At variance with influenza, in which younger individuals seem to represent a reservoir of virus and contribute to its propagation to general population, [26, 27, 28, 29, 30] SARS-CoV-2 seems to spare school age children and adolescents: clinically, they are mostly paucisymptomatic [[5]]; from the epidemiology of infection perspective, they are very rarely accounted for as the index case [[11]], indicating that not only they are largely spared from the clinical consequences of the infection, but they also are less likely to transmit it. Overall, these data suggest that spread of COVID-19 within school settings may be limited [[31], [32]]. Indeed, our data indicate that infection incidence is lower in students of any education cycle, compared to the general population. Moreover, at least in the case of elementary school children, contact tracing in schools confirms that they are less likely to transmit the virus to adults, as evidenced by a 73% lower number of secondary cases among teachers when the index case is a student (10%), compared to secondary cases elicited by a teacher index case (37%). These epidemiological data are in line with the finding that children harbor antibodies against the other common coronaviruses, and that these antibodies are cross reactive and neutralizing against SARS-CoV-2 [[7]]. Our findings are also consistent with several other reports of very limited spread of COVID-19 between children and from children to adults. In Australia (New South Wales), following COVID-19 positivity of 9 students in primary and high schools and 9 staff members, only 2 of the 735 students, and 0 of the 128 staff members with whom they had contact were identified as secondary cases [[33]]. In Ireland, during the first wave, 6 COVID-19 cases were identified in schools (three children and three adults). Among their 1155 school contacts, zero infections were recorded [[34]]. In the Netherlands, 10 COVID-19 cases aged <18 had 43 contacts, but nobody was infected, whereas 221 patients older than 18 were associated with 8.3% of infections [[35]].
Of note, we found higher rates of incidence in teachers and non-teaching staff members compared to the general population. One possible explanation for this finding is that teachers might become infected at school because of their prolonged proximity to students. However, by judging from contact tracing activity in schools of the populous province of Verona (Veneto region), secondary infections at school are rare: only 13 teachers were identified as secondary cases from 524 traced index cases. Among these rare events, frequency of secondary infections among teachers was higher when the index case was a teacher rather than a student. In the Campania region, where schools were open for 17 days (from September 24 to October 16; school week of 5 days), incidence among teachers and non-teaching staff members in the period September 12-November 7 was still higher than that in the general population. It would be difficult to ascribe this difference to 17 days of school over a total of 56 days. We also performed an important, often overlooked normalization and compared incidence among teachers from the Veneto region with incidence in the general population of similar age: incidences were comparable, and differences not significant. Thus, while incidence among teachers is similar to that in the age—matched general population, teachers are allegedly perceived at greater risk. Perhaps this perception stems from the fact that in Italy the school environment is meticulously and continuously controlled, as confirmed by our finding of very high number of tests performed for each positive case, especially when the index case is a student. This remarkable system of monitoring unveils a large proportion of perhaps asymptomatic infections among teachers, resulting in the apparently higher incidence among this type of workers. It cannot be argued that teachers and non-teaching staff members are more susceptible to infection than the general population. In fact, this increase in the incidence of test positives is not mirrored by an increase in mortality-morbidity that would mark a more susceptible population [[36]]. In sum, our analysis of data collected by the MI indicates that in Italy students are less infected than the general population and the overall protocols for contact tracing work well, questioning whether schools played a role as amplifiers of the second COVID-19 wave.
Decision makers, popular press and public opinion in Italy ascribed the second wave of COVID-19 to school reopening [[14]]. This was often accompanied by deprecating comments on “individual behavior” of adolescents especially, who would not follow the strict rules at school or outside them. However, our data suggest that this common sentiment is not evidence-based, but perhaps grounded on the temporal correlation between school opening (in September) and second wave (in October-November). Rather, our data do not identify a constant temporal association between school reopening and rise in Rt analysed on a regional basis. Because of the staggered school reopening calendar in Italy, we were well positioned to address whether there was such an association between the date of school opening and the date of reproduction number increase. Conversely, a constant association was present when we analysed the temporal distance between Rt rise and the election day, held in Italy on September 20 (and morning of 21), 2020.
Interestingly, other reports are in line with our findings: in Great Britain, incidence among staff members was higher than among students (27 cases [95% CI 23–32] per 100,000 per day among staff; 18 cases [14–24] in early-year students, 6.0 cases [4.3–8.2] in primary schools students, and 6.8 cases [2.7–14] in secondary school students); further, most cases linked to outbreaks were in staff members (154 [73%] staff vs. 56 [27%] children of 210 total cases). The median number of secondary cases in outbreaks was one (IQR 1–2) for student index cases and one (1–5) for staff index cases [[37]]. In Spain, the evolution of the global incidence does not suggest significant effects of school reopening. In most cases, there was slight if any increase in pediatric cases, consistent with the diagnostic efforts in schools [[38]]. In Germany, data collected from 53,000 schools and day-cares in autumn indicate that only circa 32 schools had more than two positives per week [[39]]. Finally, a recent report by the ECDC summarizes the available knowledge and reaches conclusions very similar to ours. While ECDC concludes with high confidence that transmission of SARS-CoV-2 can occur within school, they also note with moderate confidence that prevalence of COVID-19 within schools is influenced by the community prevalence especially when community transmission is sustained. Most importantly, transmission in schools account for a minority of all COVID-19 cases in a given country and school staff are generally at no higher risk of infection than other occupations [[40]]. ECDC recommends a variety of NPI to mitigate the risk of school COVID-19 transmission [[40]] that are even less stringent than the rules currently implemented in Italy. For example in Italy children from 6 years of age must always wear face masks at school including when sitting at their desk or playing in outdoor playgrounds [[22]], irrespective of the local epidemiological condition that WHO [[41]] and ECDC [[40]] take into consideration when advising on schools NPIs.
A current concern is that the SARS-CoV-2 variant B.1.1.7, becoming largely diffuse and predicted to display a greater Rt 42, might be more transmissible especially among children It shall be noted that the possibility that this variant become predominant because of a greater susceptibility of school age individuals (0–19) was duly took into consideration. However, modeling predicts that individuals of this age group should be twice as susceptible to the B.1.1.7 variant as compared to the wild-type virus to support its observed widespread diffusion [[42]]. Furthermore, transmission of this variant by school age individuals appears to be lower also in the real world. The most recent Public Health England report on the transmissibility of the variants of concern contains datasets of contact tracing activity performed on individuals infected with wild-type and B.1.1.7 SARS-CoV-2. The report concludes that transmissibility of B.1.1.7 is 30–35% higher than that of wild-type SARS-CoV-2 [[43]]. From this report, we extrapolated secondary infection rates stratified by age of the index case (0–19 or 20+). In the case of 0–19 years old index cases carrying wild-type SARS-CoV-2, secondary cases were reported in 279 of the 3479 contacts (8.0%) and in 317 of the 3004 contacts of an index case carrying the B.1.1.7 SARS-CoV-2 variant (10.6%). These proportions were respectively 14.1% (891 secondary cases out of 6298 contacts) and 19.7% (968 out of 4920) when the index case was 20 years and older. Thus, the increase in transmissibility of the B.1.1.7 SARS-CoV-2 variant is 39.7% if the index case is 20 years or older, and 32.5% if the person is 0–19 years old. Even with this variant, transmission by school age individuals remains therefore 46% lower than by older persons. Thus, while we were not able to investigate the role of school opening and closure in a time of widespread diffusion of B.1.1.7 SARS-CoV-2 variant, these real-world data on lower transmissibility by school age individuals support that again, 0–19 years old individuals are less prone to transmit it forward than adults.
A different question is whether closing schools is efficacious in curtailing viral spread. In some Italian regions analysed here, school closure was mandated by local authorities and eventually in certain regions by the National Government. However, this closure had no effect on the incidence of COVID-19 in the general population or in Rt decline, which had started before the mandated school closure and that continued with the same speed, irrespective of school closures in Lombardy (partial) and Campania (total). This finding is in line with a literature review of all available studies (n = 16) on the efficacy of school closures and other social distancing practices in schools in China and Hong Kong, where the rapidly implemented school closures did not substantially contribute to the control of the spread [[16]]. In Australia, by comparing data from 25 schools of different grades with those of the general population, it was found that students and school staff did not contribute to the spread of the virus more than the general population [[44]]. On the other hand, an analysis of the impact of different NPI on the reproduction number Rt across 131 countries found that school closures alone could reduce Rt by 15% (R ratio: 0.85, 95%CI: 0.66–1.10), whereas school reopening could increase it by 24% (R ratio: 1.24, 95%CI: 1.00–1.52) twenty-eight days after their implementation. However, these measured Rt changes are not statistically significant, as evidenced by the very large and overlapping confidence intervals of the R ratios [[45]]. Moreover, authors warn on the limitations of their estimates: for example, they could not consider the different precautions related to the reopening of schools taken by some countries, such as physical distancing within classrooms and masking procedures; they did not consider the impact of school holidays and the effect of reopening different school levels (e.g., elementary and middle schools). Finally, authors analysed the impact of given NPIs by comparing Rt from two arbitrarily drawn periods before and after the implementation of the given NPI [[45]]. While this approach might be more practical when comparing multiple countries, it is less informative than our analysis, performed over the whole Rt curve.
In our analyses, Rt started declining even before the implementation of any NPI in all regions analysed. These results, while perhaps surprising, are in line with findings from the group of Merler [[46]] who analysed the impact of the national March-May strict lockdown on Rt in Italy. While they concluded that this lockdown reduced Rt and brought it below 1, they admitted that the decline in Rt had started well before the national lockdown was implemented. Indeed, visual inspection of their published Rt curves confirms that this NPI did not affect the slope of Rt decline. Whether our findings can be generalized to other countries, in which the use of NPI might be less extensive than in Italy, remains unclear and admittedly requires further studies.
Of the highest importance, our study is strengthened by the several sources of data used. Longitudinal data of regional incidence of SARS-COV-2 positives subjects deposited in the public repository of the Italian Civil Protection, incidence from the Veneto Region system of COVID-19 case notification with information by age, and incidence in schools from MI with information for students, teachers and non-teaching staff members. A systematic review investigated sources of bias in observational studies trying to assess the role of school closures in the reduction of COVID-19 community transmission [[47]]. Several studies were found at risk of confounding factors and collinearity from other NPI implemented around the time of school closures. We believe that our study is a low risk of bias because we compared community transmission of SARS-CoV-2 before and after school closure/re-opening in single geographical units (regions and provinces). This approach, as commented by the authors of this review, controls for confounding from population sociodemographic factors [[47]]. We also compared transmission in different regions opening schools at different dates and this analysis is not confounded by inclusion of other NPIs because while school calendar in Italy is regionalized, NPIs are mandated nationwide, in schools and outside schools. Furthermore, we analysed several prospective cohorts. This type of study design reduces the risk of bias, as opposed to the cross-sectional study design of previous publications on this topic that analysed data at a single cut-off date. Indeed, Walsh and colleagues essentially conclude that while most studies show effects, higher quality studies tend not to [[47]], probably a consequence of the strong study design in the latter.