine efficacy, and, secondly, because the vaccine-derived immunity profiles we used were likely yet to take some time to materialise. This is due to the pace of the UK vaccine rollout and the delay between immunisation and full immune response, at which time the UK is unlikely to be under lockdown restrictions. Furthermore, it is likely that there was some immunity due to vaccination already in the population prior to 8 March, and therefore reflected in the preopening R estimates, this would impact the estimate of both the impact of further vaccine-derived immunity and school reopening.
Our descriptive analysis shows that in November, when schools were open, there was a substantial variation in contacts between children by region. We have not presented regional estimates of the impact of reopening schools on R, due to low numbers of observations between the lower-level age group aggregation used in the construction of contact matrices. However, the variation in the mean contacts points to potential geographical variation in the impact of reopening schools, which may be lower in London than in other parts of the country.
Schools have reopened since this work was carried out, providing an opportunity to reflect on how our estimates relate to epidemiological observations around the event of reopening. Based on case reports, it appears that transmission initially remained low allowing cases to continue to fall for a few weeks following reopening [
27]. This was however combined with an increase in test positivity amongst school-aged children [
28]. This suggests that transmission may have increased in this age group, but those mass testing and quarantine of infectious school children were broadly successful in curbing transmission in schools. We present these suggestions tentatively. As mentioned in the introduction of this paper, evaluation of the impact of reopening is challenging in general. In addition, there are specific events surrounding reopening in March that further complicate evaluation.
In general, real-time estimates of
R are smoothed substantially by the delay associated with the onset of symptoms and reporting of cases [
26]. Since changes in transmission are due to sharp and age-heterogeneous changes in contact, it would be expected that
R would gradually change as infections reach a new equilibrium age distribution [
29]. Also, as mentioned in the introduction, poor case ascertainment in children suggests that detection of the contribution to transmission in schools is likely to be further delayed until this change affects infections in the adult population, which are reported more reliably. This means that sharp changes in contact we expect would lead to a gradual change in
R making it difficult to associate the changes with a particular event.
More specifically, the reopening of schools in March 2021 coincided with a substantial change in testing, with large-scale home testing being rolled out, particularly amongst school-aged children. This is likely to affect the infection reporting rate. In addition, schools only opened for a short period of time before closing again for the Easter holidays (29 March 2020), where they remained closed for a further 2 weeks.
The combination of these factors makes it difficult to assess whether school outbreaks generally failed to occur due to testing, lower effective contact rates in schools than we anticipated or stochastic variation due to low prevalence in the population.
There are a number of important limitations to this work: Contacts in different settings likely contribute differently to transmission, but we assumed all contacts make equal contributions to transmission, as these differences are not well quantified in the context of control measures. If contacts at school are at lower risk than those outside of school, the impact of reopening schools would be lower. Moreover, contact survey methods are likely to be systematically biassed towards reporting a higher proportion of close contacts than incidental contacts (for example, on public transport); this may lead to an underestimate of the change in contact as restrictions are relaxed, particularly in adults who may be more likely to have this kind of contact. This in turn may overestimate the contribution of contacts of school-aged children to changes in
R over time. The age-stratified susceptibility profile is likely to change over time as natural immunity is acquired in the population. The profiles we used each reflect a single point in time. Changes in the relative immunity in children would alter the relative impact of school contacts on overall transmission. Changes in the overall immunity over time and seasonal effects on transmission are not expected to affect our main analysis, which presents an instantaneous change in reproduction number under specific contact behaviours. However, there may be some impact on our estimates of relative susceptibility as we fit
R over a period of time, we suggest that this is likely to be minimal due to the short period over which we fit and the low prevalence of infection during this period. Further, the UK’s testing capacity changed between summer 2020 and spring 2021, although not greatly within these periods. If infections in children were less likely to be identified between July and October 2020 (the period we used to estimate relative susceptibility of children) than they were in March of 2021, we may have underestimated the relative susceptibility when considering the impact on
R when schools reopened. We counter this limitation by using a wide range of estimates of relative susceptibility in children, all of which give a higher relative susceptibility than our own estimate. We assume adult contacts revert to those observed when all schools were open, which is conservative, in reality, particularly for partial reopening scenarios, adult contacts may not fully return to the same levels. Furthermore, there may also be differences in adherence to restrictions between the two lockdowns, unrelated to school closure. However, the change in adults’ contacts between the two periods was relatively small. The proportion of children in school varied over time due to exclusion-based control measures during the autumn, though the proportion attending school remained high during the November lockdown (Additional file
1: Figure S3). Contacts of children are reported by parents, which may impact their reliability, particularly in school, where parents are unlikely to witness students’ behaviour. The contact survey was conducted in this way for convenience and to allow a quick rollout. We are unaware of any previous work that has established the likely biases that arise from parent-reported contacts. It is unclear whether this would lead to systematic bias in reporting either more or fewer contacts.
Our work evaluates the impact of reopening schools on the reproduction number in England, which gives an indication of how the transmission may be affected by this change. However, there are other factors that reopening schools could introduce, such as the potential for children’s contact at school to provide routes of transmission between households, facilitating long chains of transmission that would be otherwise impossible [
30]. We are not able to capture these network effects in this analysis; however, they may play an important role in the change in epidemiology between school closure and reopening. Second, there is evidence for lower prevalence in primary school than secondary schools [
8]. Our framework has not captured these differences suggesting there may be additional factors that reduce the impact of reopening primary schools relative to secondary schools. Furthermore, additional management strategies such as mass testing of school children may have served to reduce the risk that a contact in a school results in infection compared to contacts during lockdown 2. Importantly, with the recent emergence of new variants, particularly alpha and delta (B.1.1.7 and B.1.617.2) [
31], the baseline
R will depend on the proportions of these variants as well as contact patterns. Furthermore, these proportions changed substantially over the spring period, likely altering the implications of reopening schools.
Our results suggest reopening schools under the same conditions as November 2020 would have been likely to increase R close to or above 1.0, which would stop the decrease in cases observed between January and March. However, precise estimates rely heavily on the baseline values of R and the profiles of susceptibility and infectiousness, generally assuming lower susceptibility and no greater infectiousness in children relative to adults. We advocate further evaluation of the impact of within-school measures to assess their contribution to the successful containment of school outbreaks in the weeks following reopening.
Acknowledgements
The authors wish to thank Dr Thomas House for his support with the interpretation of the ONS susceptibility estimates. We also thank the members of SPI-M for their useful discussion which helped shape the final version of this work. We would like to thank the team at Ipsos, who have been excellent in running the survey, collecting the data and allowing for the CoMix study to be implemented rapidly. Finally, we thank Katie Collis for the proofreading and excellent discussions.
The following authors were part of the Centre for Mathematical Modelling of Infectious Disease COVID-19 Working Group. Each contributed in processing, cleaning and interpretation of the data; interpreted the findings; contributed to the manuscript; and approved the work for publication: Yang Liu, Joel Hellewell, Nicholas G. Davies, C Julian Villabona-Arenas, Rosalind M Eggo, Akira Endo, Nikos I Bosse, Hamish P Gibbs, Carl A B Pearson, Fiona Yueqian Sun, Mark Jit, Kathleen O’Reilly, Yalda Jafari, Katherine E. Atkins, Naomi R Waterlow, Alicia Rosello, Yung-Wai Desmond Chan, Anna M Foss, Billy J Quilty, Timothy W Russell, Stefan Flasche, Simon R Procter, William Waites, Rosanna C Barnard, Adam J Kucharski, Thibaut Jombart, Graham Medley, Rachel Lowe, Fabienne Krauer, Damien C Tully, Kiesha Prem, Jiayao Lei, Oliver Brady, Frank G Sandmann, Sophie R Meakin, Kaja Abbas, Gwenan M Knight, Matthew Quaife, Mihaly Koltai, Sam Abbott and Samuel Clifford.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.