Association Between School Closure and Subsequent Absenteeism During a Seasonal Influenza Epidemic : Epidemiology

Secondary Logo

Journal Logo

Infectious Diseases: Original Article

Association Between School Closure and Subsequent Absenteeism During a Seasonal Influenza Epidemic

Rodriguez, Carla V.a,b; Rietberg, Kristab; Baer, Atarb; Kwan-Gett, Taob; Duchin, Jeffreya,b

Author Information
Epidemiology 20(6):p 787-792, November 2009. | DOI: 10.1097/EDE.0b013e3181b5f3ec


Nonpharmaceutical interventions, including social distancing measures, have been promoted to mitigate the impact of an influenza pandemic.1 Based on the central role children play in community influenza transmission, school closure is the centerpiece of this strategy.2 School children have been shown to have the highest age-specific attack rates for influenza during both pandemic and interpandemic periods.3 Summarizing prospective studies conducted around the 1918 pandemic, Glezen4 reported that age-specific influenza attack rates among school-aged children (5–19 years) ranged from 34.5 to 39 per 100 person-years during a pandemic year (1918) and 16.3 to 24.9 during interpandemic years (1928–1929). In comparison, rates among the elderly (≥60 years) ranged from 8.8 to 14.3 per 100 person-years during a pandemic and 16.9 to 17.9 during interpandemic periods. Moreover, age-specific patterns for school absenteeism, health-care utilization, and culture-positive isolates all illustrate a shift from school-aged children to preschool children and adults in the later stages of the epidemic.

Mathematical models suggest that school closure can be an effective tool for reducing influenza transmission when implemented early in an outbreak.5–8 But because closures are disruptive to schools and families, there are few opportunities to study this intervention, and therefore little empirical evidence to support mathematical projections.1,9 Two studies have described a reduction in respiratory diseases associated with school-based interventions,10,11 although only one of them (a study of school closure due to a teacher strike in Israel) looked specifically at impact on influenza rates. Moreover, the impact of late closures has not been evaluated. This is a matter of interest, since school closures (both in seasonal epidemics and in public health emergencies, such as SARS) often come after a marked increase in illness activity or occurrence of fatal cases has been observed.12–14

In 2007, about half the public schools in King County, Washington, cancelled their winter break because so many school days had already been lost that year due to bad weather. Coincidentally, the winter break (19–23 February) coincided with the peak of the 2006–2007 influenza season. Twelve of the 19 school districts remained open, while the remaining districts took the break, providing a rare opportunity to study the impact of school closure on disease transmission as reflected in student absenteeism rates. King County influenza surveillance data from this period show a rise in the number and proportion of influenza A isolates beginning in early January 2007, peaking in mid-February, and subsiding in mid-March. This same pattern is apparent for the proportion of children's emergency department visits for influenza-like-illness (flu; fever and cough; fever and sore throat; nonurinary sepsis; bacteremia; bronchiolitis; pneumonia; or International Classification of Disease [ICD]-9 codes 465–466 and 481–486), which accelerated rapidly in late January and peaked just before the scheduled break (Fig. 1).

Daily proportion of influenza-like-illness (ILI) visits among children presenting to King County emergency departments during a seasonal influenza epidemic, October 2006–March 2007, by age. Winter break 19–23 February 2007.

Using absenteeism as a proxy for illness, we describe absenteeism trends before and after the break. We test the hypothesis that schools on break would experience lower rates of post-break absenteeism (representing decreased influenza activity) compared with the same period for schools that remained open.


There are 224 private and 525 public schools in 19 school districts in King County, Washington. At the time of this evaluation, King County schools were required to report absenteeism to the public health department (Public Health-Seattle and King County) by phone or fax when absenteeism reached 10% or more of the school population. Enhanced surveillance was conducted to collect daily retrospective and prospective school absenteeism data from 5 February though 9 March 2007 for this evaluation. We sent participation requests to all schools except for the 102 schools in the Seattle School District that have been providing automated daily reports since March 2006 through the health department's syndromic surveillance system. Elementary school populations were of particular interest and thus actively recruited to enable a subgroup analysis. Schools with enrollment sizes of at least 250 students were also actively recruited to maximize internal validity of school measurements. Two public health officers from the Centers for Disease Control and Prevention's (CDC) Quarantine Division assisted the enrollment process by personally meeting with priority schools to facilitate participation in the enhanced surveillance, answer questions, and encourage reporting.

Data reported included date, school name, school district (for public schools), grade, number enrolled, number absent, and whether the school was open or closed on a given day. Individual schools defined the term “absent.” We were able to aggregate the schools by type (eg, high, middle, elementary) but not by grade. Schools were classified as high, middle, elementary, and other schools. Assignment of school type was based first on self-identification. Where schools did not self-identify, we considered middle schools to be those containing grades 3–12, 4–12, or 5–12 (comprehensive). We considered high school to be those containing grades 6–12 or 7–12. Any school that included pre-Kindergarten (ie, children <6 years) was classified separately (as “other”) because of the higher contact and infection rates among children in this age group.15,16 “Other” also included home schools. Reporting through a web-based system was encouraged, but reports by fax, phone, e-mail, and regular mail were also accepted.

To account for the possibility of elevated absenteeism immediately after the break as a consequence of extended vacations rather than illness, we compared the average daily percent absent for 26–28 February and 1–2 March and we found these to be very similar (mean = 0.062 [95% confidence interval {CI} = 0.059–0.065] and mean = 0.061 [95% CI = 0.058–0.065], respectively). The decision to use data for the entire 2-week post-break period enabled us to look at the most relevant outcome period (immediately after the break) and maximize the information collected.

Statistical Methods

Data collected through the enhanced surveillance system were entered into Microsoft Access and SPSS for data management and merged with data from the Seattle school district. Analysis was done using SAS v.8 (Cary, NC) and STATA version 9 (College Station, TX).

Absenteeism was described in terms of daily proportion of student body absent. We used generalized estimating equations (GEE) with Poisson distribution to evaluate whether mean absenteeism after the break (ie, 26 February–9 March 2007) differed between schools on break and those in session, adjusting for baseline absenteeism and repeated measurements by schools over time. We assumed robust standard errors and an independent correlation structure for within-school measurements. The analysis of covariance (ANCOVA) approach is appropriate because baseline absenteeism was similar between schools on break and those in session (relative risk = 1.11 [95% CI = 1.00–1.22]) and because this approach provides smaller standard errors and increased statistical power compared with an analysis of change.17 When 2 groups are the same at baseline, the difference in mean outcomes between the groups at a later time is equivalent to the change over time. Potential confounding by public/private status and school type (ie, elementary, middle, high, or other) were assessed by univariate analyses for independent associations with exposure and outcome. Covariates found to be associated with the exposure and outcome were treated as confounders and included in a multivariable analysis.

We hypothesized that a potential effect of school break on absenteeism rates would be most likely to occur among schools with a lower baseline of absenteeism, which presumably had not yet reached the apex of absenteeism for the season. Therefore, we conducted a stratified analysis by average weekly baseline absenteeism above or below the King County reporting threshold (≥10% absenteeism).

We also performed a secondary analysis restricted to elementary schools. Illness transmission in this population is of particular interest, in that absenteeism patterns in this population are thought to more closely mirror illness trends, compared with older school-aged children.18 Private schools were also analyzed separately because of the possibility that absenteeism trends in this population may be inherently different from public schools.


We contacted 715 schools, of which 522 responded. One-fifth (19%) of the public schools and one-half (50%) of the private schools did not respond. We excluded 52 schools that did not report for at least 2 days during both the baseline (5–9 February) and outcome (26 February–9 March) period. Compared with schools included in the analysis, schools that were excluded showed no clustering by district and were otherwise similar by distribution of school types, distribution of public versus private schools, and percent closed for break. Of the 470 schools available for this analysis, 265 were closed for the winter break and 205 remained open.

The sample comprised mostly elementary schools (56%). Middle schools, high schools, and other schools made up 26%, 15%, and 3% of the sample, respectively. This distribution differed little by break status. Compared with the distribution of school types in King County, elementary schools were over-sampled (they comprise 45% of all schools in King County), while all other school types were under-sampled by approximately 50%. Average enrollment ranged from 4 to 2538 children per school, with higher enrollment in public schools (median = 470 [range = 51–2538]) than private schools (218 [4–904]). Over the entire observation period, schools were closed for an average of 3.6 days (SD = 1.7; range = 0–11).

Descriptive statistics of schools by break status are provided in Table 1. Fifty-eight percent of King County schools were on break the week of February 19. Average daily baseline absenteeism (8.2%) and enrollment (495 students) in schools that took a break were similar to schools that remained open (7.4% and 506, respectively). In schools with mean baseline absenteeism below the 10% reporting threshold, there was also no difference in baseline absenteeism in schools on break versus those in session (RR = 1.04 [95% CI = 0.98–1.11]). Districts varied with regard to closure, with almost no variability within district (eTable, Private schools were more likely to remain open during the intervention period than public schools.

Baseline Characteristics of Schools by Break Status 19–23 February 2007

Data from Seattle Public Schools indicate that absenteeism had peaked and was generally on the decline by the time of the break (Fig. 2). Baseline absenteeism differed across school districts and school types, but not public/private status (eTable, The total proportion absent at baseline was greatest in high schools.

Daily proportion of student absenteeism in Seattle Schools by school type, October 2006–March 2007.

Adjusted for private schools and baseline absenteeism, post-break absenteeism did not tend to differ in schools on break compared with those in session (RR = 1.07 [95% CI = 0.96–1.20]; Table 2). Because absenteeism across the groups did not differ at baseline even after adjustment for confounding by private schools, we can also infer that the decline in absenteeism from before the break to after the break was similar across groups. In the group where we expected to see the largest effect, those with mean baseline absenteeism below the 10% reporting threshold, the change in absenteeism from before to after the break did not tend to differ in schools on break compared with those that remained in session (0.99 [0.88–1.1]).

Mean Daily Baseline Absenteeism by School Type and Public/Private Status

Among elementary schools, post-break absenteeism, adjusting for baseline absenteeism, did not differ between break groups (1.00 [0.91–1.10]; Table 3). Furthermore, there was no evidence of effect modification by average daily baseline absenteeism ≥10% (RR for interaction = 0.97 [95% CI = 0.92–1.11]). In the analysis restricted to private schools, schools on break had slightly lower post-break absenteeism as compared with schools that remained open (RR = 0.88 [95% CI = 0.70–1.12]). No differential effect was seen by baseline absenteeism ≥10%. Baseline absenteeism in elementary schools was similar by break status, as was absenteeism in private schools.

Mean Post-break Absenteeism Prevalence for 26 February to 9 March 2007 by Break Status, by Mean Baseline Absenteeism Greater Than or Equal to the 10% Reporting Threshold and by Selected School Type


In comparing schools with and without a break during the peak of a seasonal influenza outbreak, we found no effect of school closure on subsequent absenteeism. This was true regardless of baseline absenteeism of the schools. The basic study design has 2 important features not present in previous studies: (1) an appropriate temporal analysis where intervention precedes outcome; and (2) direct evaluation of intervention and outcome on the same unit of analysis: the school. Recent studies on this topic have included ecologic analysis of regional influenza trends that are concurrent with school closure periods.11,19 Outcomes that occur at the same time as the intervention limit the ability to infer a causal association.20 Furthermore, if an intervention effect occurs, it should be most pronounced in the population for which it was targeted. For example, a French modeling study found that a reduction in influenza activity after a school holiday was restricted to children (<18 years) and not adults.21

However, limitations in data collection and the timing of the intervention hampered our ability to describe an effect if one truly existed. Our experience highlights methodologic issues that should be considered in designing future evaluations of the impact of school closure on influenza transmission. First, we evaluated a natural experiment where the intervention occurred during the peak of the influenza season (Fig. 1), a suboptimal time to interrupt disease transmission. Recent findings from an ecologic analysis of nonpharmaceutical interventions during the second wave of the 1918 pandemic also suggest reduced efficacy of interventions implemented late in a pandemic.22 Even so, most calls for school closure are implemented late in the course of an epidemic. Findings from this study provide some evidence that late closures provide little effect. The US Centers for Disease Control and Prevention recommends students be dismissed from school early in a severe influenza epidemic, and sequestered at home to avoid transmitting infection in other venues.2 However, results from a North Carolina survey suggest that, despite recommendations to avoid public places, 89% of children visited at least one public venue during their school closure.23 Future evaluations should focus on early interventions to validate the utility of the CDC recommendation. Future studies could also take advantage of opportunities within the school calendar (eg, differing vacation schedules across school districts) to identify potential closure periods and possible comparison groups.

Reporting was initiated once health department officials were made aware of the differential closures. Data collection included retrospective and prospective reporting that was outside the scope of normal reporting activities for schools, representing an additional workload for school officials. A simple and quick reporting mechanism was necessary to secure the participation of the schools. While this surveillance approach enabled a speedy and relatively complete evaluation (70% of schools provided some data), it limited our ability to collect data that wasn't routinely reported. We collected total absenteeism as a proxy for influenza-related absenteeism, and did not verify the cause of absences. Without reason for absence, the measure is insensitive to environmental changes that may be responsible for nonillness related absences, such as severe weather conditions and planned holidays. Although a comparison of absenteeism in the days immediately after the break (26–28 February) and at the end of the week (1–2 March) showed no difference in overall absenteeism, we do not know if the majority of those absences were due to illness or extended vacations. Furthermore, a uniform definition for absenteeism was not imposed. This may have introduced nondifferential misclassification of the outcome, as we do not expect the definition of absenteeism to have differed by break status. The consequence of such misclassification is a potential attenuation of any real effect. Stricter outcome definitions would help ensure comparability of outcome measures and improve internal validity. If absenteeism is to be used as an outcome, reason for absenteeism should be collected on each student. However, such data collection will likely be beyond the scope of normal school operating procedures. Standardized methods for collecting absenteeism data are needed that alleviate the burden placed on limited school resources.

An important secondary finding was that high schools experienced higher baseline absenteeism rates than other school types (Table 2). Furthermore, while high schools are underrepresented in our sample, they are overrepresented in the sample of schools with baseline absenteeism above the 10% reporting threshold (2 times their distribution in the overall sample). In schools with mean baseline absenteeism at or above the 10% reporting threshold, post-break absenteeism was higher in schools on break compared with those that remained in session (RR = 1.33 [95% CI = 1.12–1.58]), controlling for private schools. Inference on the change in absenteeism from before to after the break cannot be made in this subgroup because baseline absenteeism differed by break status after adjustment for confounding by private schools (1.11 [1.02–1.20]). When high schools were removed from the stratified analysis for schools above the 10% reporting threshold, the increased risk of absenteeism associated with break was attenuated (1.24 [1.05–1.46]). We also observed that high school absenteeism continued to increase just before the start of the intervention. This pattern is inconsistent with absenteeism trends in other school types (Fig. 2). However, it is more consistent with a sustained elevation in influenza-like-illness visits among high school aged children as compared with other ages (Fig. 3). Without knowing the reason for absenteeism, it is difficult to discern whether these trends represent potential misclassification of illness-related absenteeism in the high school population, delayed spread of influenza to the older age group, or both. Excluding high schools from future studies of school closure, particularly those where absenteeism is the outcome, may improve study sensitivity.

Daily proportion of influenza-like-illness visits among children presenting to King County emergency departments during the enhanced surveillance period, 5 February–9 March 2007, by age. Winter break, 19–23 February 2007.

A lack of individual student-level data on illness and absenteeism did not allow us to control for potential confounding on the individual level. Individual-level data should be collected to understand the relationship between illness and absenteeism, and to learn about compliance with social distancing recommendations.

The lack of variability in exposure on the district level suggests some clustering. The impact of not adjusting for district-level clustering would imply even weaker inference, and does not change the results of the study.

In sum, our evaluation did not find evidence of benefit of school closures implemented during the peak of seasonal influenza activity. However, the potential utility of early school closure remains to be described. Our experience suggests that more clearly defined outcome and exposures would improve future studies. The potential benefit of school closure must be accurately described to weigh against the potentially large burden carried by such measures. Families may incur extra expenses or missed days of work, and businesses and other critical community services may be adversely affected.24,25


We thank the staff of Public Health Seattle & King County; administrators and staff at King County & Seattle School Districts; Andre D. Berro and Perry Camagon of the US Centers for Disease Control, Division of Global Migration and Quarantine for their assistance; and Bryant Karass from the University of Washington for his valuable assistance in establishing absenteeism reporting during the study period.


1.Bell DM. Non-pharmaceutical interventions for pandemic influenza, national and community measures. Emerg Infect Dis. 2006;12:88–94.
2.CDC. Interim pre-pandemic planning guidance: community strategy for pandemic influenza mitigation. Atlanta, GA: CDC; 2007.
3.Glezen WP. Herd protection against influenza. J Clin Virol. 2006;37:237–243.
4.Glezen WP. Emerging infections: pandemic influenza. Epidemiol Rev. 1996;18:64–76.
5.Ferguson NM, Cummings DA, Fraser C, Cajka JC, Cooley PC, Burke DS. Strategies for mitigating an influenza pandemic. Nature. 2006;442:448–452.
6.Germann TC, Kadau K, Longini IM Jr, Macken CA. Mitigation strategies for pandemic influenza in the United States. Proc Natl Acad Sci USA. 2006;103:5935–5940.
7.Glass RJ, Glass LM, Beyeler WE, Min HJ. Targeted social distancing design for pandemic influenza. Emerg Infect Dis. 2006;12:1671–1681.
8.Glass K, Barnes B. How much would closing schools reduce transmission during an influenza pandemic? Epidemiology. 2007;18:623–628.
9.Aledort JE, Lurie N, Wasserman J, Bozzette SA. Non-pharmaceutical public health interventions for pandemic influenza: an evaluation of the evidence base. BMC Public Health. 2007;7:208.
10.Reichert TA, Sugaya N, Fedson DS, Glezen WP, Simonsen L, Tashiro M. The Japanese experience with vaccinating schoolchildren against influenza. N Engl J Med. 2001;344:889–896.
11.Heymann A, Chodick G, Reichman B, Kokia E, Laufer J. Influence of school closure on the incidence of viral respiratory diseases among children and on health care utilization. Pediatr Infect Dis J. 2004;23:675–677.
12.Massive school closures as flu hits Hong Kong [transcript]. CTV. Available at: Accessed 2 May, 2008.
13.Ong ME. War on SARS: a Singapore experience. CJEM. 2004;6:31–37.
14.Department of Health and Human Services. Society of Toxicology. Influenza surveillance report, 2007–2008. Jacksonville, FL: Department of Health and Human Services; 2007.
15.Wallinga J, Teunis P, Kretzschmar M. Using data on social contacts to estimate age-specific transmission parameters for respiratory-spread infectious agents. Am J Epidemiol. 2006;164:936–944.
16.Xue J, Zartarian V, Moya J, et al. A meta-analysis of children's hand-to-mouth frequency data for estimating nondietary ingestion exposure. Risk Anal. 2007;27:411–420.
17.Fitzmaurice G, Laird NM, Ware JH. Applied Longitudinal Analysis. Wiley Series in Probability and Statistics. 1st ed. Vol. 1. Hoboken, NJ: John Wiley & Sons Inc; 2004.
18.Besculides M, Heffernan R, Mostashari F, Weiss D. Evaluation of school absenteeism data for early outbreak detection, New York City. BMC Public Health. 2005;5:105.
19.Cowling BJ, Lau EH, Lam CL, et al. Effects of school closures, 2008 winter influenza season, Hong Kong. Emerg Infect Dis. 2008;14:1660–1662.
20.Koepsell T, Weiss N. Epidemiologic Methods: Studying the Occurrence of Illness. 1st ed. New York: Oxford University Press Inc; 2003.
21.Cauchemez S, Valleron AJ, Boelle PY, Flahault A, Ferguson NM. Estimating the impact of school closure on influenza transmission from Sentinel data. Nature. 2008;452:750–754.
22.Markel H, Lipman HB, Navarro JA, et al. Nonpharmaceutical interventions implemented by US cities during the 1918–1919 influenza pandemic. JAMA. 2007;298:644–654.
23.Johnson AJ, Moore ZS, Edelson PJ, et al. Household responses to school closure resulting from outbreak of influenza B, North Carolina. Emerg Infect Dis. 2008;14:1024–1030.
24.Middaugh JP. Pandemic influenza preparedness and community resiliency. JAMA. 2008;299:566–568.
25.Inglesby TV, Nuzzo JB, O'Toole T, Henderson DA. Disease mitigation measures in the control of pandemic influenza. Biosecur Bioterror. 2006;4:366–375.

Supplemental Digital Content

© 2009 Lippincott Williams & Wilkins, Inc.