[Skip to Navigation]
Sign In
Figure 1.  Observed Daily COVID-19 Incidence and Mortality by Cumulative Incidence Quartile at the Time of School Closure
Observed Daily COVID-19 Incidence and Mortality by Cumulative Incidence Quartile at the Time of School Closure

COVID-19 indicates coronavirus disease 2019.

Figure 2.  Modeled Association of School Closure With Coronavirus Disease 2019 (COVID-19) Incidence and Mortality
Modeled Association of School Closure With Coronavirus Disease 2019 (COVID-19) Incidence and Mortality

The data markers indicate the national unadjusted daily rates. The lines depict aggregated national daily rates adjusted for each state’s unique set of testing and demographic characteristics on each day of the study period with 95% CIs depicted by gray lines. Six weeks after school closure, states in the lowest quartile had fewer new cases and fewer deaths compared with the states in the highest quartile. Panels A and B were adjusted for all model components retained in the incidence model (intercept: percentage of state’s population aged ≤15 years, percentage of state’s population aged ≥65 years, and US Centers for Disease Control and Prevention [CDC] social vulnerability index; before school closure: stay-at-home or shelter-in-place order, restaurant and bar closure, testing rate per 1000 residents, and urban density; after school closure: testing rate per 1000 residents, stay-at-home or shelter-in-place order, percentage of state’s population aged ≥65 years, number of nursing home residents per 1000 people, and urban density). Panels C and D were adjusted for all model components retained in the mortality model (intercept: percentage of state’s population aged ≤15 years, percentage of state’s population aged ≥65 years, and CDC social vulnerability index; before school closure: stay-at-home or shelter-in-place order, prohibition of gatherings with >10 people, restaurant and bar closure, percentage of state’s population aged ≤15 years, percentage of state’s population aged ≥65 years, number of nursing home residents per 1000 people, and urban density; after school closure: restaurant and bar closure, number of nursing home residents per 1000 people, and urban density).

Table 1.  State Characteristics by Coronavirus Disease 2019 (COVID-19) Incidence Quartile at the Time of School Closure
State Characteristics by Coronavirus Disease 2019 (COVID-19) Incidence Quartile at the Time of School Closure
Table 2.  COVID-19 Incidence and Mortality and Effect Size Associated With School Closure
COVID-19 Incidence and Mortality and Effect Size Associated With School Closure
Table 3.  Estimated Absolute Differences in COVID-19 Casesa Between Period of School Closure and Schools Remaining Open Using Linear Projectionb
Estimated Absolute Differences in COVID-19 Casesa Between Period of School Closure and Schools Remaining Open Using Linear Projectionb
Table 4.  Estimated Absolute Differences in COVID-19 Deathsa Between School Closure and Schools Remaining Open Using Linear Projectionb
Estimated Absolute Differences in COVID-19 Deathsa Between School Closure and Schools Remaining Open Using Linear Projectionb
Conversations with Dr Bauchner (30:38)

Katherine Auger, MD, MSc, of Cincinnati Children's Hospital and Julie Donohue, PhD, of University of Pittsburgh's Graduate School of Public Health discuss K-12 school reopening policy options and the way forward for the 2020-21 school year. Recorded July 29, 2020.

1.
Pan  A, Liu  L, Wang  C,  et al.  Association of public health interventions with the epidemiology of the COVID-19 outbreak in Wuhan, China.   JAMA. 2020;323(19):1915-1923. PubMedGoogle ScholarCrossref
2.
Dong  Y, Mo  X, Hu  Y,  et al.  Epidemiology of COVID-19 among children in China.   Pediatrics. 2020;145(6):e20200702.PubMedGoogle Scholar
3.
Lu  X, Zhang  L, Du  H,  et al.  SARS-CoV-2 infection in children.   N Engl J Med. 2020;382(17):1663-1665.PubMedGoogle ScholarCrossref
4.
Worby  CJ, Chaves  SS, Wallinga  J,  et al.  On the relative role of different age groups in influenza epidemics.   Epidemics. 2015;13:10-16.PubMedGoogle ScholarCrossref
5.
Cauchemez  S, Valleron  AJ, Boëlle  PY,  et al.  Estimating the impact of school closure on influenza transmission from sentinel data.   Nature. 2008;452(7188):750-754.PubMedGoogle ScholarCrossref
6.
Schanzer  D, Vachon  J, Pelletier  L.  Age-specific differences in influenza A epidemic curves.   Am J Epidemiol. 2011;174(1):109-117.PubMedGoogle ScholarCrossref
7.
Luca  G, Kerckhove  KV, Coletti  P,  et al.  The impact of regular school closure on seasonal influenza epidemics.   BMC Infect Dis. 2018;18(1):29.PubMedGoogle ScholarCrossref
8.
Chu  Y, Wu  Z, Ji  J,  et al.  Effects of school breaks on influenza-like illness incidence in a temperate Chinese region.   BMJ Open. 2017;7(3):e013159.PubMedGoogle Scholar
9.
United Nations. Policy brief: the impact of COVID-19 on children. Accessed May 11, 2020. https://www.un.org/sites/un2.un.org/files/policy_brief_on_covid_impact_on_children_16_april_2020.pdf
10.
Dong  E, Du  H, Gardner  L.  An interactive web-based dashboard to track COVID-19 in real time.   Lancet Infect Dis. 2020;20(5):533-534.PubMedGoogle ScholarCrossref
11.
US Census Bureau. American Community Survey. Accessed April 2, 2020. https://www.census.gov/programs-surveys/acs
12.
COVID Tracking Project. Data API. Accessed May 8, 2020. https://covidtracking.com/api
13.
US Centers for Disease Control and Prevention. Behavioral Risk Factor Surveillance System: 2018 data. Accessed June 15, 2020. https://www.cdc.gov/brfss/annual_data/annual_2018.html
14.
Kaiser Family Foundation. Total number of residents in certified nursing facilities. Accessed June 15, 2020. https://www.kff.org/f23fa35/
15.
US Centers for Disease Control and Prevention. Social vulnerability index. Accessed June 15, 2020. https://svi.cdc.gov
16.
Lauer  SA, Grantz  KH, Bi  Q,  et al.  The incubation period of coronavirus disease 2019 (COVID-19) from publicly reported confirmed cases.   Ann Intern Med. 2020;172(9):577-582.PubMedGoogle ScholarCrossref
17.
Colorado Department of Public Health and Environment. Get tested. Accessed March 23, 2020. https://covid19.colorado.gov
18.
State of Connecticut. Connecticut COVID-19 response. Accessed March 30, 2020. https://portal.ct.gov/Coronavirus
19.
Georgia Department of Public Health. COVID-19. Accessed March 23, 2020. https://dph.georgia.gov
20.
Indiana State Department of Health. 2019 novel coronavirus (COVID-19). Accessed March 31, 2020. https://coronavirus.in.gov
21.
Iowa Department of Public Health. Novel coronavirus acute respiratory disease: COVID-19. Accessed March 26, 2020. https://idph.iowa.gov
22.
Minnesota Department of Health. Coronavirus disease 2019 (COVID-19). Accessed March 24, 2020. https://www.health.state.mn.us/diseases/coronavirus/index.html
23.
Montana Department of Public Health and Human Services. 2019 novel coronavirus (COVID-19). Accessed March 27, 2020. https://dphhs.mt.gov
24.
New Hampshire Department of Health and Human Services. Novel coronavirus 2019 (COVID-19). Accessed March 25, 2020. https://www.nh.gov/covid19
25.
South Carolina Department of Health and Environmental Control. Coronavirus disease 2019 (COVID-19). Accessed April 3, 2020. https://www.scdhec.gov/infectious-diseases/viruses/coronavirus-disease-2019-covid-19
26.
Garg  S, Kim  L, Whitaker  M,  et al.  Hospitalization rates and characteristics of patients hospitalized with laboratory-confirmed coronavirus disease 2019.   MMWR Morb Mortal Wkly Rep. 2020;69(15):458-464.PubMedGoogle ScholarCrossref
27.
Beigel  JH, Tomashek  KM, Dodd  LE,  et al.  Remdesivir for the treatment of Covid-19.   N Engl J Med. Published online May 22, 2020. doi:10.1056/NEJMoa2007764PubMedGoogle Scholar
28.
Li  W, Zhang  B, Lu  J,  et al.  The characteristics of household transmission of COVID-19.   Clin Infect Dis. 2020;ciaa450.PubMedGoogle Scholar
29.
Gudbjartsson  DF, Helgason  A, Jonsson  H,  et al.  Spread of SARS-CoV-2 in the Icelandic population.   N Engl J Med. 2020;382(24):2302-2315.PubMedGoogle ScholarCrossref
30.
Bunyavanich  S, Do  A, Vicencio  A.  Nasal gene expression of angiotensin-converting enzyme 2 in children and adults.   JAMA. 2020;323(23):2427-2429.PubMedGoogle ScholarCrossref
31.
Zhang  J, Litvinova  M, Liang  Y,  et al.  Changes in contact patterns shape the dynamics of the COVID-19 outbreak in China.   Science. 2020;368(6498):1481-1486.PubMedGoogle ScholarCrossref
32.
Ferguson  NM, Laydon  D, Nedjati-Gilana  G,  et al. Report 9: impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare demand. Accessed July 20, 2020. https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-College-COVID19-NPI-modelling-16-03-2020.pdf
33.
Viner  RM, Russell  SJ, Croker  H,  et al.  School closure and management practices during coronavirus outbreaks including COVID-19.   Lancet Child Adolesc Health. 2020;4(5):397-404.PubMedGoogle ScholarCrossref
34.
Esposito  S, Principi  N.  School closure during the coronavirus disease 2019 (COVID-19) pandemic.   JAMA Pediatr. Published online May 13, 2020. doi:10.1001/jamapediatrics.2020.1892 PubMedGoogle Scholar
35.
Hsiang  S, Allen  D, Annan-Phan  S,  et al.  The effect of large-scale anti-contagion policies on the COVID-19 pandemic.   Nature. Published online June 8, 2020. doi:10.1038/s41586-020-2404-8 PubMedGoogle Scholar
36.
Koetsier  J. 58% of American knowledge workers are now working remotely. Accessed June 9, 2020. https://www.forbes.com/sites/johnkoetsier/2020/03/20/58-of-american-knowledge-workers-are-now-working-remotely/#517af4b13303
37.
Rosalsky  G. COVID-19 forces more people to work from home: how’s it going? Accessed June 9, 2020. https://www.npr.org/2020/05/08/852527736/covid-19-forces-more-people-to-work-from-home-hows-it-going
38.
US Centers for Disease Control and Prevention. Schools during the COVID-19 pandemic. Accessed May 15, 2020. https://www.cdc.gov/coronavirus/2019-ncov/community/schools-childcare/Schools-Decision-Tree.pdf
39.
American Academy of Pediatrics. COVID-19 planning considerations: guidance for school re-entry. Accessed July 20, 2020. https://services.aap.org/en/pages/2019-novel-coronavirus-covid-19-infections/clinical-guidance/covid-19-planning-considerations-return-to-in-person-education-in-schools/
7 Comments for this article
EXPAND ALL
Concerns About Accuracy of Given Short Study Time Period
Jessie Schwartz, RN, MPH | None
I would like to ask the authors why they decided to stop their data collection in May 2020, and if they think the massive spike in cases in the South and West this summer would change their results. It seems like the study data is too narrow to capture the results of school closures (i.e., they did not observe what happened when all schools were closed in the summer).

Do the authors worry that they may have arrived at an erroneous conclusion, and/or that these conclusions could have vast policy implications that are particularly harmful to children? />
Please see the Medscape article which discusses selection bias, sampling bias, and ascertainment bias as it relates to recent COVID-19 research (1).

Thanks for your consideration!

Signed,
A Concerned Parent of 3 Public School Children

REFERENCE

1.  "COVID-19 Data Dives: The Banana Peel of COVID-19 Research" https://www.medscape.com/viewarticle/930681
CONFLICT OF INTEREST: None Reported
READ MORE
Confidence Intervals are for Sample Data
James Gundlach, PhD, Sociology UT Austin | Professor Emeritus, Sociology, Auburn Univ
When you have population data, including count of number of actual deaths, you should not use confidence intervals. The confidence interval tells you the probable error in sample data estimating population data. When you have population data sample confidence intervals only obscure actual effects.
CONFLICT OF INTEREST: None Reported
Author Response To Comments
Katherine Auger, MD, MSc | Cincinnati Children's Hospital Medical Center
Ms. Schwartz: Thank you for your question. We were very purposeful in our study period selection. Had we extended the study past early May, it certainly would have increased our likelihood of modeling other changes (policy and behavioral) (i.e. non-school noise). Our goal was to construct models that isolated any potential effects associated with school closure to the extent possible given that we do not have data on things like increased hand-washing and masking. Thus, we focused on the immediate time period around school closure.

Dr. Gundlach: Thank you for your careful reading of our manuscript. I believe
you are referring to the "overall new cases (deaths) per 100,000" lines in Table 2. We do not have confidence intervals around those numbers because they are the actual number of observed cases (and deaths) in the entire United States during this time period (expressed as a rate per 100,000). These data are not extrapolated from population sampling--it is the US in its entirety. Yes, there is uncertainty regarding numbers of cases and deaths from COVID-19 but that is a larger issue related to testing availability and public health monitoring that is a challenge for all COVID-19 studies. 

We did want to give the reader some sense of variability by state, so we added a second set of lines on observed data in the table right below these lines "New cases (deaths) per 100,000 residents per state." In these lines we present the median and the IQR by state.

Best,
Katherine
CONFLICT OF INTEREST: None Reported
READ MORE
Shelter-in-place Household Dynamics
Jack App, MPP | Retired, Math Professor, Irvine Valley College
Did you consider that the principal impact of school closure was to force some working adults to shelter-in-place in order to provide childcare? Even the mobility of non-working parents would be impacted. Communities with families that consisted of one or two working adults and no older [>age 12] children would seem to have had a strong incentive to shelter-in-place during this period and thereby slow the spread of infection. Given that at this stage of the pandemic many fewer children than adults appear to have been infected, this de facto impact on adult behavior may have been the primary impact, not spread in or via school operations. Consequently, I am concerned that I can't find any family size or description data in your model.
CONFLICT OF INTEREST: None Reported
READ MORE
Response to Dr. App
Katherine Auger, MD, MSc | Cincinnati Children's Hospital Medical Center
Dr. App: You raise an important question, which our study team has spent much time discussing. What is the mechanism by which closing schools could potentially impact COVID-19 cases and mortalities? Closing schools disrupts child-to-child interaction, child-to-adult and adult-to-child interaction within schools, as well as adult-to-adult interaction outside of schools. Parents must alter work schedules to care for children who are no longer in schools. It is impossible to know which of these disruptions would alter coronavirus transmission (likely all do to varying degrees). However, all of these changes are part of school closure.

In our study we
examined effects associated with school closure (all of the effects including the aspects of family disruption.) We have not thought of a way to analytically separate out the possible different mechanisms listed above. We also are not sure if one should try to disentangle these mechanisms as you cannot close schools without adult disruption.

With regard to family size, household size would reflect not just the number of children in the household but also the number of adults. Household size is likely associated with crowding and poverty rates as well. In our analysis, we considered two related variables the percentage of children living in the state as well as the social vulnerability index which consists of measures of household composition and disability as well as housing type (crowding).
CONFLICT OF INTEREST: None Reported
READ MORE
Evidence on U.S. School Closures and COVID-19 Spread is Inconclusive
Charles Courtemanche, PhD | University of Kentucky
Auger and colleagues’ study used interrupted time-series analysis of state-level data and found that school closures in March 2020 were associated with lower COVID-19 incidence and mortality [1]. However, multiple other peer-reviewed studies found no significant association. An analysis of four social distancing restrictions – school closures, large gathering bans, restaurant and entertainment closures, and shelter-in-place orders – found that only the latter two were associated with significant reductions in COVID-19 cases across U.S. counties [2]. Using similar methods, an analysis of Italy, France, and the U.S. found that social distancing restrictions as a package were effective, but school closures alone were not [3]. Another study used an epidemiological model and data from 11 European nations, reaching similar conclusions [4]. Auger and colleagues did not cite two of these papers, and only briefly mentioned the third. Readers and policymakers should be aware of these studies with opposite conclusions.

Why the differing results? One possibility is the handling of other restrictions implemented at similar times as school closures. The three studies that did not find that closing schools influenced COVID-19 spread all used methods that put these policies on a “level playing field,” allowing the data to determine which interventions were most important. In contrast, interrupted time-series methods focus on a single intervention. While Auger and colleagues aimed to control for other restrictions, they did so using a step-wise procedure that selected covariates for the final model out of a list of candidates. This led to several distancing policies being excluded from the model during the post-school closure period. In effect, these other policies were omitted confounders. This approach gave school closures an advantage in emerging as the intervention with the greatest estimated impact, and this potential bias raises doubt as to whether school closures were responsible for the results. The interrupted time-series approach also forces a linear underlying trend in the pre- and post-school-closure periods, which is more restrictive than other methods.[2]

That said, the relevance of any early study for ongoing school reopening decisions is unclear. Levels of community spread are higher in many places, but awareness of preventive measures is also higher. Moving forward, similar considerations apply to future research. Since schools typically re-open in tandem with other reopening policies, researchers must be cautious about attributing changing epidemic patterns to school policies alone, particularly in models that do not treat other policies on a level playing field.

Charles Courtemanche, PhD, University of Kentucky, NBER, and IZA
Benjamin Sommers, MD/PhD, Harvard University
Aaron Yelowitz, PhD, University of Kentucky

REFERENCES

[1] Auger K, Shah S, Richardson T, et al. Association between statewide school closure and COVID-19 incidence and mortality in the US. JAMA. 2020; doi:10.1001/jama.2020.14348

[2] Courtemanche C, Garuccio J, Le A, et al. Strong social distancing measures in the United States reduced the COVID-19 growth rate. Health Affairs. 2020; doi:https://doi.org/10.1377/hlthaff.2020.00608

[3] Hsiang S, Allen D, Annan-Phan S, et al. The effect of large-scale anti-contagion policies on the COVID-19 pandemic. Nature. 2020; https://doi.org/10.1038/s41586-020-2404-8

[4] Flaxman S, Mishra S, Gandy A., et al. Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe. Nature. 2020; https://doi.org/10.1038/s41586-020-2405-7
CONFLICT OF INTEREST: None Reported
READ MORE
Response regarding other NPIs
Katherine Auger, MD, MSc | Cincinnati Children's Hospital Medical Center
To Dr. Courtemanche:

Thank you for this thoughtful comment. We conducted additional analyses to address the concern that our original models did not adequately account for non-pharmaceutical interventions (NPIs) enacted around the same time as school closure.

The original study was a population-based time series analysis of all 50 US states conducted between March 9, 2020, and May 7, 2020 incorporating a time lag for potential changes in cases and deaths to occur (1). Associations of school closure and outcomes of interest were modeled using negative binomial regression with stepwise regression.

Using original study data, sensitivity analyses
were performed to further investigate school closure and other NPIs by repeating the interrupted time series analyses excluding states that enacted either a stay at home order or closed all non-essential businesses—the two most societally restrictive policies—within 9 days of closing schools. Nine days was selected given modeling differences of 4 days between potential effects of school closure1 and other NPIs plus 5 days as the median time to onset of COVID-19 symptoms (2). Covariates from the initial models were retained (1).  All 4 NPIs (closure including non-essential business closure, stay at home orders, closure of restaurants, and gathering bans) were included as covariates in the sensitivity analyses, regardless of their inclusion in the original models. To estimate absolute differences associated with school closure, projected incidence and mortality if schools remained open were compared with the modeled incidence and mortality without school closure using a linear projection (using the same methods as in manuscript).

In total, 24 states either did not enact stay at home orders/non-essential business closure or enacted one or both orders >9 days after closing schools. States included in the sensitivity analyses had significantly lower cumulative incidence when schools closed, more children , more obesity, and lower urban density than excluded states. Included states enacted fewer non-school closure NPIs over more days than the excluded states (Table 1). The relative change in daily incidence per week associated with school closure was -76% (95%CI: -87%, -55%); the relative change in mortality per week associated with school closure was -78% (95%CI: -88%, -60%) (Table 2, Figure). Using a linear projection to estimate number of cases and deaths if schools had not closed, school closure was associated with 423.6 (95%CI: 361.8, 462.9) fewer cases per 100,000 residents over 26 days and 5.8 (95%CI: 4.5, 6.7) fewer deaths per 100,000 over 16 days (Table 3).

In sensitivity analyses of states that did not concurrently enact either stay at home orders or non-essential business closures when they closed schools, school closure was associated with fewer cases and deaths from COVID-19. The original national models capitalized on different order and timing of NPI enactment to isolate associations with school closure. These sensitivity analyses with similar results bolster the original findings that the association was related to school closure as opposed to residual confounding from NPIs.

These analyses have similar limitations as the original models. Additionally, multi-collinearity, introduced by including all NPIs, while addressed in the original model, may affect the estimates of these sensitivity analyses. Thus, our initial models presented in the manuscript better capture the association of school closure and COVID-19 nationally.

References

1. Auger KA, Shah SS, Richardson T, et al. Association Between Statewide School Closure and COVID-19 Incidence and Mortality in the US. JAMA. 2020.
2. Lauer SA, Grantz KH, Bi Q, et al. The Incubation Period of Coronavirus Disease 2019 (COVID-19) From Publicly Reported Confirmed Cases: Estimation and Application. Ann Intern Med. 2020;172(9):577-582.
3. United States Census Bureau. American Community Survey. https://www.census.gov/programs-surveys/acs. Accessed 04/02/2020.
4. Kaiser Family Foundation. Total Number of Residents in Certified Nursing Facilities. https://www.kff.org/f23fa35/. Published 2017. Accessed 06/15/2020.
5. The COVID Tracking Project. Data API. https://covidtracking.com/api. Accessed 05/08/2020.
6. Centers for Disease Control. Behavioral Risk Factor Surveillance System: 2018 Data. https://www.cdc.gov/brfss/annual_data/annual_2018.html. Accessed 06/15/2020.
7.  Centers for Disease Control. CDC Social Vulnerability Index (SVI). https://svi.cdc.gov. Accessed 04/16/2020.

Table 1. State Characteristics

 

Included in Sensitivity Analysesa

 

 

Excluded from Sensitivity Analyses

 

 

p valueb

 

Number of States

24

26

---

Cumulative incidence of COVID-19 at time of school closure

(per 100,000 residents), Median (IQR)

0.75 (0.52,1.21)

1.49 (0.81,2.88)

.02

Maximum testing rate per 1,000 residentsc, Median (IQR)

6.0 (5.0,6.0)

6.0 (5.0,8.0)

0.16

Percentage of state population < 15 years3, Median (IQR)

19.6 (18.6,20.5)

18.5 (17.5,19.2)

0.01

Percentage of state population> 65 years3, Median (IQR)

15.1 (14.1,16.1)

15.4 (14.5,16.0)

0.32

Number of nursing home residents per 1,000 people4, Median (IQR)

4.3 (3.4,5.7)

4.4 (2.7,5.3)

0.91

Percentage of state population with obesityd, Median (IQR)

33.6 (29.9,34.9)

30.0 (27.5,33.0)

0.02

Urban densitye

(urban residents per sq. mile)c

Cells include number of states and column percent in each category

<50

12 (50.0)

9 (34.6)

0.02

50-100

6 (25.0)

4 (15.4)

100-150

5 (20.8)

2 (7.7)

>150

1 (4.2)

11 (42.3)

Social vulnerability indexf, Median (IQR)

43.5 (29.8,61.3)

45.2 (40.1,52.5)

0.63

Number of other NPIs enactedg, Median (IQR)

4.0 (2.0,4.0)

4.0 (4.0,4.0)

0.001

Days between school closure and last NPI enacted, Median (IQR)

11.5 (10.0,14.5)

7.5 (5.0,9.0)

<0.001



Abbreviations: IQR- interquartile range; NPI – nonpharmaceutical interventions

a States included in sensitivity analysis enacted non-essential business closure and stay at home order at least 9 days after closing schools or did not enact statewide stay at home orders or business closures. Included states were Alaska, Arizona, Arkansas, Florida, Georgia, Iowa, Kansas, Kentucky, Louisiana, Maine, Missouri, Montana, Nebraska, New Hampshire, North Carolina, North Dakota, Oklahoma, South Carolina, South Dakota, Tennessee, Texas, Utah, Virginia, and Wyoming

b Median compared with Wilcoxon Rank Sum. Proportions compared with chi square test.

c State-level testing rate was calculated daily by dividing the cumulative number of tests performed in a state5 by the state population; modeled as a categorical variable. The minimum testing rate was <0.1 tests / 1000 residents and occurred on the first day of testing for every state. The maximum testing rate is the testing rate on the last day of the study period.

d Obesity6 defined as Body Mass Index (BMI) >30

e A state’s population density combined with the percentage of residents living in urban areas3; examined as a categorical variable

f A measure of communities’ preparedness for a natural disaster or illness outbreak by accounting for socioeconomic status, household composition, disability status, race and ethnicity composition, English proficiency, housing type and transportation. Scale is 0 to 100 with lower numbers indicating better preparedness for disaster; national median social vulnerability index is 50.7
g Number of the 4 non-school NPIs enacted ranged between 1-4. NPIs examined included: stay at home/shelter in place order, non-essential business closure, restaurant closure, and prohibition of gatherings greater than 10 people. All states enacted at least 1 non-school NPI.






Table 2. COVID-19 incidence and mortality pre-school closure, post-school closure, and effect size associated with school closure




Model






 






States in sensitivity analysis (n=24)






Incidence






Pre-school closure time period (during lag, days 1-16)




Actual new cases per 100,000 residents in the 16 days during the lag, Median (IQR)

18 (16, 27)

Unadjusted relative change per week (95% CI)

220%

(199%, 243%)

Compositea adjustedb relative change per week (95% CI)

291%

(194%, 419%)



Post-school closure time period (after lag, days 17-42)






 






Actual new cases per 100,000 residents in the 26 days after the lag, Median (IQR)




95 (75, 131)

Unadjusted relative change per week associated with school closure (95% CI)

-66%

(-70%, -63%)



Compositec adjustedb relative change per week (95%CI)






-14%






(-32%, 8%)




Adjustedb relative change per week associated with school closure (95% CI)

-76%

(-87%, -55%)



Mortality






Pre-school closure time (during lag, days 1-26)




Actual deaths per 100,000 residents in the 26 during the lag, Median (IQR)

1 (1 ,2)

Unadjusted relative change per week (95% CI)

157%

 (140%, 175%)

Compositea adjustedrelative change per week (95% CI)

186%

(150%, 228%)



Post-school closure time period (after lag, days 27-42)






Actual deaths per 100,000 residents in the 16 days after the lag




2 (1 ,3)

Unadjusted relative change per week associated with school closure (95% CI)

-65%

(-69%, -59%)



Compositec adjustedd relative change per week (95%CI)






-29%






(-47%, -6%)




Adjustedd relative change per week associated with school closure (95% CI)

-78%

(-88%, -60%)



a Composite change is the overall increase during the period before potential effects of school closure (during the lag).

b Adjusted for all four non-school NPIs and all model components retained in incidence model (intercept: percentage of state population <15 years, percentage of state population >65 years, and social vulnerability index; pre-school closure: stay at home/shelter in place order, restaurant closure, testing rate per 1,000 residents, and urban density; post-school closure: testing rate per 1,000 residents, stay at home/shelter in place order, percentage of state population >65 years, number of nursing home residents per 1,000 people, and urban density)

c Composite relative post-school change estimates are based on the linear combination of the following model parameter estimates: pre-school closure relative change, the effect associated with school closure, and all other post-school closure effects. The composite relative post-school change estimates are visualized in the post-school period in Figure.

d Adjusted for all four non-school NPIs and all model components retained in mortality model (intercept: percentage of state population <15 years, percentage of state population >65 years, and social vulnerability index; pre-school closure: stay at home/shelter in place order, prohibition of gatherings greater than 10 people, restaurant closure, percentage of state population <15 years, percentage of state population >65 years, number of nursing home residents per 1,000 people, and urban density; post-school closure: restaurant closure, number of nursing home residents per 1,000 people, and urban density).





Table 3. Estimated absolute differences in cases
a and deathsb between states with school closure and hypothetical states where schools remain open using linear projectionc.



 






 






Average stated in sensitivity analysis






 






Number of cases or deaths






(95% CI)






Incidence per 100,000 residents






With schools remaining open






Pre-school closure period (during 16 day lag, days 1-16)






30.6 (19.5 to 48.9)






In the 26 days after the lag, days 17-42






534.6 (467.5 to 658.0)






With school closure






Pre-school closure period (during 16 day lag, days 1-16)






30.6 (19.5 to 48.9)






In the 26 days after the lag, days 17-42






111.1 (71.7 to 172.8)






Difference in incidence during the 26 post-lag days between school closure and schools remaining open






423.6 (361.8 to 462.9)






Mortality per 100,000 residents






With schools remaining open






Pre-school closure period (during 26 day lag, days 1-26)






1.6 (1.0 to 2.4)






In the 16 days after the lag, days 27-42






8.3 (6.9 to 10.6)






With school closure






Pre-school closure period (during 26 day lag, days 1-26)






1.6 (1.0 to 2.4)






In the 16 days after the lag, days 27-42






2.4 (1.6 to 3.8)






Difference in deaths during the 16 post-lag days between school closure and schools remaining open






5.8 (4.5 to 6.7)






a Adjusted for all four non-school NPIs and all model components retained in stepwise incidence model (intercept: percentage of state population <15 years3, percentage of state population >65 years3, and social vulnerability index7; pre-school closure: testing rate per 1,000 residents5, and urban density3; post-school closure: testing rate per 1,000 residents5, percentage of state population >65 years3, number of nursing home residents per 1,000 people, and urban density3)

b Adjusted for all four non-school NPIs and all model components retained in stepwise mortality model (intercept: percentage of state population <15 years3, percentage of state population >65 years3, and social vulnerability index7; pre-school closure: percentage of state population <15 years3, percentage of state population >65 years3, number of nursing home residents per 1,000 people, and urban density3; post-school closure: number of nursing home residents per 1,000 people, and urban density3).

c Linear projection using slope from line tangential to the curve at the end of the lag period

d State with all characteristics set to the mean value for all covariates retained in the model.








Figure. Modeled association of school closure with COVID-19 incidence and mortality






Panel A. Daily Incidence for states included in sensitivity analyses                                                                   






Panel B. Mortality for states included in sensitivity analyses




 

Table 1. State Characteristics

 

Included in Sensitivity Analysesa

 

 

Excluded from Sensitivity Analyses

 

 

p valueb

 

Number of States

24

26

---

Cumulative incidence of COVID-19 at time of school closure

(per 100,000 residents), Median (IQR)

0.75 (0.52,1.21)

1.49 (0.81,2.88)

.02

Maximum testing rate per 1,000 residentsc, Median (IQR)

6.0 (5.0,6.0)

6.0 (5.0,8.0)

0.16

Percentage of state population < 15 years3, Median (IQR)

19.6 (18.6,20.5)

18.5 (17.5,19.2)

0.01

Percentage of state population> 65 years3, Median (IQR)

15.1 (14.1,16.1)

15.4 (14.5,16.0)

0.32

Number of nursing home residents per 1,000 people4, Median (IQR)

4.3 (3.4,5.7)

4.4 (2.7,5.3)

0.91

Percentage of state population with obesityd, Median (IQR)

33.6 (29.9,34.9)

30.0 (27.5,33.0)

0.02

Urban densitye

(urban residents per sq. mile)c

Cells include number of states and column percent in each category

<50

12 (50.0)

9 (34.6)

0.02

50-100

6 (25.0)

4 (15.4)

100-150

5 (20.8)

2 (7.7)

>150

1 (4.2)

11 (42.3)

Social vulnerability indexf, Median (IQR)

43.5 (29.8,61.3)

45.2 (40.1,52.5)

0.63

Number of other NPIs enactedg, Median (IQR)

4.0 (2.0,4.0)

4.0 (4.0,4.0)

0.001

Days between school closure and last NPI enacted, Median (IQR)

11.5 (10.0,14.5)

7.5 (5.0,9.0)

<0.001

 

Table 1. State Characteristics

 

Included in Sensitivity Analysesa

 

 

Excluded from Sensitivity Analyses

 

 

p valueb

 

Number of States

24

26

---

Cumulative incidence of COVID-19 at time of school closure

(per 100,000 residents), Median (IQR)

0.75 (0.52,1.21)

1.49 (0.81,2.88)

.02

Maximum testing rate per 1,000 residentsc, Median (IQR)

6.0 (5.0,6.0)

6.0 (5.0,8.0)

0.16

Percentage of state population < 15 years3, Median (IQR)

19.6 (18.6,20.5)

18.5 (17.5,19.2)

0.01

Percentage of state population> 65 years3, Median (IQR)

15.1 (14.1,16.1)

15.4 (14.5,16.0)

0.32

Number of nursing home residents per 1,000 people4, Median (IQR)

4.3 (3.4,5.7)

4.4 (2.7,5.3)

0.91

Percentage of state population with obesityd, Median (IQR)

33.6 (29.9,34.9)

30.0 (27.5,33.0)

0.02

Urban densitye

(urban residents per sq. mile)c

Cells include number of states and column percent in each category

<50

12 (50.0)

9 (34.6)

0.02

50-100

6 (25.0)

4 (15.4)

100-150

5 (20.8)

2 (7.7)

>150

1 (4.2)

11 (42.3)

Social vulnerability indexf, Median (IQR)

43.5 (29.8,61.3)

45.2 (40.1,52.5)

0.63

Number of other NPIs enactedg, Median (IQR)

4.0 (2.0,4.0)

4.0 (4.0,4.0)

0.001

Days between school closure and last NPI enacted, Median (IQR)

11.5 (10.0,14.5)

7.5 (5.0,9.0)

<0.001

CONFLICT OF INTEREST: None Reported
READ MORE
Original Investigation
July 29, 2020

Association Between Statewide School Closure and COVID-19 Incidence and Mortality in the US

Author Affiliations
  • 1Division of Hospital Medicine, Cincinnati Children’s Hospital Medical Center, Cincinnati, Ohio
  • 2James M. Anderson Center for Health Systems Excellence, Cincinnati Children’s Hospital Medical Center, Cincinnati, Ohio
  • 3Department of Pediatrics, University of Cincinnati College of Medicine, Cincinnati, Ohio
  • 4Pediatric Research in Inpatient Settings Network, Cincinnati, Ohio
  • 5Division of Infectious Diseases, Cincinnati Children’s Hospital Medical Center, Cincinnati, Ohio
  • 6Division of Emergency Medicine, Cincinnati Children’s Hospital Medical Center, Cincinnati, Ohio
JAMA. 2020;324(9):859-870. doi:10.1001/jama.2020.14348
Key Points

Question  Was statewide school closure associated with decreased incidence and mortality for coronavirus disease 2019 (COVID-19)?

Findings  In this US population–based time series analysis conducted between March 9, 2020, and May 7, 2020, school closure was associated with a significant decline in both incidence of COVID-19 (adjusted relative change per week, −62%) and mortality (adjusted relative change per week, −58%). In a model derived from this analysis, it was estimated that closing schools when the cumulative incidence of COVID-19 was in the lowest quartile compared with the highest quartile was associated with 128.7 fewer cases per 100 000 population over 26 days and with 1.5 fewer deaths per 100 000 population over 16 days.

Meaning  There was a temporal association between statewide school closure and lower COVID-19 incidence and mortality, although some of the reductions may have been related to other concurrent nonpharmaceutical interventions.

Abstract

Importance  In the US, states enacted nonpharmaceutical interventions, including school closure, to reduce the spread of coronavirus disease 2019 (COVID-19). All 50 states closed schools in March 2020 despite uncertainty if school closure would be effective.

Objective  To determine if school closure and its timing were associated with decreased COVID-19 incidence and mortality.

Design, Setting, and Participants  US population–based observational study conducted between March 9, 2020, and May 7, 2020, using interrupted time series analyses incorporating a lag period to allow for potential policy-associated changes to occur. To isolate the association of school closure with outcomes, state-level nonpharmaceutical interventions and attributes were included in negative binomial regression models. States were examined in quartiles based on state-level COVID-19 cumulative incidence per 100 000 residents at the time of school closure. Models were used to derive the estimated absolute differences between schools that closed and schools that remained open as well as the number of cases and deaths if states had closed schools when the cumulative incidence of COVID-19 was in the lowest quartile compared with the highest quartile.

Exposures  Closure of primary and secondary schools.

Main Outcomes and Measures  COVID-19 daily incidence and mortality per 100 000 residents.

Results  COVID-19 cumulative incidence in states at the time of school closure ranged from 0 to 14.75 cases per 100 000 population. School closure was associated with a significant decline in the incidence of COVID-19 (adjusted relative change per week, −62% [95% CI, −71% to −49%]) and mortality (adjusted relative change per week, −58% [95% CI, −68% to −46%]). Both of these associations were largest in states with low cumulative incidence of COVID-19 at the time of school closure. For example, states with the lowest incidence of COVID-19 had a −72% (95% CI, −79% to −62%) relative change in incidence compared with −49% (95% CI, −62% to −33%) for those states with the highest cumulative incidence. In a model derived from this analysis, it was estimated that closing schools when the cumulative incidence of COVID-19 was in the lowest quartile compared with the highest quartile was associated with 128.7 fewer cases per 100 000 population over 26 days and with 1.5 fewer deaths per 100 000 population over 16 days.

Conclusions and Relevance  Between March 9, 2020, and May 7, 2020, school closure in the US was temporally associated with decreased COVID-19 incidence and mortality; states that closed schools earlier, when cumulative incidence of COVID-19 was low, had the largest relative reduction in incidence and mortality. However, it remains possible that some of the reduction may have been related to other concurrent nonpharmaceutical interventions.

Introduction

The novel severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) causing coronavirus disease 2019 (COVID-19) was first identified in the US in January 2020, with subsequent spread throughout the country. In the absence of effective treatments, governors and state health officials enacted policies aimed at reducing infections through nonpharmaceutical interventions.1 The nonpharmaceutical interventions included: school closure, nonessential business closure, restaurant and bar closure, and prohibition of gatherings with more than 10 people. With limited precedent and a paucity of evidence on the effectiveness of nonpharmaceutical interventions, policies varied markedly state to state in scope and timing.

Children infected with SARS-CoV-2 may be asymptomatic or have mild symptoms indistinguishable from other common upper respiratory tract infections,2,3 allowing them to spread the virus when they feel well. Children are often key transmitters in viral epidemics like influenza4 because of spending prolonged periods in close proximity to other children at school and during physical activities. Prior studies have demonstrated an association between school closure and reduced transmission of viral respiratory illnesses.5-8 Given concerns that children represented a significant vector for SARS-CoV-2, all states closed schools despite a lack of evidence supporting the effectiveness of school closure in curbing the spread of this virus.

Schools promote child education, growth, development, and overall well-being.9 Knowing whether school closure is effective in reducing infections is critical to reduce the negative effects of continued school closure on child health if school closure is ineffective. This national study assessed the association between school closure and its timing with subsequent COVID-19 incidence and mortality, with the hypothesis that any association between school closure and incidence and mortality would be strongest in states that closed schools early when the cumulative incidence of disease was low.

Methods

The study was a population-based time series analysis of all 50 US states conducted between March 9, 2020, and May 7, 2020. This period allowed for at least 6 weeks of data collection after school closures in each state. The institutional review board at Cincinnati Children’s Hospital Medical Center deemed this study was not subject to oversight given the use of publicly available data.

Independent Variables

Associations of primary and secondary school closure (kindergarten-grade 12) and its timing with outcomes of interest were examined. Because school closure timing varied relative to disease progression in the state, we examined the cumulative incidence of COVID-19 (defined as total number of cases per 100 000 residents) grouped in quartiles by the date the school closure policy went into effect.

The analysis was performed by quartiles of cumulative incidence of COVID-19 instead of as a continuous variable because the relationships between baseline cumulative incidence and outcomes were not assumed to be linear.

Outcome Measures

Daily COVID-19 incidence and daily mortality per 100 000 residents in each state were estimated using publicly available data from the Johns Hopkins University School of Public Health.10 This source aggregates data from the US Centers for Disease Control and Prevention (CDC) as well as from state and local public health departments. In accordance with CDC guidelines, confirmed COVID-19 cases include presumptive positive cases and probable cases, and death totals include confirmed and probable cases. The denominator for the outcome measures was the state population from the 2018 American Community Survey.11

Covariates

State characteristics were included as covariates in the models to assess the independent associations with school closure. For each state, the following non–school-related nonpharmaceutical intervention covariates were considered: stay-at-home or shelter-in-place order, nonessential business closure, restaurant and bar closure, and prohibition of gatherings with more than 10 people. These nonpharmaceutical interventions were included based on the policy effective date (eTable 1 in the Supplement) plus a lag period to allow for any potential policy-related effects on daily COVID-19 incidence and mortality (eFigure in the Supplement). Potential effects associated with subsequent lifting of nonpharmaceutical interventions occurred outside the study period and thus are not included (eMethods in the Supplement).

SARS-CoV-2 testing rates varied by state and throughout the study period. To account for this variation, state-level COVID-19 testing12 (calculated daily as the cumulative number of tests per 1000 residents) was modeled as a categorical variable. State measures of urban population density (a measure of the state’s population density combined with the percentage of residents living in urban areas),11 percentage of the state’s population with obesity,13 percentage of the state’s population aged 15 years or younger,11 percentage of the state’s population aged 65 years or older,11 and the number of nursing home residents per 1000 people were included.14 The CDC social vulnerability index also was included; this index accounts for multiple factors, including socioeconomic status, household composition, disability status, race and ethnicity composition, English-language proficiency, housing type, and transportation access, to assess a community’s preparedness for a natural disaster or illness outbreak.15 Many of these factors have been associated with COVID-19 disease, mortality, or both. All covariates are described in detail in the eMethods in the Supplement.

Statistical Analysis

Interrupted time series analyses were used to compare the daily change in outcomes (daily COVID-19 incidence and mortality) before and after school closure. Acknowledging that school closure and other nonpharmaceutical interventions would not have immediate effects on COVID-19 incidence and mortality, estimates were used to determine when school-based exposure could be expected to lead to changes in COVID-19 incidence and associated mortality (eFigure in the Supplement). A time from exposure to symptom onset of 5 days was assumed per Lauer et al.16 Given the early emphasis (and some state restrictions17-25) on limiting testing to hospitalized patients, time to diagnosis was defined as time between symptom onset and hospitalization (7 days).26 For school closure, given the low documented prevalence of COVID-19 in children, an additional period was included for a child to infect an adult, assuming a child exposed at school could expose an adult prior to symptom onset and within 4 days. The analyses for the mortality outcome assumed 17 days from symptom onset to death.27 For details on lag period calculations, see the eMethods in the Supplement.

Daily COVID-19 incidence and mortality were modeled using negative binomial regression. Interactions between school closure and all included covariates were explored because school closure may affect at-risk communities differently.9 Given the large number of covariates and interactions considered, a single parsimonious model for each outcome was created selecting covariates from the primary model using a stepwise regression approach, with entry and removal criteria specified as a P value of <.20. Because factors associated with COVID-19 incidence and mortality may vary with school closure, covariate selection was completed independently during and after the lag period (eMethods in the Supplement).

Results are reported as the relative change in the outcome from week to week. Adjusted changes in both daily COVID-19 incidence and daily mortality over time are graphically displayed for all states by quartile of cumulative incidence at the time of school closure. To estimate absolute differences associated with school closure, the projected COVID-19 incidence and mortality if schools had remained open were compared with the modeled incidence and mortality with school closure. Both linear and exponential assumptions were used to project COVID-19 incidence and mortality if schools had remained open (eMethods in the Supplement). To estimate absolute differences in outcomes based on school closure timing, model parameters were used to estimate the absolute differences in the number of COVID-19 cases and deaths for a state that closed schools when the cumulative incidence of COVID-19 was in the lowest quartile compared with a state that closed schools when the cumulative incidence of COVID-19 was in the highest quartile (eMethods in the Supplement).

Analyses were performed using SAS version 9.4 (SAS Institute Inc) and 2-sided P values of <.05 were considered statistically significant.

Sensitivity Analyses Around the Lag Period

Because the COVID-19 incidence and mortality lag period estimates are based on emerging evidence, the sensitivity analyses examined the robustness of findings if the lag period was shorter (10 days for incidence and 20 days for mortality) or longer (21 days for incidence and 33 days for mortality). Detailed methods and rationale for the sensitivity ranges appear in the eMethods and eFigure in the Supplement.

Results

All 50 states closed schools between March 13, 2020, and March 23, 2020. The cumulative incidence of COVID-19 at the time of school closure ranged from 0 to 14.75 cases per 100 000 population. State characteristics by COVID-19 incidence quartile at the time of school closure appear in Table 1. There was wide variability in the testing rate per 1000 residents and in the number of nursing home residents per 1000 people. There was less variability in the percentage of the state populations for the number of nonpharmaceutical interventions enacted; 39 states enacted all 4 nonpharmaceutical interventions examined.

States in the highest quartile of cumulative incidence of COVID-19 at the time of school closure enacted multiple nonpharmaceutical interventions over a shorter period. The median time from school closure to the last enacted nonpharmaceutical intervention was 5 days (interquartile range, 2.5-8 days). In comparison, states in the lowest quartile of cumulative incidence of COVID-19 at the time of school closure enacted nonpharmaceutical interventions over a longer period. The median time from school closure to the last enacted nonpharmaceutical intervention was 12 days (interquartile range, 8-14 days; Table 1).

COVID-19 Incidence

The observed case rates of COVID-19 in each state (relative to the day of school closure by cumulative incidence) and the 16-day lag period are depicted in Figure 1A. In the unadjusted analyses during the period prior to potential effects of school closure (ie, during the lag period), the overall relative change in COVID-19 incidence per week was 220% (95% CI, 205% to 236%). The unadjusted relative change per week associated with school closure was −68% (95% CI, −70% to −66%). The unadjusted effect size associated with school closure varied by cumulative COVID-19 incidence at the time of school closure, with states in the highest quartile of cumulative COVID-19 incidence having the smallest relative effect size (Table 2).

In the adjusted analyses during the period prior to potential effects of school closure (ie, during the lag period), the relative change in COVID-19 incidence per week was 265% (95% CI, 231% to 303%; Table 2). The overall combined composite relative weekly change in COVID-19 incidence after school closure was 10% (95% CI, 1% to 18%). This composite change after school closure is a combination of the changes associated with school closure and other non–school-related changes during the period after school closure and is visually depicted as the change after school closure in Figure 2. When examining only school closure, it was associated with a relative change in COVID-19 incidence per week of −62% (95% CI, −71% to −49%; Table 2).

The states that closed early, when the cumulative incidence of COVID-19 was lowest, had the greatest relative change per week associated with school closure (−72% [95% CI, −79% to −62%]). States that were slowest to close schools and had the highest cumulative incidence of COVID-19 had a relative change per week associated with school closure of −49% (95% CI, −62% to −33%; Table 2 and Figure 2). The full model with all covariate estimates appears in eTable 2 in the Supplement. The relative change associated with school closure for COVID-19 incidence varied significantly by the testing rate per 1000 residents, by the percentage of the state’s population aged 65 years or older, by the number of nursing home residents per 1000 people, and by urban density. Information on interpreting relative weekly changes appears in the eMethods in the Supplement.

The absolute effects associated with school closure during the 26-day period after school closure (days 17-42), which were calculated using model estimates with the assumption of linear growth, yielded 638.7 cases per 100 000 that would have occurred if schools had remained open (Table 3). Compared with the 214.8 cases per 100 000 estimated from the school closure model, the absolute difference associated with school closure was 423.9 (95% CI, 375.0 to 463.7) cases per 100 000. States that closed schools late (in the highest quartile of cumulative incidence of COVID-19) had the largest absolute reduction in cases (621.7 [95% CI, 535.4 to 742.6] per 100 000). However, states that closed schools earlier (in the lowest quartile) had fewer total cases (−128.7 [95% CI, −168.7 to −74.2] per 100 000) during the period after school closure (Table 3). The absolute difference in COVID-19 incidence assuming continued exponential growth appears in eTable 3 in the Supplement.

Sensitivity Analyses Around the Lag Period

The point estimates ranged from −61% to −63% for the relative change per week associated with school closure as the lag period varied. The point estimates for the relative change for each quartile varied slightly across the lag period (eTable 4 in the Supplement).

COVID-19 Mortality

The observed death rates in each state by quartile of cumulative incidence of COVID-19 at the time of school closure appears in Figure 1B. In the unadjusted analyses during the period prior to potential effects of school closure (ie, during the lag period), the overall relative change in mortality per week was 171% (95% CI, 160% to 184%). In the unadjusted analyses, the relative mortality change per week associated with school closure was −64% (95% CI, −67% to −61%). The unadjusted effect size associated with school closure varied by COVID-19 cumulative incidence at the time of school closure, with states in the lowest quartile having the largest associated effect size (Table 2).

In the adjusted analyses during the period prior to potential effects of school closure (ie, during the lag period), COVID-19 mortality increased by 186% (95% CI, 175% to 197%) per week (Table 2). The overall combined composite relative weekly change in mortality after school closure was 2% (95% CI, −8% to 14%; Table 2). This composite change in mortality after school closure is visually depicted in Figure 2. When examining only school closure, it was associated with a relative change per week in COVID-19 mortality of −58% (95% CI, −67% to −46%). This association was greatest in states with the lowest cumulative COVID-19 incidence at the time of school closure (relative change per week of −64% [95% CI, −73% to −52%]). In comparison, states that closed schools later when cumulative COVID-19 incidence was in the highest quartile had the smallest associated relative decline in mortality (−53% [95% CI, −63% to −40%]; Table 2 and Figure 2). The full model with all covariates appears in eTable 5 in the Supplement. The relative change in mortality associated with school closure varied significantly by restaurant and bar closure and urban density.

The absolute effects associated with school closure during the 16-day period after school closure (days 27-42), which were calculated using model estimates with the assumption of linear growth, yielded 19.4 deaths per 100 000 that would have occurred if schools had remained open (Table 4). Compared with the 6.8 deaths per 100 000 estimated from the school closure model, the absolute difference associated with school closure was 12.6 (95% CI, 11.8 to 13.6) deaths per 100 000 (Table 4). States that closed schools late (in the highest quartile of COVID-19 cumulative incidence) had the largest absolute reduction in deaths (15.8 [95% CI, 13.9 to 18.1] per 100 000). However, states that closed schools earlier (in the lowest quartile) had fewer estimated total deaths (−1.5 [95% CI, −2.7 to −0.1] per 100 000) during the period after school closure (Table 4). The absolute difference in deaths assuming continued exponential growth appears in eTable 3 in the Supplement.

Sensitivity Analyses Around the Lag Period

The point estimates ranged from −55% to −61% for the relative change associated with school closure as the lag period varied. The point estimates for the relative change for each quartile varied slightly across the lag period (eTable 4 in the Supplement).

Discussion

Between March 9, 2020, and May 7, 2020, school closure in the US was temporally associated with decreased COVID-19 incidence and mortality. States that closed schools earlier (when the state’s cumulative incidence was lower) had the largest relative reduction in overall incidence and mortality.

In March 2020, states enacted multiple nonpharmaceutical interventions, including closing schools, nonessential businesses, and restaurants and bars, and prohibiting large gatherings, to curb SARS-CoV-2 spread and prevent death. Completely isolating the effects of any single nonpharmaceutical intervention is impossible because recommendations for increased handwashing, cleaning, and wearing of masks evolved simultaneously. Measured COVID-19 incidence also was affected by testing availability, which was limited early in the pandemic and varied nationally.

In this study, changes in COVID-19 incidence and mortality associated with school closure were isolated to the extent possible by adjusting for other state-enacted policies and testing rates. In adjusted models, school closure was associated with decreased COVID-19 incidence and deaths. These analyses do not incorporate the risks of school closure on child education and development or from a societal perspective. However, the analyses suggest that school closure may be effective in curbing SARS-CoV-2 spread and preventing deaths during future outbreaks.

These findings complement evolving evidence on the role of children in the transmission of SARS-CoV-2. Studies have documented lower attack rates for children,28 and children comprise a small proportion of documented infections.29 Children may be less susceptible to SARS-CoV-2 infection30; however, studies have documented viral shedding in asymptomatic children.31 Recent studies suggest school closure may have only modest effects on COVID-19 deaths.32-35 School closure in this study was associated with a −62% relative change in COVID-19 incidence per week. A decline of 62% was equivalent to 39% of the projected value with schools open. So, per week, the incidence was estimated to have been 39% of what it would have been had schools remained open. Extrapolating the absolute differences of 423.9 cases and 12.6 deaths per 100 000 to 322.2 million residents nationally suggests that school closure may have been associated with approximately 1.37 million fewer cases of COVID-19 over a 26-day period and 40 600 fewer deaths over a 16-day period; however, these figures do not account for uncertainty in the model assumptions and the resulting estimates.

The analyses presented here suggest that the timing of school closure plays a role in the magnitude of changes associated with school closure. As hypothesized, school closure in states that enacted this intervention early (when the cumulative incidence of COVID-19 was low) had greater associated relative decreases in incidence and mortality. Although these relative differences translate into smaller absolute differences associated with school closure, states that closed schools later (in the highest quartile of COVID-19 cumulative incidence) had more new cases and deaths from COVID-19 during the period after school closure. Thus, this study can inform future decisions about optimal timing for state and local officials to consider school closure to curb SARS-CoV-2 spread in the high likelihood that the pandemic continues.

The mechanism by which school closure could affect COVID-19 spread is not only through disrupting spread by or among children. School closure affects family routines, necessitating alternative childcare and modified work schedules. These changes are evident by the number of teleworkers more than doubling.36,37 The disruption in everyday life likely influenced how people engaged in group activities, traveled, and conducted business. If the primary effect associated with school closure is related to altered adult behavior, and not children spreading the virus to adults, the primary lag period considered in these analyses should be adjusted. Eliminating the 4 days for a child to adult transmission would result in a COVID-19 incidence lag period of 12 days and a mortality lag period of 22 days. In sensitivity analyses, the effect sizes associated with school closure at these shorter lag periods were similar to the primary analysis effect sizes. The degree to which the associations with school closure relate to decreased spread of SARS-CoV-2 by children or a combination of child and adult factors is unclear. Because school closure was the first nonpharmaceutical intervention in most states, the effects associated with school closure may be larger than if school closure had followed other nonpharmaceutical interventions.

It is unclear how COVID-19 spread would be affected if schools remained open while states enacted other policies to restrict movement. It is possible school-related spread may be mitigated with infection-control interventions recommended by the CDC and the American Academy of Pediatrics, including frequent handwashing, universal mask policies, physical distancing measures, and increased sanitation procedures.38,39 However, given that school closure also alters adult behavior, decreasing COVID-19 spread within schools may be inadequate as a stand-alone intervention and may require continued alteration of adult interactions.

Limitations

This study has several limitations. First, many states enacted additional nonpharmaceutical interventions concurrently with or shortly after school closure, making it impossible to fully isolate potential effects of school closure. Some nonpharmaceutical interventions, such as increased handwashing, could not be included due to lack of available data.

Second, analyses were conducted at the state level. The analyses did not account for resident travel leading to viral spread between states. Even though the study modeled state-level policies, some states had more restrictive policies locally (ie, by county). Nevertheless, these analyses are useful to understand the practical implications of state policy in containing spread.

Third, inadequate testing has impeded COVID-19 diagnosis. Testing variability was accounted for with the use of state-level testing rates as a model covariate; however, testing rates do not fully capture a state’s testing capability, infrastructure, and strictness of testing guidelines.

Fourth, the completeness and accuracy of the Johns Hopkins University database with respect to COVID-19 incidence and mortality has not been established. This data source aggregates publicly available data and accuracy may vary state to state. As with limitations in testing, inconsistencies in reporting are unavoidable limitations of all COVID-19 US population-based studies.

Conclusions

Between March 9, 2020, and May 7, 2020, school closure in the US was temporally associated with decreased COVID-19 incidence and mortality; states that closed schools earlier, when cumulative incidence of COVID-19 was low, had the largest relative reduction in incidence and mortality. However, it remains possible that some of the reduction may have been related to other concurrent nonpharmaceutical interventions.

Back to top
Article Information

Corresponding Author: Katherine A. Auger, MD, MSc, Division of Hospital Medicine, Cincinnati Children’s Hospital Medical Center, 3333 Burnet Ave, ML 9016, Cincinnati, OH 45229 (katherine.auger@cchmc.org).

Accepted for Publication: July 17, 2020.

Published Online: July 29, 2020. doi:10.1001/jama.2020.14348

Author Contributions: Dr Richardson had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Brady, Schondelmeyer, and Thomson made equal substantial contributions to the manuscript and are co-last authors.

Concept and design: Auger, Shah, Richardson, Hartley, Hall, Warniment, Bosse, Ferris, Brady, Schondelmeyer, Thomson.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Auger, Richardson, Timmons, Brady, Schondelmeyer.

Critical revision of the manuscript for important intellectual content: Shah, Richardson, Hartley, Hall, Warniment, Bosse, Ferris, Brady, Schondelmeyer, Thomson.

Statistical analysis: Auger, Richardson, Hartley, Hall.

Administrative, technical, or material support: Hartley, Warniment, Timmons, Bosse, Ferris, Thomson.

Supervision: Brady, Schondelmeyer, Thomson.

Conflict of Interest Disclosures: None reported.

Funding/Support: Funding for this work was provided by Agency for Healthcare Research and Quality awards K08HS024735 (Dr Auger), K08HS023827 (Dr Brady), K08HS026763 (Dr Schondelmeyer), and K08HS025138 (Dr Thomson) and award 5UL1TR001425-04 from the National Center for Advancing Translational Sciences, National Institutes of Health.

Role of the Funder/Sponsor: The Agency for Healthcare Research and Quality and the National Institutes of Health had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The content is solely the responsibility of the authors and does not represent the official views of the Agency for Healthcare Research and Quality, the National Institutes of Health, Cincinnati Children’s Hospital Medical Center, the Children’s Hospital Association, or the Pediatric Research in Inpatient Settings Network.

Additional Information: Drs Richardson and Hall are affiliated with the Children’s Hospital Association, Lenexa, Kansas.

References
1.
Pan  A, Liu  L, Wang  C,  et al.  Association of public health interventions with the epidemiology of the COVID-19 outbreak in Wuhan, China.   JAMA. 2020;323(19):1915-1923. PubMedGoogle ScholarCrossref
2.
Dong  Y, Mo  X, Hu  Y,  et al.  Epidemiology of COVID-19 among children in China.   Pediatrics. 2020;145(6):e20200702.PubMedGoogle Scholar
3.
Lu  X, Zhang  L, Du  H,  et al.  SARS-CoV-2 infection in children.   N Engl J Med. 2020;382(17):1663-1665.PubMedGoogle ScholarCrossref
4.
Worby  CJ, Chaves  SS, Wallinga  J,  et al.  On the relative role of different age groups in influenza epidemics.   Epidemics. 2015;13:10-16.PubMedGoogle ScholarCrossref
5.
Cauchemez  S, Valleron  AJ, Boëlle  PY,  et al.  Estimating the impact of school closure on influenza transmission from sentinel data.   Nature. 2008;452(7188):750-754.PubMedGoogle ScholarCrossref
6.
Schanzer  D, Vachon  J, Pelletier  L.  Age-specific differences in influenza A epidemic curves.   Am J Epidemiol. 2011;174(1):109-117.PubMedGoogle ScholarCrossref
7.
Luca  G, Kerckhove  KV, Coletti  P,  et al.  The impact of regular school closure on seasonal influenza epidemics.   BMC Infect Dis. 2018;18(1):29.PubMedGoogle ScholarCrossref
8.
Chu  Y, Wu  Z, Ji  J,  et al.  Effects of school breaks on influenza-like illness incidence in a temperate Chinese region.   BMJ Open. 2017;7(3):e013159.PubMedGoogle Scholar
9.
United Nations. Policy brief: the impact of COVID-19 on children. Accessed May 11, 2020. https://www.un.org/sites/un2.un.org/files/policy_brief_on_covid_impact_on_children_16_april_2020.pdf
10.
Dong  E, Du  H, Gardner  L.  An interactive web-based dashboard to track COVID-19 in real time.   Lancet Infect Dis. 2020;20(5):533-534.PubMedGoogle ScholarCrossref
11.
US Census Bureau. American Community Survey. Accessed April 2, 2020. https://www.census.gov/programs-surveys/acs
12.
COVID Tracking Project. Data API. Accessed May 8, 2020. https://covidtracking.com/api
13.
US Centers for Disease Control and Prevention. Behavioral Risk Factor Surveillance System: 2018 data. Accessed June 15, 2020. https://www.cdc.gov/brfss/annual_data/annual_2018.html
14.
Kaiser Family Foundation. Total number of residents in certified nursing facilities. Accessed June 15, 2020. https://www.kff.org/f23fa35/
15.
US Centers for Disease Control and Prevention. Social vulnerability index. Accessed June 15, 2020. https://svi.cdc.gov
16.
Lauer  SA, Grantz  KH, Bi  Q,  et al.  The incubation period of coronavirus disease 2019 (COVID-19) from publicly reported confirmed cases.   Ann Intern Med. 2020;172(9):577-582.PubMedGoogle ScholarCrossref
17.
Colorado Department of Public Health and Environment. Get tested. Accessed March 23, 2020. https://covid19.colorado.gov
18.
State of Connecticut. Connecticut COVID-19 response. Accessed March 30, 2020. https://portal.ct.gov/Coronavirus
19.
Georgia Department of Public Health. COVID-19. Accessed March 23, 2020. https://dph.georgia.gov
20.
Indiana State Department of Health. 2019 novel coronavirus (COVID-19). Accessed March 31, 2020. https://coronavirus.in.gov
21.
Iowa Department of Public Health. Novel coronavirus acute respiratory disease: COVID-19. Accessed March 26, 2020. https://idph.iowa.gov
22.
Minnesota Department of Health. Coronavirus disease 2019 (COVID-19). Accessed March 24, 2020. https://www.health.state.mn.us/diseases/coronavirus/index.html
23.
Montana Department of Public Health and Human Services. 2019 novel coronavirus (COVID-19). Accessed March 27, 2020. https://dphhs.mt.gov
24.
New Hampshire Department of Health and Human Services. Novel coronavirus 2019 (COVID-19). Accessed March 25, 2020. https://www.nh.gov/covid19
25.
South Carolina Department of Health and Environmental Control. Coronavirus disease 2019 (COVID-19). Accessed April 3, 2020. https://www.scdhec.gov/infectious-diseases/viruses/coronavirus-disease-2019-covid-19
26.
Garg  S, Kim  L, Whitaker  M,  et al.  Hospitalization rates and characteristics of patients hospitalized with laboratory-confirmed coronavirus disease 2019.   MMWR Morb Mortal Wkly Rep. 2020;69(15):458-464.PubMedGoogle ScholarCrossref
27.
Beigel  JH, Tomashek  KM, Dodd  LE,  et al.  Remdesivir for the treatment of Covid-19.   N Engl J Med. Published online May 22, 2020. doi:10.1056/NEJMoa2007764PubMedGoogle Scholar
28.
Li  W, Zhang  B, Lu  J,  et al.  The characteristics of household transmission of COVID-19.   Clin Infect Dis. 2020;ciaa450.PubMedGoogle Scholar
29.
Gudbjartsson  DF, Helgason  A, Jonsson  H,  et al.  Spread of SARS-CoV-2 in the Icelandic population.   N Engl J Med. 2020;382(24):2302-2315.PubMedGoogle ScholarCrossref
30.
Bunyavanich  S, Do  A, Vicencio  A.  Nasal gene expression of angiotensin-converting enzyme 2 in children and adults.   JAMA. 2020;323(23):2427-2429.PubMedGoogle ScholarCrossref
31.
Zhang  J, Litvinova  M, Liang  Y,  et al.  Changes in contact patterns shape the dynamics of the COVID-19 outbreak in China.   Science. 2020;368(6498):1481-1486.PubMedGoogle ScholarCrossref
32.
Ferguson  NM, Laydon  D, Nedjati-Gilana  G,  et al. Report 9: impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare demand. Accessed July 20, 2020. https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-College-COVID19-NPI-modelling-16-03-2020.pdf
33.
Viner  RM, Russell  SJ, Croker  H,  et al.  School closure and management practices during coronavirus outbreaks including COVID-19.   Lancet Child Adolesc Health. 2020;4(5):397-404.PubMedGoogle ScholarCrossref
34.
Esposito  S, Principi  N.  School closure during the coronavirus disease 2019 (COVID-19) pandemic.   JAMA Pediatr. Published online May 13, 2020. doi:10.1001/jamapediatrics.2020.1892 PubMedGoogle Scholar
35.
Hsiang  S, Allen  D, Annan-Phan  S,  et al.  The effect of large-scale anti-contagion policies on the COVID-19 pandemic.   Nature. Published online June 8, 2020. doi:10.1038/s41586-020-2404-8 PubMedGoogle Scholar
36.
Koetsier  J. 58% of American knowledge workers are now working remotely. Accessed June 9, 2020. https://www.forbes.com/sites/johnkoetsier/2020/03/20/58-of-american-knowledge-workers-are-now-working-remotely/#517af4b13303
37.
Rosalsky  G. COVID-19 forces more people to work from home: how’s it going? Accessed June 9, 2020. https://www.npr.org/2020/05/08/852527736/covid-19-forces-more-people-to-work-from-home-hows-it-going
38.
US Centers for Disease Control and Prevention. Schools during the COVID-19 pandemic. Accessed May 15, 2020. https://www.cdc.gov/coronavirus/2019-ncov/community/schools-childcare/Schools-Decision-Tree.pdf
39.
American Academy of Pediatrics. COVID-19 planning considerations: guidance for school re-entry. Accessed July 20, 2020. https://services.aap.org/en/pages/2019-novel-coronavirus-covid-19-infections/clinical-guidance/covid-19-planning-considerations-return-to-in-person-education-in-schools/
×