Improving Length of Stay by Reducing High-Flow Nasal Cannula Duration in Respiratory Illnesses

High-flow nasal cannula (HFNC) has been proposed as an intermediate level of respiratory support between standard oxygen therapy and noninvasive ventilation that can improve oxygenation and ventilation when used in a weight-based, dose-dependent fashion.^1–4 Despite multiple randomized control trials (RCTs) demonstrating that early, routine use of HFNC in the treatment of bronchiolitis does not reduce intubation rates or length of stay (LOS), HFNC use remains common.^5–8 Widespread adoption of HFNC in bronchiolitis has paradoxically been associated with increased intensive care use, increased health care costs, and prolonged LOS.^9–12 The routine use of this nonevidence based therapy that may increase costs without improving patient outcomes is an example of low-value care and has resulted in calls for deimplementation.^13–17 

Prior improvement initiatives incorporating HFNC weaning and rapid discontinuation with trials off HFNC (Holidays) have proven successful at reducing length of treatment (LOT) and LOS in bronchiolitis.^18–22 Recent studies have suggested that HFNC is commonly used in similar and often overlapping respiratory conditions such as pneumonia and asthma.^23–27 Similar to bronchiolitis, prior studies have suggested that the use of HFNC in similar respiratory conditions prolongs LOS without improving patient outcomes, representing a need for improvement efforts extending beyond bronchiolitis.^28,29 

At our institution, HFNC was used for a variety of respiratory conditions with varying initiation and weaning practices. The use of non–weight-based, subtherapeutic flow rates was common. Prolonged HFNC duration, and, therefore, prolonged LOS, prompted the formation of a multidisciplinary improvement team focused on improving the use of HFNC in children with bronchiolitis, pneumonia, or asthma in the emergency department (ED), general pediatric floor, and pediatric intensive care unit (PICU). Using quality improvement (QI) methodology, we sought to decrease HFNC LOT and LOS by 10% within 1 year of implementation.

Our improvement efforts took place at a 364-bed, quaternary-care, freestanding children’s hospital. HFNC was used to treat a variety of respiratory conditions at our institution, including bronchiolitis, pneumonia, and asthma, with the majority of HFNC used in the ED and PICU. Care team models included teaching teams with trainees, advanced practice providers, and attending physicians or nonteaching teams with attending physicians. All care teams consisted of pediatric registered nurses (RNs) and pediatric respiratory therapists (RTs).

At the start of our project, weaning occurred at the discretion of providers, and patients were typically weaned to 2 to 4 L/min of HFNC prior to discontinuation regardless of age or weight. During the fall of 2022, a surge in respiratory illnesses constrained bed availability, prompting institution-wide LOS efforts. As HFNC was identified as causing bottlenecks impacting throughput, HFNC use was targeted for improvement. Compared with previously reported HFNC use rates in bronchiolitis,⁸ our institution was noted to have a relatively low use rate. Therefore, we chose to primarily focus on HFNC weaning as opposed to initiation because we felt that this would be most impactful in our goal to reduce LOS.

In October 2022, with the support of our institution’s Center for Quality and Safety, we convened a multidisciplinary improvement team consisting of clinicians, RNs, and RTs from the ED, PICU, and inpatient floor in addition to QI experts, data scientists, and clinical informaticists. Through electronic surveys from front-line staff and improvement team discussions, we developed an Ishikawa cause-and-effect diagram (Supplemental Figure 1) and a key driver diagram (Figure 1) to identify key drivers of prolonged HFNC use.

To best target children who would benefit from standardized HFNC use, we included patients aged 1 month to 5 years receiving HFNC in the ED, general pediatric floor, or PICU for bronchiolitis, pneumonia, or asthma. This population was selected because it represented 85% of our institution’s HFNC use and represented patients who were most likely to receive HFNC for an acute respiratory illness in which standardized practices were most likely to improve LOS. Patients with comorbidities that could impact their respiratory trajectory, including prematurity, underlying lung or cardiac disease requiring baseline medications or respiratory support, anatomic airway defects, neuromuscular weakness, immunodeficiency, or severe asthma requiring continuous albuterol, were excluded.

The primary intervention was the development and implementation of an HFNC weaning algorithm and corresponding clinical decision support. Plan-Do-Study-Act cycles were used regularly throughout the improvement phase to guide the implementation and refine our interventions (Supplemental Table 1).

As nationally accepted guidelines for HFNC initiation and discontinuation are lacking, our improvement team conducted a literature search with a biomedical librarian to help inform the development of an algorithm. The team reviewed published literature regarding HFNC use and came to consensus on recommendations to incorporate in the algorithm.

The resulting evidence-informed, consensus-based algorithm included an initiation and weaning algorithm (Supplemental Figure 2). The initiation algorithm was developed to standardize the use of weight-based flow rates and allow for standardized starting points for weaning. The weaning algorithm incorporated weight-based rapid weaning and discontinuation of HFNC at weight-based flow rates with a trial on room air, referred to as a Holiday.

The progression of rapid weaning and Holiday by weight category is outlined in Table 1. As the use of subtherapeutic, non–weight-based flow rates was identified as a key driver to prolonged LOT, our algorithm recommended potentially therapeutic weight-based flow rates. Although the definition of therapeutic flow rates has not been conclusively established, weight-based flow rates of 1 to 2 L/kg in the bronchiolitis population are frequently referenced.^1–7,28 As our population included patients up to age 5 years with higher weights, and acknowledging that the dose-dependent impact of HFNC does not continue linearly with increasing weights indefinitely, we chose to stratify our flow rate recommendations by weight category. For patients weighing greater than 15 kg, flat flow rates representative of approximate weight-based flow rates were used for simplicity to improve adherence. These weight-based flow rates were inspired by a consensus-based weaning algorithm from prior RCTs studying the clinical effectiveness of HFNC in acutely ill children of various ages with varying respiratory illnesses.^3,30–32 

Once a patient met weaning eligibility, patients underwent rapid weight-based weaning in 1 or 2 steps, followed by a Holiday. Patients were eligible for a Holiday once they were tolerating the lowest weight-based flow rate for their weight category, referred to as a wean flow rate. A Holiday involved a 60-minute trial on room air in which a patient was monitored for changes in respiratory rate, work of breathing, and oxygen saturation. Holiday monitoring was a shared responsibility between RNs and RTs and consisted of a 10-minute period at bedside with a revaluation at Holiday conclusion. If respiratory status remained stable, patients remained off HFNC. If patients developed hypoxemia without increased respiratory distress, patients were trialed on supplemental oxygen via low-flow nasal cannula. If patients developed increased respiratory distress, the patient resumed HFNC at the last tolerated settings.

In March 2023, we released our HFNC algorithms to the ED and hospital medicine teams. In May 2023, the algorithms were expanded to critical care staffed teams.

Following algorithm consensus, corresponding standardized orders were created and released with the algorithms. Order sets included order panels for initiation, weaning, and Holidays in addition to communication orders encouraging RNs and RTs to notify providers once patients met weaning or Holiday eligibility. Specific responsibilities of team members during the weaning and Holiday were outlined within order comments (Supplemental Table 2). Electronic Health Record (EHR) respiratory flowsheets were modified to allow documentation of Holiday occurrence and outcomes.

Prior to algorithm release, the improvement team completed education with providers, RNs, and RTs. Educational topics included respiratory distress, HFNC physiology, overuse examples, and details of the newly created algorithms. Education was distributed to providers, RNs, and RTs through educational sessions, RT/RN huddles, hospital-wide and division presentations, and emails. During the beginning of the respiratory viral season, staff refresher educational sessions were held. To allow easy, continued access to educational materials, all materials, including educational sheets, recorded presentations, and presentation materials, were uploaded to a dedicated HFNC page on our institution’s internet.

Updated data for primary outcome measures and process measures were shared monthly with improvement team members and at physician team meetings. Distributed information included key themes identified from monthly data and areas for improvement. To obtain qualitative feedback, in-person, real-time feedback was solicited from care team members on HFNC units, and electronic surveys were distributed to care team members who were identified as having cared for eligible patients on HFNC through medical record review.

Data were collected and grouped by baseline and improvement phase relative to the release of the algorithm March 2023. During the improvement phase, data were extracted from the EHR monthly. Children between age 1 month and 5 years with an ED, observation, or inpatient encounter with an HFNC order and an International Classification of Diseases, Tenth Revision code for bronchiolitis, pneumonia, or asthma within the first 3 diagnoses (see Supplemental Table 3 for full list of codes) were included. To align with the algorithm eligibility criteria, patients meeting algorithm exclusion criteria were excluded from analyses. To ensure consistency in LOS and LOT calculations, patients transferred from another hospital were excluded.

The primary outcome measures were mean HFNC LOT and LOS. HFNC LOT was calculated as the difference in time (in hours) from the first to last recorded HFNC rate. LOS was calculated as the difference in time (in hours) from ED arrival to discharge. We chose to use ED arrival time rather than inpatient admission time to capture clinical care delivered in the ED and remove the impact of delays on inpatient transfer time during periods of high census. Process measures included rapid weaning adherence, Holiday occurrence, and mean last weight-based flow rate at HFNC discontinuation. Holiday occurrence was defined as the last HFNC flow rate equal to the algorithm-specified flow rate for weight. Balancing measures included the need for positive pressure ventilation (PPV) following a Holiday occurrence, 3-day all-cause readmission rate, and time to discharge from HFNC discontinuation. To ensure the introduction of an algorithm did not increase use, HFNC use rate was also included as a balancing measure. Supplemental Table 4 provides detailed measure definitions.

We used statistical process control (SPC) X-bar and S charts to analyze our primary outcomes, our process measure of last weight-based flow rate, and our balancing measure of time to discharge. We used an SPC p-chart for our balancing measure of HFNC use rate. We employed established rules for identifying SCV.³³ We used a run chart to monitor Holiday occurrence frequency and postimplementation rapid weaning adherence. We employed established rules for identifying nonrandom signals of change.³³ Run charts and SPC charts were generated using QI Charts (version 2.0.23, Process Improvement Products) and Microsoft Excel 2021 MSO (version 2310 build 16.0.16924.20054, Microsoft Corporation). Our balancing measures of need for PPV following a Holiday and 3-day all-cause readmission rate were monitored but not displayed on SPC charts given the rarity of the event.

This QI work was reviewed by our institutional review board and was deemed not to be human subject research and was therefore exempt from review (2023–5859, approved May 12, 2022).

A total of 430 patient encounters were included, with 281 in the baseline phase and 149 in the improvement phase (Supplemental Table 5).

SCV was noted, with mean HFNC LOT decreasing from 55.7 hours during the baseline phase to 39.7 hours following the release of the algorithm and sustained through the improvement phase (Figure 2). SCV was noted in mean LOS, which decreased from 94.1 hours in the baseline phase to 70 hours (Figure 3).

During the improvement phase, rapid weaning occurred for 45.4% of patient encounters (Supplemental Figure 3). SCV was noted for last weight-based flow rate, with mean flow rates increasing from 0.5 L/kg/min during the baseline phase to 0.9 L/kg/min (Figure 4A). For Holiday occurrences, a shift was noted, with the median frequency increasing from 10.9% during the baseline phase to 82.8% (Figure 4B).

No patients required PPV following a Holiday. There was 1 3-day all-cause readmission in the baseline phase and 1 in the improvement phase. No SCV was noted in time to discharge from HFNC discontinuation (Supplemental Figure 4). Our HFNC use rate was 18% throughout our study (Supplemental Figure 5).

Our work builds on prior studies using QI efforts to de-implement HFNC through weaning protocols,^18–22,34 but, to our knowledge, we are the first to use QI methods to discontinue HFNC at weight-based flow rates in acute respiratory illnesses in patients up to age 5 years. Our improvement interventions were associated with a reduction in HFNC LOT by 16 hours (29%) and LOS by 24 hours (25%), surpassing our initial aim. There was no concurrent increase in readmission rates, and no patients required PPV after a Holiday.

We observed a reduction in LOT shortly after the release of the algorithm. Although we did not observe a notable increase in Holiday adherence until later in our intervention, the reduction in LOT correlated with HFNC discontinuation from higher weight-based flow rates. We suspect that it took time for providers to become comfortable discontinuing HFNC at the algorithm’s high weight-based flow rates. We ultimately saw a substantial increase in Holiday adherence following monthly feedback, emphasizing the success of Holidays performed. Additionally, our interventions occurred at the conclusion of a respiratory season. This may have contributed to our success by allowing staff time to acclimate to the algorithm. By the onset of the respiratory season, our staff was more effective at following the algorithm, as evidenced by an increase in rapid weaning frequency and Holiday occurrences.

Our findings of reduced LOT and LOS contrast with the recent multicenter study by Byrd et al⁸ examining HFNC use, which did not find a significant difference in LOT or LOS between hospitals with and without a rapid HFNC weaning protocol. Weaning algorithms have been shown to vary widely,²⁶ and this difference may suggest that weaning algorithms can be effective tools if incorporating weight-based rapid weaning and Holidays. Indeed, our findings add to prior studies that have shown that weaning algorithms can be effective at reducing LOT and LOS.^{18,19,21,22,34,35} 

Although Noelck et al¹⁹ and Wiser et al³⁴ demonstrated decreases in LOS without LOT with the introduction of a weaning algorithm, our findings align with Charvat et al²¹ and Hoefert et al,¹⁸ which demonstrated reductions in both LOT and LOS. Weaning algorithms impacting LOT are important in providing patient-centric care because HFNC is associated with negative impacts on patient-reported outcomes, including disruptions in caregiver bonding, caregiver holding, and breastfeeding.³⁶ Furthermore, rapid discontinuation of HFNC furthers the delivery of high-value, patient-centric care by reducing the use of potentially subtherapeutic flow rates contributing to medical waste.

Previous QI studies implementing weaning algorithms have mostly recommended rapid discontinuation at non–weight-based, potentially subtherapeutic flow rates and have largely focused on bronchiolitis.^18,19,21,34 Similar to the recent QI collaborative by Byrd et al,²² our algorithm endorsed Holidays at higher flow rates than prior quality work. Our work builds on this by demonstrating the success of QI methods to de-implement HFNC in a broader patient population. Although the advocacy for judicious HFNC use and de-implementation efforts have centered on bronchiolitis,^18,19,21,22 HFNC use in similar and often overlapping acute respiratory illnesses offers an opportunity for broader de-implementation efforts and may curb potential overuse.^23–26 

Our study has several limitations. First, this project was completed in a single-center, quaternary-care pediatric hospital with support from our institution’s Center for Quality and Safety and in alignment with institution-wide LOS efforts, which may limit its reproducibility in other settings. Second, the institution-wide efforts to decrease LOS may have impacted our observed outcomes outside of our interventions. However, as there was no change in time to discharge from HFNC discontinuation, we suspect our decrease in LOS was related to our improvement efforts to decrease LOT. Third, our project included the ED, general pediatric floor, and PICU, which allowed for a shared mental model across teams and more seamless transitions of care. This may limit generalizability for settings with unit-specific algorithms. Fourth, although we included patients with asthma, only a small proportion of our patient population had a primary diagnosis of asthma during the improvement phase (3.4%). Additionally, although we included patients up to age 5 years, only 10% of our patient population during the improvement phase were older than age 2 years. Finally, our institution’s HFNC use for bronchiolitis is lower than what has been reported in national averages.⁸ This may have been a factor in our success, as a culture of high use has been shown to be a specific barrier in other de-implementation work.³⁷ Settings with higher use rates may need additional interventions in their de-implementation efforts.

Although our work focused on robust weaning, limiting initiation is an important aspect of HFNC de-implementation. Recent studies by Treasure et al³⁸ and Byrd et al²² demonstrated the success of applying QI methods to reduce HFNC initiation, and applying similar efforts to reduce HFNC initiation in a broader range of acute respiratory illnesses is warranted. Although our work included a broader population than bronchiolitis, our exclusion criteria often resulted in the exclusion of children with medical complexity (CMC). CMC are an important group of hospitalized patients who should also benefit from HFNC de-implementation efforts. However, this patient group may have specific and unique needs given underlying conditions that may impact their respiratory trajectory. Future directions of this work include determining the safety of efficacy of weight-based HFNC weaning algorithms in this unique population.

Through this QI initiative, we reduced mean HFNC LOT and LOS through the implementation of a weaning algorithm. Our findings demonstrate the feasibility and safety of implementing a weaning algorithm incorporating rapid weaning and discontinuation of HFNC through Holidays at weight-based flow rates in children with acute respiratory illnesses up to age 5 years.

CMC

children with medical complexity

ED

emergency department

EHR

electronic health record

HFNC

high-flow nasal cannula

LOS

length of stay

LOT

length of treatment

PICU

pediatric intensive care unit

PPV

positive pressure ventilation

QI

quality improvement

RN

registered nurse

RT

respiratory therapist

RCT

randomized control trial

SCV

special cause variation

SPC

statistical process control chart.