Association Between COVID-19 and Severity of Illness for Children With Hyperglycemic Crisis

We completed a retrospective observational cohort study of children ≤18 years old hospitalized from March 2017 to February 2022 admitted to a participating Pediatric Health Information Systems (PHIS) hospital. Hospitals were excluded if they had incomplete data submission to PHIS for the study period. We identified children with HGC using International Classification of Diseases, 10th edition, codes that included DKA, hyperglycemic hyperosmolar syndrome, and hyperosmolarity (Supplemental Table 4). The institutional review board at Children’s Hospital Los Angeles considered this project exempt from review (CHLA-21-00214).

The PHIS database contains administrative and billing data from 47 tertiary, pediatric hospitals in the United States affiliated with the Children’s Hospital Association (Lenexa, Kansas). Data quality are managed with pre- and postsubmission audits as a joint collaborative effort between the Children’s Hospital Association and the affiliated hospitals. The PHIS database contains “flags” that use Clinical Transaction Classification and International Classification of Diseases, 10th edition, to identify patients of interest. Encounter-level data includes patient demographics and date-stamped billing data for a range of clinical services including diagnostic tests and pharmaceuticals.

Admissions were grouped into 3 periods based off discharge date: pre-COVID-19 (March 2017–February 2020), COVID-19 year 1 (March 2020–February 2021), and COVID-19 year 2 (March 2021–February 2022).¹²  Hospital admissions included any patient admitted to the general care floor or ICU. Patients admitted to the ICU for a portion of their encounter were identified using the PHIS “ICU flag.” This flag does not include patients admitted to the NICU.

Indicators of patient severity of illness included admission to the ICU, hospital length of stay (LOS), cost, and interventions for neurologic compromise, mechanical ventilation, and vasoactive support. Interventions for neurologic compromise were defined as need for brain imaging, hyperosmolar therapies, neurologic medications, and electroencephalography (Supplemental Table 5). Mechanical ventilation was defined by the PHIS “mechanical ventilation” flag identifying invasive mechanical ventilation and excludes noninvasive positive pressure ventilation. Vasoactive support included continuous vasoactive infusions only. We calculated costs by multiplying adjusted total billed charges for the encounter by the respective hospital’s specific ratio of costs to charges, as reported by their hospital cost report submitted annually to the Centers for Medicare and Medicaid Services.¹³  Children with anti-glutamic acid decarboxylase 65, anti-insulin, and anti-islet cell antibodies laboratory studies were identified as children with new-onset DM because these tests are generally indicated for the initial diagnosis of DM.

Our primary outcome of interest was ICU admission. Secondary outcomes included hospital LOS in survivors and cost. In the subgroup of children admitted to the ICU, additional secondary outcomes evaluated were ICU LOS in survivors, use of invasive mechanical ventilation, interventions for neurologic compromise, vasoactive support, cost, and mortality. In the subgroup of children with HGC admitted with new DM, similar primary and secondary outcomes were analyzed.

Overall differences in hospital and ICU admissions, and demographic characteristics between the pre- and COVID-19 years were assessed for HGC patients. Continuous variables were described with medians and interquartile ranges and assessed using the Kruskal-Wallis or Wilcoxon test. Categorical variables were described with frequency and percentages and assessed with the χ² test. Post hoc pairwise P values were assessed when the overall P value was <.05 and adjusted for multiple comparisons using the Bonferroni adjustment. Multivariable, mixed effects logistic regression models were used to evaluate the association between pre- and post-COVID-19 periods and the primary and secondary outcomes. Because they were not normally distributed, LOS and cost were dichotomized using the 90th percentile.

We controlled for potentially confounding demographic variables in all multivariable models. Confounding variables were defined as those that had a P value <.1 between pre- and post-COVID-19 years in the univariate analyses or those that changed the effect estimate of time by more than 15% in either direction in multivariable modeling. All models accounted for the clustering of encounters within patient and patient within center. Analyses were performed in SAS, version 9.4.

There was a total of 3 866 549 patients admitted to 47 PHIS hospitals during the study period. Five PHIS hospitals were excluded because of incomplete data for the study period, accounting for 337 421 patients. From the remaining 42 hospitals, there were 3 529 128 total admissions during the study period, with a subset of 46 425 patients admitted with HGC. There was a total of 20 045 (43%) HGC ICU admissions, and 17 319 HGC patients (37%) were identified to have new-onset DM.

During the COVID-19 pandemic, non-HGC floor and ICU admissions decreased compared with previous years (Fig 1). There was a significant increase in the percentage of HGC admissions to the ICU when comparing COVID-19 year 1 and year 2 to the pre-COVID-19 years after adjusting for multiple comparisons (P < .001; Supplemental Table 6).

In univariate analyses, there were differences between the demographics of children that were admitted for HGC by time period for age, sex, race/ethnicity, and payer (Table 1).

In univariate analysis, there was a significant increase in the percentage of ICU admissions for hospitalized children with HGC during year 1 (48%) and year 2 (45%) of the pandemic compared with prepandemic (41%) (all P < .05; Supplemental Table 7). In multivariable modeling, children with HGC admitted during the 2 years of the COVID-19 pandemic had a higher odds of ICU admission than children admitted prepandemic: COVID-19 year 1 (odds ratio [OR], 1.31; 95% confidence interval [CI] 1.25–1.38; P < .0001) and year 2 (OR, 1.17; 95% CI, 1.11–1.22; P < .0001) after adjusting for age, sex, race/ethnicity, payer, and location (Table 2).

In univariate analyses, there were significant differences between pre-COVID-19 and each COVID-19 year for hospital LOS >4 days for survivors and cost >$16 633 (P < .001; Supplemental Table 7).

In multivariable modeling, years 1 and 2 of the COVID-19 pandemic were associated with higher odds of hospital LOS >4 days and cost >$16 663 in comparison with pre-COVID-19 after adjusting for age, sex, race/ethnicity, payer, and location (Table 2). When comparing COVID year 1 versus year 2, there were no significant differences in these outcomes.

In univariate analysis, there were differences between pre-COVID-19 versus the COVID-19 years for ICU LOS >2 days for survivors, interventions for neurologic compromise, vasoactive support, and cost >$21 337 (P < .001–0.032; Supplemental Table 7). There were no differences in use of mechanical ventilation or mortality.

In multivariable modeling, the 2 years of the COVID-19 pandemic were associated with increased ICU LOS >2 days and cost >$21 337 (P < .0001; Table 3). Year 1 of the COVID-19 pandemic was associated with increased interventions for neurologic compromise (OR, 1.13; 95% CI, 1.03–1.25; P = .008) and mechanical ventilation (OR, 1.28; 95% CI, 1.05–1.58; P = .017) in comparison with prepandemic years. Year 2 of the COVID-19 pandemic was associated with increased need for vasoactive support (OR, 1.32; 95% CI, 1.06–1.64; P = .013) in comparison with prepandemic years. There was no difference in mortality across the pandemic years. When comparing COVID year 1 versus year 2, there were no significant differences in secondary outcomes.

In the subgroup of children with new-onset DM, results were consistent for higher odds of ICU admission, hospital LOS >4 days in survivors, and cost >$19 180 in children admitted during the pandemic in comparison with the prepandemic (Supplemental Table 8). In children with new-onset DM admitted to the ICU, odds of ICU LOS >2 days and cost >$24 424 were higher during the pandemic years than the prepandemic when comparing pre-COVID-19 to COVID-19 years 1 and 2. No difference was found in other secondary outcomes for severity of illness.

In this retrospective review of the PHIS database, we identified an increase in the severity of illness for children admitted to the hospital with HGC during years 1 and year 2 of the COVID-19 pandemic in comparison with pre-COVID-19 years. Although the odds of ICU admission decreased from year 1 to year 2 of the pandemic, there remained a significant increase in comparison with the pre-COVID-19 years. Furthermore, other markers of severity of illness remained elevated during COVID-19 years 1 and 2 when compared with pre-COVID-19 years. Given that the increases in severity of HCG was sustained throughout the COVID-19 pandemic and within those with newly diagnosed DM, delayed presentation to care and medication noncompliance are not likely to account solely for these changes. When comparing HGC admissions to pediatric COVID-19 hospitalized cases, we found the pattern of HGC admissions to the floor and ICU rose during the early phase of the SARS-CoV-2 outbreak (Supplemental Fig 2).¹⁴  However, additional HGC spikes did not occur during the Δ (July 1–December 18, 2021) and ο (July 1–December 18, 2021) surges, suggesting the rise in HGC was either not solely from infection or that the association between infection and HGC differed by variant.¹⁵  Our data suggest that multiple factors may have contributed to worsen pandemic associated HCG severity of illness.

The first year of the COVID-19 pandemic was marked by a rise in DKA in children with new-onset type 1 and type 2 DM.^7,10  Research suggests that pediatric patients had delayed presentation to care early in the pandemic because of fear of exposure to COVID-19, decreased access to primary care, lack of in-person school, and other psychosocial factors that could have resulted in an initial rise in severely ill children with HGC during the COVID-19 pandemic.^16–18  Additionally, decreased number of health care providers, decreased recognition of diagnosis, rising obesity rates, and abrupt transition to telehealth likely contributed to the rise in DKA.^4,19–21  Our data do not support that the increased incidence and severity of HGC during the COVID-19 pandemic was entirely related to delayed access to care because an increased need for ICU admission was sustained from year 1 to year 2 despite changes in social distancing and schools reopening (and therefore allowing for possible earlier recognition of illness).²²  Furthermore, our findings were consistent when limited to children with newly diagnosed diabetes, a subgroup of children where noncompliance with insulin administration is not a factor. It is possible that viral factors may also play a significant role in the sustained pattern of HGC severity of illness during the pandemic, as a growing body of literature of posits. Our study did not demonstrate a significant difference in mortality during the COVID-19 pandemic, at least in part because of the overall low mortality in this patient population.

This study highlights a vulnerable pediatric population who necessitate preparedness for future pandemics. Policy and practice implications include supply chain management, access to care, and surveillance of diabetic risk factors.²³  Aside from policy implications, the increased incidence and severity of illness for HGC pediatric patients is concerning because of its association with long-term morbidity. DKA is known to alter the function and structure of the brain that can be present for up to 6 months after the presentation.^24,25  Aye et al compared long-term neurocognitive outcomes for pediatric patients presenting with moderate to severe versus mild DKA. Their study demonstrated that 1 episode of moderate to severe DKA is associated with lower cogitative scores in pediatric patients and altered brain development, which is observable up to 4 years after DKA presentation.²⁶  Multiple studies have demonstrated that pediatric patients who present in DKA at diagnosis of type 1 DM have cognitive deficits including delay in spatial memory.^24,27  Further research is needed to understand the long-term morbidity associated with higher severity of illness for children with HGC during the COVID-19 pandemic. It is important that HGC patients are followed longitudinally to identify and support morbidities from their presenting illness.

The influence of COVID-19 on pancreatic molecular pathways remains an area of active research. There is evidence of direct and systemic impact of SARS-CoV-2 on the pancreas. After analyzing autopsy tissue from patients with COVID-19, Steenblock et al found that human islet cells are permissive to infection and that multiple receptors may facilitate virus entry.²⁸  Wu et al demonstrated SARS-CoV2 can directly infect human β cells. Furthermore, SARS-CoV-2 reduces pancreatic insulin secretion and induces β-cell apoptosis.²⁹  For patients with a history of SARS-CoV-2 infection, evidence suggests that SARS-CoV-2 infection of the pancreas may promote acute and chronic pancreatic dysfunction potentially leading to new-onset diabetes.³⁰  This could be from direct insult to the β cells or as secondary injury from a systemic cytokine cascade associated with the COVID-19 infection.³¹  Mohammed et al determined that SARS-CoV-2 infection is associated with disseminated pancreatic inflammation, microthrombi, and fibrosis, which results in new-onset hyperglycemia, suggesting a possible etiology for new-onset diabetes associated with the virus.³²  In the pediatric population, emerging evidence has demonstrated that pediatric patients with a previous COVID-19 infection have a higher rate of developing type 1 and type 2 diabetes compared with patients with non-COVID-19 respiratory infections.^33–35  However, SARS-CoV2 antibodies were not routinely tested. Retrospective review of new-onset diabetes DKA cases showed that acute SARS-CoV2 infection was present in only 2% to 3% of cases.⁹  Further research is needed to understand the unique properties of the SARS-CoV-2 virus to understand why children are more likely to develop diabetes after SARS-CoV-2 infection and also present with higher severity of illness as found in our study. Together, this emerging knowledge may have public health implications related to both screening and long-term follow up for children at risk.

Our data also support an increase in median cost for hospitalized and ICU patients admitted for HGC during years 1 and 2 of the COVID-19 pandemic in comparison with pre-COVID-19. Increased encounter costs are an indication that HGC patients are potentially requiring higher resources for stabilization. This finding is in line with our outcomes demonstrating an increased need for ICU admission, interventions for neurologic compromise, vasoactive support, and mechanical ventilation. Our findings are consistent with the Children’s Hospital Association’s report of an 11% increase in the Case Mix Index of children admitted during the first year of the COVID-19 pandemic indicating an overall higher acuity of illness.³⁶ 

Aside from the impact on patient acuity, the elevation in cost during the pandemic should also be considered in conjunction with the rising costs of providing care seen at the bedside during COVID-19. The PINC AI database is a workforce dataset containing data from more than 1400 hospitals. In a recent study from October 2019 to August 2021, the PINC AI database found workforce overtime hours up 52% and temporary labor hours up 131%, which can add 50% more to an employee’s hourly rate.³⁷  Changing labor force costs are just 1 factor to be considered when analyzing patient encounter cost overtime. Other factors include inflation, supply costs, and decreased revenue streams.³⁸  Further research is needed to understand the complexity of increased severity of illness and rising costs of care for the HGC population.

The primary strength of this study is the large size of our multi-institutional cohort. However, because the data are from an administrative database, there are limitations. The PHIS network’s quality assurance collaborative is an effort between the hospitals and the Children’s Hospital Association. Despite rigorous audits, errors in coding and billing are still possible. Therefore, patients with HGC may be missing if coding was inaccurate. Although an accepted methodology, the use of a hospital’s pharmacy-specific ratio of costs to charges to calculate cost of encounters also represents a limitation in this study. A hospital’s pharmacy-specific ratio of costs to charges is an estimate of resource utilization by each institution and does not reflect reimbursement received by the institution. Another limitation is using billing data to account for a patient’s severity of illness. PHIS does not capture granular clinical data; therefore, investigators are unable to obtain common severity of illness measures such as the pediatric risk of mortality, pediatric index of mortality, or the pediatric logistic organ dysfunction score. We inferred severity of illness with charges for ICU admission, LOS, interventions for neurologic compromise, mechanical ventilation, vasoactive support, and mortality. Furthermore, variation of ICU admission practices and bed availability during the COVID-19 pandemic could influence the percent of patients with hyperglycemic crisis requiring an ICU admission. Similarly, we were unable to determine whether patients admitted with HGC were admitted with a new or chronic diagnosis of DM. We used laboratory studies clinically used to diagnose new-onset DM (glutamic acid decarboxylase 65, islet cell, and insulin antibodies) as a surrogate of a new diagnosis of DM, which may under- or overestimate the incidence of new diagnosis DM in our cohort. Finally, the cohort comprised patients from the 47 hospitals within the PHIS network, which limits the generalizability of our findings. However, we believe this limitation is mitigated by the multi-institutional national representation within the large PHIS data set.

During the first 2 years of the COVID-19 pandemic, children presenting with HGC had a higher severity of illness, demonstrated by increased ICU admission, longer hospital and ICU length of stay, increased interventions for neurologic compromise, use of mechanical ventilation and vasoactive support, and higher cost. Furthermore, the increased severity of illness remained during COVID-19 year 2 despite reduction in social restriction, school reopening, and emergence of viral variants suggesting that direct viral factors may also play a role in the pattern changes of pediatric HGC admissions.