Low-quality SRs

Chaabane et al (Unique primary studies: 100%) reported that school closures may have reduced paediatric hospitalisations.26 Nussbaumer-Streit et al’s SR (Unique: 67%) reported that adding school closures onto other measures, such as mandatory quarantine for infected students, may have reduced transmission over quarantine alone.23 Walsh et al (Unique: 64%) reported uncertain findings on school closures: half of the included primary studies at lower risk of bias found reduced community COVID-19 transmission on closing schools, while the other half reported that school closures were associated with no change in transmission. This SR further reported that school reopening was mostly not associated with increased community transmission, when community transmission was low and in-school mitigations were in place.18

Talic et al (Unique: 60%) reported that school closures were largely effective. Primary studies included in this SR were inconsistent on the efficacy of school closures, although most supported reduced community COVID-19 transmission on school closure.21 Ayouni et al’s SR (Unique: 50%) reported that school closures were associated with reduced community transmission, but also that the specific effects of school closures were difficult to disaggregate, given the other NPIs introduced at the same time.20

Critically low-quality SRs

The National Collaborating Centre for Methods and Tools (NCCMT) (Unique: 82%) reported little evidence that reopening schools increased transmission when mitigations, such as mask-usage, were in place.17 Caini et al (Unique: 87%) found that children may be less likely to transmit COVID-19 than adults, leading to limited transmission in schools.19 In meta-analyses of observational studies, Caini et al reported that onward transmission from school pupils was less likely than from adults (OR 0.26, 95% CI (0.11 to 0.63)) and school pupils may be less likely to be infected with COVID-19 (OR 0.60, 95% CI (0.25 to 1.47)).19

Viner et al 24 (Unique: 67%) is an SR from very early in the pandemic (the latest search was 19 March 2020), but predicted—from primary modelling studies—a very small beneficial effect of school closures on community COVID-19 mortality.24 The Muhammed SR reported that school closures may have been effective in reducing COVID-19 transmission (Unique: 63%). Effectiveness appeared to be dependent on infection levels in the community, and the time that closures were introduced: earlier interventions, when community levels remained relatively low, were reported to be more effective.22

Mendez-Brito et al’s SR (Unique: 60%) concluded that school closures may be the most effective NPI, and were more effective when introduced earlier: 58% of included primary studies reported reduced transmission on school closure.25 Finally, Suk et al included few unique studies (Unique: 25%), but reported that school children were rarely the index case for subsequent household transmission. Nevertheless, this SR reported that school closures may have reduced COVID-19 transmission.27

Taken together, the evidence suggests that school closures may have reduced community COVID-19 transmission, morbidity and mortality. A limitation of these data is that only one of the SRs performed quantitative meta-analysis. In most of the SRs, most of the included primary studies were unique (online supplemental table 3). The quality of the evidence was variable, with all studies of low or critically low quality (table 1), although exclusion of the SRs with critically low quality does not change the overall conclusion. The GRADE certainty for the evidence in transmission, morbidity and mortality is all very low (table 2).

Low-quality SRs

Chaabane et al (Unique: 100%) reported that students from lower socioeconomic backgrounds, and those with disabilities, may have fared worse with online learning, due to reduced access to the internet and technology at home.26 Bond et al’s SR (Unique: 96%) reported that some students were less engaged, and attendance was reduced, for virtual teaching at home—potentially due to social isolation. Furthermore, it appeared that some pupils lacked the technical skills required for the virtual learning to be effective.30

Critically low-quality SRs

Hammerstein et al (Unique: 91%) found loss of learning on closing schools, particularly affecting younger pupils and pupils from low socioeconomic backgrounds. The estimated learning loss was in the range of −0.005 to −0.05 SD per week of closures.28 Similarly, Panagouli et al (Unique: 69%) reported that loss of learning on school closures was more severe in younger pupils and pupils with special educational needs (SEN). However, the SR also reported evidence that some students appeared to gain more from virtual learning than classical classroom teaching.29

Together, these SRs suggest that learning loss occurred during school closures, and SEN students, as well as those from lower socioeconomic backgrounds, appeared to be most affected. It appears that some students engaged less with virtual teaching methods than in-person methods, and virtual teaching may have exacerbated inequalities. There was minimal overlap between primary studies included in the SRs (online supplemental table 3), but the studies were of low or critically low quality (table 1), and the GRADE certainty for this evidence base is low (table 2).

Low-quality SRs

Chaabane et al (Unique: 25%) reported increased anxiety and loneliness among students during school closures.26 Samji et al (Unique: 89%) reported that female students were worse affected, as well as older students and pupils with neurodiversity. Nevertheless, effects on mental health could be mitigated, including if students exercised more at home or had better social support networks.34

Viner et al 35 (Unique: 60%) reported increased anxiety and reduced well-being during school closures, but no change in suicide rate.35 Chai et al performed a meta-analysis of cross-sectional studies (Unique: 42%), reporting that 28% (95% CI (22% to 34%)) of Chinese students experienced mental health problems during school closures for COVID-19 – compared with best estimates (in a different population) of 17.6% (95% CI (17.4% to 17.9%)) before the pandemic.37 38 Meherali et al (Unique: 23%) reported worse effects on mental health among female students, as well as among those who spent more time on social media—usage of which also appeared to increase during school closures.33

Critically low-quality SRs

Lehmann et al (Unique: 80%) reported worsening behaviour and hyperactivity among students, and noted that students’ mental health during school closures became worse if their parents were suffering from stress. Lehmann et al caveat that unequivocal conclusions were difficult to draw, since multiple NPIs were often introduced simultaneously.36

Zhang et al’s meta-analysis of cross-sectional studies (Unique: 64%) reported that the prevalence of anxiety among Chinese students during the COVID-19 pandemic was 24% (95% CI (20% to 29%)). In subgroup analyses, Zhang et al reported that anxiety appeared to be worse in the diffusion attenuation phase of the pandemic (42%, 95% CI (35% to 50%)) than in the outbreak phase (25%, 95% CI (17% to 34%)).31 Although they did not include any primary studies unique from the above SRs, Elharake et al (Unique: 0%) also reported a decline in mental health during school closures, and reported that risk factors for this included low socioeconomic status, and having family members who work in healthcare.32

Overall, these SRs suggest that COVID-19 school closures may have increased the prevalence of mental health problems—most notably anxiety—among children. A limitation of these data is that school closures were not applied as a solitary restriction; some effects on mental health may therefore also be effects of other restrictions, or from the wider context of the COVID-19 pandemic. There was moderate overlap between primary studies, and one SR (Elharake et al) presented zero unique primary studies (online supplemental table 3). The studies were of low or critically low quality (table 1), and the GRADE certainty for this evidence base is very low (table 2).

Low-quality SRs

Chaabane et al (Unique: 67%) reported that school closures were associated with an increase in body mass index (BMI) and obesity, with predictions showing that longer closures would be associated with larger increases.26 Similarly, Viner et al 35 (Unique: 88%) reported that school closures were associated with weight gain in children, as well as with reduced exercise, increased sedentary behaviour and increased unhealthy food consumption.35

Critically low-quality SRs

Chang et al’s meta-analysis of non-randomised studies (Unique: 100%) reported that the average BMI of pupils increased by 0.77 points (95% CI (0.33 to 1.20) p=0.0006), rates of obesity increased by 1.23-fold (95% CI (1.10 to 1.37) p=0.0002), and the average increase in bodyweight was 2.67 kg (95% CI (2.12 to 3.23) p<0.00001) (noted by the authors to be a greater increase than normal average growth), following COVID-19 lockdowns.39

Overall, these SRs suggest that school closures may have increased BMI and obesity among pupils—potentially due to reduced exercise coupled with a less healthy diet at home, where students did not have access to physical education teaching or healthy school meals. Given that school closures were not applied on their own, these changes may also reflect the wider context of the pandemic. There was little overlap between primary studies (online supplemental table 3), but the SRs were of low or critically low quality (table 1), and the GRADE certainty for this evidence base is very low (table 2).

Low-quality SRs

Most of the 10 primary sleep studies included in the Viner et al 35 SR (Unique: 60%) reported reduced sleep quality during school closures, with many students developing new sleep problems and fewer students sleeping through the night.35 Although they only included one primary study on sleep, which was not unique, Chaabane et al (Unique: 0%) concluded that sleep timing (but not sleep quality) was affected during school closures.26

Critically low-quality SRs

Sharma et al (Unique: 67%) conducted a synthesis of nine primary studies and reported that, in general, students slept later and woke later during school closures. They also reported that some students had better sleep quality during school closures, but three-times this number experienced a decline in sleep quality.40 Meta-analysis found that 54% of pupils (95% CI (50% to 57%)) had sleep disturbance, and 49% (95% CI (39% to 58%)) did not achieve recommended sleep quantities during the pandemic.40

Although the quality of the SRs and their findings varied, there appears to be a trend to suggest that sleep quality reduced during school closures. This may be due to a mixture of increased anxiety and reduced physical activity. Nevertheless, there was some evidence that a subgroup of students reverted to a more natural sleeping routine, with many sleeping later and waking later. These changes could reflect school closures directly, but cannot be separated from the potential wider context of the COVID-19 pandemic. There was moderate overlap between primary studies, and one SR (Chaabane et al) presented zero unique primary studies (online supplemental table 3). The SRs were of low or critically low quality (table 1), and the GRADE certainty for this evidence base is very low (table 2).

Low-quality SRs

Viner et al 35 (Unique: 67%) described a consistent reduction in reports of child abuse during school closures, although no findings on incidence of abuse.35

Critically low SRs

Kourti et al (Unique: 80%) described that, around the world, reported cases of domestic violence against children reduced during school closures. Despite this, some studies suggested increased incidence of abuse, including increased numbers of children presenting to healthcare centres with abusive head trauma. Kourti et al suggested that this may be due to co-quarantine of perpetrators and victims.41

Together, these SRs suggest that reported cases of domestic violence may have reduced during school closures, but actual cases may have increased. There was little overlap between the primary studies in the SRs (online supplemental table 3). However, the studies were of low or critically low quality (table 1), and the GRADE certainty for the evidence base is very low (table 2).

High-quality SRs

Krishnaratne et al (Unique: 100%) reported that in-school mitigations, such as mask-wearing and isolation of positive cases, appeared to be effective in reducing community COVID-19 transmission, hospitalisation and mortality.42

Critically low-quality SRs

The NCCMT (Unique: 100%) found that mask-usage, social distancing, restricting school entrance to non-staff/non-students, stopping extracurricular activities, teaching outdoors and screening for symptoms of COVID-19, all appeared to reduce transmission in schools.17 Vardavas et al (Unique: 100%) also reported that the mitigations introduced in schools appeared effective in reducing transmission, with little transmission occurring in schools when these measures were in place.43

Overall, these SRs suggest that the mitigations implemented in schools, such as mask usage, may be effective in reducing school and community transmission of COVID-19. There was no overlap between the primary studies included in the SRs (online supplemental table 3). One SR was high quality, but the others were of critically low quality (table 1); the conclusions of the high-quality study (Krishnaratne et al) were consistent with those of the other studies. The GRADE certainties for this evidence base—for the effect of in-school mitigations on transmission, morbidity and mortality—are all very low (table 2).