Variation and Outcomes of Hospital-Level High-Flow Nasal Cannula Usage Outside of Intensive Care

Data for this study were obtained from the PHIS, an administrative database containing inpatient, emergency department, ambulatory surgery, and observation encounter-level data from pediatric hospitals affiliated with the Children’s Hospital Association (Lenexa, KS). Participating hospitals provide deidentified discharge/encounter data, including demographics, diagnoses, and procedures, and data are subjected to reliability and validity checks before being included in the database. The authors of this study initially used data from 49 hospitals.

We included patients who were <2 years old at the time of admission if they had a principal diagnosis of acute bronchiolitis (International Classification of Diseases, Tenth Revision codes J21.0, J21.1, J21.8, or J21.9) or had a principal diagnosis of acute or unspecified respiratory failure (International Classification of Diseases, Tenth Revision codes J96.00, J96.01, J96.02, J96.90, J96.91, or J96.92) and a secondary diagnosis of acute bronchiolitis as detailed above. We included encounters that were classified as either observation or inpatient status. We included patients admitted between September 1, 2018 and March 31, 2019.

This study used newly available codes for HFNC in the PHIS database. HFNC usage was defined as a patient with either a supply code (255314) or a clinical code (521172) related to HFNC within the PHIS database. This methodology has been validated at 2 hospitals with an overall sensitivity of 90.4% and specificity of 99.3%.¹⁸  We excluded all patients discharged from hospitals (n = 29) that reported HFNC utilization for <20 patients with bronchiolitis during the study period, primarily because of concerns that this indicated the PHIS HFNC codes were not valid at those sites. This left 20 hospitals for the final analysis.

For each patient, we obtained demographics including age in months at time of admission, sex, and race/ethnicity. We used PHIS flags for complex chronic conditions (CCC), including (1) premature and neonatal CCC, and (2) respiratory CCC.

We stratified individual patients into 3 exhaustive, mutually exclusive treatment groups: (1) those who were never in the ICU and never received HFNC (ICU[−]/HFNC[−]), (2) those who were never in the ICU, but received HFNC (ICU[−]/HFNC[+]), and (3) those who were at some time admitted to the ICU (ICU[+]).

Hospitals were divided into low, moderate, and high groups on the basis of the proportion of non-ICU patients who received HFNC. The non-ICU HFNC usage (NIHU) was defined as the number of patients in the ICU[−]/HFNC[+] group divided by the number of patients in both ICU[−] groups. Patients were assigned to the NIHU grouping of their admitted hospitals. The dividing points were chosen post hoc on the basis of a 3-group cluster analysis (Fig 1).

We used the PHIS-defined LOS as our primary outcome. We used the PHIS total standardized unit cost²⁰  in addition to a breakdown by cost type, including bed/nursing unit (called “other” in PHIS), clinical, imaging, laboratory, pharmacy, and supply costs. Briefly, the standardized unit cost for each item is based on the median cost for that item, derived from charges and ratios of costs to charges. Because of the right-skewed nature of LOS and cost, we applied a logarithmic transformation for these variables when performing regression. For cost values, we added a minimal offset ($1) to avoid undefined values for patients with $0 cost in a particular department when performing regression.

We graphed the proportion of patients in each of the 3 groups for each hospital (Fig 1). We performed a χ² test of independence to test for statistically significant variation among the hospitals.

We summarized patient data for the cohort using numbers and percentages. We summarized the data for patients from included versus excluded hospitals (Supplemental Table 3). We summarized the data for patients at hospitals with low, moderate, and high NIHU (Table 1). χ² test of independence was used to test for differences among respective groups.

We calculated medians (with interquartile ranges) for the outcomes of interest (LOS and cost) overall and for each NIHU group. We used a hierarchical mixed-model linear regression to determine the association of NIHU group as a fixed effect with each of the outcomes (on a logarithmic scale). We calculated the unadjusted multiplicative ratio (MR), with 95% confidence interval (CI). We also performed the same analysis after adjusting for age, sex, race/ethnicity, any CCC, prematurity or neonatal CCC, respiratory CCC, and HFNC/ICU treatment group. This decision was made to minimize the confounding effect of differing numbers of patients in each treatment group between NIHU groups.

All statistical analysis was performed in SPSS (Version 26.0; IBM Corp, Armonk, NY).

This study was deemed not human subjects research by the University of Tennessee Health Sciences Center Institutional Review Board.

Overall, 34 250 patients were identified, with 15 155 patients from 20 hospitals included and 19 095 patients from 29 hospitals excluded. There were statistically significant differences in male sex, race/ethnicity, any CCC, respiratory CCC, principal diagnosis, ICU utilization, HFNC usage, LOS, and all cost categories between included and excluded patients (Supplemental Table 3).

There were statistically significant differences in age, race/ethnicity, and ICU usage among the NIHU groups (Table 1). Hospitals with low NIHU had lower overall (non-ICU plus ICU) HFNC usage and lower HFNC usage among patients who had received ICU care than hospitals with moderate and high NIHU. There was statistically significant variation (P <.001) in the proportions of patients in the 3 groups among hospitals (Fig 1). Across all hospitals, 56.1% of patients were ICU[−]/ HFNC[−] (range 17.7%–86.8%), 26.3% of patients were ICU[−]/HFNC[+] (range 0.9%–71.3%), and 17.6% of patients were ICU[+] (range 5.5%–38.6%). Among patients who were ICU[−], there was also statistically significant variation in HFNC usage (P <.001), with NIHU averaging 31.9% (range 1.1%–80.1%) (Fig 1).

Gross LOS (Table 2) was statistically longer overall for patients in hospitals with moderate (MR 1.14; 95% CI 1.11–1.18) and high (MR 1.26; 95% CI 1.22–1.30) NIHU as compared with low NIHU. Additionally, when adjusted for demographic and clinical characteristics as well as HFNC/ICU treatment group, patients in hospitals with moderate NIHU (MR 1.03; 95% CI 1.01–1.06), and hospitals with high NIHU (MR 1.08; 95% CI 1.05–1.11) had increased LOS.

Gross total standardized cost (Table 2) was statistically higher overall for patients in hospitals with moderate (MR 1.20; 95% CI 1.16–1.25) and high (MR 1.26; 95% CI 1.22–1.31) NIHU as compared with low NIHU. Additionally, when adjusted for demographic and clinical characteristics as well as HFNC/ICU treatment group, patients in hospitals with moderate NIHU (MR 1.05; 95% CI 1.03–1.08), and hospitals with high NIHU (MR 1.05; 95% CI 1.02–1.08) had increased total standardized cost. The single largest contributor to increased cost in patients at hospitals with high NIHU was the bed/nursing unit.

This multicenter retrospective cohort study revealed that there exists wide variation in the usage of HFNC and ICU in patients with bronchiolitis across children’s hospitals. HFNC usage ranged among included hospitals from 1.1% to 80.1% among patients who did not receive intensive care. Our study also revealed variability in ICU utilization between hospitals, with included hospitals ranging from 5.5% to 38.6% of admitted patients. The overall ICU utilization, 20.5%, was lower than that previously reported between 2018 and 2019, which rose from 23.9% to 24.5% between those years,²¹  which may be due in part to the differences in ICU utilization between patients admitted to included and excluded hospitals, as discussed in the limitations below.

We analyzed the association of hospital-level NIHU with LOS and cost at 2 extremes to attempt to find bounds to its effect. At 1 extreme, assuming that all the variation in HFNC practice is due to hospital policy and practice (unadjusted analysis), we found that both LOS and total cost increased with increasing NIHU. At the other extreme, if all the variation occurred because of patient differences and that allocation was inherent to the patient (adjusted analysis), both LOS and cost had smaller, but still statistically significant, increases in patients admitted to moderate and high NIHU hospitals.

Based on the 2 analyses’ 95% CI extremes, our best estimate is that high (>32%) NIHU hospitals have 5% to 30% longer LOS. Similarly, we estimate that high NIHU hospitals have a 2% to 31% higher total cost of care. Although we believe that the true effect likely lies closer to the unadjusted analysis because of unwarranted variation, it is reasonable to assume that the actual association of hospital-level NIHU with LOS and cost is somewhere between these 2 extremes and, therefore, that high NIHU increases both the LOS and cost of care for patients.

Although these effects are small, especially for a disease with relatively short LOS and low cost, the large number of patients admitted to children’s hospitals means that even small relative changes in these values may lead to large burdens to hospitals and the health care system. In terms of the increased total cost, it is plausible that, although fewer patients in high NIHU hospitals were admitted to ICUs during their admission, this savings in cost might be overshadowed by increases in higher levels (levels IV and V) of emergency department services, as well as increases in step-down (also called progressive or intermediate) unit charges outside of the ICU in cases in which the nursing staff ratio is changed for patients receiving HFNC. Additionally, this does not evaluate indirect costs and benefits to HFNC outside of the ICU, such as (1) removing additional stress to the patient and family due to ICU admission and (2) effects on census and throughput on a hospital’s system.

This study has multiple additional limitations in its interpretation and application. As a retrospective cohort study, we were unable to assess for causality. Because of the nature of the administrative data used, there are multiple potential confounding factors for which the analysis was not adjusted, especially those that may be associated with indication bias. Although the definition we used with PHIS has been validated at 2 hospitals, there is a high likelihood of treatment misattribution for individual patients.

We used a nonrandom subset of 20 PHIS hospitals (out of 49 eligible) on the basis of the new structured billing code for HFNC, which may introduce bias in addition to the inherent limits to generalizability to care outside of children’s hospitals. There were statistically significant differences in most measured variables between patients from included and excluded hospitals. Most notably, ICU utilization was much lower (20.5% vs 30.2%) between patients at included and excluded hospitals.

Among the hospitals included in the study, there was high variability in HFNC and ICU utilization. Increased NIHU was associated with increased LOS and cost. NIHU may lead to increased overall resource utilization for children admitted with bronchiolitis.