Inter-hospital Variation in COVID-19 Era Pediatric Hospitalizations by Age Group and Diagnosis

This multicenter, retrospective study used the Pediatric Health Information System (PHIS) database, which contains deidentified clinical and administrative information from 44 not-for-profit pediatric hospitals within the Children’s Hospital Association. Data quality and validity are ensured through a multistep, collaborative effort between the Children’s Hospital Association and participating institutions before their release. The use of deidentified PHIS data were considered nonhuman subjects research by the institutional review board of the Children’s Hospital of Philadelphia.

We analyzed inpatient hospitalizations within PHIS for patients ages 1 day to 17 years. Only admissions from home were evaluated; transfers between hospitals and birth hospitalizations were excluded. Our study evaluated hospitalizations from January 1, 2017 to June 30, 2021. Given a rapid increase in severe acute respiratory syndrome coronavirus 2 infections, a national emergency was declared on March 31, 2020.⁶  At time of analysis, data through June 2021 were available. Therefore, we defined the “prepandemic era” as January 1, 2017 to February 29, 2020 and the “COVID-19 era” as March 1, 2020 to June 30, 2021.

The primary outcome was hospital-specific change in monthly hospitalizations during the COVID-19 era, compared with analogous months prepandemic. Prepandemic hospitalizations were defined as average monthly counts from the 3 preceding years to account for annual variability in seasonal infections. We stratified by PHIS-defined age categories: <30 days (newborns), 30 to 364 days (infants), 1 to 4 years (young children), 5 to 12 years (older children), 13 to 17 years (adolescents). We subdivided newborns into 1 to 14 days and 15 to 29 days, as admissions of newborns <14 days from home were likely secondary to birth-related diagnoses and birth hospitalization management.

To collect hospitalization indication, we used Clinical Classifications Software Refined (CCSR) groups corresponding to principal diagnosis code. CCSR groupings aggregate related diagnosis codes into clinically meaningful categories and are common in PHIS studies.^2,13  Patients without CCSR data were excluded from secondary analysis of admission diagnoses. Given that some CCSR codes are highly related (eg, pneumonia and other lower respiratory disease) and may be prone to miscategorization, groupings of related CCSR codes were developed where applicable (Supplemental Table 1).

Patient, hospital, and community characteristics were available through PHIS. We included insurance type, age, Census region, freestanding children’s hospital status, average prepandemic-era admission counts, and city population. Policy scores were determined from a publicly available dataset of county-level COVID-19 policy interventions.⁷  Because PHIS hospitals may serve patients from multiple counties,¹⁴  mean policy scores were calculated by state.

Descriptive statistics were calculated using χ-square tests for categorical data and analyses of variance for continuous data. Given the large sample size, standardized differences were used to test for differences between eras, with 0.1 as the significance threshold.^15,16 

Percent change in hospitalizations was calculated from the monthly COVID-19 era hospitalization count compared with the month-matched prepandemic average. Binomial probability tests determined differences between eras,¹⁷  with 2-sided P values < .05 considered significant. The hospital-specific variation in magnitude of decline in hospitalizations during the pandemic compared with prepandemic was calculated using coefficients of variation (CV), a standardized measure of variability.¹⁸  Calculated by dividing the SD by the mean, larger CV values indicate greater variation. Spearman’s test was used to determine correlation. Statistical analyses were performed using Stata 16 (College Station, TX).

In the 1 624 755 hospitalizations analyzed (1 236 074 pre-COVID era and 388 681 COVID-19 era), overall monthly hospitalizations declined (−25.3%, P < .001) in the COVID-19 era, compared with prepandemic (Table 1). Hospitalizations were relatively stable for newborns (−8.0%; P = .007 and adolescents (−6.0%, P < .001). In the 1 623 244 (>99.9%) hospitalizations with CCSR codes, there was significant decline in hospitalizations with respiratory infections (−68.4%, P < .0001) and asthma (−63.5%, P < .001). Smaller but significant declines in hospitalizations were also noted for seizures and epilepsy, chemotherapy, and congenital cardiac anomalies (Table 1).

Among adolescents, COVID-19 era monthly hospitalizations for all mental health conditions increased 9.9%, from 1023 to 1124 (P = .002) In contrast, average monthly hospitalizations of adolescents for other conditions declined or were unchanged (Supplemental Table 2). Among newborns 1 to 14 days, hospitalizations for hyperbilirubinemia, feeding disorders, respiratory conditions, and other perinatal disorders did not change, though hospitalizations for perinatal infections declined (−23.2%, P = .02) (Supplemental Table 2).

Median center-specific pandemic-era decline in overall hospitalizations was −24.4% (interquartile range [IQR] −29.2% to −19.9%), with substantial variation across centers (CV 0.41) (Fig 1). Though the decline in respiratory conditions was large (−69.5% [IQR −74.2% to −63.7%]), variation between hospitals was small (CV 0.12). In contrast, variation in the magnitude of change in hospitalizations for mental health was larger (+9.3% [IQR −13.4% to 44.0%], CV 2.58). Twenty-four hospitals (54.5%) saw an increase in mental health-related hospitalization during the COVID-19 era.

By age group, 22.7% and 29.5% of hospitals had COVID-19 era increases in hospitalizations for adolescents and newborns 1 to 14 days, respectively, whereas all hospitals had declines in other age groups (Fig 2). Thus, hospital-specific variation in COVID-19 era hospitalizations was greatest for adolescents (CV 2.30) and newborns (CV 1.98), compared with other age groups (newborns 15–29 days [CV 0.36], infants [CV 0.32], young children [CV 0.29] and older children [CV 0.42]).

Variation in pandemic-era changes in hospitalizations was not explained by hospital factors, community factors, or state-level policy score (Supplemental Table 3). Centers from the same states differed from one another with respect to change in hospitalizations (data not shown to maintain institutional anonymity). COVID-19 era change in hospitalizations was most strongly correlated with change in respiratory infections (r = 0.69, P < .001), and to a lesser extent, mental health conditions (r = 0.32, P = .037). Adolescent hospitalizations were correlated with mental health conditions (r = 0.65, P < .001) and respiratory infections (0.46, P = .002).

Among over 1.6 million hospitalizations in 44 Children’s Hospitals across the United States, hospitalizations declined markedly during the COVID-19 era. Similar to prior studies, we found a decline in overall hospitalizations and hospitalizations for respiratory infections, signaling decreased circulation of respiratory pathogens which cause bronchiolitis, pneumonia, and asthma exacerbations.^1,3,9 

Our study highlights the variation in the magnitude of COVID-19 era decline in hospitalizations. Though some variation is explained by hospital-specific changes in admissions for respiratory infections, residual variation persisted for mental health conditions. Moreover, adolescents and newborns 1 to 14 days, who were least likely to be hospitalized for respiratory conditions, had the largest between-center variation in pandemic-era changes. These findings emphasize that pandemic policies and changes in hospital practices may have had spillover effects extending beyond mitigation of respiratory pathogen transmission.

Variation in COVID-19 era decline in hospitalizations was not explained by Census region, likely because these areas are large and heterogeneous. Furthermore, there were wide differences in admission hospital rates within the same state and no correlation with state policy score, suggesting that drivers of variation act on a local level. We found no correlation of hospital specific factors with declines in overall hospitalizations or changes by age group. Given data limitations, we were unable to investigate community factors (eg, scope or adherence to COVID-related policies) or hospital-specific practices (eg, admission thresholds) that could account for the variation across hospitals.

We found marked variation across hospitals in pandemic-era changes in hospitalizations for adolescents, which was partially explained by hospitalizations for mental health conditions. Although recent reports have shown an increase in hospitalizations for mental health concerns,^10–12,19  we found that these admissions declined in some centers, suggesting that this trend may not be uniform across the country. Community-specific investigation is needed to determine if this decline was because of local protective factors or barriers in accessing necessary care.

Newborns 1 to 14 days also had large hospital-specific variation in pandemic-era hospitalization. Shifts in diagnosis prevalence may occur differentially between hospitals, as prior single-center analyses have noted COVID-19 era fluctuations in infectious and noninfectious conditions.^20,21  However, our multicenter analysis found that hospitalizations with noninfectious conditions remained stable during the pandemic, making it difficult to attribute the increased hospitalizations at 29.5% of centers studied solely to pandemic-related changes in diagnosis frequency. Pandemic-related modifications to birth hospitalization practices also differ across the nation²²  but have not been associated with increased newborn readmissions.²³  Additional hypotheses for variation across hospitals, such as differential access to outpatient services²⁴  and changing admission thresholds for the youngest patients, are plausible.

This retrospective study is subject to diagnosis misclassification, though rigorous data quality monitoring in PHIS ensures minimal error-induced bias. We aggregated the policy score at the state level to reflect the multicounty referral bases typical of many PHIS hospitals; however, even this broad classification is subject to miscategorization bias, given that some PHIS hospitals border multiple states and have even larger, regional referral bases.²⁵  This cohort does not include small pediatric hospitals and community hospitals, which may have unique hospitalization patterns because of different COVID-19 caseloads or local resource allocation. Previous work suggests that pediatric hospitals (compared with community hospitals) had higher than expected mental health care use during the pandemic²⁶ ; the association with respiratory infections and infant hospitalizations has not been examined. However, given pediatric care regionalization,^27,28  PHIS captures a large and generalizable pediatric population, which can inform national priorities and efforts.

In this novel study of hospital-specific changes in hospitalizations during the COVID-19 pandemic, we found wide variation, especially among newborns and adolescents. Variation was not explained by hospital or regional factors, indicating that hospitalization patterns may be influenced by pandemic-related factors and practices at the community or institutional level. Though some variability is attributed to COVID-19 policies and strategies affecting respiratory infections, noninfectious conditions - particularly in the youngest and oldest children - may be uniquely sensitive to pandemic-era factors that differentially influence hospitalization patterns across communities. To prepare for future pandemics and viral surges, further work should incorporate local policies, mitigation strategies, and disease burden to elucidate drivers of geographic variation, understand their downstream effects, and identify populations most vulnerable to hospitalization.