MAJOR ARTICLE

Effectiveness of a School District Closure for

Pandemic Influenza A (H1N1) on Acute
Respiratory Illnesses in the Community:
A Natural Experiment
Daphne L. Copeland,1,a Ricardo Basurto-Davila,2,b Wendy Chung,5 Anita Kurian,6 Daniel B. Fishbein,1,c
Paige Szymanowski,1,d Jennifer Zipprich,3,e Harvey Lipman,1,f Martin S. Cetron,1 Martin I. Meltzer,4 and
Francisco Averhoff1,g

Downloaded from http://cid.oxfordjournals.org/ at Kainan University on May 12, 2015

1
Division of Global Migration and Quarantine, 2Influenza Division, 3Epidemic Intelligence Service, and 4Division of Preparedness and Emerging
Infections, Centers for Disease Control and Prevention, Atlanta, Georgia; 5Dallas County Health and Human Services, and 6Tarrant County Public
Health, Fort Worth, Texas

Background. Following detection of pandemic influenza A H1N1 ( pH1N1) in Dallas/Fort Worth, Texas, a
school district (intervention community, [IC]) closed all public schools for 8 days to reduce transmission. Nearby
school districts (control community [CC]) mostly remained open.
Methods. We collected household data to measure self-reported acute respiratory illness (ARI), before,
during, and after school closures. We also collected influenza-related visits to emergency departments (EDflu).
Results. In both communities, self-reported ARIs and EDflu visits increased from before to during the school
closure, but the increase in ARI rates was 45% lower in the IC (0.6% before to 1.2% during) than in the CC (0.4%
before to 1.5% during) (RRRDuring/Before = 0.55, P < .001; adjusted ORDuring/Before = 0.49, P < .03). For households
with school-aged children only (no children 0–5 years), IC had even lower increases in adjusted ARI than in the
CC (adjusted ORDuring/Before = 0.28, P < .001). The relative increase of total EDflu visits in the IC was 27% lower
(2.8% before to 4.4% during) compared with the CC (2.9% before to 6.2% during). Among children aged 6–18
years, the percentage of EDflu in IC remained constant (5.1% before vs 5.2% during), whereas in the CC it more
than doubled (5.2% before vs 10.9% during). After schools reopened, ARI rates and EDflu visits decreased in both
communities.
Conclusions. Our study documents a reduction in ARI and EDflu visits in the intervention community. Our
findings can be used to assess the potential benefit of school closures during pandemics.
Keywords. H1N1 virus; influenza; pandemics; schools; nonpharmaceutical interventions.

Received 23 May 2012; accepted 30 August 2012; electronically published 19 During the beginning of the 2009 H1N1 pandemic,
October 2012. school closures were implemented despite a lack of
a
Present affiliation: Centers for Disease Control and Prevention, Division of
State and Local Readiness, Atlanta, Georgia. consensus on their effectiveness as a disease mitigation
b
Present affiliation: Los Angeles County Department of Public Health, Office of strategy [1]. Although previously published modeling,
Health Assessment and Epidemiology, California.
c
Present affiliation: Myanmar Research International, Santa Barbara, California. historic, and epidemiologic studies suggest that school
d
e
Present affiliation: New York Medical College, Vahalla. closure is effective in reducing influenza transmission,
Present affiliation: California Department of Public Health, Richmond.
f
Deceased.
the limitations of these studies have resulted in little
g
Present affiliation: Centers for Disease Control and Prevention, Division of agreement [2–9]. We utilized a unique opportunity,
Viral Hepatitis, Atlanta, Georgia.
Correspondence: Daphne Copeland, MD, MPH, 230 River North Dr, Atlanta, GA
during the 2009 H1N1 pandemic, to evaluate the ef-
30328 ([email protected]). fectiveness of school closure using a natural experi-
Clinical Infectious Diseases 2013;56(4):509–16 ment. Given that school closure remains an important
Published by Oxford University Press on behalf of the Infectious Diseases Society of
America 2012.
nonpharmaceutical intervention for pandemic influen-
DOI: 10.1093/cid/cis890 za planning, evaluating its effectiveness is a priority.

Effectiveness of School Closure • CID 2013:56 (15 February) • 509

BACKGROUND district B 3 classrooms from the randomly selected grade were
chosen for survey distribution by the principal of each school.
Following the emergence of 2009 pandemic influenza A H1N1 We distributed the questionnaire to parents by sending
virus ( pH1N1) in April 2009, outbreaks first occurred in the them home with students on 29 May 2009. We asked about
United States and Mexico, and rapidly spread throughout the household composition, demographics, and ARI. We asked
world [10–12]. School-aged children were disproportionately about the presence of fever, cough, sore throat, runny nose,
affected [13–17]. In the absence of vaccine, nonpharmaceutical diarrhea, and nausea/vomiting, and about the timing of illness
interventions, including hygienic and social distancing mea- onset among household members: before the school closure
sures, were key tools employed for mitigating the impact of (12 April [Easter] until 29 April), during the school closure
this novel influenza early in the pandemic [18]. (30 April–7 May) and after the schools reopened (8 May until
Dallas/Fort Worth, the fourth largest metropolitan area in survey completed). Parents returned the surveys to the school
the United States, with a population of 6.3 million, experi- by 4 June, the last day of school. The survey was provided in
enced its first cases of pH1N1 in late April 2009. Tarrant and English and/or Spanish. Trained bilingual interviewers made
Dallas counties are adjacent and in the center of the metropol- up to 3 attempts to reach all nonrespondent households
itan area, which encompasses 12 counties (Supplementary unless the phone number was incorrect or participation was
Figure 1). By 30 April, these 2 counties each had <70 laborato- refused. Phone surveys were conducted 8–23 June 2009.
ry-confirmed cases and ≤2 pH1N1 hospitalizations. After

Downloaded from http://cid.oxfordjournals.org/ at Kainan University on May 12, 2015

identifying the first few pH1N1 cases within schools, school
district A in Tarrant County, which enrolls 80 000 kindergar- Analysis of Household Survey Data
ten through 12th-grade (K-12) students (aged 5–18 years), Data were entered into a Microsoft Access database and ana-
closed its schools from 30 April to 7 May 2009. School district lyzed using Stata 11 software (StataCorp LP, College Station,
B in Dallas County, which enrolls 33 000 K-12 students, had Texas). We defined ARI as the presence of at least 2 of the fol-
few pH1N1 cases identified in their schools, but none of these lowing symptoms: fever, cough, sore throat, or runny nose. We
schools closed during the study period. There were no other calculated ARI rates for 3 time periods: before the school
large-scale school closures in these counties. These events pro- closure, during the school closure, and after the schools re-
vided an opportunity to compare the impact of an early large- opened, adjusting for the different lengths of each time period.
scale school closure on acute respiratory illness rates. We compared the differences in the changes in rates of ARI in
the IC and the CC during these time periods. Individuals who
did not assign a time period to their symptoms (n = 147) were
excluded from analysis.
METHODS
We used the difference-in-differences (DiD) method to esti-
mate the impact of the school closure on ARI [19]. The
We conducted a household survey of families with children
formula below describes an unadjusted DiD for changes in
enrolled in school district A (intervention community [IC])
ARI from the period before the intervention to the period
and school district B (control community [CC]) to compare
during the intervention:
the rates of self-reported acute respiratory illness (ARI). To
determine if there was an impact on the broader community,
DiD ¼ ðARICC;During  ARICC;Before Þ  ðARIIC;During
we also analyzed chief complaint data from emergency depart-
ments (EDs) in the region.  ARIIC;Before Þ
This investigation was determined to be an evaluation of a In this formula, ARI is the average ARI rate in a given school
public health response and thus nonresearch, and not requir- district (either IC or CC) and for a given time period (before,
ing review by an institutional review board. during, or after the school closure). The DiD estimator
removes biases in second-period comparisons between the in-
tervention and control groups that could be the result of pre-
Household Survey viously existing differences between them, as well as biases
We surveyed all schools (K-12) in each of the 2 districts (109 from comparisons over time in the intervention group that
schools in school district A, 32 schools in school district B). For could be the result of trends independent of the intervention.
each school, we randomly chose 1 grade to sample, using a sys- We calculated the DiD estimator for changes in ARI rates
tematic random cluster sample methodology. Because of the dif- from before to during and from during to after the school clo-
ference in size between the districts, only 1 classroom from the sures. The null hypothesis that DiD is equal to zero was exam-
randomly selected grade was chosen for survey distribution by ined using the Wald test. This analysis was conducted for the
the principal of each school in school district A, while in school entire sample and stratified by age group (0–5 years, 6–18

510 • CID 2013:56 (15 February) • Copeland et al

years, and ≥19 years). As a measure of the relative impact of acute hospital beds in Tarrant and Dallas counties. Patient’s
the intervention, we calculated the percentage difference residence zip codes (ie, postal codes) were specified for each
between the observed ARI rate in the intervention community visit. Patient chief complaints were not verified by physicians,
during the school closure and the ARI rate that would have and discharge diagnoses were not reported as part of this syn-
been observed had ARI rates increased as they did in the dromic surveillance system.
control community (Supplementary Methods 1). We extracted from ESSENCE self-reports of influenza using
In addition, we used mixed-effects logistic regressions, for keywords (flu, influenza), and patient reports of symptoms of
the entire sample and stratified by age group, to model the influenza-like illness (ILI; fever plus cough and/or sore
probability of ARI as a function of individual characteristics, throat). We defined extracted cases as EDflu visits. The inter-
household characteristics, school district of residence, and vention community (ICED) included 15 zip codes entirely
time period. In this regression model, the coefficient of inter- served by school district A and 4 zip codes with an area
est was the interaction between the school district and time mostly covered (>50% of the geographic area) by school dis-
period indicators. This coefficient allowed us to estimate the trict A. All the remaining zip codes in Tarrant and Dallas
odds of ARI in the intervention community relative to the counties were considered the control community (CCED), in-
odds in the control community during the closure, adjusted cluding 9 zip codes with <50% of their geographic area served
for differences in ARI rates that existed before the closure and by school district A. Patients who reported residence zip codes
for changes in ARI rates unrelated to the closure. Separate lo- outside of Tarrant and Dallas counties were excluded.

Downloaded from http://cid.oxfordjournals.org/ at Kainan University on May 12, 2015

gistic regressions were estimated comparing changes in the
probability of ARI from before to during and from during to Analysis of Emergency Department Data
after the school closure. In order to account for clustering and As in the analysis of the household survey, we defined 3 time
sampling design, these regressions included random coeffi- periods: before the closure, during the closure, and after
cients for individuals, households, and schools. We tested for schools reopened. Although the actual closure dates were 30
multicollinearity between the variables included in the regres- April–7 May 2009, we added 2 consecutive school days to the
sion using variance inflation factors and variance decomposi- periods before and during the closure to account for the incu-
tion analysis (Supplementary Methods 2). bation of disease; thus, in our analysis, we used 2–12 May as
Finally, because childcare centers in both areas were not the closure period [21].
closed during the study period, we conducted a separate anal- We calculated daily rates of EDflu visits in each community
ysis in which we added another control group within the in- by dividing the number of EDflu visits by the total number of
tervention community. We compared individuals in the ED visits. We compared differences in rates of EDflu visits in the
intervention community who lived with only school-aged chil- ICED with the CCED in each period. Because our data collection
dren (no children aged 0–5 years in the same household) with does not involve random sampling and the data come from EDs
(1) those in the IC who lived with young children (0–5 years) whose hospitals have 96% of the acute hospital beds in the 2
and (2) those living with only school-aged children in the CC. counties, no statistical analysis is needed for comparisons.
In this analysis, we estimated a mixed-effects logistic regres-
sion for the entire sample, similar to that described in the par- RESULTS
agraph above. We added to this regression model a term with
a triple interaction between school district, time period, and Household Survey
an indicator variable for individuals who lived with a child Response and Demographic Characteristics
aged 5 or younger (in addition to the full set of main effects A smaller proportion of households responded in the IC
and interactions between the terms in the triple interaction) (1187/2725 [44%]) than in the CC (1155/1944 [59%])
(Supplementary Methods 3). (P < .001). There were also small differences in the distribution
of race/ethnicity and age that also reached statistical signifi-
Emergency Department Data cance (P < .05). Data were available for analysis for 5188
We assessed the association of school closure with self-report- household members in the IC and 4842 in the CC (Table 1).
ed ARIs in the broader community by analyzing chief com-
plaint data from EDs. These were collected by the Electronic Acute Respiratory Illnesses in Household Members
Surveillance System for the Early Notification of Community- Overall, 619 individuals met the ARI case definition; 238 of
Based Epidemics (ESSENCE) from 1 January to 31 May 300 (79%) in the IC and 234 of 319 (73%) in the CC provided
2009 [20]. These data ( provided by the Southwest Center for illness onset dates and thus were included in the analysis. The
Advanced Public Health Practice, Tarrant County Public frequency of symptoms and distribution of symptom combi-
Health) come from EDs whose hospitals have 96% of the nations for ARI can be found in Supplementary Table 1 and

Effectiveness of School Closure • CID 2013:56 (15 February) • 511

Table 1. Characteristics of Survey Participants for Self-Reported Acute Respiratory Illness

Interventiona Control P Value (Cramer’s V)b

Participation (households) 1187/2725 (44%) 1155/1944 (59%) <.001
Data collection
Self-administered survey 884 (75%) 984 (85%) <.001
Phone interview 303 (25%) 171 (15%)
Spanish language (n = 2342) 441 (37%) 427 (37%) .927
Race/ethnicity
Hispanic 708 (63%) 701 (65%) <.001 (0.135)
White 200 (18%) 183 (17%)
African-American 189 (17%) 115 (11%)
Other 33 (3%) 84 (8%)
Median household size (n = 2229) 4.5 4.0 .096
Economically disadvantagedc 69% 73%
Household members 5188 4842
Female (n = 9767) 2728 (54%) 2482 (53%) .298
Age distribution (n = 9873)

Downloaded from http://cid.oxfordjournals.org/ at Kainan University on May 12, 2015

0–5 years 645 (13%) 535 (11%) .035 (0.026)
6–18 years 2325 (45%) 2117 (45%)
≥19 years 2152 (42%) 2099 (44%)
a
Intervention community included respondents from school district A households that were sampled; the control community includes respondents from school
district B households that were sampled.
b
P values were calculated using χ2 tests; Cramer’s V was used to test strength of association for variables with multiple categories; it ranges from 0 to 1. Values
close to zero indicate a weak association.
c
Data on economically disadvantaged individuals from total school population (Texas Education Agency 2007–2008).

Supplementary Table 2. In both communities, self-reported

ARIs increased from before to during the school closure, but
the increase in ARI rates was 45% lower in the IC (0.6%
before to 1.2% during) than in the CC (0.4% before to 1.5%
during); RRRDuring/Before = 0.55, P < .001) (Figure 1, Table 2).
Overall, the DiD in ARI rates from before to during the school
closure in the IC was −0.47 percentage points compared with
the CC (P = .046), representing an ARI rate in the IC during
the school closure that was 29% lower than it would have been
had ARI increased as it did in the CC. This effect varied by
age (Table 2). The DiD in ARI rates from during the closure
to after school reopening showed no significant differences
between the IC and the CC (Table 2).
In the logistic regression analysis, the adjusted odds ratio
(AOR) of reporting ARI during the school closure in the IC
Figure 1. Weekly rates of acute respiratory illness among survey par-
was 51% lower (0.49, P = .03) than in the CC, controlling for ticipants, by period of onset of first symptom and community of resi-
differences between the 2 communities in the odds of ARI dence, 12 April–23 June 2009. Acute respiratory illness was defined as
before the school closure. This effect also varied by age the presence of at least 2 of the following symptoms: fever, cough, sore
(Table 3). We found no statistically significant difference in throat, or runny nose. The intervention community included respondents
from households in school district A; the control community included re-
the odds of ARI between the 2 communities when comparing
spondents from households in school district B. P values are from Wald
the period during the closure to after the schools reopened tests of the difference-in-differences estimator for the change in rates of
(Supplementary Table 3). We did not find evidence of multi- acute respiratory illness, either from before the closure to during the
collinearity in our regression models (Supplementary Results). closure, or from during the closure to after schools reopened.

512 • CID 2013:56 (15 February) • Copeland et al

Table 2. Weekly Rates of Acute Respiratory Illnessa (ARI) in the Intervention and Control Communities Before the Closure, During the
Closure, and After the Schools Reopenedb, and Difference in Differences for Weekly Rates of ARI in the Intervention Versus the Control
Communities From Before to During the School Closure and From During the Closure to After the Schools Reopenedc

Before School Closure During School Closure After School Reopening

(April 12–29) (April 30–May 7) (May 8–June 23)

ARI IC CC P Value IC CC P Value IC CC P Value

All ages 0.6% 0.4% .036 1.2% 1.5% .194 0.4% 0.7% .002
0–5 years 1.0% 0.7% .349 2.3% 2.1% .812 0.7% 0.9% .272
6–18 years 0.6% 0.5% .531 1.6% 2.1% .220 0.5% 0.8% .012
≥19 years 0.5% 0.2% .042 0.5% 0.8% .159 0.3% 0.5% .068
DiD, Before to During School Closure DiD, During Closure to After Reopening
(P Value)d (P Value)
All ages − 0.47 (.046) + 0.05 (.819)
0–5 years − 0.12 (.890) − 0.48 (.576)
6–18 years − 0.56 (.171) + 0.18 (.657)
≥19 years − 0.56 (.030) + 0.15 (.549)
Abbreviations: ARI, acute respiratory illness; CC, control community; DiD, difference in differences; IC intervention community.

Downloaded from http://cid.oxfordjournals.org/ at Kainan University on May 12, 2015

a
ARI was defined as the presence of at least 2 of the following symptoms: fever, cough, sore throat, or runny nose.
b
The IC includes respondents from households in school district A; the CC includes respondents from households in school district B.
c
DiD is equal to the change in the ARI rate in the IC from the first period to the second period minus the change in the ARI rate in the CC between the same 2
periods.
d
P values were calculated using the Wald test.

Table 3. Results of Mixed-Effects Logistic Regression of Acute Respiratory Illness on Individual and Household Characteristics, Time
Period, and Community of Residencea,b, 12 April–23 June 2009

All Ages, OR (P Value) Ages 0–5, OR (P Value) Ages 6–18, OR (P Value) Ages ≥19, OR (P Value)
Age group
0–5 Reference
6–18 0.76 (.183)
≥19 0.29 (<.001)
Female 1.19 (.234) 0.67 (.339) 1.07 (.737) 1.89 (.027)
Spanish-speaking 0.44 (<.001) 0.44 (.083) 0.51 (.009) 0.40 (.008)
Total children in household 0.87 (.121) 1.16 (.344) 0.77 (.015) 0.87 (.300)
Total adults in household 0.83 (.157) 0.70 (.234) 0.79 (.117) 1.02 (.933)
Child aged 0–5 in household 1.07 (.783) 0.95 (.861) 1.76 (.112)
Time period
Before closure Reference Reference Reference Reference
During the closure 3.84 (<.001) 3.79 (.012) 4.20 (<.001) 3.30 (.002)
School district
Control Reference Reference Reference Reference
Intervention 1.49 (.141) 1.39 (.600) 1.22 (.540) 1.96 (.098)
Intervention × during closure 0.49 (.030) 0.64 (.900) 0.61 (.223) 0.26 (.032)

Abbreviation: OR, odds ratio.

a
The outcome in these models is an indicator variable equal to 1 if the individual reported symptoms of acute respiratory illness during the corresponding period,
adjusting for the different length of each time period. Acute respiratory illness was defined as the presence of at least 2 of the following symptoms: fever,
cough, sore throat, or runny nose.
b
Intervention represents individuals with residence in area covered by school district A; control represents individuals with residence in area covered by school
district B.

Effectiveness of School Closure • CID 2013:56 (15 February) • 513

Effect of Preschool-Aged Children in the Household 0–5 years) in the IC vs CC. We also observed the greatest
In a separate within-group analysis, we found that in the IC, impact among school-aged children when analyzing EDflu
before the school closure, there were no significant differences visits.
(AOR = 1.08, P = .881) between individuals who lived with Our study was subject to some limitations. Outcome mea-
young children aged 0–5 years and those who did not live sures were based on self-reported ARI, not laboratory-con-
with children of this age. However, during the school closure, firmed influenza. The circulation of influenza may have been
the odds of reporting ARI increased 3-fold (AOR = 3.05, different between the communities in our study. However, the
P = .029) for individuals in the IC who lived with young chil- 2 populations are geographically adjacent and have similar de-
dren, compared with others in the IC who did not live with mographics. Moreover, during the first 4 months of 2009,
children aged 0–5 years. In the same analysis, focusing on in- before school closure, patterns of EDflu visits were similar.
dividuals who live with only school-aged children (ie, no chil- Given the low prevalence of ARI reported in the community,
dren 0–5 years), the AOR of ARI for these individuals in the the spike in EDflu visits seen in late April may have been due
IC was 72% lower (0.28, P = .001) during the school closure to many factors, such as media-generated concern about
relative to the CC (Supplementary Table 4). pH1N1. This pattern was seen in many places during the
emergence of pH1N1. During this time, 4.9% of clinician-
Emergency Department Data
confirmed ILIs tested positive for pH1N1 at the largest pediat-
The percent of daily EDflu visits in the ICED and the CCED ric hospital in Dallas [22]. (This study used a modified defini-

Downloaded from http://cid.oxfordjournals.org/ at Kainan University on May 12, 2015

were similar to each other from January through April 2009; tion for ILI; fever >38°C and 2 signs or symptoms of viral
this time period includes a seasonal influenza peak in Febru- infection; the Centers for Disease Control and Prevention
ary (Supplementary Figure 2). There was a significant increase defines ILI as temperature >100°F [>37.8°C] plus cough or
in the percent of EDflu visits in both the ICED and the CCED sore throat.) School districts did not exactly match zip codes;
in late April, coincident with significant media attention about thus, 13 zip codes included a mixture of students from school
pH1N1 (Figure 2). district A and other school districts that remained open. This
The percentage of EDflu visits was similar in the ICED and potential bias, however, is likely to increase the chances of
CCED in the period before the school closure (Supplementary finding no difference between the 2 communities. Relying on
Table 5). The percent of EDflu visits in the ICED increased school principals to select classrooms within randomly chosen
from 2.8% before to 4.4% during the school closure, whereas grades could have also introduced biases, but we cannot assess
in the CCED the percentage of EDflu visits increased >2-fold: either the degree or direction of such biases. The household
from 2.9% to 6.2%. Among children aged 6–18 years, the per- survey was distributed 3 weeks following the reopening of
centage of EDflu visits in the ICED remained constant (5.1% schools in school district A introducing the potential for recall
before school closure; 5.2% during), whereas in the CCED the bias. Response rates were low and differed between the inter-
percentage of EDflu visits doubled from 5.2% to 10.9%. The vention and control communities, raising the potential for
difference between the 2 communities during the closure in participation bias. There were some small differences in re-
rates of EDflu visits for “all ages” can be explained mostly by spondent characteristics; however, the DiD and the multivari-
the difference in rates among 6–18-year-olds. The percentage ate analyses control for these differences. Onset dates for
of EDflu visits decreased and returned close to baseline in both illness were not known for 21% of respondents in the IC and
communities after the schools reopened (Figure 2). 27% of respondents in the CC. Collecting data from the field
immediately following an outbreak may have resulted in the
survey data containing biases, but the data related to the ED
DISCUSSION visits showed similar effects. It is possible that there were
biases that affected both the survey and ED data; however, we
This natural experiment provides evidence that supports the are not aware of any factors in the community that could have
effectiveness of early school closure in reducing ARIs in the affected both sources of data in the same manner. We did not
community during the period of closure. Two major strengths measure the impact of closures on school absenteeism follow-
of this study are having a comparison group and 2 indepen- ing reopening of schools. We did not investigate the concur-
dent sources of data. Importantly, this closure was implemented rent use of other mitigation measures in these communities.
at a time of low influenza activity and for only 8 consecutive In addition, although there were differences between commu-
days; yet, we found a reduction in ARIs and in EDflu visits nities for both ARI and EDflu visits, these effects were mea-
in the IC compared with the CC. We found an even greater sured during time of public concern. However, this would
impact on ARI reduction when the analysis was limited to only affect the results if the level of concern differed between
households with school-aged children only (ie, no children communities.

514 • CID 2013:56 (15 February) • Copeland et al

 Downloaded from http://cid.oxfordjournals.org/ at Kainan University on May 12, 2015
Figure 2. Percentage of daily influenza-related emergency department visits (EDflu) in the intervention community vs the rest of the region, 15 March–
31 May 2009, by age group. EDflu visits were defined as self-reports of influenza using keywords (flu, influenza), and patient reports of symptoms of
influenza-like illness (fever plus cough and/or sore throat). Patient chief complaints were not verified by physicians, and discharge diagnoses were not
reported. Intervention included patient zip codes in school district A. The rest of the region included zip codes in the remaining area of Tarrant and
Dallas counties. Patients who reported residence zip codes outside of Tarrant and Dallas counties (n = 11 470; 2.2% of the sample) were excluded.
School closure dates for analysis, adjusted for influenza incubation period and weekend, are shaded in gray (2–12 May 2009). Actual school closure
dates were 30 April–7 May 2009. Abbreviations: ED, emergency department; ILI, influenza-like illness.

The effectiveness of school closure as an intervention for reader. The posted materials are not copyedited. The contents of all sup-
plementary data are the sole responsibility of the authors. Questions or
reducing influenza illnesses during a pandemic is a complex
messages regarding errors should be addressed to the author.
issue to study. Our study documents a reduction in ARI and
EDflu visits in the IC compared with the CC during school
Notes
closure. Our findings can be used to assess the potential
benefit of school closures during influenza pandemics, and are Acknowledgments. We thank the following individuals for their assis-
tance with this study: Centers for Disease Control and Prevention: Maria
consistent with previously published modeling, historic, and
Cano, John Neatherlin, Joaquin Rueda, Julie Thwing, and Marta Guerra
epidemiologic studies [2–9, 23, 24] .While this natural experi- for their contributions during data collection; Isaac McCullum and Yecai
ment adds to our knowledge of the effectiveness of school Liu for their assistance with data management and data analysis; and the
closure, additional research is needed to better understand the many volunteers who contributed to data collection. Tarrant County
Public Health: Diana Cervantes, Micky Moerbe, and Sandra Parker for
long-term benefits, societal acceptability, and costs of this in- their contributions during data collection. Dallas County Health and
tervention. Future studies should consider prospectively mea- Human Services: John Carlo for his assistance with editing. We also thank
suring school closure impact using laboratory-confirmed the Fort Worth Independent School District for their assistance with data
collection, the Southwest Center for Advanced Public Health Practice and
influenza. Tarrant County Public Health for their assistance with data collection,
and the Texas State Department of Health for their support in this public
health response.
Supplementary Data Author contributions. D. L. C., H. L., D. B. F., W. C., A. K., J. Z.,
F. A., M. S. C., and M. I. M. designed the study; W. C., A. K., D. B. F.,
Supplementary materials are available at Clinical Infectious Diseases online J. Z., D. L. C., and P. S. gathered the data; R. B.-D., H. L., and M. I. M.
(http://www.oxfordjournals.org/our_journals/cid/). Supplementary materi- provided the statistical analysis; D. L. C., W. C., A. K., D. B. F., J. Z.,
als consist of data provided by the author that are published to benefit the M. S. C., F. A., and P. S. assisted with the analysis and interpretation of

Effectiveness of School Closure • CID 2013:56 (15 February) • 515

data; D. L. C., R. B.-D., F. A., M. I. M., and M. S. C. had full access to all 10. Chan M. World now at the start of 2009 influenza pandemic. Avail-
of the data in the study and take responsibility for the integrity of the data able at: http://www.who.int/mediacentre/news/statements/2009/h1n1_
and the accuracy of the data analysis; D. L. C., R. B.-D., F. A., M. S. C., pandemic_phase6_20090611/en/index.html. Accessed 7 May 2010.
and M. I. M. drafted the manuscript. All authors provided critical revision 11. Update: novel influenza A (H1N1) virus infections—worldwide, 6
of the manuscript and participated in the decision to publish the May 2009. MMWR Morb Mortal Wkly Rep 2009; 58:453–8.
manuscript. 12. Update: infections with a swine-origin influenza A (H1N1) virus—
Disclaimer. The findings and conclusions in this article are those of United States and other countries, 28 April 2009. MMWR Morb
the authors and do not necessarily represent the views of the CDC. Mortal Wkly Rep 2009; 58:431–3.
Financial support. This study was conducted as a public health re- 13. Fraser C, Donnelly CA, Cauchemez S, et al. Pandemic potential of a
sponse, with collaboration among the CDC, Tarrant County Public strain of influenza A (H1N1): early findings. Science 2009; 324:
Health, and Dallas County Health and Human Services. 1557–61.
Potential conflicts of interest. All authors: No reported conflicts. 14. Dawood FS, Jain S, Finelli L, et al. Emergence of a novel swine-origin
All authors have completed and submitted the ICMJE Form for Disclo- influenza A (H1N1) virus in humans. N Engl J Med 2009;
sure of Potential Conflicts of Interest. No authors had conflicts of interest, 360:2605–15.
including financial interests, relationships, or affiliations relevant to the 15. Swine-origin influenza A (H1N1) virus infections in a school—
subject of this study during the time this work was completed. New York City, April 2009. MMWR Morb Mortal Wkly Rep 2009;
58:470–2.
16. 2009 pandemic influenza A (H1N1) virus infections—Chicago, Illi-
References nois, April–July 2009. MMWR Morb Mortal Wkly Rep 2009;
58:913–8.
1. Cauchemez S, Ferguson NM, Wachtel C, et al. Closure of schools 17. Kwan-Gett TS, Baer A, Duchin JS. Spring 2009 H1N1 influenza out-
during an influenza pandemic. Lancet Infect Dis 2009; 9:473–81. break in King County, Washington. Disaster Med Public Health Prep

Downloaded from http://cid.oxfordjournals.org/ at Kainan University on May 12, 2015

2. Ferguson NM, Cummings DA, Fraser C, Cajka JC, Cooley PC, Burke 2009; 3(suppl 2):S109–16.
DS. Strategies for mitigating an influenza pandemic. Nature 2006; 18. Centers for Disease Control and Prevention. Interim pre-pandemic
442:448–52. planning guidance: community strategy for pandemic influenza miti-
3. Vynnycky E, Edmunds WJ. Analyses of the 1957 (Asian) influenza gation in the United States. Available at: http://pandemicflu.gov/
pandemic in the United Kingdom and the impact of school closures. professional/community/community_mitigation.pdf. Accessed 23
Epidemiol Infect 2008; 136:166–79. April 2010.
4. Markel H, Lipman HB, Navarro JA, et al. Nonpharmaceutical inter- 19. Meyer B. Natural and quasi-experiments in economics. J Bus Econ
ventions implemented by US cities during the 1918–1919 influenza Stat 1995; 13:151–61.
pandemic. JAMA 2007; 298:644–54. 20. Lombardo JS, Burkom H, Pavlin J. ESSENCE II and the framework
5. Cauchemez S, Valleron AJ, Boelle PY, Flahault A, Ferguson NM. Esti- for evaluating syndromic surveillance systems. MMWR Morb Mortal
mating the impact of school closure on influenza transmission from Wkly Rep 2004; 53(suppl):159–65.
sentinel data. Nature 2008; 452:750–4. 21. Cauchemez S, Donnelly CA, Reed C, et al. Household transmission of
6. Wu JT, Cowling BJ, Lau EH, Ip DKM, Ho L-M, Tsang T. School 2009 pandemic influenza A (H1N1) virus in the United States. N Engl
closure and mitigation of pandemic (H1N1) 2009, Hong Kong. Emerg J Med 2009; 361:2619–27.
Infect Dis 2010; 16:533–41. 22. Chang ML, Evans A, Jodan-Villegas A, et al. Respiratory viruses in
7. Glass K, Barnes B. How much would closing schools reduce transmis- an urban children’s hospital emergency department identified during
sion during an influenza pandemic? Epidemiology 2007; 18:623–8. the influenza a (H1N1) pandemic. Pediatr Emerg Care 2011; 28:
8. Hens N, Ayele GM, Goeyvaerts N, et al. Estimating the impact of 990–7.
school closure on social mixing behaviour and the transmission of 23. Earn DJ, He D, Loeb MB, Fonseca K, Lee BE, Dushoff J. Effects of
close contact infections in eight European countries. BMC Infect Dis school closure on incidence of pandemic influenza in Alberta,
2009; 9:187. Canada. Ann Intern Med 2012; 156:173–81.
9. Heymann AD, Hoch I, Valinsky L, Kokia E, Steinberg DM. School 24. Yu H, Cauchemez S, Donnelly CA, et al. Transmission dynamics,
closure may be effective in reducing transmission of respiratory border entry screening, and school holidays during the 2009 influenza
viruses in the community. Epidemiol Infect 2009; 137:1369–76. A (H1N1) pandemic, China. Emerg Infect Dis 2012; 18:758–66.

516 • CID 2013:56 (15 February) • Copeland et al