The relation between school holidays and transmission of influenza: modelling analysis of seasonal influenza data from England and Wales

Abstract

BackgroundSchool closure is often regarded as a possible intervention during influenza outbreaks. Previous modelling studies have fitted transmission models to seasonal influenza time-series data from several countries and concluded that school holidays reduce transmission, but whether these findings occur elsewhere is unclear. We aimed to estimate how school holidays affect influenza transmission in England and Wales. MethodsWe quantified influenza transmission with the contact parameter—ie, the per-head rate of effective contact (contact sufficient to allow transmission if it occurs between a susceptible and an infectious individual). We used two methods to quantify the effects of school holidays on the contact parameter for influenza, on the basis of GP consultation data from England and Wales (1967–68 to 2007–08) for periods during which consultation rates for influenza-like illness were above an established threshold for denoting influenza circulation. First, we fitted a susceptible–infectious–recovered (SIR) model to the data (all age groups combined) for each influenza year to estimate the percentage by which the contact parameter differed during school holidays compared with during term time, for different assumptions about the infectious period. The proportion of infections reported, the number of infectious individuals present at the start of each season, and the proportion of the population that was susceptible at the start of each season were also estimated by fitting to the data. 90% plausible ranges for the change in the contact parameter were estimated by bootstrapping. Second, we estimated weekly values of the contact parameter, separately for 5–14-year-olds and all ages combined, with a formula derived from simple mass action models (the estimated number of infectious individuals divided by the product of the estimated number of susceptible and infectious individuals in the previous week). We calculated the percentage difference between mean values for holidays and term time in each year and explored its sensitivity to the proportion of infections reported and the proportion of individuals who were susceptible at the start of each season. Heterogeneity in the estimates was assessed with the I2 statistic and, when appropriate, estimates were summarised by random effects meta-analysis. FindingsFitting to the consultation data produced heterogeneous yearly estimates of the effects of school holidays on influenza transmission. The difference in the contact parameter comparing school holidays with term time ranged from a 47% reduction (90% range 45–48) to a 71% increase (66–76). Meta-analysis of the estimates from mass action models suggested that the contact parameter among 5–14-year-olds was 17% (95% CI 9–25) lower during holidays than during term time, with some heterogeneity. Estimates for individual years ranged from an 84% reduction to a 41% increase, and were robust to changing assumptions about the proportion of infections reported and the proportion susceptible at the beginning of each influenza season. There was no summary evidence that the contact parameter for all ages combined differed between holidays and term time. The changes in the contact parameter estimated with the two methods were correlated (correlation coefficient 0·6–0·7, dependent on the assumed duration of the infectious period). InterpretationSchool holidays were associated with reductions in contact between school-aged children, although results varied between years. The size of the reduction was similar to that estimated in previous studies. Estimates did not depend on the assumption about the proportion of infections reported, suggesting that secular changes in consultation behaviour are unlikely to affect results. Limitations include the absence of age structure in the transmission model and the absence of control for other factors (eg, humidity) that might affect transmission. School closure could be a useful intervention during a pandemic. FundingCJ was supported by a Research Training Fellowship from the NIHR for this research.