Prevalence of SARS-CoV-2 in Children Identified by Pre-procedural Testing at Five US Children’s Hospital Systems

Abstract

Background:

The burden of SARS-CoV-2 in asymptomatic children was initially presumed to be high, which influenced hospital, school, and childcare policies. Before vaccines were widely available, some hospitals implemented universal pre-procedural SARS-CoV-2 PCR testing on asymptomatic patients. Understanding SARS-CoV-2 prevalence in asymptomatic children is needed to illuminate the diversity of viral characteristics and to inform policies implemented during future pandemics.

Methods:

Data was extracted from patient records of outpatient children who were pre-procedurally tested for SARS-CoV-2 from five US hospital systems between March 1, 2020 and February 28, 2021. Prevalence was determined from positive test results. Adjusted odds ratios (AOR) were calculated using mixed logistic regression with site as random effect.

Results:

This study analyzed 93,760 pre-procedural SARS-CoV-2 test results from 74,382 patients and found 2693 infections (3.6%) from 2889 positive tests (3.1%). Site-specific prevalence varied across sites. Factors modestly associated with infection included being uninsured (AOR, 1.76; 95% CI, 1.45, 2.13), publicly insured (AOR, 1.17; 95% CI, 1.05, 1.30), Hispanic (AOR, 1.78; 95% CI, 1.59–1.99), Black (AOR, 1.22; 95% CI, 1.06–1.39), elementary school age (5–11 years AOR, 1.15; 95% CI, 1.03–1.28), or adolescent (12–17 years AOR, 1.26; 95% CI, 1.13, 1.41).

Conclusions:

SARS-CoV-2 prevalence was low in outpatient children undergoing pre-procedural testing, a population that was predominantly asymptomatic at the time of testing. This study contributes evidence that suggests undetected infection in children likely did not play a predominant role in SARS-CoV-2 transmission during the early pre-vaccine pandemic period when the general population was naïve to the virus.

INTRODUCTION

During the first year of the SARS-CoV-2 pandemic, there was considerable concern that children were asymptomatic drivers of infection, and infections were thought to be under-detected due to lack of testing.^1,2 Because SARS-CoV-2 was a novel virus and little was known about the role of children in spread, corollary assumptions were made based on influenza transmission patterns. Up to 43% of children become infected with influenza during a typical influenza season, and children play a major role in influenza transmission in households and schools.^3–5 As the burden of SARS-CoV-2 in children was unknown at the time, it was considered to be potentially similar to that of influenza. Schools and childcare centers were consequently closed in efforts to contain the virus and protect older adults in the community; however, these closures were highly contentious and varied across the country.⁶ Children’s hospitals initially postponed non-urgent procedures and later instituted pre-procedural SARS-CoV-2 testing, requiring patients to be tested even if they did not have COVID-19 symptoms.

Because there was a shortage of testing supplies and testing was concentrated on those with symptoms, the burden of asymptomatic infection in children was not well studied during that unique time period when the global population was naïve to SARS-CoV-2. We therefore used outpatient pre-procedural testing results from five children’s hospitals across the nation as a proxy for systematically estimating the prevalence of infection in asymptomatic children.

This study assessed the period prevalence of SARS-CoV-2 infection in children using pre-procedural tests performed from March 2020 through February 2021. The study also examined associations between SARS-CoV-2 infection and patient demographic characteristics, including individual insurance status, community median income level, age, race, and ethnicity. Our finding that asymptomatic children were uncommonly infected contributes evidence that silent infections in children did not drive SARS-CoV-2 transmission in the pre-vaccine era of the pandemic.

METHODS

Data collection and participants

Five pediatric hospital systems in five cities (Seattle, WA; Kansas City, MO; Cincinnati, OH; Pittsburgh, PA; and Houston, TX) participated in this study. During the study period, hospital policies directed patients to be tested once within 48–72 hours before the scheduled procedure, even if they denied having COVID-19 symptoms. Patients were required to obtain a negative SARS-CoV-2 test within 48–72 hours of their scheduled procedure. All hospitals used an in-house RT-PCR assay to detect SARS-CoV-2.

We requested electronic medical record (EMR) data of pediatric outpatients <18 years who had pre-procedural SARS-CoV-2 test results from March 1, 2020 through February 28, 2021. Patients admitted for one day or classified as observation were included. To limit the study population to asymptomatic patients, inpatients, and emergency department patients were excluded. Data variables extracted from the EMR included testing date by MMWR calendar week, test results, sex, age at time of testing, race, ethnicity, insurance type, and residential zip code. Records that did not include a test result, testing date, or age were excluded. Race and ethnicity data were self-reported by patients at the time of patient registration or at health care encounter. Patients could identify their race as Alaska Native, American Indian, Asian, Black, Native Hawaiian, Pacific Islander, White, or multiple races. Patients could also identify their ethnicity as Hispanic or Non-Hispanic. These data were included for descriptive purposes to assess the representativeness of the study population because health care utilization may vary by race or ethnicity.^7,8 Institutional Review Boards at all institutions approved or issued an exempt determination for this study with the University of Pittsburgh acting as the coordinating site (STUDY20120010). This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Definitions

Due to the sensitivity of RT-PCR, a patient could have multiple positive tests associated with a single infection because patients who tested positive were required to reschedule their procedures and repeat testing a few weeks later. Some patients presented for multiple procedures and were tested each time. All sites except Cincinnati provided de-identified unique patient IDs for each test record. All test records from Cincinnati were assumed to be from unique patients.

An infection was defined as a positive SARS-CoV-2 pre-procedural test result separated by more than 13 weeks from any preceding or subsequent positive SARS-CoV-2 preprocedural test result associated with the same patient. Prevalence was defined as the number of infections during the study period per total population of unique patients.

Statistical analysis

Descriptive statistics were used for demographic information and SARS-CoV-2 test result counts. Median income by zip code was obtained from the US Census.^9,10 Simple logistic regression models were used to calculate the unadjusted odds ratio (OR) between demographic characteristics and infection status. A mixed logistic regression model with site as random effect was used to calculate the adjusted OR (AOR), accounting for variability at the site level. A multivariable logistic regression was used to determine AORs for associations stratified by site. Missingness was assumed to be missing completely at random. Data analyses were conducted using SAS version 9.4 (SAS Institute Inc). Sensitivity analyses were run by excluding Cincinnati data and income data.

RESULTS

Descriptive characteristics of study population

A total of 93,760 pre-procedural SARS-CoV-2 tests from 74,382 patients collected from March 1, 2020 to February 28, 2021 were included in this study. The overall prevalence of SARS-CoV-2 infection was 3.6% (2693 of 74,382), and 3.0% (2240 of 74,382) of patients were infected at least once during the study period (Table 1). Of all tests, 3.1% (2889 of 93,760) were positive. Prevalence varied across sites: Pittsburgh, 0.7% (38 of 5519); Seattle 0.7% (68 of 9853), Cincinnati, 1.5% (98 of 6633); Kansas City, 2.6% (333 of 12,638); Houston, 5.4% (2156 of 39,739). The number of tests collected each week was relatively consistent during the study period (Fig. 1A), and percent positivity peaked during the end of December 2020 (Fig. 1B).

Table 1.

Patient characteristics by SARS-CoV-2 infection status

		SARS-CoV-2 infection status
	Total	At least 1 infection during study period	Never infected during study period	Unadjusted OR (95%CI)	Adjusted ORa (95%CI)
	(N=74382)	(N=2240 [3.01%])	(N=72142 [96.99%])
Site, n (col%, row%)b
Cincinnati	6633 (8.9)	98 (4.4, 1.5)	6535 (9.1, 98.5)	2.42 (1.64, 3.58)
Seattle	9853 (13.2)	47 (2.1, 0.5)	9806 (13.6, 99.5)	0.77 (0.50, 1.20)
Houston	39739 (53.4)	1784 (79.6, 4.5)	37955 (52.6, 95.5)	7.58 (5.39, 10.66)
Kansas City	12638 (17.0)	277 (12.4, 2.2)	12361 (17.1, 97.8)	3.62 (2.53, 5.17)
Pittsburgh	5519 (7.4)	34 (1.5, 0.6)	5485 (7.6, 99.4)	ref
Age, years				1.02 (1.01, 1.03)
Median (IQR)	6 (2, 12)	7 (3, 12)	6 (2, 12)
Range	0, 17	0, 17	0, 17
Age group, n (col%, row%)
0–4	30138 (40.5)	796 (35.5, 2.6)	29342 (40.7, 97.4)	ref	ref
5–11	24755 (33.3)	776 (34.6, 3.1)	23979 (33.2, 96.9)	1.19 (1.08, 1.32)	1.15 (1.03, 1.28)
12–17	19489 (26.2)	668 (29.8, 3.4)	18821 (26.1, 96.6)	1.31 (1.18, 1.45)	1.26 (1.13, 1.41)
Gender, n (col%, row%)
Male	41694 (56.1)	1210 (54.0, 2.9)	40484 (56.1, 97.1)	ref	ref
Female	32675 (43.9)	1029 (46.0, 3.1)	31646 (43.9, 96.9)	1.09 (1.00, 1.18)c	1.06 (0.97, 1.16)
Missing	13	1	12
Race, n (col%, row%)
Asian	3051 (4.5)	62 (3.0, 2.0)	2989 (4.5, 98.0)	0.66 (0.51, 0.86)	0.93 (0.71, 1.22)
Black	10181 (15.0)	343 (16.8, 3.4)	9838 (14.9, 96.6)	1.11 (0.99, 1.25)	1.22 (1.06, 1.39)
White	53500 (78.7)	1624 (79.4, 3.0)	51876 (78.7, 97.0)	ref	ref
Otherd	1245 (1.8)	16 (0.8, 1.3)	1229 (1.9, 98.7)	0.42 (0.25, 0.68)	1.06 (0.62, 1.78)
Missing	6405	195	6210
Ethnicity, n (col%, row%)
Hispanic	19571 (27.4)	1030 (47.9, 5.3)	18541 (26.7, 94.7)	2.52 (2.31, 2.75)	1.78 (1.59, 1.99)
Not Hispanic	51913 (72.6)	1119 (52.1, 2.2)	50794 (73.3, 97.8)	ref	ref
Missing	2898	92	2807
Insurance, n (col%, row%)
Private / Both	34571 (46.7)	779 (34.9, 2.3)	33792 (47.0, 97.7)	ref	ref
Public	36975 (49.9)	1292 (58.0, 3.5)	35683 (49.7, 96.5)	1.57 (1.44, 1.72)	1.17 (1.05, 1.30)
Self-Pay	2520 (3.4)	158 (7.1, 6.3)	2362 (3.3, 93.7)	2.90 (2.43, 3.46)	1.76 (1.45, 2.13)
Missing	316	11	305
Income, USD e				0.96 (0.94, 0.97)
Median (IQR)	69196 (52793, 92026)	65105 (50118, 85803)	69500 (53003, 92096)
Range	8618, 250000	19250, 210577	8618, 250000
Income quintiles, n (col%, row%)
1st quintile (<$50,118)	14738 (20.0)	556 (24.9, 3.8)	14182 (19.8, 96.2)	1.42 (1.24, 1.62)	0.99 (0.85, 1.15)
2nd quintile ($50,118-$62,125)	14693 (19.9)	462 (20.7, 3.1)	14231 (19.9, 96.9)	1.17 (1.02, 1.35)	1.02 (0.88, 1.19)
3rd quintile ($62,126-$77,260)	14810 (20.1)	475 (21.3, 3.2)	14335 (20.0, 96.8)	1.20 (1.05, 1.37)	1.07 (0.92, 1.24)
4th quintile ($77,261-$98,541)	14737 (20.0)	339 (15.2, 2.3)	14398 (20.1, 97.7)	0.85 (0.74, 0.99)f	0.88 (0.75, 1.03)
5th quintile (>=$98,542)	14759 (20.0)	397 (17.8, 2.7)	14362 (20.1, 97.3)	ref	ref
Missing	645	11	634

^aAdjusted OR from mixed logistic model with site as random effect

^bColumn percentages are shown for total data, and column and row percentages are shown for SARS-CoV-2 infection status.

^cAssociation is significant when Cincinnati data are excluded (OR, 1.1; 95% CI, 1.01–1.2) in the unadjusted model.

^dPatients who indicated American Indian, Alaska Native, Native Hawaiian, Pacific Islander, or more than one category were classified as other race due to small sample sizes.

^eUnit of income is $10,000

^fAssociation is not significant when Cincinnati data are excluded (OR, 0.87; 95% CI, 0.75–1.01) in the unadjusted model.

Figure 1.

SARS-CoV-2 test result counts (A) and percentage of tests with positive detection (B) by week

The median (IQR) age of patients was 6 (2,12) years and 56.1% were male. Houston contributed 53.4% of the total patients, followed by Kansas City (17.0%), Seattle (13.2%), Cincinnati (8.9%), and Pittsburgh (7.4%). The majority of patients identified as White (White 78.7%, Black 15.0%, Asian 4.5%, other race 1.8%) and non-Hispanic (72.6%) (Table 1). Most patients had health insurance coverage (public 49.9%, any private insurance (private / both) 46.7%), though 2520 (3.4%) were self-paying / uninsured individuals. Using residential zip codes, the median (IQR) community income of patients was $69,196 ($52793, $92026) (Table 1).

Factors associated with SARS-CoV-2 infection

Unadjusted analysis of all sites in aggregate

In the unadjusted model, SARS-CoV-2 infection was significantly associated with age, race, ethnicity, insurance type, and community income. Compared with 0–4-year-olds, school-age children and adolescents had significantly higher odds of infection (e.g., 12–17 years OR, 1.31, 95% CI, 1.18–1.45) (Table 1). Asian patients and those in other racial groups had lower odds of infection compared with White patients, though the association was no longer significant when adjusted for other covariates (Asian OR, 0.66; 95% CI, 0.51–0.86; other OR, 0.42; 95% CI, 0.25–0.68). Hispanic patients had greater odds of infection than non-Hispanic patients (OR, 2.52; 95% CI, 2.31–2.75) (Table 1).

Individuals who self-paid had higher odds of infection compared to those with any private insurance (OR, 2.90; 95% CI, 2.43–3.46). Though not as marked as those who self-paid, patients with public insurance also had greater odds of infection compared to those with any private insurance (OR, 1.57; 95% CI, 1.44–1.72) (Table 1).

Median income of patient residential zip codes was used to examine an association between infection and community income. For every $10,000 increase in income, the odds of infection decreased by 4% (OR, 0.96; 95% CI, 0.94–0.97) (Table 1), reflected in the modest downward trend of proportion infected as community median income increased (Fig. 2A). The lowest income quintile (<$50,118) had the highest proportion of infections (3.8%, 556 of 14,738) (Table 1). Higher income quintiles had lower proportions of infections (4^th quintile, $77,260-$98,541, 2.3% (339 of 14737); 5^th quintile, ≥$98,542, 2.7% (397 of 14759). Thus, the lowest income quintile had higher odds of infection compared to the highest income quintile (OR, 1.42; 95% CI, 1.24, 1.62). The second quintile ($50,118-$62,125) and third quintile ($62,126-$98,541) had higher odds of infection compared to the 5^th quintile (2^nd quintile OR, 1.17; 95% CI, 1.02–1.35; 3^rd quintile OR, 1.20; 95% CI, 1.05–1.37) (Table 1). When stratified by insurance status, proportion infected among the self-pay group was consistently higher than in the other groups across income levels, and private/both consistently had the lowest proportion infected (Fig. 2B).

Figure 2.

Proportion of patients infected by SARS-CoV-2 by residential median community income level: overall population (A) and stratified by insurance type (B)

Income bins <$30,000 and ≥$130,000 were collapsed due to low number of patients.

Adjusted analysis of all sites in aggregate

When considering all patients in aggregate, age, race, ethnicity, and insurance status were independently associated with acquiring at least one SARS-CoV-2 infection. Older children had higher odds of infection compared to children 0–4 years (e.g., 12–17 years: AOR, 1.26; 95% CI, 1.13–1.41). Black patients had higher odds of infection compared to White patients (AOR, 1.22; 95% CI, 1.06–1.39), and Hispanic patients had higher odds of infection than non-Hispanic patients (AOR, 1.78; 95% CI, 1.59–1.99). Patients who self-paid and those with public insurance had higher odds of infection compared to those with any private insurance (Self-pay AOR, 1.76; 95% CI 1.45–2.13; public AOR, 1.17; 95% CI, 1.05–1.30) (Table 1).

Site-specific findings

Because Houston contributed 53.4% of the total patient population (Table 1), we stratified the analysis by site (see Table, Supplemental Digital Content 1). Most Hispanic patients in this study were from Houston (80.6%, 15,770 of 19,571) and 41.6% of Houston patients were Hispanic (see Table, Supplemental Digital Content 1). Nevertheless, when stratifying results by site, Hispanic patients at every site except Seattle had significantly higher odds of infection than non-Hispanic patients; Pittsburgh’s Hispanic population was too small to analyze (Cincinnati AOR, 7.62; 95% CI, 4.15–13.98; Houston AOR, 1.67; 95% CI, 1.48–1.89; Kansas City AOR, 1.84; 95% CI, 1.3–2.6). Children 12–17 years also had significantly higher odds of infection than children 0–4 years in Cincinnati (AOR, 1.94; 95% CI, 1.12–3.37), Houston (AOR, 1.15; 95% CI 1.01–1.30), and Kansas City (AOR, 1.85; 95% CI, 1.35–2.54) (see Table, Supplemental Digital Content 1).

Sensitivity analysis

Because zip code median income does not necessarily reflect the individual patient’s person income, we conducted a sensitivity analysis excluding zip code median income levels from the model. We also ran an analysis excluding Cincinnati data because unique patient identifiers were not available, and the prevalence of infection rose from 3.6% to 3.8% (2595 of 67,749). The ORs, AORs, and 95% CIs were very similar regardless of whether zip code median income or Cincinnati data were included in the model.

DISCUSSION

This 12-month retrospective cross-sectional study revealed a low prevalence of SARS-CoV-2 infection (3.6%) in outpatient children tested before undergoing a procedure and a low percent positivity of all pre-procedural tests (3.1%). This suggests that the level of SARS-CoV-2 infection in asymptomatic or pre-symptomatic children was low during the early phase of the pandemic when the general population was naïve to the virus.

During the study period from March 2020 to February 2021, before any vaccines were widely available, all hospitals in this study implemented universal pre-procedural testing regardless of the absence of symptoms. Under hospital practices at the time, patients with acute respiratory symptoms may have been asked to reschedule their outpatient procedures and to delay pre-procedural testing until they no longer had symptoms. SARS-CoV-2 testing on symptomatic patients should not have been classified as pre-procedural and therefore were excluded from this study. Nevertheless, some patients may still have had acute respiratory symptoms at the time of testing or may have been pre-symptomatic. Because this study included nearly 100,000 test records across five hospital systems, it was not feasible to confirm the absence of symptoms at the time of testing or during a longitudinal follow up period using medical chart review. However, other pre-procedural studies on smaller populations reported that only 3–12% of patients tested pre-procedurally were symptomatic at the time of testing.^11–13 Therefore, we interpret the SARS-CoV-2 pre-procedural prevalence from this study to represent the upper bound of the prevalence of infection in asymptomatic children.

To our knowledge, this is the largest study of pre-procedural SARS-CoV-2 prevalence, analyzing almost 100,000 test records from five hospitals over one year from all clinical specialties. The overall (3.6%) and site-specific prevalence (0.7–5.4%) observed in this study are consistent with the low prevalence (0 to <1%) reported by other pre-procedural studies^11–16 that encompassed shorter study periods or smaller populations as well as the low frequency of asymptomatic SARS-CoV-2 infection in children observed by surveillance studies of households with children and children attending childcare centers (3–4%).^17–20 Similarly, studies that performed SARS-CoV-2 testing on all hospitalized children and all asymptomatic children presenting to the emergency department showed that only 1–4% of asymptomatic children tested positive.^21,22 Though some of these studies highlighted that a substantial proportion of SARS-CoV-2 cases were asymptomatic (33–86%), the total number of asymptomatic cases detected was small, resulting in only 0.5–4.4% positivity among asymptomatic children, consistent with our findings.^{11,13,17–20,22}

In contrast to SARS-CoV-2, previous studies have found that other respiratory viruses are very commonly found in asymptomatic children, with detection of at least one virus in 27% to 42% of asymptomatic children.^23–25 Considering the high proportion of infection in asymptomatic children for some respiratory viruses, it was reasonable that hospitals and schools were concerned about asymptomatic SARS-CoV-2 infection at the beginning of the pandemic when less was known about the burden of SARS-CoV-2 infection in children. However, this study adds to previous work that illustrates the diverse characteristics of respiratory viruses. For example, rhinovirus was detected in 32% of asymptomatic children, but human metapneumovirus and respiratory syncytial virus were uncommonly detected.^23,24,26 Therefore, the presumption that children would be major spreaders of SARS-CoV-2 as they are for influenza did not fully account for the varied transmission patterns of respiratory viruses and was not, in retrospect, supported by our findings.

Our exploratory analysis also revealed that race, ethnicity, and insurance status were modestly associated with higher odds of SARS-CoV-2 infection. Though investigating the underlying factors was outside the scope of this study, these findings were consistent with other studies that proposed the disparities of SARS-CoV-2 infection were complex and due to multiple structural and systemic factors such as limited resources, barriers to health care access, and discrimination.^27–32 In our study, Hispanic patients had 78% greater odds of infection compared to non-Hispanic patients, and Black patients had 22% higher odds of infection compared to White patients. Those who self-paid, whom we interpret to lack insurance coverage, had 76% greater odds of infection, while those with public insurance had 17% greater odds of infection compared to those with private insurance. We used individual-level insurance, race, and ethnicity data, and our findings were consistent with other studies that used SARS-CoV-2 test results from the general population and community-level estimates for insurance, race, and ethnicity from US Census data.^27–32 Contrary to studies that found that lower family income was associated with higher infection rates,^27,28,31 our results showed that the association between income level and infection was no longer significant after adjusting for all covariates, including insurance status. We interpret this to indicate that insurance status is the stronger effect. This difference could be because the other studies either did not adjust for insurance status^27,31 or used community-level estimates for insurance status in their model.²⁸ Another factor is that these studies, including this study, used median household income by zip code, which does not necessarily reflect individual household income. For example, we observed some instances where individuals with public insurance lived in zip codes with very high median incomes. This incongruence could obscure the effect of income in our analysis.

Strengths and Limitations

The strengths of this study include the large sample size across five cities in geographically different locations in the US, which included pre-procedurally tested outpatients from all clinical specialties with individual-level insurance status data. The study period spanned 12 months and covered the first three waves of the pandemic when the global population was naïve to the virus, before vaccines were widely available.

This study has some limitations due to its cross-sectional design. Without longitudinal follow-up, we could not determine whether patients were truly asymptomatic or pre-symptomatic at the time of testing. Because of the large sample size, it was not feasible to confirm the absence of symptoms from the patients’ health records. Therefore, the prevalence reported in this study could be an overestimate of the true prevalence of infection in asymptomatic children. One site was not able to provide de-identified unique patient identifiers for each test record; however, excluding that site from the analysis had very little effect on results. Missingness for any variable for all sites in aggregate was <9%, but race information was missing for 20% of Seattle test records. The racial distribution of Seattle patients in this study was similar to that of the general population in the city, except that the percentage of White patients was higher (72.1% vs 63.6%) and the percentage of Asian patients was lower (10.2% vs 16.8%) than in the general population.³³ None of the Seattle racial group ORs were statistically significant, but they may therefore slightly deviate from the true effect estimate. Because individual income or residential address was not available, zip code median income was used instead of income by census tract. The study population was drawn from children who needed outpatient procedures, which limits the generalizability of this study.

CONCLUSIONS

This study confirms that the overall prevalence of SARS-CoV-2 infection in asymptomatic children was low in the pre-vaccine era. Factors modestly associated with infection included being uninsured, publicly insured, Hispanic, Black, school-aged, or adolescent. When considered with a previous study that showed that asymptomatic infected children rarely transmitted the virus,¹⁹ this study contributes evidence that suggests asymptomatic infection in children likely did not play a predominant role in SARS-CoV-2 transmission during the early pandemic period. Because school and childcare closures have substantial detrimental effects on childhood health and development,^34–37 investigation and factual findings regarding children being major spreaders of disease must be prioritized early so that appropriate policies are instituted. Virus-specific characteristics should be considered when evaluating the need to implement universal pre-procedural testing and infection prevention policies for future emergent viruses.

Supplementary Material

Supplemental Digital Content (Including Legend)

Supplemental Digital Content 1. Table