Introduction

Over 3 million people around the world have lost their lives from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2, or COVID-19) as of May 2021. The United States alone has reported over 580,000 deaths [1]. A critical bottleneck in containing the virus is understanding its transmission dynamics. Although much effort has been expended on population-level seroprevalence surveys, scant data exist to understand COVID-19’s potential effect on the US public school system’s students and staff. Existing data are sobering: ten days into its reopening, an Israeli high school experienced a major outbreak [2]. Work on the pediatric transmission of COVID-19 are conflicting; in multiple studies conducted in Southwest Germany, Ireland, and Northern France, children under 10 were found to have had little effect on the spread of the virus, while a study in Chile claimed that elementary students were more likely to contract the virus relative to secondary students [3–6]. Here, we determine the seroprevalence of COVID-19 in the Lake Central School Corporation (LCSC), a public school system located in suburban Indiana, US. The LCSC consists of 1 high school (grades 9–12; 3,261 students), 3 middle schools (grades 5–8; 2,859 students), and 6 elementary schools (grades kindergarten-4, 3,467 students), and transitioned to fully virtual instruction on March 16^th 2020 [7]. Although there is enormous heterogeneity within the US public school system, LCSC staff’s demographics are broadly representative of national statistics. The LCSC has 16.6 students per teacher, similar to the United States average of 16 students per teacher [8, 9]. The median age of public school employees in the United States is 41, while the median age of the Lake Central employee population lies at 48 [10, 11]. Notably, previous reports have established that advanced age is highly associated with COVID-19 hospitalization, further underscoring the threat of morbidity and mortality within this community relative to other school districts [12]. Given the lack of specificity of COVID-19 symptoms and that mild and asymptomatic cases of COVID-19 may go undocumented, antibody-based seroprevalance studies are required to estimate population-level exposure to SARS-CoV-2, although it should be noted that immunity to SARS-CoV-2 via such antibodies has yet to be firmly established [13].

Materials and methods

Results and discussion

Of the eligible 1261 staff members, 753 (60%) participated in the study. The LCSC staff comprises 1060 (84%) women; the participation group had 635 (85%) women; age demographics for the study population compared to that of the total staff were similarly representative (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of study population and total staff population.

https://doi.org/10.1371/journal.pone.0243676.t001

22 individuals tested positive for anti-SARS-CoV-2 IgG antibodies, representing 2.9% seroprevalence (95% CI: 1.8% - 4.4%). After correcting for the reported sensitivity and specificity of the IgG antibody test, the seroprevalence is estimated as 1.7% (90% Credible Interval: 0.27% - 3.3%) [20].

In the univariate analyses, having a previous positive COVID-19 test (RR 29.5, 95% CI: 14.3–60.4, p<0.0001) or contact with a COVID-19 case (RR 6.86, 95% CI: 3.04–15.5), p<0.0001) were found to be statistically significantly associated with seropositivity, after Bonferroni correction (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Univariate analysis.

https://doi.org/10.1371/journal.pone.0243676.t002

No causal relationship between self-reported mask-wearing history and seropositivity could be identified, after controlling for six confounders available in our data set, identified using a causal diagram (S1 Table) (RR 0.63, 95% CI 0.16–4.3). This result was maintained when missing data were replaced, rather than excluded (S2 Table). In the multivariable model employing backward-stepwise feature elimination, a previous positive COVID-19 test (OR 48 95% CI 3.9–600, p = 0.003) or contact with a COVID-19 case (OR 5.6, 95% CI 2.1–15.1, p = 0.001) were found to be the most predictive factors for seropositivity (S3 Table); similar results were observed when missing data were replaced, rather than excluded (S4 Table). To interrogate the specific interaction between contact history and seropositivity, the dataset was subsetted to the vast majority of individuals without a previous positive COVID-19 test and complete data (n = 727). In this population, a logistic regression analysis including the predictors mask wearing, travel history, and symptom history, a positive contact history conferred a 7-fold increase in the odds of seropositivity (p<0.0001) (Table 3). The odds increased from 1:54 without such a contact to 1:7 with contact—or the probability of seropositivity increased from 1.8% to 12%. Similar results were observed when replacing, rather than excluding, missing data (S5 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Multivariable model.

https://doi.org/10.1371/journal.pone.0243676.t003

This study provides the necessary baseline for future longitudinal monitoring of COVID-19 transmission through a representative US public school system. With a demographically representative 60% staff participation rate, these data offer an unprecedented view into the seroprevalence of the target population of a Midwest US public school system. The estimated 1.7% seroprevalence of COVID-19 antibodies among LCSC staff is far below that reported in metropolitan areas throughout the US at the time of testing, underscoring the large and continued risk of COVID-19 infection within this community upon exposure [20]. As a point of comparison, Lake County, Indiana, the county within which the LCSC resides, had 4,985 reported cases of COVID-19, representing 1.0% of the county’s population [21].

Although the effect of mask wearing on seropositivity, utilizing a causal diagram to identify potential confounders, did not achieve statistical significance, this result should not be interpreted as a lack of efficacy in the use of masks given the overwhelming quality and quantity of data supporting their continued use, and is likely due to lack of power [22]. Both univariate and multivariable analyses demonstrate that a previous positive COVID-19 test or a history of contact with a COVID-19 patient were associated with seropositivity. The 7-fold increase in odds of seropositivity for individuals with a positive contact history is especially noteworthy amongst the vast majority (n = 727) of the individuals without a previous positive COVID-19 test and complete data. Although these relationships are associational, the combination of low staff seroprevalence and the strong positive association of seropositivity with contact history highlights a high risk group for which the importance of aggressive contact tracing would be efficient and which would minimize the transmission of COVID-19, as well as continued protective procedures during school hours such as mask and face shield wearing, and social distancing [23].

This study has several limitations. Although data were collected from staff employed at 11 different sites (10 schools and the transportation facility), only the LCSC was involved in the study, limiting the generalizability of this work. Neither ethnicity nor income data were collected, precluding analysis of these variables’ previously demonstrated associations with COVID-19 positivity [24]. We rely on study participants to self-report variables, excluding antibody status and blood type, via questionnaire. Some questions may be insufficiently granular, such as the binary mask wearing variable, and participants may also make errors filling out the questionnaire. Insufficient granularity is likely to make groups defined by the variable more similar to each other, while reporting errors are likely to be random and unrelated to the outcome, antibody status. Thereby both will be expected to create bias towards the null, meaning that the results reported in this study are conservative. The low baseline seroprevalence (22 positive tests in the 753-person cohort) prevents the identification of more subtle, but potentially real, associations among the collected variables and seropositivity. Additionally, the lack of county-level seroprevalence data prevents comparison to the broader population outside of the school district. Given that not all staff elected to participate in the study, some level of volunteer bias is possible. The high participation rate (60.0%), highly representative sample population (Table 1) and the wide availability of COVID-19 testing outside of this study, however, all limit the potential for volunteer bias to drive the observed results [25]. Finally, data was not collected on students in the corporation, preventing the investigation of a potential link between staff or student positivity and transmission within or between these two co-exposed groups.

The reopening of US schools poses a potential risk of COVID-19 transmission to both staff and students. The low pre-opening seroprevalence, in combination with the advanced age of a significant fraction of the LCSC staff, may increase the number and severity of cases should an outbreak occur within this school system, or other school systems sharing its demographic characteristics. Teachers should consider deploying novel teaching strategies that limit the amount of non-distanced interactions. Administrators and legislators should allocate the required resources to implement and maintain robust personal protective measures as recommended by the CDC, including contact-tracing, masks, face shields, and social distancing to protect the lives of both the children of the school district and those charged with educating them [26].

Supporting information

Acknowledgments

We thank all of the LCSC staff members who participated in this study. We also thank Jana L. Lacera, Director, IRB/Bio-Ethics, who graciously volunteered both her and her organization’s resources; the nurses from Franciscan WorkingWell for collecting participant samples; Dr. Marc Lipsitch for his invaluable guidance, feedback and encouragement; and the LCSC for enabling this work.