Severity of Acute COVID-19 in Children <18 Years Old March 2020 to December 2021

EHR data were obtained from institutions participating in PEDSnet (pedsnet.org), a national network of pediatric health systems that share clinical data and conduct observational studies, clinical trials, and population surveillance.¹⁴  Institutions included: Children’s Hospital of Philadelphia, Cincinnati Children’s Hospital Medical Center, Children’s Hospital of Colorado, Ann & Robert H. Lurie Children’s Hospital of Chicago, Nationwide Children’s Hospital, Nemours Children’s Health System (a Delaware and Florida health system), Seattle Children’s Hospital, and Stanford Children’s Health. EHR data were aggregated for health care encounters provided in outpatient, inpatient, and emergency department (ED) or urgent care settings. Annually, these PEDSnet institutions provide services to 3.3% of the nation’s children (2.4 million patients).

The PEDSnet COVID-19 Database Version 2022-01-06 was used for this study. All historical EHR data from institutions were extracted for patients who met selection criteria. Institutional source data were standardized to the PEDSnet common data model, an extension of the Observational Medical Outcomes Partnership common data model, described in detail elsewhere.¹⁵  The Children’s Hospital of Philadelphia’s institutional review board designated this study as not human subjects research and waived the need for consent. We conducted a 22-month retrospective cohort study of patients aged <18 years with a positive SARS-CoV-2 viral test (either nucleic acid amplification test or the polymerase chain reaction test) from March 1, 2020, to December 30, 2021. Children with multisystem inflammatory syndrome were included if they had a positive viral test. The cohort entrance date for patients was the day when the positive viral test was done. Four institutions extracted data in mid-December and the others in early January 2022. Institutions validated the count of tested patients and test-positive patients with organizational registries.

Illness severity refers to level of physiologic derangement and instability, and it can be graded from low to high.¹⁶  Most COVID-19 severity classifications use hospitalization and ICU level of care as indicators.¹⁷  We developed a new acute COVID-19 severity classification for this study on the basis of EHR data and included concepts, such as COVID-19–related health conditions from previous severity classifications.^17–21  However, we elected to exclude hospitalization as a criterion because children may be tested as part of universal inpatient screening and found to be positive for SARS-CoV-2 incidentally, which would not reflect illness severity. Severity classification had 4 levels: asymptomatic, mild (COVID-19–related symptoms), moderate (moderately severe COVID-19–related health conditions), and severe (unstable COVID-19 health conditions). Symptoms and health conditions were identified using clinically cogent sets of SNOMED CT diagnosis codes, examined 7 days before and 13 days after the day of viral test positivity, creating a 21-day episode of care. A patient was assigned to the highest level of severity for which they met at least 1 criterion.

Symptom clusters included abdominal pain, anorexia, cough, diarrhea, fever/chills, headache, loss of taste or smell, fatigue, myalgia, nasal congestion, nausea, sore throat, respiratory symptoms, and vomiting. We also included a diagnosis of uncomplicated COVID-19 that occurred up to 13 days after the positive viral test, but not on the day of testing. Conditions indicative of moderate severity included acute bronchitis, bronchiolitis, dehydration, gastroenteritis, pneumonia, and use of intravenous fluids during the first 6 hours of an ED visit. Unstable conditions reflective of severe illness were acute respiratory distress syndrome, acute kidney injury, acute liver failure, death, encephalopathy/encephalitis, myocarditis, pericarditis, respiratory failure, sepsis, shock, and thromboembolism. Patients with evidence of ICU level care, invasive or noninvasive mechanical ventilation, or vasopressor/inotropic support were also considered to have severe illness (SNOMED and RxNorm code sets are freely available and downloadable at https://github.com/PEDSnet/COVID19_Severity).

The exposure variable was calendar time. Patients were assigned to a calendar month on the basis of the day when their first positive SARS-CoV-2 test was done. Other covariates included age, sex, race or ethnicity (Hispanic based on ethnicity and other non-Hispanic categories based on race), institution, and chronic condition comorbidities. We used the Pediatric Medical Complexity Algorithm (PMCA) Version 2.0²²  to categorize children as having no chronic condition, noncomplex chronic condition, or complex chronic condition comorbidities. We considered diagnoses up to 3 years before cohort entrance. Children were assigned to the complex chronic condition category if they had conditions that affected ≥2 body systems (eg, type 1 diabetes with end-organ complications) or a progressive chronic condition (eg, muscular dystrophy), or continuous dependence on technology for ≥6 months (eg, dialysis, tracheostomy with ventilator assistance).²³  We also used the PMCA to aggregate related diagnoses according to the body system involved for cardiac, immunologic, malignancy, mental health, and renal disorders. Asthma, diabetes, prematurity, sickle cell disease, and trisomy 21 were identified using SNOMED CT diagnostic terms. As indicators of severity for preexisting conditions, patients with asthma were on the basis of whether they were hospitalized for asthma up to 12 months before cohort entrance, and malignancy was stratified according to active treatment (ie, receipt of cancer chemotherapy up to 3 months before cohort entrance). Obesity was defined as presence of an age-sex–standardized BMI z-score ≥95th percentile based on weight measured at the time of SARS-CoV-2 testing and height within 60 days of cohort entrance.

We contrasted the study sample with patients whose SARS-CoV-2 viral test was negative on demographic characteristics and chronic disease comorbidities. Significance testing for proportions was done using the χ² test and for means using analysis of variance. A critical value of 0.01 was used because of the large sample size.

To evaluate the construct validity of the new severity classification, we tested the hypotheses that increasing severity of illness would be associated with higher use of COVID-19–related medications (eg, remdesivir, systemic steroids, etc; see Supplemental Table 4 for full list), receipt of a COVID-19 diagnosis code, and hospitalization during the 21-day episode of illness. We used this 21-day observation window because our focus was on acute COVID-19 only, which excludes postacute and chronic COVID-19 syndromes. We also counted the number of outpatient visits during 27 days after test positivity. Finally, we tested the hypothesis on the basis of previous research^24,25  that children with complex chronic conditions were more likely to be represented in the higher severity categories.

We computed the monthly number of acute COVID-19 cases on the basis of each patient’s cohort entrance date, overall and by severity category. These counts were plotted with total cases on one y-axis and moderate or severe cases on the other y-axis. To evaluate change in severity over time, we plotted the proportions of cases that were assigned to the moderate or severe severity levels.

In unadjusted and adjusted analyses, we examined the risk of moderate or severe illness by demographic, clinical, and calendar time predictors. The sample was children enrolled in the study cohort July 1, 2020, or later, when outpatient testing become widespread in PEDSnet institutions. The adjusted analyses employed logistic regression and the generalized estimating equation with robust estimators to address the clustering of cases by institution. Because the unadjusted risk of moderate to severe illness did not substantively change from July 2020 to February 2021, we grouped those months, set them as the referent, and allowed each month thereafter to be an indicator variable. Other predictors included demographics and chronic condition comorbidities.

Across the 9 health systems, 866 914 patients were tested for SARS-CoV-2 infection. Of this total, 70 862 (8.2%) had no previous encounter with the health systems. The study cohort comprised 82 798 patients (9.6% of those tested) with a positive SARS-CoV-2 viral test. The number of tests done peaked in September 2021 (n = 66 728). The test positivity rate varied by calendar time, with the highest rate in December 2021 (12.7%), probably as a result of the emergence of the omicron variant. Compared with test-negative patients, those who were test-positive were more likely to be adolescent, female, Black/African-American, and Hispanic (Table 1).

There were 54 948 patients (66.4%) classified as asymptomatic, 22 303 (26.9%) mild, 3781 (4.6) moderate, and 1766 (2.1%) severe. Among patients classified as mild, the 5 most common symptoms were fever and chills (39.6%), cough (30.6%), sore throat (21.5%), nasal congestion (11.5%), and headache (9.4%). In contrast to adults, loss of taste and smell was an infrequent symptom for children. For those classified as moderately ill, gastroenteritis and dehydration (26.2%), bronchitis (19.9%), pneumonia (9.2%), and bronchiolitis (7.6%) were the most common conditions. We also included in the moderate category patients who received intravenous fluids within the first 6 hours of an ED encounter (54.7%). For the severe group, respiratory failure (34.8%), acute respiratory distress syndrome (39.4%), mechanical ventilation (invasive and noninvasive combined) (34.0%), and admission to the ICU (46.0%) were the most common conditions accounting for assignment to the severe category (see Supplemental Table 5 for the complete distributions of symptoms and conditions by severity level).

Severity of illness had monotonic and positive relationships with receipt of COVID-19–related medications, a COVID-19 diagnosis, outpatient visits, and hospitalization (Table 2). We found an increasing proportion of patients with complex chronic conditions across the severity levels. Among the severely ill patients, 41.5% had a complex chronic condition and the majority were treated with a COVID-19–related medication. These findings provide strong support for the validity of the classification typology.

COVID-19 cases peaked and declined during the 2020–2021 winter months and, again, during the alpha variant phase (March–June 2021), with a third rise during the delta variant phase (July–December 2021, Fig 1). As total cases rose or declined, so, too, did the number of moderate to severe cases. On the other hand, the proportion of moderate and severe cases remained stable from July 2020 to April 2021, even as the number of cases rose and fell (Fig 2). Severity peaked in June 2021 and declined during the delta variant phase (July–September 2021).

Infants and those of Hispanic ethnicity were at the highest risk of moderate to severe illness (Table 3). Nearly all the chronic condition comorbidities were associated with severe illness, including diabetes, obesity, cancer (with markedly increased risk for those actively treated), prematurity, and asthma with a history of previous hospitalization. In a second regression analysis that controlled for institution and used the same demographics but replaced the specific chronic conditions with a 3-category chronic condition grouping, the risk of moderate or severe illness was markedly increased for complex chronic conditions (adjusted odds ratio [aOR] 4.2; 95% confidence interval [CI], 3.9–4.6) but not for noncomplex chronic disease (aOR, 1.1; 95% CI, 1.0–1.2).

Children are at risk for acquiring SARS-CoV-2 infection, and although most cases are mild, about 1 in 16 will contract moderate or severe illness that may require hospitalization or require intensive care, mechanical ventilation, and/or treatment with antiviral medications. We and others^24,25  have demonstrated that children with complex chronic comorbidities, such as cancer that is actively treated, are particularly vulnerable to SARS-CoV-2 infection. Our data provide no evidence for an increased risk of severe COVID-19 associated with the delta variant, which predominated in the United States beginning July 2021. Nonetheless, because the absolute number of children with severe illness increases in tandem with increases in overall cases of COVID-19, the count of moderate or severely ill patients did increase during the delta variant phase. Additional observation is needed to evaluate the effect of the omicron variant on illness severity. The best indicator of the burden of COVID-19 that may require advanced medical care is overall community rates of the disease.

Our severity of illness classification did not rely on hospitalization and demonstrated excellent validity, a strength of this study. We found a transient rise in severity during the alpha variant phase of the pandemic in the United States (March–June 2021), with declines during the delta variant phase (July–December 2021). Of note, this trend in severity parallels observed rises in prevalence of other respiratory viruses, including respiratory syncytial virus, metapneumovirus, and parainfluenza virus.²⁶  We speculate that the observed increase in severity may have been because of, at least in part, respiratory illness caused by other viruses concurrently with SARS-CoV-2. In the absence of per-case testing for other viral infections, however, this hypothesis must be approached with caution, as the time period also comprises a number of other potentially significant changes, such as season, increased social activity, and shifts from school to summer schedules in children. The peak in severity in May to June 2021 also coincides with a relaxation of universal masking recommendations by the Centers for Disease Control and Prevention and the issuance of the emergency use authorization for Pfizer/BioNTech mRNA vaccination of children aged 12 to 17 years. It is possible that these 2 events changed behavior and risk perception for some children who were more susceptible to severe disease.

Although this study provides valuable insight into COVID-19 behavior over time in a large population of children with demographic, geographic, and medical diversity, our findings must be interpreted with awareness of several limitations. First, our entry criteria included a positive SARS-CoV-2 test, which provides high confidence that infection occurred, but is subject to ascertainment bias. This is significant because the majority of test-positive patients were classified as asymptomatic and may have undergone testing because of exposure, or clearance for group activities or medical procedures. Furthermore, our testing data are from member health systems, and the availability of testing at retail sites, schools, and home test kits has shifted the pattern of test location over time. Although PEDSnet includes member institutions across the country, they are large pediatric medical centers, and their experiences may not be representative of other types of community or general hospitals.

Second, given test availability and testing practices, we expect that some tests were performed incidental to other medical care, which led us to exclude hospitalization as a severity criterion. It is therefore difficult in some cases to ascertain what diagnoses or interventions are primarily attributable to COVID-19, represent exacerbation of underlying morbidity by SARS-CoV-2 infection, or are essentially independent of the infection. We considered outcomes close in time to positive viral testing to be related to the infection. This may lead to overestimation of COVID-19 severity and of risk for severe disease associated with chronic medical conditions that may themselves require significant supportive care. Despite this, we observed low rates of severe illness. We would also not expect this effect to occur differentially across time periods, making it unlikely that it confounds our time trend analyses. As the duration of the pandemic increases, it may also be possible to address this problem through longitudinal study of within-patient effects.

Finally, attribution of trends we observe is further complicated by limitations in available data. Discrete EHR data typically reflect disorders to a greater extent than patients’ experiences, such as symptoms, which are underreported as encounter diagnoses; hence, our data underestimate the proportion of children with mild illness (ie, symptoms only). Practice patterns may also limit our ability to evaluate biologic hypotheses at the patient level. For example, national surveillance provides information on prevalence of other respiratory viruses, but patient-level testing to establish coinfection is limited and is unequally applied across severity levels, making it infeasible to assess risk because of coinfection. Our data also do not include important social determinants of health not linked to health care utilization, which may mask risk factors for both COVID-19 and comorbid conditions affected by disparities in access to care or other socioeconomic factors.

We have constructed and provided evidence for the validity of a novel EHR-based severity classification for COVID-19 in children. Applying this typology to children at 9 large health systems across the United States, we have shown that risk for more severe illness has not changed substantially as different strains of SARS-CoV-2 have become prevalent. Rather, risk factors are primarily presence of complex chronic conditions and infancy. These findings underscore the need to reduce case rates overall, highlighting the importance of timely testing and deployment of vaccines in children for pediatric benefit, as well as to reduce transmission to adults. Further research will also be needed to better understand risk factors (ie, both biologic and socioeconomic) and protective factors (eg, vaccines) for severe disease, and to better define the long-term burden of infection in children. Collaborative consortia such as PEDSnet provide opportunities to engage a large and diverse population of patients, clinical experts, and researchers to elucidate COVID-19 biology, improve clinical care, and better understand rare outcomes in children.

COVID-19

coronavirus disease 2019

ED

emergency department

EHR

electronic health record

PMCA

Pediatric Medical Complexity Algorithm

SARS-CoV-2

severe acute respiratory syndrome coronavirus 2