Throughout the pandemic, children with COVID-19 have experienced hospitalization, ICU admission, invasive respiratory support, and death. Using a multisite, national dataset, we investigate risk factors associated with these outcomes in children with COVID-19.
Our data source (Optum deidentified COVID-19 Electronic Health Record Dataset) included children aged 0 to 18 years testing positive for COVID-19 between January 1, 2020, and January 20, 2022. Using ordinal logistic regression, we identified factors associated with an ordinal outcome scale: nonhospitalization, hospitalization, or a severe composite outcome (ICU, intensive respiratory support, death). To contrast hospitalization for COVID-19 and incidental positivity on hospitalization, we secondarily identified patient factors associated with hospitalizations with a primary diagnosis of COVID-19.
In 165 437 children with COVID-19, 3087 (1.8%) were hospitalized without complication, 2954 (1.8%) experienced ICU admission and/or intensive respiratory support, and 31 (0.02%) died. We grouped patients by age: 0 to 4 years old (35 088), and 5 to 11 years old (75 574), 12 to 18 years old (54 775). Factors positively associated with worse outcomes were preexisting comorbidities and residency in the Southern United States. In 0- to 4-year-old children, there was a nonlinear association between age and worse outcomes, with worse outcomes in 0- to 2-year-old children. In 5- to 18-year-old patients, vaccination was protective. Findings were similar in our secondary analysis of hospitalizations with a primary diagnosis of COVID-19, though region effects were no longer observed.
Among children with COVID-19, preexisting comorbidities and residency in the Southern United States were positively associated with worse outcomes, whereas vaccination was negatively associated. Our study population was highly insured; future studies should evaluate underinsured populations to confirm generalizability.
As of November 2022, more than 15 million children have been infected with COVID-19, representing 18.3% of US cases.1 Early in the pandemic, information on pediatric clinical characteristics and outcomes were limited, possibly because of infrequent testing and interventions (eg, school closures). As variants emerged, reports of children with severe COVID-19 increased, accompanied by a rise in multisystem inflammatory syndrome in children.2,3
Early data suggested that children with chronic medical conditions suffered worse outcomes.4–13 Yet hospitalization, ICU admission, intensive respiratory support, and death rates were difficult to assess because of inconsistent illness severity and outcome definitions.1,14 Small sample sizes and minimal geographic diversity limited generalizability of clinical and demographic factors associated with poor outcomes in children with COVID-19.4–13 Additionally, COVID-19 vaccination rates in children vary by geography and age, necessitating exploring the impact of vaccination across ages.15–24
Harnessing a large, multisite, national, US-based electronic health record (EHR)-based dataset, we evaluated which clinical and demographic factors (including variant and vaccination) are associated with severe outcomes in children with COVID-19.
Methods
Study Design
In this observational, retrospective study using the Optum COVID-19 EHR dataset (Optum Inc., Eden Prairie, Minnesota), we evaluated the probability of severe outcomes within 90 days of diagnosis as a function of clinical and demographic factors in children with COVID-19. Ninety days was chosen to be inclusive given the rarity of severe outcomes in children with COVID-19. The primary outcome was all-cause, ordinal severity with 3 categories: no hospitalization; hospitalization without intensive care (hospitalization henceforth); or a composite including ICU admission, intensive respiratory support, or death (critical illness or death). These categories were chosen because of insufficient data on items such as hypoxia and radiography in our dataset, and because of rarity of death.
Study Population
We included children aged 0 to 18 years testing positive for COVID-19 on antigen or polymerase chain reaction test between January 1, 2020, and January 20, 2022. We also identified children hospitalized with a primary diagnosis of COVID-19 to compare children admitted “for COVID” to children “with COVID.” We divided our cohort into 0- to 4-, 5- to 11-, and 12- to 18-year-old groups because clinical differences, vaccine availability differences during the study period, and consistency with previous literature on pediatric COVID-19.3,25,26
Data Source
Optum’s deidentified COVID-19 EHR dataset is a national low-latency pipeline leveraging data from inpatient and ambulatory EHRs, practice management systems, and other internal systems.27 The dataset captures approximately 4.5 million adults and children tested for COVID-19 older than age 2 years, derived from more than 700 hospitals and 7000 clinics in the United States. Unique identifiers link encounters to patients. An independent statistical expert certified the data’s deidentification. This release includes patients with a diagnosis of COVID-19 or acute respiratory illness after January 1, 2020, and/or COVID-19 testing (positive or negative result) between January 1, 2020, and January 20, 2022.
Extracted Data Elements
For each patient, we used the first positive test for COVID-19 to define the diagnosis date and extracted diagnosis International Classification of Diseases-10 codes, severity outcomes, clinical characteristics, and demographic factors.
Hospitalization and ICU admission were identified using encounter location. Intensive respiratory support (high-flow nasal cannula, nonintensive positive pressure ventilation, and mechanical ventilation) was identified using Form UB-04 Revenue Codes (31500, 94002, 31502, 94660, 94640, 31605, 31603) or current procedure terminology/health care common procedure coding system codes (A4615, A4620, A4619, E0562). Death was identified by death date. Out-of-health-system death information (eg, death certificates) was not available.
Optum deidentification removed birth month and date; thus, age was estimated as year of first COVID-19 diagnosis minus birth year. Sex options were male or female. Race and ethnicity were included not as biologic differences but to evaluate effects of social constructs often associated with poorer access to health care on outcomes. Options were race (African American, Asian, Caucasian, and Other/Unknown) and ethnicity (Hispanic, Not Hispanic, and Unknown). We created 1 variable for race and ethnicity: patients identified as Hispanic were classified as such or classified by their reported race.28 Insurance options were commercial, public (Medicaid, Medicare, or Medicare Advantage), other, uninsured, and unknown. If multiple insurances were documented at COVID-19 diagnosis, the option with greater coverage was hierarchically selected (eg, for a patient documented with Medicaid and uninsured, Medicaid was selected). Hospital/clinic region (region henceforth) was defined using US census (Midwest, Northeast, South, West, and other/unknown).
Body mass index (BMI) was reported from the first COVID-19 diagnosis or the most recent BMI recorded within 6 months of the first COVID-19 diagnosis. BMI was excluded for 0- to 4-year-old children because of Centers for Disease Control and Prevention recommendations against using BMI-for-age in children younger than aged 2 years and high missingness in this group.29 Comorbidity data documented before COVID-19 diagnosis were aggregated using a validated index predicting hospitalization risk, referred to as pediatric comorbidity index (PCI) score.30 PCI score is a continuous variable representing increased hospitalization risk in children using International Classification of Diseases-10 codes.
Vaccination categories were: at least 1 vaccine received (documented dose of messenger RNA vaccine before COVID-19 diagnosis) or no known vaccination because of rarity of vaccination in this cohort and period. Vaccinations were extracted from vaccination encounter or self-report. No external data sources on vaccination were available. Predominant variant at diagnosis was determined using public data by region and date from outbreak information.31,32
Statistical Analysis
We identified missingness in sex (0.1%), insurance type (4.7%), and BMI (17.9%). We performed random forest imputation using the missRanger R package with 16 candidate nonmissing values and noncandidate variables to sample from in predictive mean matching steps and 100 trees to account for missingness while reducing bias and increasing precision.33–36
We described severity outcomes, clinical factors, and demographic factors in the study population. To explore associations with severe outcomes, we fit multivariable ordinal logistic regressions using proportional odds for each age group. Models for 5- to 11- and 12- to 18-year-old patients evaluate association of outcomes with age, gender, BMI, PCI score, vaccination status, predominant variant, race/ethnicity, region, and insurance status. The 0- to 4-year-old model uses these variables except BMI and vaccination status because of insufficiently granular age data and unavailability of COVID-19 vaccines for this age group during the study period, respectively. We fit models including interaction terms between region and insurance and region and race/ethnicity to adjust for confounding between race, ethnicity, and insurance status with region. Continuous variables (age, PCI score, and BMI) were modeled using restricted cubic splines with 3 knots for nonlinear relationships.37 All analysis was performed using R version 4.1.2 (R Core Team, Vienna, Austria) and rms R package, α = 0.05.37,38 Model significance (compared with null model and between models) was assessed using likelihood ratio χ2 tests. Discrimination significance was measured by concordance indices. We report odds ratios (ORs) and 95% confidence intervals (95% CI) for each factor. For factors modeled by restricted cubic splines, ORs compare 25th and 75th percentiles (interquartile range).
A secondary analysis was done using only severity outcomes (hospitalization, ICU admission, intensive respiratory support, or death) in which admissions had a primary encounter diagnosis of COVID-19 to attempt to account for differences between admissions “for COVID” and admissions “with COVID” detected on routine screening.
Results
In our dataset, 165 437 children tested positive for COVID-19 through January 20, 2022. Most children were 12 to 18 years old (45.7%) followed by 5 to 11 years old (33.1%) and 0 to 4 years old (21.2%) (Fig 1).
We report outcomes, clinical, and demographic factors in Table 1. Our cohort was primarily Caucasian (49.7%), followed by Hispanic (18.1%), African American (9.2%), and Asian (1.8%). Sex was split between male (50.2%) and female (49.7%). Children were primarily in the Midwest (32.8%), followed by Northeast (22.7%), South (21.9%), and West (6.5%). Most children (82.5%) had commercial insurance, followed by public insurance (10.3%).
. | 0–4 Year Olds (N = 35 088) . | 5–11 Year Olds (N = 75 574) . | 12–18 Year Olds (N = 54 775) . | Overall (N = 16 5437) . |
---|---|---|---|---|
Race/ethnicity, n (%) | ||||
African American | 3090 (8.8) | 5107 (9.3) | 6974 (9.2) | 15171 (9.2) |
Asian | 735 (2.1) | 1105 (2.0) | 1139 (1.5) | 2979 (1.8) |
Caucasian | 13 608 (38.8) | 27 094 (49.5) | 41 533 (55.0) | 82 235 (49.7) |
Hispanic | 8353 (23.8) | 9552 (17.4) | 12 061 (16.0) | 29 966 (18.1) |
Other/unknown | 9302 (26.5) | 11917 (21.8) | 13 867 (18.3) | 35 086 (21.2) |
Sex, n (%) | ||||
Female | 16 185 (46.1) | 26 704 (48.8) | 39 252 (51.9) | 82 141 (49.7) |
Male | 18 814 (53.6) | 27 996 (51.1) | 36 246 (48.0) | 83 056 (50.2) |
Missing | 89 (0.3) | 75 (0.1) | 76 (0.1) | 240 (0.1) |
Insurance category, n (%) | ||||
Commercial | 27 501 (78.4) | 45 476 (83.0) | 63 570 (84.1) | 136 547 (82.5) |
Other | 679 (1.9) | 1111 (2.0) | 1461 (1.9) | 3251 (2.0) |
Public | 4511 (12.9) | 5342 (9.8) | 7141 (9.4) | 16994 (10.3) |
Uninsured | 121 (0.3) | 259 (0.5) | 526 (0.7) | 906 (0.5) |
Missing | 2276 (6.5) | 2587 (4.7) | 2876 (3.8) | 7739 (4.7) |
Region, n (%) | ||||
Midwest | 8693 (24.8) | 16 971 (31.0) | 28 673 (37.9) | 54 337 (32.8) |
Northeast | 8244 (23.5) | 13 791 (25.2) | 15 575 (20.6) | 37 610 (22.7) |
South | 9255 (26.4) | 10 829 (19.8) | 16 151 (21.4) | 36 235 (21.9) |
West | 1947 (5.5) | 3912 (7.1) | 4950 (6.5) | 10 809 (6.5) |
Other/unknown | 6949 (19.8) | 9272 (16.9) | 10 225 (13.5) | 26 446 (16.0) |
Predominant variant, n (%) | ||||
Original | 8042 (22.9) | 10 012 (18.3) | 17 390 (23.0) | 35 444 (21.4) |
Alpha | 8585 (24.5) | 11 485 (21.0) | 19 922 (26.4) | 39 992 (24.2) |
Delta | 6705 (19.1) | 15 510 (28.3) | 17 083 (22.6) | 39 298 (23.8) |
Omicron | 5067 (14.4) | 9002 (16.4) | 11 826 (15.6) | 25 895 (15.7) |
Unknown/no predominant variant | 6689 (19.1) | 8766 (16.0) | 9353 (12.4) | 24 808 (15.0) |
BMI | ||||
Mean (SD) | 18.1 (5.56) | 19.5 (5.48) | 24.7 (6.77) | 21.7 (6.79) |
Median (IQR) | 17.0 (10.1–54.9) | 17.8 (10.1–54.9) | 23.0 (10.3–54.9) | 19.9 (10.1–54.9) |
Missing, n (%) | 9934 (28.3) | 9326 (17.0) | 10 422 (13.8) | 29 682 (17.9) |
PCI Score | ||||
Mean (SD) | 2.00 (2.60) | 2.28 (2.81) | 3.08 (3.71) | 2.59 (3.25) |
Median (IQR) | 1.00 (0.0–27.0) | 1.00 (0.0–29.0) | 2.00 (0.0–31.0) | 2.00 (0.0–31.0) |
Vaccination status, n (%) | ||||
No known vaccination | 35 088 (100) | 51 405 (93.8) | 63 330 (83.8) | 149 823 (90.6) |
At least 1 vaccine received | 0 (0) | 3370 (6.2) | 12 244 (16.2) | 15 614 (9.4) |
Outcome, n (%) | ||||
All other outcomes | 32 853 (93.6) | 53 463 (97.6) | 73 049 (96.7) | 159 365 (96.3) |
Hospitalization without critical care | 854 (2.4) | 583 (1.1) | 1650 (2.2) | 3087 (1.9) |
Hospitalization with critical care | 1365 (3.9) | 726 (1.3) | 863 (1.1) | 2954 (1.8) |
Deceased | 16 (0.05) | 3 (0.005) | 12 (0.02) | 31 (0.02) |
. | 0–4 Year Olds (N = 35 088) . | 5–11 Year Olds (N = 75 574) . | 12–18 Year Olds (N = 54 775) . | Overall (N = 16 5437) . |
---|---|---|---|---|
Race/ethnicity, n (%) | ||||
African American | 3090 (8.8) | 5107 (9.3) | 6974 (9.2) | 15171 (9.2) |
Asian | 735 (2.1) | 1105 (2.0) | 1139 (1.5) | 2979 (1.8) |
Caucasian | 13 608 (38.8) | 27 094 (49.5) | 41 533 (55.0) | 82 235 (49.7) |
Hispanic | 8353 (23.8) | 9552 (17.4) | 12 061 (16.0) | 29 966 (18.1) |
Other/unknown | 9302 (26.5) | 11917 (21.8) | 13 867 (18.3) | 35 086 (21.2) |
Sex, n (%) | ||||
Female | 16 185 (46.1) | 26 704 (48.8) | 39 252 (51.9) | 82 141 (49.7) |
Male | 18 814 (53.6) | 27 996 (51.1) | 36 246 (48.0) | 83 056 (50.2) |
Missing | 89 (0.3) | 75 (0.1) | 76 (0.1) | 240 (0.1) |
Insurance category, n (%) | ||||
Commercial | 27 501 (78.4) | 45 476 (83.0) | 63 570 (84.1) | 136 547 (82.5) |
Other | 679 (1.9) | 1111 (2.0) | 1461 (1.9) | 3251 (2.0) |
Public | 4511 (12.9) | 5342 (9.8) | 7141 (9.4) | 16994 (10.3) |
Uninsured | 121 (0.3) | 259 (0.5) | 526 (0.7) | 906 (0.5) |
Missing | 2276 (6.5) | 2587 (4.7) | 2876 (3.8) | 7739 (4.7) |
Region, n (%) | ||||
Midwest | 8693 (24.8) | 16 971 (31.0) | 28 673 (37.9) | 54 337 (32.8) |
Northeast | 8244 (23.5) | 13 791 (25.2) | 15 575 (20.6) | 37 610 (22.7) |
South | 9255 (26.4) | 10 829 (19.8) | 16 151 (21.4) | 36 235 (21.9) |
West | 1947 (5.5) | 3912 (7.1) | 4950 (6.5) | 10 809 (6.5) |
Other/unknown | 6949 (19.8) | 9272 (16.9) | 10 225 (13.5) | 26 446 (16.0) |
Predominant variant, n (%) | ||||
Original | 8042 (22.9) | 10 012 (18.3) | 17 390 (23.0) | 35 444 (21.4) |
Alpha | 8585 (24.5) | 11 485 (21.0) | 19 922 (26.4) | 39 992 (24.2) |
Delta | 6705 (19.1) | 15 510 (28.3) | 17 083 (22.6) | 39 298 (23.8) |
Omicron | 5067 (14.4) | 9002 (16.4) | 11 826 (15.6) | 25 895 (15.7) |
Unknown/no predominant variant | 6689 (19.1) | 8766 (16.0) | 9353 (12.4) | 24 808 (15.0) |
BMI | ||||
Mean (SD) | 18.1 (5.56) | 19.5 (5.48) | 24.7 (6.77) | 21.7 (6.79) |
Median (IQR) | 17.0 (10.1–54.9) | 17.8 (10.1–54.9) | 23.0 (10.3–54.9) | 19.9 (10.1–54.9) |
Missing, n (%) | 9934 (28.3) | 9326 (17.0) | 10 422 (13.8) | 29 682 (17.9) |
PCI Score | ||||
Mean (SD) | 2.00 (2.60) | 2.28 (2.81) | 3.08 (3.71) | 2.59 (3.25) |
Median (IQR) | 1.00 (0.0–27.0) | 1.00 (0.0–29.0) | 2.00 (0.0–31.0) | 2.00 (0.0–31.0) |
Vaccination status, n (%) | ||||
No known vaccination | 35 088 (100) | 51 405 (93.8) | 63 330 (83.8) | 149 823 (90.6) |
At least 1 vaccine received | 0 (0) | 3370 (6.2) | 12 244 (16.2) | 15 614 (9.4) |
Outcome, n (%) | ||||
All other outcomes | 32 853 (93.6) | 53 463 (97.6) | 73 049 (96.7) | 159 365 (96.3) |
Hospitalization without critical care | 854 (2.4) | 583 (1.1) | 1650 (2.2) | 3087 (1.9) |
Hospitalization with critical care | 1365 (3.9) | 726 (1.3) | 863 (1.1) | 2954 (1.8) |
Deceased | 16 (0.05) | 3 (0.005) | 12 (0.02) | 31 (0.02) |
Public insurance category includes Medicaid, Medicare, and Medicare Advantage. Critical care represents ICU admission or use of intensive respiratory support, defined as high-flow nasal cannula, nonintensive positive pressure ventilation, and mechanical ventilation. IQR, interquartile range.
Within 90 days of diagnosis with COVID-19, 2985 (1.8%) children were hospitalized with ICU admission, intensive respiratory support, or death; 3087 (1.9%) children were hospitalized without critical care; and 159 635 (96.3%) children were not hospitalized. The 0- to 4-year-old children experienced the highest incidence of hospitalization with critical care of all age groups. A total of 1.9% of unvaccinated children experienced the composite outcome of hospitalization with ICU admission (1.0%), intensive respiratory support (0.98%), or death (0.02%). No child receiving at least 1 vaccine dose died within 90 days of diagnosis.
Overall model likelihood ratio χ2 tests were statistically significant (P < .001) in all models (Supplemental Table 2). Overall model concordance indices were 0.807 for 0- to 4-, 0.881 for 5- to 11-, and 0.876 for 12- to 18-year-old patients.
Figure 2 depicts ORs and 95% CIs for the multivariable ordinal regression model between factors and severe outcomes in 0- to 4-year-old children with COVID-19. Independent factors most associated with ICU admission, intensive respiratory support, or death were residency in the Southern United States compared with the Midwest (OR, 3.61; 95% CI, 3.15–4.15), PCI score (OR, 2.59; 95% CI, 2.28–2.93), and African American compared with Caucasian race (OR, 1.44; 95% CI, 1.23–1.68). Residency in the Northeast United States compared with the Midwest (OR, 1.3; 95% CI, 1.09–1.54) and Hispanic ethnicity (OR, 1.16; 95% CI, 1.03–1.29) were positively associated with severe outcomes. Compared with the original strain, the Alpha variant was more associated with severe outcomes (OR, 1.25; 95% CI, 1.13–1.39). The Delta (OR, 0.47; 95% CI, 0.4–0.56) and Omicron (OR, 0.26; 95% CI, 0.2–0.34) variants were negatively associated with severe outcomes. Female sex (OR, 0.82; 95% CI, 0.75–0.9) and 1-year-old age compared with 3-year-old age (OR, 0.58; 95% CI, 0.54–0.63) were negatively associated with severe outcomes. We produced a partial-effects plot to predict age effects holding other variables constant, showing highest relative risk in infants with a nonlinear decrease as age approaches 2 years (Supplemental Fig 5).
Figure 3 depicts ORs and 95% CIs for the multivariable ordinal regression model between factors and severe outcomes in 5- to 11-year-old patients with COVID-19. Independent factors most associated with ICU admission, intensive respiratory support, or death were residency in the Southern United States compared with the Midwest (OR, 7.16; 95% CI, 5.94–8.63), PCI score (OR, 4.58; 95% CI, 3.78–5.55), and African American compared with Caucasian race (OR, 1.85; 95% CI, 1.52–2.24). Asian compared with Caucasian race (OR, 1.69; 95% CI, 1.18–2.42), residency in the Northeast United States compared with the Midwest (OR, 1.67; 95% CI, 1.32–2.11) and Hispanic ethnicity (OR, 1.65; 95% CI, 1.43–1.9) was positively associated with severe outcomes. Female sex (OR, 0.83; 95% CI, 0.74–0.94), 10-year-old age compared with 6-year-old age (OR, 0.78; 95% CI, 0.69–0.88), and receiving at least 1 vaccine dose (OR, 0.48; 95% CI, 0.33–0.71) were negatively associated with severe outcomes. Compared with the original strain, Delta (OR, 0.26; 95% CI, 0.21–0.32) and Omicron (OR, 0.11; 95% CI, 0.08–0.16) variants were negatively associated with severe outcomes.
Figure 4 depicts ORs and 95% CIs for the multivariable ordinal regression model between factors and severe outcomes in 12- to 18-year-old patients with COVID-19. Independent factors most associated with ICU admission, intensive respiratory support, or death were PCI score (OR, 6.67; 95% CI, 5.72–7.78), residency in the Southern United States compared with the Midwest (OR, 5.8; 95% CI, 5.17–6.52), and Asian compared with Caucasian race (OR, 1.97; 95% CI, 1.45–2.69). Hispanic ethnicity (OR, 1.93; 95% CI, 1.74–2.14), African American compared with Caucasian race (OR, 1.76; 95% CI, 1.53–2.03), residency in the West United States compared with the Midwest (OR, 1.58; 95% CI, 1.23–2.03), and residency in the Northeast United States compared with the Midwest (OR, 1.31; 95% CI, 1.1–1.56) were positively associated with severe outcomes. Compared with the original strain, Delta (OR, 0.27; 95% CI, 0.23–0.31) and Omicron (OR, 0.08; 95% CI, 0.06–0.11) variants were negatively associated with severe outcomes. Receiving at least 1 vaccine dose was negatively associated with severe outcomes (OR, 0.59; 95% CI, 0.51–0.68).
Models with linear interactions between region and insurance type and region and race/ethnicity were significant and had similar discrimination per C indices (Supplemental Table 3). Although these factors do interact, the models demonstrate independently significant region, insurance type, and race/ethnicity effects after accounting for interactions.
Supplemental Figs 6, 7, and 8 depict ORs and 95% CIs for the secondary analysis of multivariable ordinal regressions between risk factors and hospitalizations with a primary encounter diagnosis of COVID-19 in 0- to 4-, 5- to 11-, and 12- to 18-year-old patients, respectively. We report outcomes, clinical, and demographic factors for this analysis in Supplemental Table 4. Limiting criteria for hospitalization, ICU admission, intensive respiratory support, or death to only such outcomes with a primary encounter diagnosis of COVID-19 reduced the number of children experiencing these outcomes from 6072 in our initial analysis to 2232 in the secondary analysis. Most factors associated with ICU admission, intensive respiratory support, and death are shared between children hospitalized with COVID-19 and those hospitalized for COVID-19 as the primary encounter diagnosis; however, some are not. In 0- to 4-year-old patients, Hispanic ethnicity, Alpha variant, and were no longer associated with severe outcomes. Residence in the US South was negatively associated with severe outcomes. In 5- to 11-year-old patients, Asian race, residence in the US South, and residence in the US West were no longer associated with severe outcomes. In 12- to 18-year-old patients, residence in the US South and residence in the US West were no longer associated with severe outcomes.
Discussion
In this nationwide cohort of 165 437 children with COVID-19, several variables were positively associated with severe outcomes. Across all ages, residency in the US South, pre–COVID-19 comorbidities, African American race, Hispanic ethnicity, and predominant variant at test positivity were positively associated with severe outcomes. Infancy and male sex were positively associated with severe outcomes in the 0- to 4-year-old group. Asian race and lack of vaccination were positively associated with severe outcomes in the 5- to 18-year-old group.
Hospitalization and mortality rates in our dataset were consistent with American Academy of Pediatrics–reported rates in US children, and our analysis substantiates previous findings that infancy, non-Caucasian race, Hispanic ethnicity, and preexisting comorbidities are associated with severe outcomes in COVID-19.1,4–10 Proportionally more children in our 0- to 4-year-old cohort experienced severe outcomes compared with the older age groups. The greatest risk for severe outcomes was in those aged younger than 2 years old, contrasting previous observations that young infants were protected from severe COVID-19.39,40 The Delta and Omicron variants were negatively associated with ICU admission, intensive respiratory support, and death compared with the original strain in our cohort, conflicting with studies showing increased hospitalizations and severe infections resulting from the Omicron and Delta variants.2,3 This may reflect our highly insured cohort, resulting in increased health care access that may have reduced severe outcomes as the pandemic progressed and more effective treatments were identified.
Residency in the Southern United States was a significant, independent factor positively associated with severe outcomes, even after accounting for interactions with insurance status and race/ethnicity, aligning with another published regional analysis.41 This may be explained by poverty rates and reduced access to social programs and health care, but the reason for this disparity is unknown.41,42 Our results confirm that vaccination is negatively associated with severe outcomes in 5- to 18-year-old patients, supporting increased vaccination rates in children, particularly in vulnerable regions and populations.
Strengths and Limitations
This study has several strengths. First, longitudinal data available throughout the pandemic allowed for assessment through shifts in variant predominance, infection rates, and vaccine eligibility. Second, presence of many variables from the EHR data allowed inclusion of a wide range of clinical and demographic patient characteristics. Third, the dataset produced sufficient sample sizes to include uncommon severe outcomes in children with COVID-19, allowing for many clinical variables without exceeding allowable statistical degrees of freedom, and to describe incidence of COVID-19 in partially or fully vaccinated children.
Like any observational study using a large EHR dataset, our study suffers limitations. First, loss of birth date and month allowed only for age estimation. Though the estimation was consistent, this limits prediction in individual patients. However, our findings of a nonlinearly increased severe outcome rate in youngest patients supports clinical practice.
Second, our dataset period limits understanding vaccination effects. Vaccines were available to children during a short period in our dataset (from December 2020 for 16- to 18-year-old patients, May 2021 for 12- to 15-year-old patients, and November 2021 for 5- to 11-year-old patients). Despite this, vaccination was independently associated with less severe outcomes. Our study cutoff occurred before Omicron BA.4, BA.5, and other subvariants, and only contained vaccinations reported from a clinical encounter or self-report, potentially leading to underreporting. However, we believe our findings bolster vaccine outreach, especially among the most vulnerable.
Third, some variables were coded as “other/unknown.” We treated “other/unknown” as a separate category rather than missing to account for potential bias in these categories. Fourth, few children experienced severe outcomes in our cohort. Though fortunate, this limited our ability to evaluate severe outcomes individually. Fifth, vaccine types, polymerase chain reaction test names, and procedure codes were documented differently between sites. Computational and manual methods were used to include these, but some information could have been missed.
Sixth, outcomes were limited by their availability in the dataset. Patients hospitalized outside the Optum-associated network were not recorded. Because the database is amalgamated by an analytics company associated with a national insurer, only 10% of our cohort had public insurance, which is well below the reported 37.5% in US children.43 Thus, our results may not be generalizable among patients with low access to care and emerging treatments. Future studies using Medicaid data or multisite data from public hospitals may address this gap. Seventh, clinical site and EHR system heterogeneity in our study prevented use of standard-of-care risk scores (eg, Pediatric Risk of Mortality, Pediatric Logistic Organ Dysfunction, Pediatric Early Warning Signs) to confirm infection severity.
Finally, some patients in our study were likely admitted “with COVID” identified on routine screening on admission for reasons other than “for COVID,” leading to inclusion of patients whose outcomes may not have been related to COVID-19. Our secondary analysis of patients with a primary encounter diagnosis of COVID-19 attempts to account for this. However, we acknowledge the limitations of EHR-designated primary encounter diagnosis to represent the true primary clinical diagnosis given the complexity and variability in primary diagnosis documentation.
Conclusions
Our study demonstrates population level associations between demographic and clinical factors and severe outcomes in pediatric COVID-19. Our findings demonstrate the protective effect of vaccines, increased association with severe outcomes among infants, and disparities related to race, ethnicity, insurance, and geography. We identified populations in need of outreach for vaccination and other preventive efforts.
Ms Ho and Dr Turer conceptualized and designed the study, conducted the analyses, drafted the initial manuscript, and reviewed and revised the manuscript; Drs Most, Perl, Radunsky, Hanna, and Thakur conceptualized and designed the study and reviewed and revised the manuscript critically for important intellectual content; Mr Diaz, Ms Casazza, Dr Saleh, and Ms Pickering designed and conducted analyses and reviewed and revised the manuscript critically for important intellectual content; Drs Medford and Lehmann contributed substantially to conception and design of the study, acquisition of data, supervised the analyses, and reviewed and revised the manuscript critically for important intellectual content; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
FUNDING: This work was supported by the UT Southwestern Medical Center’s Clinical Informatics Center. The funder/sponsor did not participate in the work.
CONFLICT OF INTEREST DISCLOSURES: Dr Medford is funded through the Texas Health Resources Clinical Scholar Program, the Centers for Disease Control and Prevention (grant U01CK000590) and has received research funding through Verily Life Sciences and the Sergey Brin Family Foundation. All other authors have indicated they have no potential conflicts of interest to disclose.
Comments