Results
Absenteeism due to ILI: Across the 5 academic years, the mean tallies of a-ILI for the two-week periods before winter and spring breaks were 130.4 (range 51-262) and 151.4 (range 69-275), respectively. The mean a-ILI tallies for the periods after winter and spring breaks were 82 (range 33-152) and 49.6 (range 33-74), respectively. Comparatively, the two weeks before pseudo-breaks had an a-ILI average of 106 (range 43-200), and the two weeks after pseudo-breaks had a-ILI average of 100.8 (range 71-131). The grade distributions of a-ILI are displayed in Figure 3, showing higher levels of a-ILI reported among students in 4K and elementary schools, in comparison to middle and high schools.
Crude association between school breaks and a-ILI: The two-week after-break period was associated with a statistically significant decrease in the odds of a-ILI compared to the two-week before-break period. The CMH test estimated an odds ratio of 0.679 (95% CI: 0.600-0.769; p<0.001) following winter breaks and 0.327 (95% CI: 0.283-0.378; p<0.001) following spring breaks. The crude a-ILI counts for each school year, occurring before versus after breaks, are depicted in Table 2. Differences in a-ILI proportions in the two weeks before and after each true break varied every school year (Figure 4). While several of the yearly school breaks had a clear difference in the a-ILI proportions, not every yearly break displayed a difference.
Adjusted association between school breaks and a-ILI: In the regression models, the estimated a-ILI over the two-week period after a break was nearly half the amount of that in the period before a break. The estimated proportional change following a break was 0.483 (95% CI: 0.347-0.673; p<0.001) for winter break and 0.488 (95% CI: 0.327-0.730; p<0.001) for spring break. The weekly community MAI count was also strongly associated with a-ILI (p≤0.001). No statistical significance was detected in the change in linear or quadratic time components for before vs. after breaks (Table 3).
The models produced estimates of daily mean a-ILI for the ten days before and after each break, based on the mean weekly community MAI counts and the mean student enrollment of 3,749 in OSD (Figure 5). Although the behavior of time remained similar in the ten days before and after each break, the model consistently estimated an overall reduction in the amount of a-ILI in the periods following breaks compared to the periods before breaks.
The assessment for the association between a-ILI in the periods before and after breaks was found to be significant. The null model, which consisted of a removal of the two-week period indicator and its interactions with linear and quadratic time, yielded a X2 statistic value of 125.9 on 3 degrees of freedom (p<0.001) in the winter break analysis and 102.4 (p<0.001) in the spring break analysis. This indicated that the inclusion of the period indicator in the model was associated with a statistically significant amount of variation, after accounting for linear and quadratic passage of time and the weekly community ILI count.
Pseudo-breaks as a control: There was consistently no statistically significant difference observed in a-ILI in the two-week periods before and after the pseudo-break when school was actually in session. The unadjusted association between the two-week period after the pseudo-break and the risk for change in a-ILI estimated an odds ratio of 0.985 (95% CI: 0.872-1.11; p=0.839). The changes in proportions of a-ILI before and after each pseudo-break vary throughout the five years (Figure 4). The LRT for removal of the two-week period indicator and its interactions with linear and quadratic time yielded a X2 statistic value of 4.8 on 3 degrees of freedom (p=0.189), indicating that how the period indicator was included in the model was not associated with a statistically significant amount of variation in a-ILI. All covariates included in the pseudo- break model were non-statistically significant (Table 3). In Figure 5, the estimated daily ILI means predicted by the model displayed no clear level of change in absenteeism counts for before versus after a pseudo-break.
Conclusions Over a 5-year period of enhanced monitoring of cause-specific absenteeism, from September 2014 through June 2019, a nearly 50% reduction in a-ILI was observed consistently in the two-week periods immediately following scheduled winter and spring breaks with durations of 9 to 16 days, as compared to the two weeks immediately preceding these breaks. We found a strong association between the period indicator and a-ILI in regression models. This implies that the regular scheduled school breaks produce a significant acute effect on a-ILI. Such an effect has high biological plausibility: (a) if schools are primary centers of influenza transmission and acceleration, and (b) given that the time period spans approximately 2.8 to 4.4 serial intervals for influenza24.
The scale of the proportional differences in a-ILI associated with each break in Figure 4 appears to reflect the timing of peak influenza circulation and annual seasonal peak across Wisconsin (Figure 2). For example, during the 2014/2015 and 2017/2018 school years, there was relatively widespread circulation before the commencement of winter break, with the seasonal peak occurring in late December and early January25. Thus, winter break appeared to have a larger impact on reducing a-ILI than spring break in these years. Conversely, in 2015/2016, 2016/2017, and 2018/2019, widespread circulation occurred later in the season with the peak between February and March25, explaining the more profound difference in a-ILI following spring break. This observation emphasizes the importance of the timing of a school closure on the potential impact on influenza risk.
The absence of significant findings for the pseudo-breaks lends credence to the true school breaks being an actual causal mechanism to reduce a-ILI, particularly with the lack of association between pseudo-breaks and reductions in a-ILI and weekly community MAI. Although the changes in a-ILI after the pseudo-break for any given year in Figure 4 may appear to be significant, the changes are inconsistent with three years (2014/2015, 2015/2016, and 2018/2019) having higher a-ILI following the pseudo-break and two years (2016/2017 and 2017/2018) having lower a-ILI after the pseudo-break.
Other results from ORCHARDS—specifically data generated through home visits to a subset of K-12 students who had to miss school due to an acute respiratory illness—complement the findings from this analysis on school breaks20. Over the five school years (2014-2019), 79% of participants with acute respiratory infections reported missing school because of their illness; 65% of these students who were absent tested positive for influenza or another non-influenza respiratory viral infection, and more than half thought a classmate or friend was the likely source of infection20. Thus, the ORCHARDS results support the concept that within-school transmission drives community-wide outbreaks, and that well-timed school breaks (or, alternatively, short-term transitions to distance learning of equivalent duration as a winter or spring break) can reduce influenza or other respiratory virus transmission.
This assessment has several limitations. First, findings based on the models used are suggestive of an association, but do not necessarily imply a causal relationship. The assessment periods occurring before and after the planned breaks are—by definition—ordered through time; therefore, any temporal effect during this same period that may impact influenza may result in confounding. Second, there is some violation in the assumption of independence of observations in both the adjusted and unadjusted analyses. Since the data used in this assessment were de-identified and a-ILI was measured by counts, it is likely that individual students contributed multiple, sequential absences to the a-ILI counts, thereby altering the independence of daily counts. Third, because parents self-report absences through the absentee line, there is potential that a-ILI numbers are underestimated because of underreporting by parents. Fourth, results generated from OSD over five influenza seasons (2014-2019) may not be generalizable to other locations and populations, for markedly different influenza seasons, or over different academic calendars in terms of school break timing relative to local influenza outbreak peaks. Fifth, we used a-ILI as a proxy for influenza. Whereas we have demonstrated a significant association between influenza virus infection and a-ILI, we have also shown that influenza type and subtype have differential effects on a-ILI20. Finally, although community data on MAI were used in an attempt to represent the underlying community risk, the models are imperfect as they do not capture the entirety of the relationship between underlying community level risk and the risk in schools. It is possible that the period indicator is representing differential community-level risk behaviors during before- vs. after-break periods.
Although reports documenting the effect of school closures on reduced influenza transmission exist, there remains a lack of consensus on its effectiveness. The majority of current literature has assessed the impact of reactive school closures during an influenza pandemic26-33. Differences in the timing of implementation and length of closure during the pandemic may explain why studies have found variable results from reactionary closures.
Results from these analyses are consistent with findings from other studies looking at the role of scheduled breaks on ILI34,35. A study in South Korea observed an immediate 27-39% reduction in influenza transmission during the break period, with a 6-23% reduction in overall transmission following spring break34. Another study found school closures to prevent or delay up to 42% of potential influenza cases among school-age children35. Although we measured a-ILI as the outcome in this analysis, previous studies have suggested that observed a-ILI can adequately represent changes in community influenza36. Moreover, we have previously demonstrated a significant association between a-ILI and influenza in ORCHARDS20. Furthermore, several studies have proposed that regular school closures may mitigate community impact by changing social mixing patterns37-39.
Overall, the findings from these analyses support the hypothesis that planned K-12 school breaks of moderate duration (9-16 days) reduce influenza transmission. Our finding is consistent with the results of the modeling studies which explored the impact of different timing and durations of the school closures during influenza pandemics29, as well as with the conclusions of observational studies of school holidays’ effect on influenza transmission in other countries12,40. The identified impact occurs in the short term and does not imply a long-term effect on an annual seasonal influenza epidemic; however, such short-term effect may be helpful for targeted suppression of influenza activity to reduce pressures on local health care systems during the local disease surges. Additional analyses investigating the impact of well-timed shorter breaks, both planned and unplanned, on a-ILI may determine an optimal duration for brief school closures to effectively suppress community transmission of influenza.