[Skip to Navigation]
Sign In
Figure 1.  Analysis Overview
Analysis Overview

ACS indicates US Census Bureau American Community Survey; CDC, US Centers for Disease Control and Prevention; COVID-19, coronavirus disease 2019; MMWR, Morbidity and Mortality Weekly Report; and SSA, US Social Security Administration.

Figure 2.  Estimated Years of Life Lost
Estimated Years of Life Lost

Center dots indicate medians; thin lines, 91% probability intervals; thick lines, 50% probability intervals; and shaded areas, probability density of the estimated years of life lost under each condition.

Table.  Years of Life Lost Due to COVID-19 Deaths Through June 3, 2020
Years of Life Lost Due to COVID-19 Deaths Through June 3, 2020
1.
Shekerdemian  LS, Mahmood  NR, Wolfe  KK,  et al; International COVID-19 PICU Collaborative.  Characteristics and outcomes of children with coronavirus disease 2019 (COVID-19) infection admitted to US and Canadian pediatric intensive care units.   JAMA Pediatr. 2020;174(9):868-873. doi:10.1001/jamapediatrics.2020.1948PubMedGoogle Scholar
2.
Zachariah  P, Johnson  CL, Halabi  KC,  et al; Columbia Pediatric COVID-19 Management Group.  Epidemiology, clinical features, and disease severity in patients with coronavirus disease 2019 (COVID-19) in a children’s hospital in New York City, New York.   JAMA Pediatr. 2020;e202430. doi:10.1001/jamapediatrics.2020.2430PubMedGoogle Scholar
3.
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. doi:10.1056/NEJMoa2006100PubMedGoogle Scholar
4.
Park  YJ, Choe  YJ, Park  O,  et al; COVID-19 National Emergency Response Center, Epidemiology and Case Management Team.  Contact tracing during coronavirus disease outbreak, South Korea, 2020.   Emerg Infect Dis. 2020;26(10):2465-2468. doi:10.3201/eid2610.201315PubMedGoogle Scholar
5.
Christakis  DA.  School reopening: the pandemic issue that is not getting its due.   JAMA Pediatr. Published online May 13, 2020. doi:10.1001/jamapediatrics.2020.2068PubMedGoogle Scholar
6.
US Department of Education Office of Planning, Evaluation, and Policy Development. Evaluation of evidence-based practices in online learning: a meta-analysis and review of online learning studies. Published 2010. Accessed October 19, 2020. https://www2.ed.gov/rschstat/eval/tech/evidence-based-practices/finalreport.pdf
7.
Fitzpatrick  B, Berends  M, Ferrare  J, Waddington  R.  Virtual illusion: comparing student achievement and teacher and classroom characteristics in online and brick-and-mortar charter schools.   Educational Researcher. 2020;49(3)161-175. doi:10.3102/0013189X20909814Google Scholar
8.
Freedberg L. Sal Khan says distance learning can't fully replace in-person instruction. EdSource. Published August 3, 2020. Accessed October 19, 2020. https://edsource.org/2020/sal-khan-says-distance-learning-cant-fully-replace-in-person-instruction/637541
9.
Illinois State Board of Education. Remote learning recommendations during COVID-19 emergency. Published March 27, 2020. Accessed October 19, 2020. https://www.isbe.net/Documents/RL-Recommendations-3-27-20.pdf
10.
Barnum M, Bryan C. America’s great remote-learning experiment: what surveys of teachers and parents tell us about how it went. Chalkbeat. Published June 26, 2020. Accessed October 19, 2020. https://www.chalkbeat.org/2020/6/26/21304405/surveys-remote-learning-coronavirus-success-failure-teachers-parents
11.
Educators for Excellence. Voices from the virtual classroom: a survey of America’s teachers on COVID-19-related education issues. Accessed October 19, 2020. https://e4e.org/sites/default/files/voices_from_the_virtual_classroom_2020.pdf
12.
Malkus  N, Christensen  C. School district responses to the covid-19 pandemic: round 5, plateauing services in America’s schools. American Enterprise Institute. Published May 22, 2020. Accessed October 19, 2020. https://www.aei.org/research-products/report/school-district-responses-to-the-covid-19-pandemic-round-5-plateauing-services-in-americas-schools/
13.
 Reopening K-12 Schools During the COVID-19 Pandemic: Prioritizing Health, Equity, and Communities. Bond EC, Dibner K, Schweingruber H, eds. Washington, DC: The National National Academy of Science, Engineering, and Medicine; 2020.
14.
Jaume  D, Willen  A.  The long-run effects of teacher strikes: evidence from Argentina.   J Labour Econ. 2019;37(4):1097-1139. doi:10.1086/703134Google Scholar
15.
Belot  M, Webbink  D.  Do teacher strikes harm educational attainment of students?   Labour. 2010;24(4):391-406. doi:10.1111/j.1467-9914.2010.00494.xGoogle Scholar
16.
Johnson  D.  Do strikes and work-to-rule campaigns change elementary school assessment results.   Canadian Public Policy. 2011;37.4:479-494. doi:10.3138/cpp.37.4.479Google Scholar
17.
Caldwell  W, Jeffreys  L.  The effect of teacher strikes on student achievement: new evidence.   Government Union Review. 1983;4.1:40-58.Google Scholar
18.
Hernandez  DJ. Double jeopardy: how third-grade reading skills and poverty influence high school graduation. The Annie E. Casey Foundation. Published January 1, 2012. Accessed October 19, 2020. https://www.aecf.org/resources/double-jeopardy
19.
Conti  G, Heckman  J, Urzua  S.  The education-health gradient.   Am Econ Rev. 2010;100(2):234-238. doi:10.1257/aer.100.2.234PubMedGoogle Scholar
20.
Conti  G, Heckman  JJ.  The developmental approach to child and adult health.   Pediatrics. 2013;131(suppl 2):S133-S141. doi:10.1542/peds.2013-0252dPubMedGoogle Scholar
21.
Lleras-Muney  A.  The relationship between education and adult mortality in the United States.   Rev Econ Stud. 2005;72(1):189-221. doi:10.1111/0034-6527.00329Google Scholar
22.
Mazumder  B.  Does education improve health? A reexamination of the evidence from compulsory schooling laws.   Economic Perspectives. 2008;32(2).Google Scholar
23.
COVID-19 planning considerations: guidance for school re-entry. American Academy of Pediatrics. Updated August 19, 2020. Accessed October 19, 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/
24.
Husereau  D, Drummond  M, Petrou  S,  et al; CHEERS Task Force.  Consolidated Health Economic Evaluation Reporting Standards (CHEERS) statement.   BMJ. 2013;346:f1049. doi:10.1136/bmj.f1049PubMedGoogle Scholar
25.
Arias  E.  United States life tables, 2017.   Natl Vital Stat Rep. 2019;68(7):1-66.PubMedGoogle Scholar
26.
Clark  D, Royer  H.  The effect of education on adult mortality and health: evidence from Britain.   Am Econ Rev. 2013;103(6):2087-2120. doi:10.1257/aer.103.6.2087PubMedGoogle Scholar
27.
Gathmann  C, Jürges  H, Reinhold  S.  Compulsory schooling reforms, education and mortality in twentieth century Europe.   Soc Sci Med. 2015;127:74-82. doi:10.1016/j.socscimed.2014.01.037PubMedGoogle Scholar
28.
Grytten  J, Skau  I, Sørensen  R.  Who dies early? education, mortality and causes of death in Norway.   Soc Sci Med. 2020;245:112601. doi:10.1016/j.socscimed.2019.112601PubMedGoogle Scholar
29.
Meghir  C, Palme  M, Simeonova  E.  Education and mortality: evidence from a social experiment.   American Economic Journal: Applied Economics. 2018;10(2):234-256. doi:10.1257/app.20150365Google Scholar
30.
van Kippersluis  H, O’Donnell  O, van Doorslaer  E.  Long run returns to education: does schooling lead to an extended old age?   J Hum Resour. 2009;4:1-33.PubMedGoogle Scholar
31.
Oliver  A.  A normative perspective on discounting health outcomes.   J Health Serv Res Policy. 2013;18(3):186-189. doi:10.1177/1355819613485671PubMedGoogle Scholar
32.
Haacker  M, Hallett  TB, Atun  R.  On discount rates for economic evaluations in global health.   Health Policy Plan. 2020;35(1):107-114.PubMedGoogle Scholar
33.
Heinzerling  L.  Regulatory costs of mythic proportions.   Yale Law J. 1998;107(7):1981-2070. doi:10.2307/797416Google Scholar
34.
Landefeld  JS, Seskin  EP.  The economic value of life: linking theory to practice.   Am J Public Health. 1982;72(6):555-566. doi:10.2105/AJPH.72.6.555PubMedGoogle Scholar
35.
Malcolm  D, Roseboom  J, Clark  C, Fazar  W.  Application of a technique for research and development program evaluation.   Oper Res. 1959;7(5):646-669. doi:10.1287/opre.7.5.646Google Scholar
36.
New York City Department of Health and Mental Hygiene (DOHMH) COVID-19 Response Team.  Preliminary estimate of excess mortality during the COVID-19 outbreak—New York City, March 11-May 2, 2020.   MMWR Morb Mortal Wkly Rep. 2020;69(19):603-605. doi:10.15585/mmwr.mm6919e5PubMedGoogle Scholar
37.
Auger  KA, Shah  SS, Richardson  T,  et al.  Association between statewide school closure and COVID-19 incidence and mortality in the US.   JAMA. 2020;324(9):859-870. doi:10.1001/jama.2020.14348PubMedGoogle Scholar
38.
Courtemanche  C, Garuccio  J, Le  A, Pinkston  J, Yelowitz  A.  Strong social distancing measures in the United States reduced the COVID-19 growth rate.   Health Aff (Millwood). 2020;39(7):1237-1246. doi:10.1377/hlthaff.2020.00608PubMedGoogle Scholar
39.
Berchick  E.  Most uninsured were working-age adults.  US Bureau of the Census. Published September 12, 2018. Accessed December 23, 2020. https://www.census.gov/library/stories/2018/09/who-are-the-uninsured.html
40.
Higgins  JPT, Thomas  J, Chandler  J, Cumpston  M, Li  T, Page  MJ, Welch  VA.  Cochrane Handbook for Systematic Reviews of Interventions. 2nd ed. John Wiley and Sons; 2019.
41.
Hsiang  S, Allen  D, Annan-Phan  S,  et al.  The effect of large-scale anti-contagion policies on the COVID-19 pandemic.   Nature. 2020;584(7820):262-267. doi:10.1038/s41586-020-2404-8PubMedGoogle Scholar
42.
Dorn  E, Hancock  B, Sarakatsannis  J, Viruleg  E. COVID-19 and student learning in the United States: the hurt could last a lifetime. McKinsey & Co. Published June 1, 2020. Accessed July 21, 2020. https://www.mckinsey.com/industries/public-sector/our-insights/covid-19-and-student-learning-in-the-united-states-the-hurt-could-last-a-lifetime
43.
Eubank  S, Eckstrand  I, Lewis  B, Venkatramanan  S, Marathe  M, Barrett  CL.  Commentary on Ferguson, et al., “Impact of Non-pharmaceutical Interventions (NPIs) to Reduce COVID-19 Mortality and Healthcare Demand”.   Bull Math Biol. 2020;82(4):52. doi:10.1007/s11538-020-00726-xPubMedGoogle Scholar
44.
Ferguson  N, Laydon  D, Nedjati-Gilani  G, et al. Report 9: impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare demand. MRC Centre for Global Infectious Disease Analysis. Published March 16, 2020. Accessed October 19, 2020. https://www.imperial.ac.uk/mrc-global-infectious-disease-analysis/covid-19/report-9-impact-of-npis-on-covid-19/
45.
Roser  M, Ortiz-Ospina  E. Global education. Our World in Data. 2020. Accessed December 8, 2020. https://ourworldindata.org/global-education
46.
Nadworny  E. ‘Losing a generation’: fall college enrollment plummets for 1st-year students. NPR. Published December 17, 2020. Accessed December 21, 2020. https://www.npr.org/2020/12/17/925831720/losing-a-generation-fall-college-enrollment-plummets-for-first-year-students
47.
Education Week staff. Coronavirus and learning: what’s happening in each state. Education Week. Published April 3, 2020. Accessed July 21, 2020. http://blogs.edweek.org/edweek/campaign-k-12/2020/04/coronavirus_and_schools_state_guide.html?cmp=SOC-SHR-FB
48.
Viboud  C, Miller  M, Olson  D, Osterholm  M, Simonsen  L.  Preliminary estimates of mortality and years of life lost associated with the 2009 A/H1N1 pandemic in the US and comparison with past influenza seasons.   PLoS Curr. 2010;2:RRN1153-RRN1153. doi:10.1371/currents.RRN1153PubMedGoogle Scholar
3 Comments for this article
EXPAND ALL
Methodological Concerns
Ilya Kashnitsky, PhD | Interdisciplinary Centre on Population Dynamics, University of Southern Denmark, Odense, Denmark
The paper by Christakis, Van Cleve, and Zimmerman (CVZ) [1] contains numerical, statistical, and methodological flaws. As the arguments presented therein have immediately influenced public debate and policy decisions around the globe, a delineation of these is necessary to maintain the fidelity of the debate around the complicated benefits and drawbacks around school closure to reduce the spread of COVID-19.

The meta-analysis performed by the authors [1] drives a key figure within the paper - a 25% reduction in relative risk of death across the lifespan per year of schooling. The Lleras-Muney (2005) figure is the largest reduction in
this paper, which reports that the estimates varying between 0.42-0.67 were not statistically distinguishable. Mazumder (2008), has a transcription error: the value in the meta-analysis by CVZ is 0.65 (-0-1.3), instead of the correct 0.89 (0.743-1.034). Further, it was corrected in a 2010 erratum which brought the correct value to 1.028. The number for Clark and Royer (2013) reads 1 (0.94-1.07) but should properly be 1.022 (1.01-1.034).

Additionally, the paper [1] specifies the use of an inverse-variance model with extra weight to US studies which is mathematically impossible given the confidence bounds of the included research: a study with confidence bounds (0.08-0.77) for relative risk is weighted 2,790x higher than a study with bounds (0.94-1.07).

Correcting the errors and running an inverse-variance model [2] results in the RRR decreasing from 25% to 3%. This reverses the main finding of the paper, with school closures then costing ~0.65 million YLL, compared to the estimate for COVID-19 of 2.93 million.

The assumptions in the paper do not bear weight. First, an Argentinian study conducted on children in the 70s and 80s during teacher strikes [3], which explicitly warned against using its results causally, is taken as directly representative of US children learning from home in 2020. Second, an assumption is made that lost years of schooling can be directly translated into lost years of life. Even if a causal relationship is assumed, educational attainment does not explain all the variance in longevity, which is the implicit assumption of the deterministic translation used in the paper. Without this assumption, the connection of schooling to YLL would fall apart, as it is otherwise a host of factors causing the reduction in risk of death.

Additional errors in the paper are detailed in our preprint, and we encourage readers to check our re-estimates themselves running the code we provide [2]. The estimates of YLL in this study are not accurate, the assumptions are wrong, and the errors reverse the main results of the paper. We suggest the editors and authors consider correcting and/or retracting the paper.

Gideon Meyerowitz-Katz, MPH
Ilya Kashnitsky, PhD

REFERENCES:

[1] Christakis DA, Van Cleve W, & Zimmerman FJ (2020). Estimation of US Children’s Educational Attainment and Years of Life Lost Associated With Primary School Closures During the Coronavirus Disease 2019 Pandemic. JAMA Network Open, 3(11), e2028786.

[2] Meyerowitz-Katz G, & Kashnitsky I (2020). Comparing bad apples to orange soda: Flaws and Errors in an Estimation of Years of Life Lost Associated With School Closures and COVID-19 deaths by Christakis, Van Cleve, and Zimmerman. https://doi.org/10.31219/osf.io/9yqxw

[3] Jaume D & Willén A. Journal of Labor Economics 37, 1097–1139 (2019). https://doi.org/10.1086/703134
CONFLICT OF INTEREST: None Reported
READ MORE
Response to Concerns, Clarifications, and Corrections
Dimitri A. Christakis, MD, MPH | Department of Pediatrics, Seattle Children’s Research Institute, University of Washington, Seattle
Mr Meyerowitz-Katz and Dr Kashnitsky raise questions about our study (1), including concerns about numerical errors in our Supplement, computational errors in our models, and the validity of assumptions used in our analysis.

Our study was based on 2 assumptions: interrupted primary school instruction is associated with reduced school performance and eventual educational attainment, and educational attainment is associated with longevity. We attempted to find reasonable estimates of the strength of each association in the literature and calculated years of life lost (YLL) based on those estimates.

All studies included in our analysis published multiple estimates of effect
size, and we selected one estimate to include in our analysis based on the result we judged to be the most methodologically defensible. A corrected Supplement reports the original effect sizes the studies used and provides additional detail about the weights and sources of the estimates. The study by Lleras-Muney (2) reported 14 estimates; of which we chose the instrumental variables estimate with a complete set of controls. This is the largest estimate in that study but we chose it for its methodological rigor (see the corrected Supplement to our article) (1). We reviewed the study by Mazumder (3) and have not identified any transcription errors; the Mazumder erratum did not apply to the estimate used in our study (see the corrected Supplement) (1). Of the more than 30 estimates reported in this study, we chose the result that is most conceptually similar to the other studies included in our analysis. Meyerowitz-Katz and Kashnitsky prefer an estimate of .89 which was subsequently corrected to 1.028. Those estimates weigh mortality after age 80, and calculations based on this result in a much lower estimate for the association between education and mortality, because at that age survival is likely due to other endogenous or random factors. The estimate from the study by Clark and Royer (4) was correctly reported, but the 95% CI should have been 0.92-1.07 instead of 0.94-1.07 as originally reported.

We thank Meyerowitz-Katz and Kashnitsky for pointing out an error in our calculation of the average effect size using an inverse-variance model with heavier weighting for US studies. We did make a computational error in calculating the weights, and in correcting this error, we noticed a second one. Given the significant heterogeneity among studies, a formal weighted meta-analytic approach was not appropriate.

Meyerowitz-Katz and Kashnitsky suggest using the closest analogue possible among published studies, a principle we endorse. This principle would suggest limiting the studies to those from the US where possible. However, only 2 of the 7 studies in our analysis were from the US, and the estimate we used was more conservative than our preferred specification from each of these 2 US-based studies. The universe of available studies is weighted toward Scandinavian studies, which have large sample sizes, and over 95% of the combined sample of the studies we used was from Norway and Sweden. However, the mortality-education association in the 2 countries, both of which have robust social protections, strong vocational programs, meaningful income equality, and universal health care, is unlikely to be a good model for the association between education and mortality in the US, which does not have these structural supports. Therefore we have corrected the article to present 2 estimates for potential YLL associated with school closure, using separate weighted relative risk estimates from US-based and European-based studies for the association between lost education and mortality.

Third, with respect to the validity of the assumption for association between primary school instruction and educational attainment, no randomized trials have evaluated the effect of school interruption. The strongest studies (of which there are very few, and none for US populations with long term follow-up) used an instrumental variables approach. The study from Argentina on which we based this portion of our analysis was of high quality (5). Moreover, the Argentine health and educational context is similar to the US. Life expectancy at birth is currently estimated to be 79.1 years in the US and 77.2 years in Argentina. The expected educational attainment for US children is 16.5 years; in Argentina it is 17.4 years (6). The effect size identified in Argentina was smaller than that found in a study in Belgium or in short-term studies in the US and Canada as we noted in our article (1). As to the concern about the study being based on schooling interruptions that occurred decades ago, that would be the case for any study assessing the long-term outcomes associated with interrupted schooling. Meyerowitz-Katz and Kashnitsky also point out that education does not explain all of the variance in longevity. We agree with this assertion, and it is the reason we used regression coefficients that estimate an association after controlling for possible confounders.

In their preprint, Meyerowitz-Katz and Kashnitsky (7) make a number of other assertions. They note that our analysis did not attempt to account for COVID-related morbidity, but they did not acknowledge the substantial morbidity associated with school closures and subsequent reduced educational attainment. In addition, they take issue with our assumption that remaining life expectancy of all people over 85 is the same as people who are 85. Actuarily, people over age 85 have on average 3-5 years remaining. Life expectancy does not diminish until they reach age 95 (8). Because the mortality of COVID victims occurs at older ages on average than the mortality associated with school closures, any bias introduced by this assumption is a conservative bias. Meyerowitz-Katz and Kashnitsky also claim that our estimates are overstated because baseline mortality risk is decreasing over time, but in fact it is not for the US (6).

We endeavored to make reasonable and well-justified assumptions for the many parameters necessary in this model. Different analyses might plausibly take different approaches to these assumptions, with different results. We endeavored both in the original article and in the corrected article and corrected Supplement that accompanies this response to be transparent about these assumptions, our rationale for making them, and their implications for the results.
We appreciate the opportunity to present these corrected analyses and apologize to readers and editors for any confusion that may have resulted from the computational errors in the original version of our published article. We continue to believe that our model is both useful and important in reframing the discussion to include the possible damage being done to children by keeping schools closed. Our objective was and remains that those seeking to protect public health consider the potential effects of their decisions on the current and future health of children.

References

1. Christakis DA, Van Cleve W, Zimmerman FJ. Estimation of US children’s educational attainment and years of life lost associated with primary school closures during the coronavirus disease 2019 pandemic. JAMA Netw Open. 2020;3(11):e2028786. doi:10.1001/jamanetworkopen.2020.28786

2. Lleras-Muney A. The relationship between education and mdult Mortality in the United States. The Review of Economic Studies. 2005;72(1):189-221.

3. Mazumder B. Does education improve health? A reexamination of the evidence from compulsory schooling laws. Economic Perspectives. 2008;32(2).

4. Clark D, Roayer H. The effect of education on adult mortality and health: evidence from Britain. The American Economic Review. 2013;103(6):2087-2120.

5. Jaume D, Willen A. The long-run effects of teacher strikes: evidence from Argentina. J Labour Econ. 2019;37(4):1097-1139. doi:10.1086/703134

6. Roser M, Ortiz-Ospina E. Global Education; Our World in Data 2020; https://ourworldindata.org/global-education. Accessed Dec 8th 2020

7. Meyerowitz-Katz G, Kashnitsky I. Comparing bad apples to orange soda: flaws and errors in an estimation of years of life lost associated with school closures and COVID-19 deaths by Christakis, Van Cleve, and Zimmerman. OSF Preprints. December 1, 2020. https://doi.org/10.31219/osf.io/9yqxw
CONFLICT OF INTEREST: Member, Children and Screens Advisory Board (Christakis)
READ MORE
The Role of School in COVID-19 Exaggerated
Chao Wang, PhD | Kingston University
This paper discussed an important issue on the costs and benefits of school closure during the COVID-19 pandemic.

One critical limitation of this paper is their calculation of "how many deaths might have occurred... had primary schools remained open". To calculate this, they used a mortality ratio of 2.85 estimated from Auger et al. (2020) as the maximum and a ratio of 1 as the minimum. However, this is unreliable: as Auger et al. stated, "many states enacted additional nonpharmaceutical interventions concurrently with or shortly after school closure, making it impossible to fully isolate potential effects of school closure". />
More recent studies have indicated school is at no greater risk (indeed even lower risk) than the community. For example, Public Health England (2020) found "Seropositivity rates in students and staff were similar and not associated with school attendance during the lockdown". Ludvigsson et al. (2021) found schoolteachers had a lower mortality risk (relative risk=0.43) than other occupations (excluding health care workers). Ladhani et al. (2021) found "In secondary schools, SARS-CoV-2 infection, seropositivity and seroconversion rates were similar in staff and students, and comparable to local community rates." The latest ONS School Infection Survey (ONS, 2021) found "prevalence of infection among pupils sampled in school was consistently lower than prevalence of infection among children in the wider community across all time periods".

The growing evidence seems to suggest the authors' assumption of mortality ratio is greatly exaggerated. There seems little or even an increased risk of school closure in terms of COVID-19.

References

Auger, K. A., Shah, S. S., Richardson, T., Hartley, D., Hall, M., Warniment, A., Timmons, K., Bosse, D., Ferris, S. A., Brady, P. W., Schondelmeyer, A. C., & Thomson, J. E. (2020). Association Between Statewide School Closure and COVID-19 Incidence and Mortality in the US. JAMA, 324(9), 859–870. https://doi.org/10.1001/jama.2020.14348

Ladhani, S. N., Ireland, G., Baawuah, F., Beckmann, J., Okike, I. O., Ahmad, S., Garstang, J., Brent, A. J., Brent, B., Walker, J., Aiano, F., Amin-Chowdhury, Z., Letley, L., Flood, J., Jones, S. E. I., Kall, M., Borrow, R., Linley, E., Zambon, M., … Ramsay, M. E. (2021). SARS-CoV-2 infection, antibody positivity and seroconversion rates in staff and students following full reopening of secondary schools in England: A prospective cohort study, September–December 2020. EClinicalMedicine, 37, 100948. https://doi.org/https://doi.org/10.1016/j.eclinm.2021.100948

Ludvigsson, J. F., Engerström, L., Nordenhäll, C., & Larsson, E. (2021). Open Schools, Covid-19, and Child and Teacher Morbidity in Sweden. New England Journal of Medicine, 384(7), 669–671. https://doi.org/10.1056/NEJMc2026670

Office for National Statistics. (2021). COVID-19 Schools Infection Survey, England: Round 6, June 2021. https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/covid19schoolsinfectionsurveyengland/round6june2021

Public Health England. (2020). Prospective active national surveillance of preschools and primary schools for SARS-CoV-2 infection and transmission in England, June 2020. https://www.gov.uk/government/publications/covid-19-surveillance-in-school-kids-skids-pre-and-primary-schools

CONFLICT OF INTEREST: None Reported
READ MORE
Original Investigation
Public Health
November 12, 2020

Estimation of US Children’s Educational Attainment and Years of Life Lost Associated With Primary School Closures During the Coronavirus Disease 2019 Pandemic

Author Affiliations
  • 1Department of Pediatrics, Seattle Children’s Research Institute, University of Washington, Seattle
  • 2Department of Anesthesiology and Pain Medicine, University of Washington, Seattle
  • 3Department of Health Policy & Management, Center for Health Advancement, Fielding School of Public Health, University of California, Los Angeles, Los Angeles, California
JAMA Netw Open. 2020;3(11):e2028786. doi:10.1001/jamanetworkopen.2020.28786
Key Points

Question  Based on the current understanding of the associations between school disruption and decreased educational attainment and between decreased educational attainment and lower life expectancy, is it possible to estimate the association between school closure during the coronavirus disease 2019 pandemic and decreased life expectancy of publicly educated primary school–aged children in the United States?

Findings  This decision analytical model found that missed instruction during 2020 could be associated with an estimated 13.8 (95% CI 2.5-42.1) million years of life lost based on data from US studies and an estimated 0.8 (95% CI 0.1-2.4) million years of life lost based on data from European studies. This estimated loss in life expectancy was likely to be greater than would have been observed if leaving primary schools open had led to an expansion of the first wave of the pandemic.

Meaning  These findings suggest that the decision to close US public primary schools in the early months of 2020 may be associated with a decrease in life expectancy for US children.

Abstract

Importance  United States primary school closures during the 2020 coronavirus disease 2019 (COVID-19) pandemic affected millions of children, with little understanding of the potential health outcomes associated with educational disruption.

Objective  To estimate the potential years of life lost (YLL) associated with the COVID-19 pandemic conditioned on primary schools being closed or remaining open.

Design, Setting, and Participants  This decision analytical model estimated the association between school closures and reduced educational attainment and the association between reduced educational attainment and life expectancy using publicly available data sources, including data for 2020 from the US Centers for Disease Control and Prevention, the US Social Security Administration, and the US Census Bureau. Published peer-reviewed studies (2 US studies, 5 European studies) were identified that provided estimates of the relative risk (RR) of annual mortality related to educational attainment, which were weighted and applied to the most recent life table data to obtain YLL estimates across the life course. Direct COVID-19 mortality and potential increases in mortality that might have resulted if school opening led to increased transmission of COVID-19 were also estimated.

Main Outcomes and Measures  Years of life lost.

Results  A total of 24.2 million children aged 5 to 11 years attended public schools that were closed during the 2020 pandemic, losing a median of 54 (interquartile range, 48-62.5) days of instruction. Missed instruction was associated with a mean loss of 0.15 (95% credible interval [CI], 0.08-0.22) years of final educational attainment for boys and 0.12 (95% CI, 0.06-0.19) years for girls. Summed across the population, based on the RR from US studies, an estimated 13.8 million (95% CI, 2.5-42.1) YLL may be associated with school closures. Summed across the population, based on the RR from the European studies, an estimated 0.8 million (95% CI, 0.1 -2.4) YLL may be associated with school closures. The Centers for Disease Control and Prevention reported a total of 88 241 US deaths from COVID-19 through the end of May 2020, with an estimated 1.50 million (95% CI, 1.23-1.85 million) YLL as a result. Had schools remained open, an estimated 4.4 million (95% CI 2.29-6.41,) YLL could have been expected as a result, based on results of studies associating school closure with decreased pandemic spread. Comparing the full distributions of estimated YLL under both “schools open” and “schools closed” conditions, based on the US studies and the European studies, the analysis observed a 98.9% probability and a 26.3% probability, respectively, that school opening would have been associated with a lower total YLL than school closure.

Conclusions and Relevance  In this decision analytical model of years of life potentially lost under differing conditions of school closure, based on the US studies, the analysis favored schools remaining open. Future decisions regarding school closures during the pandemic should consider the association between educational disruption and decreased expected lifespan and give greater weight to the potential outcomes of school closure on children’s health.

Introduction

In early 2020, school closures were widely instituted across the United States as a coronavirus disease 2019 (COVID-19) containment strategy. The rationale for closures was 2-fold. First, at least initially, the risks that the virus posed to children were unclear but worthy of precaution. Second, it was assumed that children might represent important vectors for disease spread even if they were themselves unaffected or asymptomatic. Both of these considerations appeared to justify the harm of missed education in order to minimize the population-level risk of disease. In the ensuing months, data have emerged indicating that COVID-19 infection poses significantly less direct risk to children to adults.1,2 While the scientific evidence on transmission of SARS-CoV-2 by children remains in flux, recent studies indicate that young children (<10 years) appear less likely to serve as vectors for COVID-19 transmission.3,4 Although the risks of keeping schools open drove decisions made in the early phases of the pandemic, the probable harm to children associated with school closure were less explicitly discussed.5 The public debate has pitted “school closures” against “lives saved,” or the education of children against the health of the community. Presenting the tradeoffs in this way obscures the very real health consequences of interrupted education.

These consequences are especially dire for young children. There is little reason to believe that virtual learning environments can be effective for primary school–aged children. A meta-analysis6 of 99 experimental studies included only 5 conducted in school-aged children, and they were primarily in fifth through eighth grade. The meta-analysis concluded that “the mean effect size [for online learning] is not significant for the seven contrasts involving K-12 students.”6(pXV) That so few studies have even been conducted in this age group is also telling. A recent study comparing Indiana children in grades 3 through 8 who switched from brick-and-mortar to virtual schooling “experienced large, negative effects in math and [English/language arts] that were sustained across time”.7(p170) Sal Khan, a widely respected innovator in the field of distance learning, reported that distance learning approaches do not work for younger students.8 Furthermore, it is not clear how much access to remote instruction primary school–aged children actually received during the spring of 2020. For example, in its March 2020 guidelines for districts, the Illinois Department of Education recommended that primary school children have a maximum of 60 to 120 minutes per day in remote learning, representing a fraction of a regular school day.9 In 2 national surveys, teachers of all grades reported that only 60% of their students were regularly engaging in distance learning at all, and only 27% of teachers took attendance.10-12 Accordingly, it is reasonable to infer that primary school–aged children received minimal meaningful instruction beyond what is being delivered by their parents or other caregivers at home.13 It is not surprising, then, that the National Academies of Science, Engineering, and Medicine’s report on school openings13 concluded that districts should make returning primary school children to in-person classes a priority.

Evidence suggests that missing school has adverse effects on eventual educational attainment. A longitudinal study of teacher strikes in Argentina revealed that disrupted schooling lowered graduation rates, total educational attainment, and subsequent income.14 An educational reform in Belgium differentially affected Flemish-speaking and French-speaking parts of the country and resulted in strikes of approximately 60 days in the French-speaking part of the country against none in the Flemish-speaking part. Using this natural experiment in a difference-in-difference framework, economists estimated the long-term effects of these strikes on educational attainment to be a 5.8% reduction in total years of educational attainment, a somewhat larger effect than that identified in Argentina.15 Studies of prolonged strike in the United States and Canada are lacking, but even short-term strikes were found to result in diminished test scores.16,17 One US report18 found that the single best predictor of high-school graduation was fourth-grade reading test scores: 23% of children who are not reading at grade level by the end of third grade will not graduate high school, compared with 9% of those who are. The risks are even greater for low-income Black or Hispanic students: 33% of those not reading at grade level will not graduate from high school. These educational impairments are in turn consequential for mortality: the quality and quantity of education received today have considerable effects on life expectancy.19-22

The American Academy of Pediatrics’ policy statement on school reopening suggested that science drive decision-making.23 Doing so requires a better-informed estimate of the tradeoffs being considered. The primary objective of this study was to model the expected years of life lost (YLL) in association with primary school closures in early 2020 and to compare them to potential YLL had schools remained open.

Methods

This decision analytical model consisted of an analysis of publicly available data and an estimate of the relative risk (RR) of annual mortality, derived from the existing literature. Our analysis and report follows the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) reporting guideline.24 As an analysis of publicly available data, our institutions considered this work exempt from institutional review board review.

Our goal was to compare the number of expected YLL due to COVID-19 in the United States under 2 different scenarios (Figure 1). Scenario 1, observed from January 1 to May 30, 2020, consisted of the closing of primary schools during the early phase of the COVID-19 pandemic in the United States, resulting in expected YLL from 2 primary pools: premature deaths from COVID-19 and shortened life expectancy due to decreased educational attainment among children. Scenario 2 was unobserved and was based on a counterfactual decision to allow primary schools to remain open, with a potentially increased number of deaths and years of life directly lost due to COVID-19 if school opening led to increased pandemic spread.

Statistical Analysis
Scenario 1 Estimation

To estimate the YLL due to COVID-19 under conditions of school closure, we required an estimate of direct mortality from COVID-19 during the early phase of the pandemic (see Scenario 2 Estimation) and an estimate of the YLL associated with primary school closure. This latter estimate was constructed in 2 stages: we first estimated the association between school closures and educational attainment, and then we estimated the association between reduced educational attainment and mortality risk. For stage 1, we used the results of a carefully constructed econometric analysis of quasi-random school closures due to teachers’ strikes in Argentina. These results—which use within-province differences in strike exposure across birth cohorts and within-cohort differences in strike exposure across provinces as sources of exogenous variation—provide a reasonable estimate of the association between missed education during primary school and total years of educational attainment.14 This analysis suggests that 10 days of missed school is associated with a reduction in final educational attainment of 0.0262 (SE, 0.0064) years for boys and 0.0217 years (SE, 0.0062) years for girls.

To apply this analysis to the US, we used the 2018 1-year estimates from the US Census Bureau American Community Survey (ACS) and counted all children between the ages of 5 and 11 (inclusive) who were enrolled in public school. Because life expectancy varies by race/ethnicity, we categorized children as White, Black, Hispanic, or Other. Children in the Other category were assigned the average life expectancy (nonconditioned on race/ethnicity), whereas other races/ethnicities were assigned specific life expectancies based on life tables for these groups available from the US Centers for Disease Control and Prevention (CDC).25 State-specific missed days of education were estimated by reviewing published school closure orders and school calendars from school districts in the capital of each state and counting weekdays only. We assumed that final educational attainment reported by respondents aged 25 to 29 years in the 2018 ACS (again stratified by sex, race/ethnicity, and state) would be representative of eventual attainment of children attending primary school in 2020.

Prior research in economics has estimated the association between decreased educational attainment and life expectancy using a variety of causal inference techniques.21,22,26-30 Using the method described in the eAppendix in the Supplement, we attempted to estimate a weighted average effect of these estimates. However, with this analytic approach of the 7 studies that provided data (2 US studies, 5 non-US studies), a single estimate could not be derived because of high heterogeneity of settings, methods, and data (eAppendix in the Supplement). Therefore, we used 2 estimates in our model, the weighted RR of annual mortality from the 2 US studies21,22 (RR, 0.541; 95% CI, 0.166-0.916) and from the 5 non-US (all European) studies26-30 (RR, 0.976; 95% CI, 0.910-1.042). It should be noted that although the RRs between the US studies and the European studies were different, the estimates from different studies within the US and within Europe were similar. These RRs were then applied to the most recent life table published by the CDC25 to obtain YLL estimates across the life course. Uncertainty estimates were generated by a Monte Carlo simulation, treating the associations between school closure and education and between education and life expectancy as truncated normal and normally distributed variables, respectively.

In cost-benefit analyses, it is customary to apply a discount factor to costs and benefits that occur in the future so that they can reasonably be compared to current values. Although discounting of future costs is reasonable, how and even whether to discount future health benefits is controversial, with no clear consensus in the literature.24-27 Yet because the association of missed education with YLL of today’s children occurs so far in the future, discounting is consequential to the analysis. We therefore summarized the YLL associated with decreased educational attainment under 3 different annual discounting scenarios: no discount, 0.5%, and 3%, with the understanding that the choice of the appropriate discount rate represents a value judgment (see Discussion).31-34 The simulation results are reported for the nation as a whole in terms of total estimated YLL.

Scenario 2 Estimation

Estimation of YLL under unobserved conditions (ie, US primary schools being left open with a potential for increased spread of COVID-19) necessarily required explicit modeling of a range of potential scenarios. In order to model parameter uncertainty, we again performed a Monte Carlo simulation, this time using a Program Evaluation Research Task (PERT) distribution.35 The PERT distribution is a modification of the beta distribution with 2 favorable characteristics for this type of analysis. First, it allows for explicit specification of the minimum, maximum, and most likely (modal) values of the parameter. Second, it assigns greater probability weight to the modal value (approximately 4-fold greater) while still incorporating the minimum and maximum values into the resultant distribution. Like all beta distributions, it is positive-bounded.

As a separate calculation, we used CDC data on patient ages and numbers of COVID-19 deaths observed in early 2020 as the source of our mortality estimates and merged these data with sex-specific actuarial life tables published by the US Social Security Administration to estimate the lost years of life expectancy incurred by deaths in each age group, assuming that deaths occurred in the middle of the 10-year span and that 85 was the maximum age of death.25 There is broad acceptance that CDC mortality estimates for COVID-19 are an underestimate; we corrected for this bias by modeling a multiplicative ratio of YLL using a PERT distribution. We defined the minimum ratio as 1 (excess deaths being unlikely to be negative), used CDC estimates of excess mortality observed in New York state in early 2020 to define a modal ratio of 1.22, and doubled the modal estimate as our maximum.36 This method provided a distributional estimate of US YLL directly related to COVID-19 observed in 2020.

The next question we considered was how many deaths might have occurred (and how many YLL would have resulted) had primary schools remained open. Again, substantial uncertainty exists regarding the degree to which school closures affect the spread of COVID-19, with some authors4 suggesting that young children contribute very little to the spread of SARS-CoV-2 and others suggesting that school closures might play a substantial role in the pandemic’s spread.37 Auger and colleagues37 recently published an estimate that school closures may have prevented 40 600 US deaths during the early phase of the pandemic, with a mortality rate of 19.4 deaths per 100 000 population under school opening and 6.8 per 100 000 under school closure. In contrast, Courtemanche and colleagues38 estimated a nonsignificant daily decrease in mortality associated with school opening. Using a PERT distribution, we used the Auger mortality ratio (2.85) as our maximum estimate. Given the belief on the part of some investigators that young children do not measurably influence disease spread, we used a mortality ratio of 1 as our minimum estimate. To define a plausible modal estimate, we used the average of the Auger and Courtemanche mortality ratios, using a baseline schools-closed mortality rate in the Courtemanche estimate (which was not provided) equal to that in Auger. This yielded a modal mortality ratio of 1.93.

We summarized our data with 95% credible intervals (95% CIs) drawn from our Monte Carlo estimates, constructed distributional summary plots, and finally estimated the probability that YLL associated with COVID-19 in 2020 were greater under primary school closure than had schools remained open by enumerating the proportion of Monte Carlo draws where this event was observed. All statistical analyses and graphs were constructed using R, version 3.6.2 (R Foundation).

Results

Using the 2018 ACS, we estimated that approximately 24.2 million children aged 5 to 11 years attended public schools that were closed during the 2020 pandemic (11.4 million White, 4.3 million Hispanic, 3.6 million Black, and 4.9 million Other). Across all US states, we estimated that public schools were closed for a median 54.0 days as a result of COVID-19 (IQR, 48-62.5 days). The mean (SD) final years of educational attainment reported by 25- to 29-year-old respondents in the 2018 ACS was 13.7 (2.1).

We estimated that primary school–aged boys could lose, approximately 0.15 (95% credible interval, 0.08-0.22) final years of education as a result of school closure (whereas girls could lose 0.12 (95% CI, 0.05-0.19) years. Based on the 2 US studies of the education-related RR of mortality, these losses in education were associated with a mean loss of life of 0.67 (95% CI, 0.12-2.09) years for boys and 0.45 (95% CI, 0.08-1.52) years for girls. Undiscounted, these estimated losses in life related to missed education summed to a median estimate of 13.8 (95% CI, 2.5-42.1) million YLL. Under conditions of 0.5% annual discounting, YLL were estimated at 10.9 (95% CI, 2.0-33.5) million. Under 3% annual discounting, YLL were estimated at 3.8 (95% CI, 0.7-11.9) million. Based on the 5 European studies of the education-related RR of mortality, losses in education related to school closures were associated with a mean loss of life of 0.38 (95% CI, 0.00-0.12) years for boys and 0.03 (95% CI, 0.00-0.09) years for girls. Undiscounted, these estimated losses in life due to missed education summed to a median estimate of 0.8 (95% CI, 0.1-2.4) million YLL. Under conditions of 0.5% annual discounting, YLL were estimated at 0.6 (95% CI, 0.0-1.9) million. Under 3% annual discounting, YLL were estimated at 0.2 (95% CI, 0.0-0.7) million.

The CDC reported a total of 88 241 US deaths from COVID-19 through May 30, 2020. Using Social Security Administration life tables, we estimated that these deaths resulted in 1 146 136 YLL (Table). Adjusting for potential undercounting of COVID-19 deaths led us to estimate that US deaths due to COVID-19 in early 2020 generated 1.50 million (95% CI, 1.23-1.85 million) YLL. Applying the counterfactual procedure described in our methods, we further estimated that had primary schools remained open and, in doing so, permitted the pandemic to spread, an estimated total of 2.97 million (95% credible interval, 1.88-4.30 million) YLL might have been observed, with 1.47 million (95% credible interval, 0.45-2.59 million) YLL associated with schools remaining open.

We also compared the YLL in each scenario: the scenario in which the United States experienced deaths due to COVID-19 as well as decreased predicted life expectancy due to decreased educational attainment associated with school closures (ie, the path followed by the United States in early 2020 of primary schools closed) or an alternate, counterfactual scenario in which the country had left schools open, with a possible increase in transmission of COVID-19. The results of this analysis, including modifications of these estimates by discounting future deaths, using estimates from the US-based studies and the European-based studies for the association between lost education and mortality, are depicted in Figure 2. When comparing the distribution of probable YLL under these “primary schools open” and “primary schools closed” scenarios using US-based studies, we estimated a 98.9% probability that nondiscounted YLL would be greater under school closure than had schools remained open (98.4% under 0.5% discounting and 89.6% under 3% discounting). Using European-based studies, the estimated probability was 26.3% (17.9% under 0.5% discounting and 1.6% under 3% discounting).

Discussion

In this decision analytical model, based on estimates of the RR of mortality associated with educational attainment derived from US studies, we estimated that there was a 98.9 % probability that the decisions to close US primary schools in March of 2020 could be associated with more eventual YLL than would be observed if these schools had remained open, even if schools remaining open had led to a substantial increase in the rate of death observed during the early phase of the pandemic. Based on estimates of RR derived from European studies, the probability was 26.3%.

This difference in these estimates provides important insights. More than 95% of the combined sample of the European studies we used in our analysis was from Norway and Sweden. However, the mortality-education association in these 2 countries, both of which have robust social protections, strong vocational programs, meaningful income equality, and universal health care, is unlikely to be a good model for the association between education and mortality in the US, which does not have these structural supports. The Norwegian study,28 for example, explicitly tested the relationship between education and health care utilization and found no association. While the 2 US studies did not explicitly test this link, there is a well-known association between educational attainment and medical care access in the US.39

There are also methodological differences between the US-based and European-based studies. The US-based studies may provide an estimate of the education-mortality link that is too large, especially if, for example, the US does develop better social protections in the future. The Cochrane collaboration recommends thinking qualitatively about the sources of such heterogeneity.40 Following that advice, the best estimate of YLL may lie somewhere in the middle between the US-based studies and the European-based studies. As a referent, the midpoint between the 2 estimates is 7.3 million YLL.

Our analysis was based on the decreased life expectancy associated with a decrease in the mean years of educational attainment that could be expected to occur as a result of the disruption in schooling during a critical period of educational development. These potential associations of school closure with child health remain hidden at present, as the shortening of lifespans that today’s primary school children could experience would not manifest until long after the pandemic is over. The public discourse on reopening schools has focused on lives saved in the present and largely neglected the years of life that may be lost in the future because of decisions made today.

We believe that the debates over opening schools have often lacked nuance, portraying schools monolithically—as all opened or all closed—without regard for potential safeguards that could reduce transmission within schools. Furthermore, to our knowledge, very little discussion has considered the very different outcomes and costs associated with school closure for children at different ages and developmental stages. Based on the existing evidence regarding the limited role primary school–aged children play in transmission of COVID-19 and the heavy burden of decreased educational exposure on their health, we believe that restoring access to in-person primary school education should be an immediate national priority, even while the country awaits a vaccine.4,41

There are several ways in which our estimates of YLL are conservative. First, we focused only on interruptions to primary school. We did this because distance learning has been demonstrated to be largely ineffective for this age group, and there are well-controlled longitudinal data to model the association between primary school disruption and final educational attainment. Nevertheless, we believe it highly probable that school closures in early 2020 will also ultimately depress high school graduation rates (and therefore may influence lifespan) for older children as well. A recent McKinsey report42 estimated that 2% to 9% of high school students could drop out of school because of COVID-19–related school absences. These losses in educational attainment and resultant increased mortality are not included in our estimates.

Second, we adjusted the number of COVID-19–related deaths in early 2020 to account for underreporting, thus favoring the schools-closed scenario. Third, we used a well-constructed estimate of increased mortality associated with school opening to inform our estimate of potential increased pandemic spread associated with school opening. In fact, in a recent study, the net contribution of school closures to virus spread was found to be close to zero.41 Fourth, whereas other models have been criticized for their complexity and opacity, our model is simple and transparent.43,44 Fifth, our estimates do not consider quality of life or disability associated with decreased educational attainment, which would be difficult to assess. However, it stands to reason that by not graduating high school, children may experience lower wages and the attendant detriments to quality of life for the duration of their lives.

Limitations

Our study has a number of limitations. First, we based our estimates of decreased educational attainment associated with school closures on data from a single study performed in Argentina. The Argentine health and educational context is similar to the US. Life expectancy at birth is currently estimated to be 79.1 years in the US and 77.2 years in Argentina. The expected educational attainment for US children is 16.5 years; in Argentina it is 17.4 years.45 Although we believe this assumption to be justified, the extent to which these data can be applied to a US population is unknown.

Second, our results are reliant on the assumptions that the disruptions in Argentinian schooling observed during teacher strikes in the 1983 to 2014 mimic those of school closures due to COVID-19 and that these disruptions will have similar effects on eventual educational attainment. In our opinion, these assumptions are reasonable, in part, based on a recent US study that confirms that falling behind in primary school has a significant negative association with rates of high school graduation.18 In fact, the effects are already being demonstrated among high school students, as higher education enrollment declined by 3.6% in Fall 2020, doubling the decline observed in 2019.46 Students in the United States also have some access to distance learning (not present in the Argentinian example), and it can be argued that US children received at least some form of education during COVID-19–related closures. The effectiveness of distance learning for primary school–aged children is highly questionable. Moreover, in 28 states representing 48% of children in the United States, distance learning was not mandated, and many children received none.11,47 As for parental teaching, it is probable that Argentinian parents likewise provided some level of instruction to their children during the prolonged strikes.

Third, the estimates of YLL directly due to COVID-19 in our study are dependent on the population prevalence of SARS-CoV-2 during that time. Since the end of May 2020, some communities have seen increases in disease prevalence, and a few have seen decreases. The association between school opening and population disease prevalence remains uncertain, and school opening could lead to greater transmission of disease among both adults and children. The most recently available data suggests that children younger than 10 years (consistent with the age we used in our model) transmit the virus considerably less readily than adults, suggesting that community prevalence may not be as important a factor in this population.4 Our analysis, however, presupposes that leaving schools open would be associated with a substantial increase in pandemic spread. Despite this obviously undesirable potential outcome, very likely borne to some extent by teachers and elderly relatives, the net YLL due to COVID-19 (at least during the early pandemic) appears to have favored schools remaining open.

Fourth, our estimates are sensitive to the discount rate applied to future health outcomes. Discounting for conditions that affect young children is a matter of considerable controversy and represents the values that society places on children vs adults. Many argue against discounting of any kind, some authors argue for 0.5% annual discount rates, and others argue for 3% or even greater.31,33,34 As a matter of pure math, a discount rate of 3% heavily biases against any outcome that will occur in the distant future. In fact, any intervention that affects children over the entirety of their lifespans will be penalized by a high discount rate. Prior studies examining death from H1N1 influenza selected discount rates of zero.48 We presented 3 different rates to highlight that prior beliefs about future event value will influence policy changes. Our belief is that the future health outcomes of children deserve our full attention and consideration today and that a discount of 0 or 0.5% is most appropriate. However, our findings suggest that even under a 3% discount rate, the probability that life years were saved owing to school closures may essentially be a toss-up.

Fifth, this study used estimates of the relationship between education and mortality derived from a literature that is necessarily imperfect. Even if it were possible or ethical to randomize children to more or less education and observe mortality risk years later, such a study would lack external validity, as this relationship is deeply contextual to the policy, social, and economic environments of the time and place. The studies used to estimate this relationship vary widely. We attempted to address this limitation by providing 2 distinct estimates that likely bracket the range, using weighted estimates from 2 US-based studies and 5 European-based studies.

This study has potentially important implications. The results of this decision analytical model suggest that the attempt to save lives by closing schools may not have resulted in a net savings when considering the potential harms associated with this intervention. This lack of intergenerational health equity is as unjust and deserves careful societal consideration. Another implication is that the losses being experienced among children are unlikely to be equitably distributed across the boundaries of gender, socioeconomic status, and race/ethnicity. Both Argentinian and US data suggest the adverse influence of educational interruption is greater for low-income children and for boys, suggesting that the potential outcomes of school closures may be felt more substantively by vulnerable populations in the United States.14,18 Those outcomes are compounded by the fact that there was demonstrably less engagement in distance learning among low-income minority children: schools serving predominantly Black and Hispanic students reported that only 60% to 70% were participating in remote education on a regular basis, and only one-third were participating daily.11,42,47 In addition, to preserve intergenerational equity, the costs of future life-years lost for young children today must be factored into decision-making regarding school openings and potential future closings. During the COVID-19 pandemic, school closures in the US may have contributed to possible adverse future consequences for children to protect the health of older adults. During a pandemic, this may well be an ethically defensible tradeoff, but only if resources are invested to reverse the potential harms to health and education that may be associated with this strategy.

Conclusions

In this decision analytical model of years of life potentially lost under differing conditions of school closure, based on the US studies, the analysis favored schools remaining open. Future decisions regarding school closures during the pandemic should consider the association between school closures and life expectancy of children.

Back to top
Article Information

Accepted for Publication: October 14, 2020.

Published: November 12, 2020. doi:10.1001/jamanetworkopen.2020.28786

Correction: This article was corrected on January 8, 2021, to fix computational errors and an incorrect analysis.

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2020 Christakis DA et al. JAMA Network Open.

Corresponding Author: Dimitri A. Christakis, MD, MPH, Department of Pediatrics, Seattle Children’s Research Institute, 2001 Eighth Ave, Suite 400, Seattle, WA 98101 (dimitri.christakis@seattlechildrens.org).

Author Contributions: Dr Van Cleve 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.

Concept and design: All authors.

Acquisition, analysis, or interpretation of data: Van Cleve, Zimmerman.

Drafting of the manuscript: All authors.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: All authors.

Administrative, technical, or material support: Christakis.

Supervision: Christakis, Zimmerman.

Conflict of Interest Disclosures: Dr Christakis reported serving as a member of the Children and Screens Advisory Board. No other disclosures were reported.

References
1.
Shekerdemian  LS, Mahmood  NR, Wolfe  KK,  et al; International COVID-19 PICU Collaborative.  Characteristics and outcomes of children with coronavirus disease 2019 (COVID-19) infection admitted to US and Canadian pediatric intensive care units.   JAMA Pediatr. 2020;174(9):868-873. doi:10.1001/jamapediatrics.2020.1948PubMedGoogle Scholar
2.
Zachariah  P, Johnson  CL, Halabi  KC,  et al; Columbia Pediatric COVID-19 Management Group.  Epidemiology, clinical features, and disease severity in patients with coronavirus disease 2019 (COVID-19) in a children’s hospital in New York City, New York.   JAMA Pediatr. 2020;e202430. doi:10.1001/jamapediatrics.2020.2430PubMedGoogle Scholar
3.
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. doi:10.1056/NEJMoa2006100PubMedGoogle Scholar
4.
Park  YJ, Choe  YJ, Park  O,  et al; COVID-19 National Emergency Response Center, Epidemiology and Case Management Team.  Contact tracing during coronavirus disease outbreak, South Korea, 2020.   Emerg Infect Dis. 2020;26(10):2465-2468. doi:10.3201/eid2610.201315PubMedGoogle Scholar
5.
Christakis  DA.  School reopening: the pandemic issue that is not getting its due.   JAMA Pediatr. Published online May 13, 2020. doi:10.1001/jamapediatrics.2020.2068PubMedGoogle Scholar
6.
US Department of Education Office of Planning, Evaluation, and Policy Development. Evaluation of evidence-based practices in online learning: a meta-analysis and review of online learning studies. Published 2010. Accessed October 19, 2020. https://www2.ed.gov/rschstat/eval/tech/evidence-based-practices/finalreport.pdf
7.
Fitzpatrick  B, Berends  M, Ferrare  J, Waddington  R.  Virtual illusion: comparing student achievement and teacher and classroom characteristics in online and brick-and-mortar charter schools.   Educational Researcher. 2020;49(3)161-175. doi:10.3102/0013189X20909814Google Scholar
8.
Freedberg L. Sal Khan says distance learning can't fully replace in-person instruction. EdSource. Published August 3, 2020. Accessed October 19, 2020. https://edsource.org/2020/sal-khan-says-distance-learning-cant-fully-replace-in-person-instruction/637541
9.
Illinois State Board of Education. Remote learning recommendations during COVID-19 emergency. Published March 27, 2020. Accessed October 19, 2020. https://www.isbe.net/Documents/RL-Recommendations-3-27-20.pdf
10.
Barnum M, Bryan C. America’s great remote-learning experiment: what surveys of teachers and parents tell us about how it went. Chalkbeat. Published June 26, 2020. Accessed October 19, 2020. https://www.chalkbeat.org/2020/6/26/21304405/surveys-remote-learning-coronavirus-success-failure-teachers-parents
11.
Educators for Excellence. Voices from the virtual classroom: a survey of America’s teachers on COVID-19-related education issues. Accessed October 19, 2020. https://e4e.org/sites/default/files/voices_from_the_virtual_classroom_2020.pdf
12.
Malkus  N, Christensen  C. School district responses to the covid-19 pandemic: round 5, plateauing services in America’s schools. American Enterprise Institute. Published May 22, 2020. Accessed October 19, 2020. https://www.aei.org/research-products/report/school-district-responses-to-the-covid-19-pandemic-round-5-plateauing-services-in-americas-schools/
13.
 Reopening K-12 Schools During the COVID-19 Pandemic: Prioritizing Health, Equity, and Communities. Bond EC, Dibner K, Schweingruber H, eds. Washington, DC: The National National Academy of Science, Engineering, and Medicine; 2020.
14.
Jaume  D, Willen  A.  The long-run effects of teacher strikes: evidence from Argentina.   J Labour Econ. 2019;37(4):1097-1139. doi:10.1086/703134Google Scholar
15.
Belot  M, Webbink  D.  Do teacher strikes harm educational attainment of students?   Labour. 2010;24(4):391-406. doi:10.1111/j.1467-9914.2010.00494.xGoogle Scholar
16.
Johnson  D.  Do strikes and work-to-rule campaigns change elementary school assessment results.   Canadian Public Policy. 2011;37.4:479-494. doi:10.3138/cpp.37.4.479Google Scholar
17.
Caldwell  W, Jeffreys  L.  The effect of teacher strikes on student achievement: new evidence.   Government Union Review. 1983;4.1:40-58.Google Scholar
18.
Hernandez  DJ. Double jeopardy: how third-grade reading skills and poverty influence high school graduation. The Annie E. Casey Foundation. Published January 1, 2012. Accessed October 19, 2020. https://www.aecf.org/resources/double-jeopardy
19.
Conti  G, Heckman  J, Urzua  S.  The education-health gradient.   Am Econ Rev. 2010;100(2):234-238. doi:10.1257/aer.100.2.234PubMedGoogle Scholar
20.
Conti  G, Heckman  JJ.  The developmental approach to child and adult health.   Pediatrics. 2013;131(suppl 2):S133-S141. doi:10.1542/peds.2013-0252dPubMedGoogle Scholar
21.
Lleras-Muney  A.  The relationship between education and adult mortality in the United States.   Rev Econ Stud. 2005;72(1):189-221. doi:10.1111/0034-6527.00329Google Scholar
22.
Mazumder  B.  Does education improve health? A reexamination of the evidence from compulsory schooling laws.   Economic Perspectives. 2008;32(2).Google Scholar
23.
COVID-19 planning considerations: guidance for school re-entry. American Academy of Pediatrics. Updated August 19, 2020. Accessed October 19, 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/
24.
Husereau  D, Drummond  M, Petrou  S,  et al; CHEERS Task Force.  Consolidated Health Economic Evaluation Reporting Standards (CHEERS) statement.   BMJ. 2013;346:f1049. doi:10.1136/bmj.f1049PubMedGoogle Scholar
25.
Arias  E.  United States life tables, 2017.   Natl Vital Stat Rep. 2019;68(7):1-66.PubMedGoogle Scholar
26.
Clark  D, Royer  H.  The effect of education on adult mortality and health: evidence from Britain.   Am Econ Rev. 2013;103(6):2087-2120. doi:10.1257/aer.103.6.2087PubMedGoogle Scholar
27.
Gathmann  C, Jürges  H, Reinhold  S.  Compulsory schooling reforms, education and mortality in twentieth century Europe.   Soc Sci Med. 2015;127:74-82. doi:10.1016/j.socscimed.2014.01.037PubMedGoogle Scholar
28.
Grytten  J, Skau  I, Sørensen  R.  Who dies early? education, mortality and causes of death in Norway.   Soc Sci Med. 2020;245:112601. doi:10.1016/j.socscimed.2019.112601PubMedGoogle Scholar
29.
Meghir  C, Palme  M, Simeonova  E.  Education and mortality: evidence from a social experiment.   American Economic Journal: Applied Economics. 2018;10(2):234-256. doi:10.1257/app.20150365Google Scholar
30.
van Kippersluis  H, O’Donnell  O, van Doorslaer  E.  Long run returns to education: does schooling lead to an extended old age?   J Hum Resour. 2009;4:1-33.PubMedGoogle Scholar
31.
Oliver  A.  A normative perspective on discounting health outcomes.   J Health Serv Res Policy. 2013;18(3):186-189. doi:10.1177/1355819613485671PubMedGoogle Scholar
32.
Haacker  M, Hallett  TB, Atun  R.  On discount rates for economic evaluations in global health.   Health Policy Plan. 2020;35(1):107-114.PubMedGoogle Scholar
33.
Heinzerling  L.  Regulatory costs of mythic proportions.   Yale Law J. 1998;107(7):1981-2070. doi:10.2307/797416Google Scholar
34.
Landefeld  JS, Seskin  EP.  The economic value of life: linking theory to practice.   Am J Public Health. 1982;72(6):555-566. doi:10.2105/AJPH.72.6.555PubMedGoogle Scholar
35.
Malcolm  D, Roseboom  J, Clark  C, Fazar  W.  Application of a technique for research and development program evaluation.   Oper Res. 1959;7(5):646-669. doi:10.1287/opre.7.5.646Google Scholar
36.
New York City Department of Health and Mental Hygiene (DOHMH) COVID-19 Response Team.  Preliminary estimate of excess mortality during the COVID-19 outbreak—New York City, March 11-May 2, 2020.   MMWR Morb Mortal Wkly Rep. 2020;69(19):603-605. doi:10.15585/mmwr.mm6919e5PubMedGoogle Scholar
37.
Auger  KA, Shah  SS, Richardson  T,  et al.  Association between statewide school closure and COVID-19 incidence and mortality in the US.   JAMA. 2020;324(9):859-870. doi:10.1001/jama.2020.14348PubMedGoogle Scholar
38.
Courtemanche  C, Garuccio  J, Le  A, Pinkston  J, Yelowitz  A.  Strong social distancing measures in the United States reduced the COVID-19 growth rate.   Health Aff (Millwood). 2020;39(7):1237-1246. doi:10.1377/hlthaff.2020.00608PubMedGoogle Scholar
39.
Berchick  E.  Most uninsured were working-age adults.  US Bureau of the Census. Published September 12, 2018. Accessed December 23, 2020. https://www.census.gov/library/stories/2018/09/who-are-the-uninsured.html
40.
Higgins  JPT, Thomas  J, Chandler  J, Cumpston  M, Li  T, Page  MJ, Welch  VA.  Cochrane Handbook for Systematic Reviews of Interventions. 2nd ed. John Wiley and Sons; 2019.
41.
Hsiang  S, Allen  D, Annan-Phan  S,  et al.  The effect of large-scale anti-contagion policies on the COVID-19 pandemic.   Nature. 2020;584(7820):262-267. doi:10.1038/s41586-020-2404-8PubMedGoogle Scholar
42.
Dorn  E, Hancock  B, Sarakatsannis  J, Viruleg  E. COVID-19 and student learning in the United States: the hurt could last a lifetime. McKinsey & Co. Published June 1, 2020. Accessed July 21, 2020. https://www.mckinsey.com/industries/public-sector/our-insights/covid-19-and-student-learning-in-the-united-states-the-hurt-could-last-a-lifetime
43.
Eubank  S, Eckstrand  I, Lewis  B, Venkatramanan  S, Marathe  M, Barrett  CL.  Commentary on Ferguson, et al., “Impact of Non-pharmaceutical Interventions (NPIs) to Reduce COVID-19 Mortality and Healthcare Demand”.   Bull Math Biol. 2020;82(4):52. doi:10.1007/s11538-020-00726-xPubMedGoogle Scholar
44.
Ferguson  N, Laydon  D, Nedjati-Gilani  G, et al. Report 9: impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare demand. MRC Centre for Global Infectious Disease Analysis. Published March 16, 2020. Accessed October 19, 2020. https://www.imperial.ac.uk/mrc-global-infectious-disease-analysis/covid-19/report-9-impact-of-npis-on-covid-19/
45.
Roser  M, Ortiz-Ospina  E. Global education. Our World in Data. 2020. Accessed December 8, 2020. https://ourworldindata.org/global-education
46.
Nadworny  E. ‘Losing a generation’: fall college enrollment plummets for 1st-year students. NPR. Published December 17, 2020. Accessed December 21, 2020. https://www.npr.org/2020/12/17/925831720/losing-a-generation-fall-college-enrollment-plummets-for-first-year-students
47.
Education Week staff. Coronavirus and learning: what’s happening in each state. Education Week. Published April 3, 2020. Accessed July 21, 2020. http://blogs.edweek.org/edweek/campaign-k-12/2020/04/coronavirus_and_schools_state_guide.html?cmp=SOC-SHR-FB
48.
Viboud  C, Miller  M, Olson  D, Osterholm  M, Simonsen  L.  Preliminary estimates of mortality and years of life lost associated with the 2009 A/H1N1 pandemic in the US and comparison with past influenza seasons.   PLoS Curr. 2010;2:RRN1153-RRN1153. doi:10.1371/currents.RRN1153PubMedGoogle Scholar
×