“For COVID” or “With COVID”: Classification of SARS-CoV-2 Hospitalizations in Children

This retrospective cohort study included patients <18 years of age with SARS-CoV-2 infection hospitalized between May 10, 2020, (when universal screening of all admissions began) and February 10, 2021, at a university-based, quaternary care children’s hospital in Northern California. Our aim was to define a cohort that most represented reporting of COVID-19 hospitalizations to the Santa Clara County Department of Public Health (DPH). The patients in our cohort were obtained from an organizational dashboard that captured hospitalized patients placed in isolation for SARS-CoV-2. Patients with multisystem inflammatory syndrome in children (MIS-C) that would not have been placed in isolation were obtained from the infection prevention and control department, which reports all MIS-C cases and all patients who test positive for SARS-CoV-2 on admission to the children’s hospital to the county DPH.

Infection was established by detection of SARS-CoV-2 RNA via real-time reverse transcriptase polymerase chain reaction. Patients who tested positive in the emergency department or at an outside hospital previous to admission or transfer were reported to the DPH by the testing facility per mandatory reporting requirements. Readmissions within 90 days were not reported, even if tested positive because this is within the known persistent viral shedding period.⁷  If patients tested positive >90 days from their original positive result at the time of admission, they were reported as having a new positive result.

We used 2 approaches for classification of SARS-CoV-2 hospitalizations. In the first, we used a previously published categorization system and classified patients as either asymptomatic, mild or moderate, severe, critical, or the Centers for Disease Control and Prevention (CDC) definition of MIS-C.^8,9  Patients were considered asymptomatic if they did not have symptoms described by the CDC to be consistent with COVID-19.¹⁰  Mild or moderate severity applied to patients who had symptoms attributable to COVID-19 but did not require supplemental oxygen. Patients requiring new or increased oxygen support from the baseline but not new positive-pressure ventilation were classified as severe. Patients who received invasive or noninvasive mechanical ventilation or with sepsis or multiorgan failure were classified as critical.⁹  Initiation or uptitration of high-flow nasal cannula (HFNC) requires intensive care admission at our institution and was considered noninvasive positive-pressure ventilation, so patients requiring HFNC were classified as having critical illness. We also recorded COVID-19 specific treatments received, as an additional way to capture the severity of illness.

Because the above severity classification approach does not indicate whether SARS-CoV-2 infection caused the hospitalization, we also categorized the hospitalization as likely or unlikely to have been caused by SARS-CoV-2. For this categorization, we reviewed admission and discharge medical documentation and attempted to discern whether the hospitalization would have occurred in the absence of SARS-CoV-2 infection (Supplemental Figure 1). Patients initially admitted for non–COVID-19 reasons who developed new or worsening COVID-19 symptoms that would have ultimately driven hospital admission were labeled as likely. If there was an uncertain link between SARS-CoV-2 and cause of admission (eg, brief resolved unexplained events (BRUEs), increased seizure frequency), we erred toward an attribution of likely, especially if no alternative explanation existed. For patients whose hospitalizations were unlikely to be caused by SARS-CoV-2, the severity of illness was determined by symptoms attributable to COVID-19, not to an alternative illness. For example, a patient with a severe traumatic brain injury who underwent mechanical ventilation for airway protection would not be labeled as critical unless there was also lung disease consistent with COVID-19.

Given multiple reports of disparities in COVID-19, we opted to collect information on race and ethnicity, as documented in the electronic medical record. Immunocompromised was defined as currently receiving chemotherapy, having an oncologic diagnosis, or having a history of a solid organ or hematopoietic stem cell transplant. Other causes for immune compromise were not included.

The first 25 charts were first reviewed independently by all 4 investigators, then as a group to solidify our methodology for the likely versus unlikely designation for cause of admission. These 25 subjects were excluded from the κ calculation. The remainder of the charts were then reviewed independently by 2 investigators. We used a κ statistic and a weighted κ to calculate interrater agreement for the reason for admission (likely versus unlikely categorization) and severity of illness, respectively. Disagreements were resolved by consensus after discussion among all 4 investigators. Descriptive statistics were calculated and reported as medians with interquartile ranges and proportions, as appropriate. The investigation was approved by the institutional review board.

The characteristics of our 117-patient cohort are depicted in Table 1. Most (83; 70.9%) identified as Latino. A total of 19 (16.2%) were immunocompromised, and 31 (26.5%) required ICU admission attributable to COVID-19 or MIS-C. None died during the study period, although 1 immunocompromised patient died shortly after study conclusion from respiratory complications of COVID-19.

For classification of the severity of illness, there was 100% agreement between reviewers for MIS-C designation (yes or no). For the remaining 4 categories, there was 93.3% agreement between reviewers (κ 0.82). For determination of COVID-19 as the likely cause of admission, there was 89% agreement between reviewers (κ = 0.78). Examples of disagreements included patients with new or worsening seizures, patients with BRUEs, and a patient who had fever and diarrhea with a stool polymerase chain reaction positive for Salmonella.

The tabulation of illness severity and likelihood of SARS-CoV-2 causing the hospitalization is depicted in Table 2; 46 (39.3%) were asymptomatic, 33 (28.2%) had mild or moderate disease, 9 (7.7%) had severe illness, and 15 (12.8%) had critical illness. Of the 15 with critical illness, the maximum respiratory support was invasive ventilation for 5, positive-pressure ventilation for 4, and HFNC for 3. None received vasopressor support. A total of 14 (12%) patients had MIS-C; 1 underwent invasive ventilation, and 5 received vasopressor support. Three of the patients recovered with no specific treatments for MIS-C. Most patients received intravenous immunoglobulin alone or with methylprednisolone or aspirin. One patient received anakinra.

SARS-CoV-2 was likely to be the cause of admission in 64 (54.7%; 95% confidence interval 45.2%–63.9%) patients and unlikely in 53 (45.3%; 95% CI 36.1%–54.8%) patients. Common alternative diagnoses in hospitalizations deemed unlikely to be caused by SARS-CoV-2 are summarized in Table 3.

Our findings reveal that most children hospitalized with SARS-CoV-2 have asymptomatic or mild or moderate disease, and nearly one-half of these hospitalizations were not caused by infection from the virus itself. Additionally, we demonstrate that interrater reliability for an existing and novel pediatric COVID-19 hospitalization classification system is adequate but not perfect. Part of the challenge in attributing patient symptoms to SARS-CoV-2 is that most symptoms are nonspecific and common features of other disease processes. Additionally, we are still learning about clinical manifestations of infection. We may ultimately learn, for example, that SARS-CoV-2 could play a role in common conditions such as appendicitis, diabetic ketoacidosis, and some manifestations of mental illness, in which case we may have underestimated the proportion of likely SARS-CoV-2–related hospitalizations.^11,12 

Alternatively, several of the hospitalizations categorized as likely to be caused by SARS-CoV-2 had additional diagnoses that made the role of COVID-19 unclear (eg, other documented respiratory viral infections, worsening seizures, and BRUEs), which may lead to overestimation of the proportion of likely COVID-19 hospitalizations. Nonetheless, despite the imperfect precision, our findings illustrate how reliance on reported hospitalization rates in children may lead to an inflated sense of disease burden in this population.

At the start of the pandemic, children were considered to be minimally impacted and represented a small proportion of reported infections.¹³  As time has elapsed, we are learning that infection and transmission rates in children, although lower than that of adults, are non-negligble.¹⁴  Rates of infection in children have increased in parallel with community prevalence.^15,16  The pediatric inpatient COVID-19 experience has been different from adults with regards to severity and volume; however, reports of increased COVID-19 hospitalizations in children have captured news headlines, leading to concerns over school reopening^2,17  and may influence policy decisions. Our findings reveal that such decisions should account for the fact that reported hospitalization rates lead to overestimation of the COVID-19 disease burden in children considerably.

Our study is limited by the single-center, retrospective design, which has an impact on generalizability. However, our proportion of asymptomatic patients is similar to other investigations,^1,18,19  suggesting that our proportion of “unlikely” might align similarly. We had to make subjective assessments about the disease severity and cause of hospitalization from documentation in the electronic medical record. Although our interrater reliability was adequate, there were challenges inherent to this approach.

We acknowledge the departments of Infection Prevention and Control and the Enterprise Analytics at Lucile Packard Children’s Hospital for information on reporting to the DPH and development of the dashboard, respectively. We thank Monika Roy, MD, from the Santa Clara County Department of Public Health, for providing our team with further insight into the COVID-19 hospitalization reporting process.