Impact of High Flow Nasal Cannula on Resource Utilization in Bronchiolitis

Bronchiolitis is the most prevalent indication for hospitalization of children <24 months of age in the United States, accounting for ∼ 17% of hospitalizations in this age group from 2000 to 2016. In 2016, inpatient care for bronchiolitis cost an estimated $734 million.¹  Without effective pharmacologic therapies, supportive care is the mainstay of management of this viral lower respiratory tract illness. The American Academy of Pediatrics discourages use of low-value interventions that add to cost without improving outcomes.²  Improving health care value for bronchiolitis includes limiting ineffective interventions, minimizing cost, and decreasing length of stay (LOS), ideally without increasing readmissions.

High flow nasal cannula (HFNC) is increasingly used to support children with respiratory distress because of bronchiolitis. In one analysis, 23 of 41 children’s hospitals had ward-based protocols for HFNC.³  Proposed mechanisms for its efficacy include: decreased respiratory anatomic dead space, induction of positive end expiratory pressure, improved oxygenation, secretion hydration, and washout of carbon dioxide-rich reservoirs.^4–7  Although it does not appear to modulate disease course, evidence is mixed as to its effect on utilization.^8–10  Its adoption in the ICU has led to decreased utilization of invasive measures such as central venous catheters.¹¹  However, its use in inpatient units has not led to decreased duration of supplemental oxygen therapy or LOS.^8,10,12,13 

The effect of HFNC on health care utilization is not well understood. Single-center studies have suggested cost-effectiveness of HFNC on general pediatric units, one in comparison with HFNC use in the ICU¹⁴  and the other in comparison with standard nasal cannula.¹⁵  Another found no difference in 30-day readmissions before and after implementation of an HFNC guideline.¹²  However, multicenter database studies have been limited because of the inability to identify HFNC use. The objectives of the current study were to (1) validate a novel combination of clinical transaction category (CTC) codes for determining HFNC usage in the Pediatric Health Information System (PHIS)¹⁶  at 2 hospitals, (2) measure the use of such codes across PHIS-participating hospitals, and (3) compare health care costs, LOS, ICU utilization, and readmissions between patients identified as receiving or not receiving HFNC within hospitals utilizing this method.

This multicenter, retrospective, cross-sectional study of children hospitalized with bronchiolitis used PHIS, an administrative and billing database of 51 tertiary care pediatric hospitals in the United States affiliated with Children’s Hospital Association (Lenexa, Kansas). Patient data are de-identified; however, encryption of identifiers allows for tracking of patients across encounters. Hospital billing codes are mapped to a common set of CTC codes, which are categorized into laboratory, pharmacy, imaging, supplies, clinical services, and other charges (including room charges). This study was approved by the institutional review board at each center participating in chart review.

Children aged 1 to 24 months with index hospitalizations for bronchiolitis at PHIS-participating hospitals from January 2018 to March 2020 were eligible for inclusion. The age range was chosen for consistency with the American Academy of Pediatrics clinical practice guidelines for bronchiolitis.²  We identified bronchiolitis hospitalizations using All Patients Refined Diagnosis Related Groups (APR-DRG) 138. Patients admitted to the ward or ICU were included. We excluded patients hospitalized in June to August 2018 and 2019 because of the low incidence of bronchiolitis and concerns those hospitalizations captured “off-season” would more likely reflect atypical practices. We excluded those transferred from another inpatient facility because of the inability to capture resource use throughout the entire care episode. Finally, we excluded patients with complex chronic conditions as described by Feudtner.¹⁷ 

HFNC use was defined in PHIS using 2 CTC codes (255314 supply code and 521172 clinical code), which were chosen to represent supplies and services necessary to deliver HFNC. Patients were considered to have received HFNC if either code was present. Validity testing of this definition was performed at the authors’ 2 institutions to ensure accuracy across sites. Local electronic medical record (EMR) and PHIS data were compared with identify concordance/discordance in identification of HFNC use from January 2018 to April 2019. Chart review was performed to resolve any differences between the EMR and PHIS. HFNC use was based on documentation within progress notes, vital signs, respiratory therapy notes, and/or orders. Discordant pairs resolved in favor of PHIS (ie, PHIS identified HFNC use, whereas local data report did not) were considered correct identification of HFNC use.

We then applied these CTC codes to the 51 PHIS-participating hospitals and stratified hospitals to those utilizing codes for <1% or ≥1% of patients hospitalized with bronchiolitis. We excluded those reporting <1% HFNC utilization. This decision was made early during cohort development in response to absence of CTC codes at several hospitals to remove hospitals not reliably utilizing billing codes to describe HFNC use.

We examined demographic characteristics, including age (1–2 months, 3–6 months, 7–12 months, 13–24 months), sex, race/ethnicity (non-Hispanic white, non-Hispanic Black, Hispanic, Asian, Other), payer (private/commercial, government, self-pay, other), and median household income quartile (based on zip code of residence reported in hospital records). We examined patient characteristics, including route of admission (ie, direct versus through the admitting hospital’s emergency department) and APR-DRG severity of illness (SOI) level. APR-DRG SOI is a classification system based on diagnosis codes that accounts for intensity of resource use, severity of illness, and mortality risk.¹⁸ 

Outcomes included patient-level standardized cost (US dollars), LOS, ICU utilization, and all-cause hospital readmissions at 3, 7, and 14 days. Costs of hospitalization included those from the admitting hospital’s emergency department, if any, and the inpatient stay. In PHIS, standardized costs reflect the resources billed during the encounter and remove the high interhospital variation in item costs. We calculated standardized costs using billing data from PHIS as described by Keren et al.¹⁹  In addition to total costs, we examined costs based on service line, including laboratory, pharmacy, imaging, supplies, clinical, and other costs (eg, room and board). To ensure cost information was included in geometric mean estimates for patients whose care did not include a service line, we assigned $0.01 to such services. We also evaluated cost incurred on the first calendar day of hospitalization (from time of presentation to midnight; day 0 cost) and mean cost per 24 hours of hospitalization (daily cost). We examined readmissions at multiple points to capture return visits associated with worsening disease and/or complications of the index admission. To address any confounding effects of ICU care on resource utilization, we performed subanalysis of the above measures, excluding patients requiring ICU care at any time during hospitalization.

Categorical variables were summarized using frequencies and percentages; continuous variables were summarized using mean with standard errors for normally distributed variables and geometric means with 95% confidence intervals (CI) for log-normally distributed variables. Categorical demographic characteristics and unadjusted probability of readmission were compared between patients who received HFNC and those who did not using an χ² test for association; normally, distributed patient demographics and log-normal unadjusted outcomes were compared using a student’s independent t test. Multivariable, generalized, linear, mixed models were used to compare probabilities of HFNC use, probabilities of readmission, and costs. For HFNC use and readmission, models were based on binomial distribution with a logit link, and for costs, models assumed log-linear distribution. All models included a random intercept for each hospital to control for clustering of patients within a hospital and included adjustment for sex, age, race/ethnicity, payer type, APR-DRG SOI, route of admission, median household income quartile, and US Census region. Adjustment variables were chosen a priori. All statistical analyses were performed using SAS v.9.4 (SAS Institute, Cary, NC), and P values <.05 were considered statistically significant.

We evaluated EMR data of 1105 patients at 2 hospitals included in PHIS from January 2018 to April 2019 (Table 1). In total, 1047 children had PHIS codes for HFNC consistent with their EMR. Chart review demonstrated that 4 children were discordantly positive for HFNC in PHIS and negative in the EMR, and 54 children were discordantly negative for HFNC in PHIS and positive in the EMR. The combination of HFNC CTC codes demonstrated a sensitivity of 90.4% and specificity of 99.3% for identification of HFNC usage.

Of the 51 PHIS-participating hospitals, 24 had CTC codes (based on billing codes) for HFNC utilization for ≥1% of patients admitted with bronchiolitis. Of the remaining 27 hospitals, only 1 reported any HFNC use (<1%). The other 26 reported 0 occurrences of HFNC use via their CTC codes.

A total of 34 049 children were hospitalized with bronchiolitis at the 24 PHIS-hospitals reporting >1% HFNC use during the study period (Table 2). Of those, 27.1% received HFNC, with utilization ranging from 1.7% to 74% across hospitals. A greater percentage of children aged 13 to 24 months and those of non-Hispanic Black, Asian, or other race/ethnicity received HFNC. APR-DRG SOI was significantly greater in patients receiving HFNC at 2.17 vs 1.37 (P < .001). Most patients were admitted through the institution’s emergency department, and those admitted through the emergency department were more likely to have received HFNC than those admitted directly to the inpatient unit (83.4% vs 16.6% P < .001).

Children admitted through the emergency department (adjusted odds ratio [aOR] 1.35 [95% CI 1.23 − 1.47]), male children (aOR 1.16 [95% CI 1.09 − 1.23]), children >13 months (aOR 1.22 [95% CI 1.12 − 1.33]) and non-Hispanic Black children (aOR 1.22 [95% CI 1.11 − 1.34]) were more likely to receive HFNC (Supplemental Table 4). The characteristic most associated with HFNC treatment was APR-DRG SOI.

In unadjusted analyses, children receiving HFNC incurred total costs $5623 greater on average than those who did not (P < .001) (Supplemental Table 5). Spending on laboratory, medications, imaging, supplies, clinical, and other costs were greater in the HFNC group (P < .001 for each). The mean daily cost and day 0 cost were greater in the HFNC group (P < .001 for each). In a subanalysis excluding children who received ICU care, total, daily, and day 0 costs remained greater in the HFNC group (P < .001 for each) (Supplemental Table 6).

Adjusting for clinical and demographic characteristics, total costs remained greater for the HFNC group ($7054 [95% CI $6191−$8037] vs $4544 [95% CI $3990−$5175]; P < .001) (Table 3). Daily cost also remained greater for the HFNC group ($2922 [95% CI $2534−$3370] vs $2613 [95% CI $2266−$3013]; P < .001). When subdivided by cost category, adjusted daily cost was greater in the HFNC group for laboratory, medications, imaging, supplies, clinical, and other costs (P < .001 for each) (Fig 1). Other costs, which included room and board, accounted for most of the total cost in both groups. When those with ICU care were excluded from adjusted analyses, total, daily, and day 0 costs remained greater for the HFNC group (P < .001 for each) (Supplemental Table 7).

In unadjusted analyses, the mean LOS was longer for children receiving HFNC (67.6 vs 33.5 hours; P < .001) (Supplemental Table 5). A greater proportion of patients in the HFNC group required intensive care (P < .001). When ICU patients were removed from unadjusted analysis, LOS remained longer for the HFNC group (59.4 vs 32.8 hours; P < .001) (Supplemental Table 6). After adjusting for clinical and demographic characteristics, LOS for the HFNC group remained longer than that of the no-HFNC group (57.6 vs 41.6 hours; P < .001) (Table 3). This difference remained significant after removing ICU patients from adjusted analysis (Supplemental Table 7).

Within unadjusted analyses, there were fewer readmissions within 3 days of discharge among those treated with HFNC (0.8% vs 2.0%; P < .001) (Supplemental Table 5). This difference persisted at 7 days (1.3% vs 2.5%; P < .001) and 14 days (2.7% vs 3.2%; P = .038). Removing ICU patients from unadjusted analysis, 3- and 7-day readmissions remained fewer for the HFNC group (Supplemental Table 6). With adjustment for clinical and demographic characteristics, odds of readmission remained lower in the HFNC group at 3 days (aOR 0.52 [95% CI 0.39,0.70]; P < .001) and 7 days (aOR 0.60 [95% CI 0.47,0.76]; P < .001) (Table 3) (Fig 2). Adjusted analysis found no difference between the HFNC (2.2%) and no-HFNC (2.5%) groups at 14 days (P = .124). When ICU patients were excluded from adjusted analysis, readmissions remained lower in the HFNC group at 3 days (P < .001) and 7 days (P = .001). Again, there was no difference at 14 days (P = .111) (Supplemental Table 7).

In this cross-sectional analysis of children hospitalized for bronchiolitis, a combination of CTC codes in PHIS identified utilization of HFNC with high sensitivity and specificity at 2 institutions as validated by chart review. When applied to PHIS-participating hospitals, 24 of 51 used these codes for ≥1% of patients admitted for bronchiolitis, likely highlighting significant hospital-level variation in billing rather than in HFNC use. At included hospitals, HFNC was associated with increased resource utilization.

Billing and coding practices differ across institutions, introducing a potential source of confounding.^20,21  For the 24 hospitals using billing codes we applied to capture relevant encounters, our methodology reliably identified children who did or did not receive HFNC. Twenty-seven hospitals were excluded for <1% utilization, yet a proportion of these have institutional clinical practice guidelines supporting HFNC on the general wards, suggesting more frequent use. Although PHIS codes used in this study cannot capture HFNC utilization at these hospitals, their exclusion increased homogeneity among included hospitals, improving the validity of our results among institutions that do use these codes. Future work is needed to examine billing practices to enhance generalizability of our results.

At the included hospitals, children who received HFNC had greater overall costs and longer LOS compared with those who did not receive HFNC. These differences remained after adjustment for clinical and demographic characteristics. The HFNC group had a smaller proportion of readmissions at 3 and 7 days, but by 14 days, this difference was no longer significant. Although ICU admissions for bronchiolitis have increased substantially in the last decade,²²  the differences we found in cost, LOS, and readmissions persisted within a subanalysis where patients requiring ICU care were removed. These findings highlight the complexities of balancing risks and benefits of HFNC. Although HFNC was associated with decreased likelihood of readmission within 1 week, this must be weighed against the possibility of greater costs during the index admission.

Patients treated with HFNC had greater APR-DRG SOI scores and incurred greater costs than those who did not receive HFNC. Even after adjusting for patient characteristics, including SOI, median costs remained $2510 greater in the HFNC cohort. This difference was driven by other costs, which includes room and board, and therefore are impacted by LOS. Cost for laboratory, medications, and imaging were also greater in the HFNC group, suggesting providers are more likely to order additional diagnostic tests or medications generally discouraged in the management of bronchiolitis. Differences in costs between the 2 groups remained even when examining costs per 24 hours, which suggests greater costs associated with HFNC are not because of increased LOS but because of factors unattributable to room and board. Greater daily cost was observed among the HFNC cohort in every service line. These findings differ from 2 previous studies, suggesting improved cost-effectiveness of HFNC compared with conventional oxygen delivery; however, those studies estimated costs by employing decision analysis models rather than verifiable cost comparisons.^14,15 

LOS was longer for the HFNC cohort. A 16-hour difference in LOS is significant given the already short standard for bronchiolitis. Of the prospective comparisons between HFNC and conventional oxygen delivery, only 2 have demonstrated significant differences in LOS favoring HFNC, though both had small sample sizes.^23,24  Larger prospective randomized studies did not find differences in time to wean off supplemental oxygen or in LOS.^8,25  Although we recognize that using retrospective administrative data to make assertions about clinical outcomes is subject to multiple sources of confounding, the longer LOS we observed in children treated with HFNC persisted after adjustment for APR-DRG SOI and after accounting for patients treated in the ICU, suggesting the influence on LOS cannot be fully explained by illness severity or treatment setting.

Interestingly, readmissions in the HFNC cohort were lower over the first week after discharge, despite greater APR-DRG SOI relative to the no-HFNC cohort. It could be those children presenting earlier in the disease course appeared less ill, and therefore were not treated with HFNC. These children may have been discharged from the hospital before the peak of illness, then readmitted within the following week. Considering the typical duration of bronchiolitis, this effect would not be expected at 14 days, matching what was observed in this data set. An alternative explanation for the lower readmission rate in the HFNC group relates to their longer LOS. Survey data have demonstrated a dearth of standardized practices across institutions for weaning HFNC.^26,27  Without guidelines, providers may be uncertain about when to wean or discontinue HFNC. If patients are discharged later in the course of illness, they may be less likely to require readmission.

Several limitations must be noted. APR-DRG SOI, used as a proxy for clinical illness severity, depends on diagnostic and procedural codes. Although APR-DRG SOI assignment is not directly affected by the billing codes used to identify HFNC use, it does not fully capture clinical illness severity, and future studies should seek to adjust for more clinically relevant measures, particularly at hospital presentation. The billing codes are only applicable to PHIS and can only be used to evaluate data from 2018 onward when these codes were first implemented. Finally, we only captured readmission to the same institution; thus, readmissions may have been underestimated if patients presented to different hospitals.

Among included hospitals, HFNC use based on PHIS codes was associated with greater resource utilization, including longer LOS and greater overall cost relative to no HFNC use. However, billing codes are inconsistently used among PHIS hospitals. Additional evaluation of hospital billing practices is needed to fully deploy these codes at a broader set of hospitals to investigate the value of HFNC as it relates to medical resource utilization and patient outcomes. A greater understanding of this risk-benefit balance will help providers make informed decisions regarding HFNC in the management of bronchiolitis.

We thank Isaac Jonas and Britney Byars for assisting with the data reports.