Effects of School Closures, 2008 Winter Influenza Season, Hong Kong

In winter 2008, kindergartens and primary schools in Hong Kong were closed for 2 weeks after media coverage indicated that 3 children had died, apparently from influenza. We examined prospective influenza surveillance data before, during, and after the closure. We did not find a substantial effect on community transmission.


The Study
We reviewed prospective surveillance data on infl uenza and infl uenza-like illness activity during the 2008 winter infl uenza season. We then considered the effects of the school closures on community transmission.
As Surveillance data from different settings before, during, and after the period of school closures are shown in the Figure. Laboratory isolation of infl uenza viruses in children (panel A) and adults (panel B) show that the infl uenza season began in January, rose to a peak in late February, and was already waning by the time the decision was made to close schools, as temperatures and relative humidity were increasing (8). Infl uenza circulation has remained at a low baseline level since schools reopened in early April. Absenteeism rates in sentinel childcare centers and primary schools gradually rose to maximums of 7.9% and 3.5%, respectively, before the school closures and returned to low levels after the closures (data not shown). Similarly, infl uenza consultation rates at public and private outpatient clinics (panel C) peaked before the closures and generally refl ected the reference laboratory data, except for a dip during Chinese New Year, when many sentinel practices were closed.
When compared with the infl uenza seasons of the preceding 9 years, the 2008 winter infl uenza season was moderately severe in terms of outpatient consultations (online Appendix Figure, (6). The elderly appeared to have been less affected, with no clear rate increases noted by febrile sentinel surveillance in elderly care homes and generally low infl uenza-related admission rates in this age group (data not shown).
Panel E of the Figure shows daily estimates of the effective reproductive number, or R t , based on a simple method (9) that we applied to daily interpolations of the laboratory and outpatient sentinel data. We used a Weibull model for the serial interval with mean of 3.6 days and standard deviation of 1.6 days, based on data from a recent community study (10). The effective reproductive number on day t can be interpreted as the average number of new persons infected by an infector who had symptom onset on day t. Therefore, a reproductive number >1 implies that an epidemic will grow in the short term, whereas a number <1 implies that an epidemic will die out. These trends, in particular the lack of any apparent negative infl ection point during the entire 2-week period of school closure, suggest that the effect of the intervention was not substantial. Trends in estimated R t were similar if serial intervals of mean 2.5 or 2.0 days were assumed.

Conclusions
Although we can only speculate, given the limitations of an uncontrolled natural experiment on the population level, routine surveillance data did not detect a large effect from the school closures. In particular, we noted a decline in laboratory isolations of infl uenza viruses that preceded the intervention and the lack of association between school closures and R t . In fact, sentinel data may not accurately represent the incidence of infl uenza in the underlying population because, for example, other cocirculating upper respiratory viruses contribute to overall infl uenza-like illness consultation rates. Laboratory data, however, should be less affect-ed, and extra testing in response to the heightened awareness of infl uenza activity might have artifactually lowered the positivity rate. The epidemic curves generated from the surveillance data showed a decline in cases that may have naturally concluded without any intervention. We note the diffi culty of making inferences directly from changes in epidemic curves because changes in the epidemic curve may lag behind changes in the underlying transmission dynamics by at least 1 serial interval, as has previously been shown for severe acute respiratory syndrome (9,11). Although the estimates of R t (panel E) are crude, the estimated values of 1.2-1.5 during the rising phase of the 2008 winter epidemic in Hong Kong are slightly lower than previous estimates for interpandemic infl uenza (12,13), perhaps because of the low time-dependent resolution of the weekly aggregation of surveillance data. We emphasize that our results must be interpreted with caution; in particular, infl uenza might have continued to circulate for a longer period had the school closures not been implemented. Furthermore, notwithstanding our tentative null fi ndings, some previous reports have demonstrated that school closures may be effective at mitigating infl uenza seasons. For example, a study showed signifi cant reduction in respiratory infections during school closures in Israel (14), and a recent model estimated that school holidays prevent 16%-18% of seasonal infl uenza cases in France (12).
We acknowledge that our assessment has the benefi t of hindsight, whereas at the time the decision was made to close schools it might well have been unclear from surveillance data that the infl uenza season was only moderate and might have already been in natural decline. Although daily hospital admissions data were available in real time from a new integrated computer system and therefore did show the decline, this system only refl ected serious illness. However, outpatient sentinel data, which are more indicative of overall infl uenza activity in the general community, were available with an ≈7-day lag; reports of laboratory reference data lagged even further. If public health decisions are to be made on the basis of prospective surveillance, these systems must be improved to refl ect real-time or near realtime reporting and analysis. One possibility in Hong Kong would be to use the wealth of data from rapid infl uenza tests in hospitals, now that >1,000 rapid tests are conducted every month on most newly admitted patients with pneumonia or respiratory symptoms. Furthermore, although most local surveillance data are aggregated ( Figure), the spread of infl uenza likely varies according to population subgroup. For example, infl uenza infections in children cause considerable illness and death, and it is often hypothesized that children are affected generally earlier in epidemics because of the higher transmission rates (15). Therefore, justifi cation is strong for local authorities to begin collecting and among all children's specimens that were submitted to the World Health Organization (WHO) reference laboratory at Queen Mary Hospital (most specimens are referred from local hospitals). B) Proportion of infl uenza A and B isolations (by date of collection) among all adult patients' specimens that were submitted to the WHO reference laboratory at Queen Mary Hospital. C) Weekly infl uenza-like illness (ILI) (defi ned as fever plus cough or sore throat) consultation rates in sentinel networks of outpatient clinics in the private (GP) and public (GOPC) sectors. D) Weekly rates of public hospital admissions in young children (<4 years) with a principal diagnosis of infl uenza (International Classifi cation of Diseases, 9th revision, code 487), where the denominator is the general population of the same age. E) Daily estimates of the effective reproductive number based on the laboratory and sentinel outpatient data. Source for panels B-D: (7).
reporting timely age-specifi c community surveillance in sentinel and laboratory networks.