Abstract
Introduction School closures are associated with significant negative consequences and may exacerbate inequalities. They were implemented worldwide to control SARS-CoV-2 in the first half of 2020, but their effectiveness remains uncertain. This review summarises the empirical evidence of their effect on SARS-CoV-2 community transmission.
Methods The study protocol was registered on Prospero (ID:CRD42020213699). On 12 October 2020 we searched PubMed, Web of Science, Scopus, CINAHL, the WHO Global COVID-19 Research Database, ERIC, the British Education Index, and the Australian Education Index. We included empirical studies with quantitative estimates of the effect of school closures/reopenings on SARS-CoV-2 community transmission. We excluded prospective modelling studies and intra-school transmission studies. We performed a narrative synthesis due to data heterogeneity.
Results We identified 3,318 articles, of which ten were included, with data from 146 countries. All studies assessed school closures, and one additionally examined re-openings. There was substantial heterogeneity between studies. Three studies, including the two at lowest risk of bias, reported no impact of school closures on SARS-CoV-2 transmission; whilst the other seven reported protective effects. Effect sizes ranged from no association to substantial and important reductions in community transmission.
Discussion Studies were at risk of confounding and collinearity from other non-pharmacological interventions implemented close to school closures. Our results are consistent with school closures being ineffective to very effective. This variation may be attributable to differences in study design or real differences. With such varied evidence on effectiveness, and the harmful effects, policymakers should take a measured approach before implementing school closures.
Evidence before this study A previous systematic review, published by some of us in April 2020, found good evidence that school closures are effective for the control of influenza, but limited evidence of effectiveness for coronavirus outbreaks. At the time there was no available empirial evidence from the COVID-19 pandemic.
Added value of this study This study is the first systematic review of the empirical evidence from observational studies of the effect of school closures and reopenings on community transmission of SARS-CoV-2. We include 10 studies, covering 146 countries. There was significant heterogeneity between studies. Some studies reported large reductions in incidence and mortality associated with school closures, however, studies were at risk of confounding and collinearity, and studies at lower risk of bias reported no association.
Implications of all the available evidence The evidence is consistent with either no effect, or a protective effect of school closures. With such varied evidence on effectiveness, and the harmful effects, policymakers should take a measured approach before implementing school closures.
Introduction
School closures have been a common strategy to control the spread of SARS-CoV-2 during the COVID-19 pandemic. By 2 April 2020, 172 nations had enacted full closures or partial ‘dismissals’, affecting nearly 1·5 billion children2. However, school closures have significant negative consequences on children’s wellbeing and education, which will impact on life chances and long-term health3,4. Closures may exacerbate existing inequalities. Children in higher income families may have better opportunities for remote learning.
Moreover, whilst the role of non-pharmaceutical interventions (NPIs) collectively in limiting community spread is established, the specific contribution of school closures remains unclear. Observational studies suggest that school-aged children, particularly teenagers, play a role in transmission to peers and bringing infection into households8, although the relative importance compared to adults remains unclear9. Younger children appearless susceptible to infection and may play a smaller role in community transmission, compared with older children and adults10. Whilst some modelling studies have suggested that school closures can reduce SARS-CoV-2 community transmission5, others disagree6,7.
A rapid systematic review published in April 2020 found only limited evidence of the effectiveness of school closures in controlling the spread of coronaviruses.1 However, this study was undertaken very early in the pandemic and included no observational data on SARS-CoV-2. Several empirical studies on the effects of school closures on SARS-CoV-2 community transmission have been published since the April review, but there has been no systematic review of these studies. Here, we synthesise the empirical published and grey literature on the impact of closing or reopening schools on COVID-19 incidence, hospitalisation, and mortality.
Methods
The study protocol for this systematic review is registered on Prospero (ID:CRD42020213699).
Inclusion and Exclusion Criteria
We included any empirical study which reported a quantitative estimate of the effect of school closure or reopening on community transmission of SARS-CoV-2. We considered ‘school’ to include early years settings (e.g. nurseries or kindergartens), primary schools, and secondary school, but excluded further or higher education (e.g. universities). Community transmission was defined as any measure of community infection rate, hospital admission rate, or mortality attributed to COVID-19.
We included studies published in 2020 only. We included pre-prints, peer-reviewed and grey literature. We did not apply any restriction on language, but all searches were undertaken in English. We excluded prospective modelling studies and studies in which the assessed outcome was exclusively transmission within the school environment rather than wider community transmission.
Search strategy
We searched PubMed, Web of Science, Scopus, CINAHL, the WHO Global COVID-19 Research Database (including Medrxiv), ERIC, the British Education Index, and the Australian Education Index, searching title and abstracts for terms related to SARS-CoV-2 AND terms related to schools or NPIs. To search the grey literature, we searched Google. Full details of the search strategy are included in Appendix A. No restrictions on dates were placed and all searches were undertaken on 12 October 2020.
Data extraction and risk of bias assessment
Article titles and abstracts were imported into the Rayyan QCRI webtool11. Two reviewers independently screened titles and abstracts, retrieved full texts of potentially relevant articles, and assessed eligibility for inclusion (SW assessed all articles; AC, SR and VB each assessed one third).
Two reviewers independently extracted data and assessed risk of bias. Data extraction was performed using a pre-agreed extraction template which collected information on publication type (peer-reviewed or pre-print), country, study design, exposure type (school closure or re-opening), setting type (primary or secondary), study period, unit of observation, confounders adjusted for, other NPIs in place, analysis method, outcome measure, and findings. We used the Cochrane Risk of Bias In Non-randomised Studies of Interventions (ROBINS-I) tool12 to evaluate bias.
Discrepancies were resolved by discussion in the first instance and by a third reviewer if necessary.
Data synthesis
Given the heterogeneous nature of the studies, prohibiting meta-analysis, a narrative synthesis was conducted.
Results
We identified 3,318 studies (Figure 1). After removing 372 duplicates, 2,946 unique records were screened for inclusion. We excluded 2,814 records at the title or abstract stage, leaving 132 records for full text review. Ten of these met the inclusion criteria.
Included studies are described in Table 1. All studies 13–22 reported the effect of school closures on community transmission of SARS-CoV-2, and one study21 additionally examined school re-opening. All studies used data from national Government sources or international data repositories, and reported on the first half of 2020.
Six studies reported data from a single country or region: the USA13,14,17,20 (n=4), Japan15(n=1), and Jerusalem, Israel21 (n=1). The remaining four reported data from multiple countries, of which two16,19 provided estimates for an overall worldwide effect of school closures, and two provided estimates for three individual countries each (one study22France, Italy, USA; the other18 Argentina, Italy, South Korea). All studies were ecological, and used a state (USA), regional, or national unit of analysis.
Five studies13,15,17,20,21 specified that both primary and secondary schools were included (children aged 5 or 6 to 18); the others did not specify school type. No study provided independent estimates of the effect of closing either primary or secondary schools only.
Six studies specifically sought to estimate an effect of school closures on SARS-CoV-2 transmission.13–15,17,18,21 The remaining four studies primarily sought to estimate the effect of NPIs (but reported an independent estimate for school closures within their analysis).16,19,20,22
Several analytic approaches were used, including: various types of regression models (n-7),13,14,16–20,22 time series analysis with Bayesian inference (n=1),15 comparison to a synthetic control group derived from data from comparable countries (n=1),18 and presentation of an epidemic curve21.
In most instances of school closures, other NPIs were introduced at or around the same time and potentially confounded the estimate. One study17 dealt with this by selecting US states that closed schools first and left a gap before implementing other NPI measures. Whilst another study18 took a similar approach, choosing countries (South Korea, Italy and Argentina) that shut schools early relative to national lockdown; these countries had significant other NPIs in place at the time of school closure. Four studies13,14,22,23 used statistical adjustment to control for other interventions. Four studies15,16,20,21 did not account for other NPIs. Some studies also adjusted for other potential confounders, such as population factors (e.g. proportion of population aged ≥65, population density and testing regimes).
Regarding outcomes, eight studies13–17,19,21,22 reported effects on incidence, and four studies13,17,18,20 used mortality data (one of which17 additionally reported hospitalisation rates). The assumed lag period from school closure to changes in incidence rate varied between seven and 20 days, with longer time periods of 26 to 28 days generally assumed for mortality.
Risk of bias of the studies is summarised in Table 2: two studies were found to be at low risk of bias14,22, two at moderate risk13,17, five at serious risk15,16,18–20 and one at critical risk21.
Table 3 reports study findings. Seven studies13,16–21 reported that closing schools was associated with a reduction in incidence or mortality rates, whilst three14,15,22 found no association. There was significant heterogeneity in the reported effect size of closing schools, ranging from precise estimates of no effect, to approximately halving the incidence and mortality rates13. The two studies with the lowest risk of bias14,22 reported no effect of school closures on transmission.
At a country level, four studies13,14,17,20 exclusively reported data from the USA, and one further study22 reported an independent effect size for the USA. The results from these studies are discordant, with two studies reporting null effects14,22, and the other studies reporting large preventative effects.13,17,20 Two studies reported effect estimates for Italy, one being preventative18 and one22 tending towards a non-significant preventative effect. Single estimates that were preventative were observed for the following countries: Argentina18, Israel21, and South Korea18; with single estimates of no association for France22 and Japan15.
Of the eight studies that reported an effect on incidence, five13,16,17,19,21 were preventative and three14,15,22 had no effect. Only one study17 reported an effect on hospitalisation, which was preventative. All four of the studies13,17,18,20 that reported an effect on mortality reported a preventative effect.
Narrative Synthesis
We identified three study designs: within-area before-after comparisons, pooled multiple-area before-after comparisons, and pooled multiple-area cross-sectional comparisons.
Within-area before-after comparisons
Five studies15,17,18,21,22 compared community transmission of SARS-CoV-2 before and after school closure/re-opening for single geographical units. This approach controls for confounding from population sociodemographic factors.
Of these, two studies sought to adjust for other NPIs.
Hsiang et al.22 (low risk of bias) used a reduced form of econometric regression to compare changes in incidence in six countries (China, France, Iran, Italy, USA and South Korea) before and after NPI implementation. Other key NPIs and testing regimes were adjusted for. Effect sizes for school closures were only reported for France, Italy and the USA; but (and without explanation) not for China, Iran or South Korea. The authors report a null effect of school closures on growth rate of SARS-CoV-2 incidence, with narrow confidence intervals for France and the USA, but a regression coefficient suggestive of a non-significant preventative effect in Italy (−0·11 (95% CI −0·25, 0·03)).
Neidhofer et al.18 (serious risk of bias) used a difference in difference comparison to estimate reduction in deaths in the 18 days post-school closure in Argentina, Italy and South Korea; compared with synthetic controls derived from the weighted average of epidemic curves from countries that closed schools later. This method indirectly adjusted for some confounders by selecting the most comparable countries with regards to both sociodemographic features and the number of SARS-CoV-2 deaths at the time of closure (Argentina 2, South Korea 22, Italy 80). The authors reported a 63%-90% reduction, 21%-35% reduction, and 72%-96% reduction in the daily average COVID-19 deaths in Argentina, Italy and South Korea respectively. The small number of cumulative deaths in Argentina and South Korea at the start of the study period made reliable extrapolation of mortality trends to inform the control units unlikely.
The other three studies did not analytically adjust for other NPIs.
Matzinger et al.17 (moderate risk of bias) identified the three US states which introduced school closures first, and with a sufficient lag before implementing other measures to assess their specific impact. They plotted incidence rates on a log2 scale and identified points of inflexion in the period after school closure. This assumes exponential growth in the absence of interventions, which may not have occurred given changes to testing regimes. The doubling time of new cases in Georgia slowed from 2·1 to 3·4 days one week after closing schools. Similar results were observed in Mississippi (1·4 to 3·4 days) and Tennessee (2·0 to 4·2 days). The authors also noted inflexion points for hospitalisations and mortality, although numerical changes were not reported. Tennessee showed a slowing in hospitalisations after one week, and deaths another week later; whereas Mississippi shows a slowing of both at the same time (after one week) – the authors do not comment on this discrepancy. Georgia lacked early hospitalisation data to make such a comparison. Iwata et al.15 (serious risk of bias) used time series analysis with Bayesian Inference to estimate the effects of school closures on SARS-CoV-2 incidence in Japan, reporting a null effect. Whilst growth in cases was observed during the study period, the number of cases remained low (<100 cases per day). Publicly available data24 shows implementation of mass gathering bans occurred with school closures, and foreign travel bans were already in place. Stein-Zamir et al.21 (critical risk of bias) reported an age-stratified epidemic curve of SARS-CoV-2 incidence in Jerusalem, with identification of the timing of school closures and re-openings. They show a large reduction in incidence starting one week after schools closed, with proportional reductions across all age groups; and a resurgence in case numbers around two weeks after schools were gradually re-opened, predominantly driven by younger age groups. There is no adjustment for other NPIs, though school closures were implemented alongside mass gathering bans and other social distancing rules; whilst school re-openings coincided with lifting hospitality and retail restrictions, and relaxing mass gathering bans25. Mass testing of a single secondary school was undertaken as part of an outbreak investigation, and the sharp increase in the number of new cases amongst young people is almost entirely accounted for by the cases identified by this.
Pooled multiple-area before-after comparisons
Three studies13,14,20 reported data on multiple geographical units, and then pooled the results into one unified estimate of effect using regression analysis.
One study had a low risk of bias and reported a null effect. Courtemanche et al.14 used a fixed effects model (which accounts for inter-area sociodemographic differences) to estimate the effect of school closures on SARS-CoV-2 incidence in US counties. They adjusted for relevant NPIs and testing regime confounders, and reported a null effect of school closures on growth rate applying a lag of either 10 or 20 days.
Two studies had a higher risk of bias due to a lack of adjustment for confounding NPIs, and reported preventative effects.
Auger et al.13 (moderate risk of bias) used interrupted time series analysis to calculate the rate of change in SARS-CoV-2 incidence and mortality in US states, and then used negative binomial regression to combine effect sizes into one pooled national estimate. Stepwise regression was used to build models, excluding covariates with P values >0·20, resulting in exclusion of several NPIs and testing regime data from their models. They estimated that school closures reduced incidence and mortality by c.60%.
Yehya et al.20 (serious risk of bias) also used negative binomial regression to combine the observed effects in US states, with COVID-19 mortality as the outcome measure. Relevant sociodemographic differences between states were accounted for as confounders in the multivariable model. However, they did not adjust for the effect of other NPIs. They estimated that school closures reduced COVID-19-related deaths by 5% per day.
Pooled multiple-area cross-sectional comparisons
Two studies16,19 considered countries from around the world in a cross-sectional design in which NPIs were considered as binary variables on a specific date: in place or not in place, and the cumulative incidence to that point was compared to the number of new cases of COVID-19 over a subsequent follow-up period; countries were then compared using regression analysis to elicit independent effect sizes for individual policies including school closures. This approach reduces bias from different testing regimes over time and between countries. However, the use of a single cut-off date for whether school closure was in place means that that the effects of long-standing and more recent school closures were pooled. Both studies reported preventative effects of school closures on SARS-CoV-2 incidence (Juni et al.16:23% relative reduction in the incidence rate, Wong et al.19: 50% relative reduction).
Juni et al.16 (serious risk of bias) used an exposure cut-off date of 20 March 2020 with a ten-day lag period and seven-day follow-up period. The authors adjusted for a comprehensive set of sociodemographic and geographical confounders (see Table 3) but did not adjust for the effect of other NPIs because they were implemented around the same time as school closures. Wong et al.19 (serious risk of bias) used a cut-off data of 31 March with a 14-day lag period and a 14-day follow-up period. The authors only adjusted for potential sociodemographic confounding from gross domestic product and population density. The authors did adjust for the presence of other NPIs using the Stringency Index, but this does not include relevant measures such as social distancing rules or mask wearing.
Discussion
We identified ten studies that provided a quantitative estimate of the impact of school closures on community transmission of SARS-CoV-2. The studies spanned a range of countries and were heterogenous in design. Findings ranged from no association to a 62% relative reduction in incidence and mortality rate13. The studies at lowest risk of bias reported no association14,22 (figure 2), whilst those with a higher risk of bias generally reported preventative effects. An exception was a paper by Matzinger et al.17 which focused on US states that implemented school closures first and without co-interventions, and reported a two-fold increase in the time for cases to double one week after school closures.
A major challenge with estimating the ‘independent’ effect of school closures is disentangling their effect from other NPIs occurring at the same time. Most studies tried to account for this, but it is unclear how effective these methods were. In direct correspondence one author reported that adjustment for other NPI was not possible due to clustering.16Even where adjustment occurred there is a risk of residual confounding, which likely overestimated preventative associations; and collinearity (highly-correlated independent variables meaning that is impossible to estimate specific effects for each) which could bias results towards or away from the null. Four studies did not specifically seek to estimate an effect size for school closures, instead studying school closures as an example of NPIs. These studies may not have specified the model in an optimal way to estimate effects of school closures. The divergent results for the USA, highlight these problems and may suggest that methodological differences are an important cause in the variation of the findings.
The strength of this study is that it draws on empirical data from actual school closures during the COVID-19 pandemic and includes data from 146 countries. By necessity, we include observational rather than randomised controlled studies, as understandably no jurisdictions have undertaken such trials. We were unable to meta-analyse due to study heterogeneity. We were unable to examine differences between primary and secondary schools as no studies distinguished between them, despite the different transmission patterns for younger and older children.
The studies are not able to distinguish between the direct and indirect effect of school closures. Indirect effects might include parents staying at home (reducing workplace contacts), and the signalling effect that closing schools sends to the general population to be cautious and reduce social contacts. Whilst some studies reported effects on mortality, it was not always clear whether the specified timeframe was appropriate. Interventions affecting children would be expected to have a longer lag than other interventions: to allow time for impacts on infections in older adults and ultimately mortality. Data are also lacking from low-income countries, where sociocultural factors may produce different effects of school closures on transmission to high income settings, leaving a substantial gap in the evidence base.
Our estimates describe the impact of school closures policies early in the year. School re-opening, with substantial infection prevention measures in place, may have a very different effect on community transmission. Where school re-openings have occurred but other NPIs have remained, less biased estimates of effect may be possible. Data from school holidays should also be considered for future high-quality natural experiment studies. In addition, none of the included studies used mobility or genomic sequencing of viral strains which may have allowed for a mechanistic understanding of how school closures effect community transmission patterns.
The variability in findings from our included studies are likely to reflect issues with study design. However, this may also suggest that there is no single effect of school closures on community transmission and that contextual factors may modify the impact of closures in different countries and over time. If the purpose of school closures is reduction in social contacts among children, the level of social mixing between children that occurs outside school once schools are closed is likely to be a key determinant of their effect at reducing community transmission. This will be influenced by other NPIs, and other key contextual factors including background prevalence of infection, age of children affected, as well as sociodemographic and cultural factors.
Different countries have adopted different approaches to controlling COVID-19. In the first wave of the pandemic school closures were common, and in some places one of the first major social distancing measures used. In contrast, the UK Government’s strategy for managing the second wave has prioritised keeping educational institutions open. With such varied findings and quality of evidence on the effect of school closures on limiting community transmission of SARS-CoV-2, and given the harmful effects of school closures3,4, policymakers and governments need to take a measured approach before implementing school closures in response to rising infection rates.Other evidence, such as the harms of school closures and transmission patterns in children should be considered alongside the evidence presented here when making decisions about school closures. Less damaging measures such as effective test, trace and isolate regimes in schools, as well as enhanced hygiene and social distancing measures should be considered as alternatives to school closures. This work also underscores the need for a robust and systematic approaches to the evaluation of all interventions deployed in a pandemic, not just those readily amenable to randomisation.
Data Availability
All included articles are publicly available.
Author Contributions
SW, CW, CBo, RV and OM designed the review protocol. SW, AC, SR and VB screened articles for inclusion, assessed risk of bias, and performed data extraction. SW and OM drafted the manuscript. All authors commented on the final manuscript.
Role of the funding source
There is no direct funding for this study. The funding bodies who support the researchers involved in this work had no role in study design, data collection, data analysis, data interpretation, or writing the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.
Declaration of interests
The authors declare no conflicts of interest
Acknowledgements
Appendix A Search Strategy
Search date: 12/10/20
PubMed
Search Title/Abstract:
(coronavirus[mh] OR Coronavirus Infections[mh] OR coronavirus*[tw] OR “COVID-19”[tw] or “2019-nCoV”[tw] or “SARS-CoV-2”[tw]) AND (Schools[mh:noexp] OR schools, nursery[mh] OR “Child Day Care Centers”[mh] OR “Nurseries, Infant”[mh] OR school*[tiab] OR preschool*[tiab] OR “pre-school*”[tiab] OR nurser*[tiab] OR kindergarten*[tiab] OR “day care”[tiab] OR daycare[tiab] OR “education setting*”[tiab] OR “educational setting*”[tiab] OR NPI*[tiab] OR “non-pharmaceutical intervention*”[tiab])
Web of Science
TS=(coronavirus* OR “COVID-19” OR “2019-nCoV” OR “SARS-CoV-2”)
AND
TS=(school* OR nurser* OR preschool* OR “pre-school*” OR kindergarten* OR “day care” OR daycare OR “education setting*” OR “educational setting*” OR NPI* OR “non-pharmaceutical intervention*”)
Scopus
TITLE-ABS-KEY ((coronavirus* OR “COVID-19” OR “2019-nCoV” OR “SARS-CoV-
2”) AND (school* OR nurser* OR preschool* OR “pre-school*” OR kindergarten* OR “day care” OR “daycare” OR “education setting*” OR “educational setting*” OR NPI* OR “non-pharmaceutical intervention*”)) AND (LIMIT-TO (PUBYEAR, 2020))
CINAHL (via HDAS)
((coronavirus* OR “COVID-19” OR “2019-nCoV” OR “SARS-CoV-2”) AND (school* OR nurser* OR preschool* OR “pre-school*” OR kindergarten* OR “day care” OR “daycare” OR “education setting*” OR “educational setting*” OR NPI* OR “non-pharmaceutical intervention*”)).ti,ab [DT 2020-2020]
WHO Global COVID-19 Research Database
(tw:(school*)) OR (tw:(nurser*)) OR (tw:(“pre-school*”)) OR (tw:(preschool*) OR (tw:(kindergarten*)) OR tw:(“day care”) OR tw:(“daycare”) OR tw:(“education setting*”) OR tw:(“educational setting*”) OR tw:(NPI*) OR tw:(“non-pharmaceutical intervention*”))
Including: WHO COVID Database, MedRxiv. Title, abstract, subject. 2020.
ERIC
Coronavirus OR “COVID-19” or “2019-nCoV” or “SARS-CoV-2”
British Education Index
Coronavirus OR “COVID-19” or “2019-nCoV” or “SARS-CoV-2”
Australian Education Index
Coronavirus OR “COVID-19” or “2019-nCoV” or “SARS-CoV-2”
Grey Literature Search, Google
First 100 hits on google search, limiting to PDF files, up to ‘last year’. Search: “COVID-19” OR “coronavirus” OR “school” OR “education”