When to Transfer: Predictors of Pediatric High Flow Nasal Cannula Failure at a Community Hospital Free

We performed a retrospective cohort study of children aged 0 to 18 years who were hospitalized and started on HFNC in the community hospital emergency department or inpatient unit as a part of clinical care over 16 months (July 2018–October 2019). We excluded children with chronic medical conditions (ie, chronic lung disease, cardiac or neuromuscular disease). The project was reviewed by the university’s institutional review board and deemed exempt from institutional review board approval.

This study was conducted at a community hospital in the Mountain West with 14 inpatient beds and no PICU. The hospital is staffed by pediatric emergency medicine providers, pediatric hospital medicine providers, and a dedicated pediatric respiratory therapist. At night, the inpatient unit is staffed by a pediatric resident in-house or by a pediatric hospital medicine provider who is available over the phone. The community hospital does not have a colocated ICU, but is affiliated with a large quaternary care urban teaching hospital with an ICU ∼1 hour away by car. The community hospital and quaternary care center are at 5400 to 5800 feet of elevation.

Acutely ill patients with symptoms of lower respiratory tract disease were started on HFNC on the basis of hypoxia and work of breathing not responsive to LFNC or simple face mask according to our institution’s HFNC policy. On the basis of literature identifying elevated partial pressure of carbon dioxide (pCO2) as a risk factor for deterioration,^10,11  a venous, arterial, or capillary blood gas (most commonly venous) was performed at the time of HFNC initiation. Per hospital protocol, all patients were started on 100% fraction of inspired oxygen (FiO2) upon HFNC initiation, regardless of pCO2, and at maximum flow level allowable on the floor for their age. Per hospital protocol, maximum flow levels were determined on the basis of patient age (Supplemental Table 2). For example, the maximum allowable flow on the floor for a patient aged 12 months would be 8 L. These initiation settings were chosen to provide patients with the highest chance of improvement within a 2-hour period of observation. Settings are then titrated down as able on the basis of vital signs and work of breathing.

The institutional protocol called for patients to be automatically transferred from the community to the quaternary center (ie, interfacility transfer) if they met requisite demographic criteria (<6 months of age, <37 weeks’ gestation if <1 year of age, or with medical comorbidities), if pCO2 was >50 mmHg, or if they required respiratory support that surpassed the maximum age-based liters of flow (Supplemental Table 2), noninvasive ventilation, or mechanical ventilation. Patients could also be transferred at the provider’s clinical discretion (ie, if the provider felt that the patient’s condition was likely to escalate and need ICU-level care). Because HFNC failure was determined retrospectively and decisions made in real time could not predict the eventual patient outcome, not all transferred patients experienced HFNC failure. However, all patients with eventual HFNC failure were transferred because, by definition, they required ICU-level care, which was not available at the community hospital.

Our outcome of interest was HFNC failure, defined as need for ICU-level care at our institution for 1 of the following reasons:

1. receipt of HFNC that surpassed the maximum age-based liters of flow allowed on the floor (Supplemental Table 2);

2. noninvasive positive pressure ventilation (eg, continuous positive airway pressure, bilevel positive airway pressure); or

3. intubation and mechanical ventilation.

The primary outcome definition was unrelated to interfacility transfer, because not all transferred children ultimately needed ICU-level care (but all patients who needed ICU-level of care were transferred and met our definition of HFNC failure). Those who were not transferred and those who were transferred but remained at or below our maximum age-based liters of HFNC (even if they were transferred to the ICU) were categorized as “no HFNC failure.”

To address generalizability given other centers’ use of weight-based HFNC flow rates, a secondary analysis was performed redefining HFNC failure as any of the following:

1. receipt of HFNC >2 L/kg based on standard flow rates described in the literature^12,13 ;

2. noninvasive positive pressure ventilation (eg, continuous positive airway pressure, bilevel positive airway pressure); or

3. intubation and mechanical ventilation.

Weight-based flow was calculated using the highest HFNC flow rate identified after initiating HFNC therapy.

We collected patient demographics (eg, age, gestational age, race), and clinical characteristics (eg, asthma history, discharge diagnosis, days of illness, change in vital signs after HFNC initiation, minimum oxygen saturation [spO2], pCO2 result) from the electronic medical record (EMR). Although race is a social and not biological construct, it was included to identify potential health disparities by race that may suggest structural or institutional racism. Age categories (0–3 months, 3–6 months, 6–12 months) were created to help identify which age groups are at higher risk of failure. We collected vital sign data on the basis of previous work identifying worsening or persistently abnormal respiratory rate (RR) and heart rate (HR), with initiation of HFNC as potential predictors of failure.^14–16  Vital signs included RR and HR documented pre-HFNC initiation (vital signs taken right before initiation, up to 2 hours previous) and post-HFNC initiation (vital signs taken 1–2 hours after). We evaluated the change in HR and RR after HFNC initiation, and whether the post-HFNC vitals improved or remained normal (defined as <90th percentile for age) or worsened or remained abnormal (>90th percentile for age). For HR and RR, we created 2 mutually exclusive categories “improved or remained normal” and “worsened or remained abnormal” (Supplemental Fig 2). Diagnoses were identified using the discharge summary and classified as bacterial pneumonia, viral lower respiratory tract infection (LRTI) (eg, bronchiolitis, viral pneumonia), and/or asthma/reactive airway disease. Study subjects could have >1 diagnosis (eg, bronchiolitis and superimposed bacterial pneumonia).

Patients were identified within the EMR on the basis of HFNC use. All clinical data were extracted from the EMR by manual chart review. Demographic data were extracted from the EMR via an automated report.

In bivariable analyses, demographic and clinical factors were compared between those with and without HFNC failure using Pearson’s χ² or Fisher’s exact tests to compare proportions or a Wilcoxon rank sum test to compare medians. Multivariable analysis was performed using Poisson regression, with robust error variance to calculate the adjusted relative risk (aRR) of HFNC failure for each variable in the model. Poisson regression with robust error variance was chosen to allow us to calculate relative risks, rather than odds ratios, with improved accuracy of the SE with increasing outcome incidence.¹⁷  Variables with P values <.2 in bivariable analyses were included in the model. Weight was not included in the model because of collinearity with age. pCO2 was excluded from the multivariable model because very few (n = 6) patients had a pCO2 great>50, which is the threshold that previous studies have used to evaluate pCO2 as a predictor of HFNC failure.¹⁰  Furthermore, all patients with a pCO2 >50 failed, which created statistical issues with model performance because of 0 variability for the pCO2 covariate. There also was not a clinically relevant pCO2 cutoff identified via area under the curve analysis (Supplemental Fig 3), and pCO2 as a continuous variable was not considered because of fluctuating failure rates across quartiles of pCO2. We created a forest plot to compare the aRRs. Data were analyzed using SAS version 9.4 software (Cary, NC). All statistical tests were performed with a level of significance of α = .05.

In the cohort of 195 children, 65% (n = 126) were >12 months of age (range 0 months–8 years; median 17 months), almost all had a viral LRTI (96%, n = 188), and the median spO2 at presentation was 87% (Table 1). Some patients had a diagnosis of bacterial pneumonia (19%, n = 37) and asthma (26%, n = 51), often concurrently with a viral LRTI. In the overall cohort, 51% (n = 99) experienced HFNC failure. Fifteen percent (n = 15) of those who failed did so because of needing HFNC flow greater than allowable on the floor only, 82% (n = 81) failed because of the need for noninvasive ventilation, and 3% (n = 3) failed because of the need for mechanical ventilation. Seventy-five percent (n = 146) of all patients required interfacility transfer during the study period. HFNC failure occurred in 68% of patients who were transferred. Patient characteristics between those with and without HFNC failure are shown in Table 1.

In unadjusted analysis, when compared with the group without HFNC failure, those who did have HFNC failure were younger and more likely to have a diagnosis of asthma (32%, n = 32, in the failure group versus 20%, n = 19), an HR that worsened or remained abnormal (46%, n = 45, in the failure group versus 24%, n = 23), an RR that worsened or remained abnormal (72%, n = 71, in the failure group versus 48%, n = 46), and a pCO2 >50 (8% in the failure group versus 0%). The group with HFNC failure had a lower median weight (10.2 kg [interquartile range 7.1–12.0] vs 11.1 kg [interquartile range 8.9–12.8]). Additional clinical covariates were not significantly different between groups (Table 1). Although pCO2 was significant in bivariate analysis, there were only 6 patients with pCO2 >50 and 68 patients with pCO2 <50 who failed HFNC.

After adjusting for age, asthma codiagnosis, change in HR, and change in RR in multivariable analysis (on the basis of bivariable results in Table 1), the aRR of failure was higher in all age groups of patients aged <12 months as compared with those aged >12 months (Fig 1). Younger children (3–5 months of age) had an 85% higher risk of failure compared with older patients (aRR 1.85, confidence interval [CI] 1.34–2.54). Patients with a diagnosis of an asthma exacerbation had a 39% higher risk of failure compared with those without (aRR 1.39, CI 1.03–1.88). Patients whose RR deteriorated or remained abnormal had a 73% greater risk of failure as compared with those who did not (aRR 1.73, CI 1.24–2.41). Patients whose HR deteriorated or remained abnormal had a 47% higher risk of failure (1.47, CI 1.14 –1.90).

In secondary analysis performed to assess HFNC failure outcome defined as need for HFNC greater than 2 L/kg, noninvasive positive pressure ventilation or intubation and mechanical ventilation 43% (n = 84) of patients failed. There was no significant difference in risk of failure between these 2 groups. Predictors of failure were similar, except there was no longer a significant difference in risk of failure in the youngest age group (patients aged 0–2 months), likely because of small sample size (Supplemental Table 3).

In this study, we found that patients who were younger, had asthma, and did not have an improved RR or HR after HFNC initiation were more likely to experience HFNC failure. There was no relationship between viral LRTI diagnosis, bacterial pneumonia diagnosis, day of illness, or initial spO2 and HFNC failure. Although pCO2 >50 had a high specificity for identifying HFNC failure, it was rare, and many patients with pCO2 <50 failed HFNC (ie, poor sensitivity), rendering pCO2 an unreliable screening tool in our cohort. The secondary analysis yielded very similar results, suggesting that predictive values for HFNC failure are likely to be applicable regardless of HFNC escalation policies. Our study is one of the first studies to examine the use of HFNC at a community hospital site without an on-site PICU in the United States.

Our findings are consistent with previous literature in other settings which identified that a reduction in tachycardia^14,16  and/or improvement in RR^15,16  after HFNC initiation has been shown to be associated with decreased rates of ICU transfer and intubation. Contrary to previous literature, we did not find any relation between initial spO2 and HFNC failure.⁹  Previous literature has also demonstrated that patients with a higher FiO2 requirement are also more likely to fail.⁸  One similar study describing the use of HFNC at a community site in Canada¹⁸  found that patients with increased FiO2 requirement and respiratory syncytial virus (RSV) positivity were at higher risk of failure; however, the population was significantly older (average age 3.41 years versus 17 months in our study). FiO2 data were not included in our study because all patients were started on 100% FiO2 upon HFNC initiation per hospital protocol. Studies have also identified the presence of RSV as an additional risk factor for HFNC failure.¹⁸  RSV status was not included in our study because respiratory testing was not routinely sent in all patients presenting with bronchiolitis, consistent with American Academy of Pediatrics guidelines, which advise against laboratory testing in routine care of patients with bronchiolitis.¹⁹ 

Previous studies have indicated that initial elevated pCO2 levels were associated with failure of HFNC.^10,11  However, unlike those studies, pCO2 was not a helpful predictor of failure. pCO2 was found to a be a poor marker of failure risk given low sensitivity, and because of these findings, obtaining pCO2 is no longer a part of our hospital protocol. Providers should consider the utility of obtaining pCO2 values in patients on HFNC, if purely to help stratify risk, because many patients who ultimately fail have a normal pCO2. In this scenario, obtaining pCO2 is not aligned with providing high-value care.

Early identification of risk factors for HFNC failure is critical, particularly for community hospitals. Not transferring a patient who needs higher level of care until they are critically ill and needing an urgent transfer raises concerns for patient safety and is distressing to families. On the other hand, unnecessary transfers, which are not uncommon from community sites, are also distressing to families, costly, and do not maximize the potential of community hospitals.^20–22  Current literature on pediatric transfer of high-acuity patients, such as those with acute respiratory insufficiency or failure, is lacking. Literature on low-acuity pediatric patients suggests that unnecessary transfers are costly and many come from community sites. In the adult population, interhospital transfer has been shown to be especially distressing to families when the timing of the transfer is unexpected.²³  The results of this study are poised to guide community hospitals in creating a transfer protocol that would allow for timely transfer to both reduce unnecessary transfer and facilitate timely transfer when necessary.

Overall, the results of this study help to provide general guiding principles to frame HFNC protocol development, such as monitoring of HR and RR within 2 hours of HFNC initiation to assess response to HFNC treatment and consideration of age and asthma status, which can be tailored to the unique environment of each hospital (eg, staffing model, proximity to ICU). Immutable characteristics, such as age and asthma status, should be considered in conjunction with vital sign data. Given the higher rate of failure in younger children and those with asthma, hospitals might have a lower threshold to transfer these patients if their vital signs are borderline. Creating an effective transfer protocol that identifies patients who are unlikely to be stabilized with floor-appropriate levels of HFNC is important to facilitate safe transfer before further deterioration and to avoid costly unnecessary transfers of patients who could have stayed at the initial hospital. Additionally, a protocol helps to provide clarity for and between providers, including emergency department providers and hospitalists, outlining objective data that do or do not support the need for transfer.

This study is subject to a few limitations. This was a retrospective study performed at a single community hospital site staffed by pediatricians, which may limit the generalizability of our findings. HFNC initiation guidelines at our institution are vague and largely left to provider clinical discretion, which could also limit generalizability; however, this is also the case at the majority of institutions nationwide.²⁴  This study did not include variables found to be significant in other studies (RSV status, FiO2 requirement) given practice differences in obtaining this data and HFNC initiation policies. However, this limitation is somewhat diminished because national guidelines recommend against RSV testing. Although the majority of patients had a viral LRTI, there were patients with additional diagnoses (often superimposed on a viral LRTI), such as asthma and bacterial pneumonia. These patients may have been receiving additional treatments such as bronchodilators or steroids, which may have had an impact on their vital signs, thus confounding the determination of their response to HFNC. Additionally, our definition of HFNC was specific to our maximum age-based allowable flow rates on the floor, which is specific to our institution. We have addressed this limitation by conducting a secondary analysis of failure rates with weight-based flow maximums to improve generalizability to other settings. In our secondary analysis, predictors of failures were similar; however, patients aged 0 to 2 months had similar risk of failure as compared with patients aged >12 months, suggestive that low flow rates in our study may be contributing to higher rates of failure in the youngest population. Finally, given literature demonstrating HFNC overuse, it is possible that some patients were inappropriately put on HFNC who would not benefit from the effects.²⁵  However, we believe that our population reflects patients who are routinely put on HFNC in everyday clinical practice, and thus our findings can be used to identify patients at risk (or not at risk) for HFNC failure.

In summary, we found that patients who are younger, have asthma, and do not show improvement in RR or HR after HFNC initiation were more likely to experience HFNC failure in a cohort of children who required HFNC at a community hospital. Future studies should expand upon this work to evaluate factors associated with HFNC failure across multiple sites to facilitate the development of a generalizable HFNC policy that allows children to safely receive HFNC, avoid unnecessary transfers, and optimize high-value care.

We thank Benjamin Bernier, Sarah Moultrie, and Kathryn Walsh, MD, for their role in chart review and data collection for the data used in this study.