We conducted a retrospective population-based cohort study using the Healthcare Cost and Utilization Project’s (HCUP) State Inpatient Databases (SIDs) from 21 geographically disperse U.S. states in 2001, 2004, 2010, 2016, and 2019 ( Supplementary Table 1 ). The SIDs include inpatient records for all discharges from non-federal acute care hospitals within each state. We included states that submitted SIDs to HCUP with revenue codes allowing identification of ICU care, and we followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline (19). The study was determined to be exempt from human subject review by the Seattle Children’s Hospital Institutional Review Board.

We included all children aged 0-17 years, excluding those in major diagnostic categories 14 (delivering a baby) and 15 (newborns and other neonates) and those admitted to rehabilitation and psychiatric hospitals. We identified ICU care using revenue codes and included admissions to any non-neonatal ICU in the analyses.

To evaluate representation compared to the general ICU population, we obtained patient characteristics including age, sex, race, ethnicity, and insurance status as reported by individual hospitals. We used International Classification of Diseases, Ninth and Tenth Revision (ICD-9 and ICD-10) codes to identify diagnoses, comorbid conditions, organ failures, and procedures ( Supplementary Table 2 ). We categorized admissions into those involving hematologic malignancies, solid malignancies, and HCTs, as well as specific oncologic diagnoses and oncologic versus non-oncologic indications for HCT. We classified comorbidities using the Pediatric Complex Chronic Conditions classification system (20), excluding codes for hematologic or oncologic comorbidities. Technology dependence per the classification system includes presence of devices such as tracheostomies, gastrostomies, colostomies, ventricular shunts, dialysis access, and pacemakers. Metabolic comorbidities included endocrine, amino acid metabolism, lipid metabolism, and storage disorders. We identified organ failures using ICD-9 and ICD-10 codes per previously published algorithms (21, 22) and considered patients to have multiple organ dysfunction syndrome (MODS) if they had two or more dysfunctional organ systems. We estimated total hospital costs using hospital-specific cost-to-charge ratios adjusted to 2019 dollars using the Consumer Price Index (23).

We summarized categorical variables using percentages and continuous variables using medians and interquartile ranges (IQR). We determined the population-based rate of inpatient and ICU admissions per 1000 children using U.S. census data from each included year and calculated incidence rate ratios (IRR) for each year relative to 2001. We used generalized linear Poisson regression to estimate the relative risk (RR) of each categorical variable in each year relative to 2001 and linear regression to estimate mean change in each year relative to 2001 for continuous variables. We used Cuzick’s test of trend to determine overall trends from 2001-2019. We conducted all analyses using Stata version 17 (StataCorp LLC, College Station, TX).

In all study years combined, there were 2,157,991 pediatric hospital admissions, of which 12.8% (n=275,656) included ICU care. There were 82,988 admissions for children with oncologic diagnoses (3.8% of hospital admissions) and 7,381 admissions for children who had received an HCT (0.3% of hospital admissions) ( Table 1 ). A total of 13.9% (n=11,517) of oncologic admissions and 23.7% (n=1,749) of HCT admissions included ICU admission.

Among all oncology and HCT patients, age did not change substantially over time with a median age of 4.3 years (IQR 2.4-7.4) in 2019 ( Table 1 ). The percentage of non-Hispanic White patients decreased from 67.1% in 2001 to 58.5% in 2019, while the percentage of Hispanic patients increased from 9.8% in 2001 to 17.5% in 2019. In contrast, Hispanic patients in the non-oncology ICU population only increased from 12.5% to 16.1%. The use of public insurance among oncology and HCT patients admitted to an ICU doubled from 22.7% in 2001 to 45.5% in 2019 (RR 1.99, 95% CI 1.79-2.21) ( Table 1 ).

From 2001 to 2019, children with oncologic diagnoses comprised an increasing percentage of hospital admissions, rising from 3.4% in 2001 to 4.4% in 2019, and the percentage of HCT admissions increased from 0.2% in 2001 to 0.4% in 2019. ( Table 1 ). ICU admission prevalence among hospitalized patients remained similar for patients with either hematologic or solid malignancies, while it decreased for HCT admissions from a peak of 26.2% in 2010 to 19.5% in 2019 (p=0.002).

In contrast, the population-based rate of combined oncology and HCT admissions decreased from 1.10 to 0.95 hospitalizations per 1000 children per year during that same period (IRR 0.86 [95% CI 0.84-0.88]). The population-based rate of ICU admissions for oncology and HCT patients decreased from 0.16 to 0.13 ICU admissions per 1000 children per year from 2001-2019 (IRR 0.83 [95% CI 0.78-0.89]) ( Table 1 ).

Of 520 hospitals in the dataset that admitted children with oncologic conditions or HCTs across the five sample years, only 75 (14.4%) were dedicated children’s hospitals. Of the 180 hospitals that admitted pediatric oncology and HCT patients to the ICU, only 66 (36.7%) were children’s hospitals. There were 55 ICUs that provided care to >25 oncology or HCT patients in any given year, and of those, 38 (69.1%) were in a children’s hospital.

Patients were increasingly admitted to ICUs in children’s hospitals over the course of the study period, rising from 62.5% to 92.9% of oncology admissions (p<0.001) and from 71.0% to 88.8% of HCT admissions from 2001-2019 (p<0.001) ( Supplementary Table 3 ). The prevalence of ICU admission for oncology patients declined from 14.1% in 2001 to 13.3% in 2019 (p=0.003) in children’s hospitals and increased from 13.6% to 14.5% in general hospitals. Among HCT patients, the prevalence of ICU admissions decreased from 33.7% in 2001 to 18.1% in 2019 in children’s hospitals and increased from 14.0% to 40.5% over the same period in general hospitals. Oncology and HCT patients treated in children’s hospitals were of similar age to patients treated in general hospitals but had higher prevalence of non-oncologic and non-hematologic comorbid conditions with higher in-hospital mortality and hospitalization costs ( Supplementary Table 3 ).

The prevalence of chronic non-oncologic and non-hematologic comorbidities among children admitted for an oncologic diagnosis increased from 44.3% in 2001 to 69.1% in 2019 (RR 1.56 [95% CI 1.46-1.66]) ( Figure 1 ). Children with HCTs had a greater increase in prevalence of chronic comorbidities from 41.4% to 80.5% (RR 1.94 [95% CI 1.61-2.34]) ( Table 2 ). In 2019, the most common types of comorbidities among patients with hematologic malignancies were metabolic (35.5%), renal (21.7%), and cardiovascular (15.4%), while patients with solid malignancies most commonly had neuromuscular comorbidities (44.4%). The most common comorbidities among patients with HCTs were metabolic (33.2%) and gastrointestinal (31.9%). Each group had high rates of technology dependence, including 29.8% of patients with hematologic malignancies, 41.5% of patients with solid malignancies, and 52.1% of HCT patients ( Supplementary Table 4 ).

By 2019, 63.7% of oncology patients admitted to an ICU had at least one organ failure, an increase from 36.6% in 2001 (RR 1.74 [95% CI 1.61-1.88]) ( Table 2 ). Patients with hematologic malignancies had the highest prevalence of organ failure at 76.8% in 2019, while patients with solid malignancies had the greatest rise over time from 29.9% in 2001 to 55.2% in 2019 (RR 1.85 [95% CI 1.66-2.06]). Prevalence of organ failure among HCT admissions increased from 49.7% to 69.5% (RR 1.40 [95% CI 1.18-1.66]). Neurologic failure was the most common organ failure for all subgroups and was highest among patients with hematologic malignancies (41.7%). Cardiovascular failure was the second most common organ failure among patients with hematologic malignancies, doubling from 18% to 36.5% from 2001-2019, while respiratory failure was the second most common organ failure among patients with solid malignancies or HCTs ( Supplementary Table 5 ).

From 2001-2019, the percentage of admissions with MODS increased from 9.5% to 33.3% for oncology patients (RR 3.52, 95% CI 2.97-4.18) and from 12.4% to 39.7% for HCT patients (RR 3.20, 95% CI 2.09-4.89). MODS prevalence was highest among patients with hematologic malignancies at 48.9% in 2019 ( Table 2 ). Use of mechanical ventilation did not change over time, while both hematologic malignancy and HCT admissions had increasing prevalence of dialysis use, peaking at 7.1% and 7.9% respectively in 2019. The prevalence of extracorporeal membrane oxygenation support (ECMO)was highest among patients with hematologic malignancies, increasing from 0.3% in 2001 to 1.2% in 2019 ( Supplementary Table 6 ). ECMO use ranged from 0% to 1%, depending on the year, in patients with HCT.

The median hospital LOS remained unchanged for all groups between 2001 and 2019, with the longest LOS among patients with hematologic malignancies. Despite stable LOS, hospitalization costs increased in each group. Hematologic malignancy admissions had the highest median cost, reaching $51,895 in 2019, and HCT admissions had the largest increase from $23,622 in 2001 to $40,958 per admission in 2019 (p<0.001) ( Table 3 ).

Hospital mortality for oncology admissions overall ranged from 4.1-5.7% in each included year, with no change from 2001 to 2019. Mortality declined for admissions with hematologic malignancies from 9.4% in 2001 to 6.2% in 2019 (RR 0.66 [95% CI 0.45-0.97]) while mortality for solid malignancy admissions remained unchanged, ranging from 2.5-4.9% in each year. Mortality for HCT admissions declined from 12.4% in 2001 to 7.2% in 2019 (RR 0.58 [95% CI 0.33-1.03]) ( Table 3 ).

Among hematologic malignancies, hospitalized patients with acute lymphoblastic leukemia (ALL) and lymphoma had an increasing ICU admission prevalence, rising from 8.8% to 10.4% of admissions for patients with ALL and from 8.5% to 10.2% of admissions for patients with lymphoma from 2001-2019. Among solid tumors, brain tumors had the largest increase in ICU admission prevalence from 28.9% to 35.3% of admissions from 2001-2019, while the prevalence of ICU admissions for patients with neuroblastomas decreased from 19.4% to 13.0%. Among HCT patients, ICU admission prevalence was higher for patients with non-oncologic HCTs than oncologic HCTs. ICU admission prevalence declined over time for patients with non-oncologic HCTs from 27.4% in 2001 to 22.5% in 2019 and remained similar for oncology-related HCTs ( Supplementary Table 7 ).

The prevalence of MODS increased from 2001-2019 across all oncologic diagnoses evaluated, with the highest prevalence of MODS observed among patients with acute myeloid leukemia (AML), with an increase from 24.4% to 63.3%, and oncologic HCTs, with an increase from 13.6% to 44.0% ( Supplementary Figure 1 ). MODS prevalence was lowest among patients with nerve tumors (9.5% in 2019) and brain tumors (19.6%).

Despite the increase in MODS, mortality decreased for patients with AML from 14.6% of ICU admissions in 2001 to 8.5% in 2019, while mortality remained similar from 2001-2019 for patients with other oncologic diagnoses including ALL (7.6% to 7.1%), brain tumors (5.1% to 5.0%), and neuroblastoma (5.9% to 3.9%). Among patients with HCTs, mortality was lower for non-oncologic HCT patients than oncologic HCT patients throughout the time period, but with a greater relative decline for oncologic HCT patients from 15.5% in 2001 to 9.2% in 2019 ( Supplementary Table 7 ).