This systematic review of the effects of school closures on influenza outbreaks extends previous reviews2
to include published experiences from the 2009 pandemic. The results suggest that school closure can reduce transmission of pandemic12
influenza among schoolchildren. Many datasets, however, show no clear effect of school closure. As noted by some authors,20
this may sometimes have been because schools shut late in the outbreak (often close to or after the peak).
In some studies, incidence increased when schools reopened.5
This apparent reversibility provides evidence that school closure can cause reductions in influenza incidence. In two of the studies of seasonal influenza which showed reversibility,7
no additional interventions (beyond usual seasonal interventions) were used. In many other datasets, multiple interventions were used, so the specific effects of school closures are difficult to isolate.
In 2009, several countries closed schools while in others, planned holidays coincided with outbreaks. Several datasets from this pandemic strengthen support for school closure as an intervention; however, others illustrate that benefits are not guaranteed and that timely closure may be challenging. The sensitivity of the 2009 pandemic to school closures probably reflects the high attack rates in children compared with adults; outbreaks in which children are less affected might be less sensitive to school closure.
Studies presenting age-stratified data suggested that the effects of school closure on transmission were greater among children than adults. Few studies stratified children further, for example, into primary and secondary school students. Older children might socialise more than younger children during school closures, so closing primary schools may have a greater effect on transmission than closing secondary schools (eg, in Hong Kong in 2009, primary schools were closed proactively while secondary schools closed if cases occurred among their students12
The long-term effects of closing schools are unclear, as relatively few studies presented substantial data after schools reopened. For example, school closure could result in multiple peaks, potentially involving more cases than would otherwise have occurred.8
However, a study published since this review was conducted estimated that case numbers in Alberta, Canada, could have been up to twice as high as those seen if schools had not closed for planned holidays.66
It is difficult to compare reactive versus proactive closures, different durations of closure and local versus national closures as studies typically differed in several respects.
Some studies have concluded that reopening schools after holiday periods can accelerate epidemic growth (eg during the 1957 67
pandemics). These studies were beyond the scope of this review of the effects of closing schools during outbreaks, but they suggest that extending school holidays might delay the spread of an epidemic beginning during a break.
Results from analyses of seasonal influenza may not be directly applicable to a pandemic. Schools were often closed for planned holidays rather than in response to the outbreaks; contact patterns may differ between reactive school closures70
Extrapolating from previous pandemics may also be problematic. Modelling studies72–74
have predicted that school closures will have the greatest effects if transmission occurs mainly among children. The importance of children in transmission has varied between pandemics;75
in 2009, attack rates were higher in children than in adults, probably because of pre-existing immunity in older individuals.76
Viral virulence will also influence individuals’ responses to school closure and other interventions, for example, spontaneous social distancing during a mild pandemic may be less dramatic than occurred in 1918. Changes in household size, contact patterns, children's behaviour and school systems since 1918, 1957 and 1968 may also limit the generalisability of experiences from these pandemics. As noted in a study of the 1918 pandemic in Connecticut, reverse causality may occur when comparing rates in cities which closed schools to those in cities which did not, if closure was a response to a particularly severe local outbreak.10
One limitation of the datasets is that ascertainment may have changed during the outbreaks, due to changes in surveillance and care-seeking behaviour. Increases in ascertainment during an outbreak could obscure any reductions in incidence during school closures (eg in one study, enhanced surveillance began the day the school closed55
). Conversely, the proportion of patients who undergo virological testing may be reduced late in an outbreak, and in some settings (eg New Zealand14
) patients with ILI were discouraged from consulting GPs during the 2009 pandemic. The estimated proportion of influenza cases that were reported in Hong Kong declined to ~5% of its original value during the move from containment to mitigation during the 2009 pandemic.12
In England, the introduction of the National Pandemic Flu Service telephone helpline coincided with the school holiday, and was estimated to have reduced the probability of GP consultation for adults with ILI from 16% to 1.8%.22
Case definitions may not always have been well suited to detecting any effect of school closure. For example, school absenteeism is a relatively non-specific measure, while laboratory specimens frequently represent severe infections (eg in the elderly, who may have little contact with children and therefore be relatively unaffected by school closure).
Influenza transmission is influenced by factors besides contact in schools, including temperature and absolute humidity (AH).77–80
Two studies which assessed the role of AH during the 2009 pandemic did not find strong evidence that it affected transmission.24
The two waves seen in the UK in 2009 could be explained by changes in contact patterns during school holidays.28
In a modelling study of data from Alberta, Canada, the best-fitting model included effects of temperature and school holidays on transmission, and predicted that if schools had not closed, the outbreak would have been restricted by temperature effects but would still have been 2.1 times larger than was observed in the province as a whole (1.38 and 1.54 times in the cities of Calgary and Edmonton, respectively).66
A study of the interplay between school calendars, AH and population susceptibility in enhancing influenza transmission concluded that high AH may prevent influenza outbreaks.78
However, if a sufficiently high proportion of the population is susceptible, outbreaks can occur even when AH is high; the opening of schools may enhance transmission.78
Taken together, these studies suggest that contact in schools is not the only determinant of influenza transmission, but it is one influential (and modifiable) factor.
Previous studies have estimated the effects of public-health interventions using transmission models.8
The development of such models is complicated for the datasets reviewed here, and would not necessarily have provided conclusive insight into the impact of school closures. For example, many factors are unknown and would need to be estimated or assumed for each dataset, such as the basic reproduction number, proportion of infections that were reported, effect of other interventions and the proportion of individuals who were immune at the start of the outbreak.
The review was limited to published studies, which could potentially introduce publication bias. However, many of the studies identified did not aim to evaluate the effects of school closure on transmission, so publication bias appears unlikely. This is supported by the apparent lack of an effect of school closure in many of the studies (including some of those which did specifically assess school closure as an intervention). A further limitation is that most papers were screened (and all data were extracted) by a single reviewer. Foreign language papers were excluded, but in most cases it was clear from the title and/or abstract (available in English) that the papers were not relevant to this review.