High-Flow Nasal Cannula: The Challenge of Studying What Cannot Be Measured Free

As the number 1 reason for hospitalization of infants, bronchiolitis is a commonly encountered and studied pediatric illness.¹  Despite decades of research, the recommended treatment remains supportive care.¹  Delivery of heated and humidified air through a high-flow nasal cannula (HFNC) to support work of breathing has become common after initial single-center observational studies suggested HFNC might provide some positive pressure and reduce mechanical ventilation rates.^2–4  However, these studies were poorly adjusted for confounding factors and focused mainly on outcomes in patients already in the ICU, rather than on clinical outcomes when HFNC was used outside the ICU.^2–4  Since that time, randomized controlled trials comparing early use of HFNC to standard nasal cannula failed to demonstrate improvements in clinical outcomes.^5–7  Furthermore, the widespread adoption of HFNC outside the ICU has coincided with paradoxical increases in ICU utilization, noninvasive ventilation rates, and costs among hospitalized patients with bronchiolitis.^8–10  The inability to identify HFNC use in administrative data sets remains a significant limitation, hindering our ability to investigate these population-level associations and understand the impact of HFNC on patient outcomes and resource utilization in actual practice. In this issue of Hospital Pediatrics, Biggerstaff et al aimed to overcome this challenge using data from the Pediatric Health Information System (PHIS) database.¹¹ 

Specifically, the authors aimed to develop and validate a method for identification of HFNC use in the PHIS database and then use their method to compare outcomes for hospitalized patients with bronchiolitis who did and did not receive HFNC. High-flow nasal cannula use was defined by the presence of 1 of 2 clinical transaction category (CTC) codes, representing the supplies and services necessary for HFNC delivery. This method was validated using chart review at 2 hospitals and demonstrated a sensitivity of 90.4% and specificity of 99.3%. However, 26 of 51 PHIS hospitals failed to report these codes for any patients hospitalized with bronchiolitis, whereas 1 single hospital reported a HFNC utilization rate of less than 1% and was excluded. After exclusion of these 27 hospitals, HFNC utilization rates at the remaining 24 hospitals ranged from 1.7% to 74.0% of all patients hospitalized with bronchiolitis. Within hospitals that used the CTC codes, Biggerstaff et al found that patients treated with HFNC had greater total hospitalization costs, greater daily costs, and longer total hospital length of stay (LOS) after adjusting for clinical and demographic characteristics.¹¹  Children treated with HFNC experienced lower 3- and 7-day readmission rates; however, the authors speculate that this was due to their longer LOS, rather than a direct effect of HFNC, given that 14-day readmission rates were similar between the 2 groups.

While identification of HFNC in the PHIS database would greatly improve researchers’ ability to study resource utilization and outcomes in hospitalized children with bronchiolitis, this study has several notable limitations to consider. First, although the selected CTC codes demonstrated high sensitivity and specificity at the 2 hospitals at which the codes were validated, HFNC use at those institutions was identified in 46% of children hospitalized with bronchiolitis. In contrast, it seems unlikely that this identification strategy would have performed similarly well had validation occurred at the hospital with an apparent utilization rate of 1.7%. Given the widespread use of HFNC for bronchiolitis at tertiary children’s hospitals like those included in the PHIS database, the assumption that a utilization threshold of greater than or equal to 1% signified reliable reporting of the relevant CTC codes lacks face validity and likely introduces significant misclassification bias. Second, only 24 of 51 participating PHIS hospitals were included in this study. As the PHIS database represents only 20% of pediatric hospitalizations in the United States, reducing the cohort further significantly limits the generalizability of the study’s results. In addition, this is compounded by the reliance on CTC codes, which although based on billing data, are created and maintained solely for use in the PHIS database. Finally, as with any observational study, there are likely underlying differences between patients treated with HFNC and those who were not. The authors attempted to account for these differences by adjusting for patient characteristics including age and insurance status, and by using All Patient Refined-Diagnosis Related Group Severity of Illness index scores as a proxy for disease severity. However, adjusting for All Patient Refined-Diagnosis Related Group Severity of Illness index score is problematic as it is highly reliant on diagnosis and nonoperating-room procedure codes, which may be affected by the use of HFNC, a limitation acknowledged by the authors.¹⁰  Therefore, it is possible that the significant association observed between HFNC use and higher resource utilization actually represents an underestimate of the true effect given the potential for overadjustment due to coding bias. Taken together, these limitations make interpretation of the study’s results challenging, especially given the questionable ability to accurately measure the primary exposure.

Whereas there was a great deal of optimism that HFNC would improve clinical outcomes in patients with bronchiolitis (eg, improved ICU transfer and mechanical ventilation rates, decreased length of time on oxygen therapy, and decreased hospital LOS), 3 high-quality randomized control trials have failed to demonstrate improved outcomes apart from possible benefit when HFNC was used as a rescue therapy in the minority of patients who failed initial treatment with standard oxygen therapy.^5–7  And whereas studies have shown that HFNC is generally safe, there are other potential harms to consider with its widespread use, including the costs of equipment, additional nursing and respiratory therapist tasking and monitoring, delays in oral feeding, and increased ICU utilization with its spread onto the general wards.⁸  Biggerstaff et al’s results align with existing literature on the association between HFNC and more intense resource utilization, and importantly, the authors found further evidence of widespread adoption of HFNC in bronchiolitis, with the therapy used in 46% of patients at the validation sites and up to 74% of patients at 1 institution.¹¹  Although bronchiolitis severity could not be reliably assessed in this study, such high utilization suggests that many children with mild to moderate bronchiolitis are being preemptively treated with HFNC rather than reserving it as a rescue therapy. These findings highlight the importance of local and national deimplementation efforts, as HFNC use in patients with mild to moderate bronchiolitis likely represents low-value care.

Like steroids, albuterol, and hypertonic saline before it, HFNC is the latest therapy for bronchiolitis to be implemented despite little high-quality evidence supporting its use. Luckily, as evidence accumulated showing these earlier therapies to be ineffective for most children with bronchiolitis, researchers and quality improvement experts were able to leverage administrative databases for benchmarking and deimplementation efforts.^12,13  However, the widespread adoption of HFNC poses a unique problem, because our ability to track its use at scale is limited. Furthermore, efforts such as the ongoing HI-FLO (High-flow Interventions to Facilitate Less Overuse) project through the AAP Value in Inpatient Pediatrics Network, which aims to decrease the initiation and hasten weaning of HFNC in bronchiolitis, would be less administratively burdensome and more sustainable if data were readily available in administrative databases. Given the potential implications for health care spending and patient outcomes, it is no surprise that identification of HFNC within large administrative databases has become a research target.

Given the current uncertainty around the benefit of HFNC, multiple routes of study are needed. First, high-quality prospective studies are necessary to identify which subset of children hospitalized with bronchiolitis are likely to benefit from HFNC and establish criteria for when initiation of HFNC should take place. These should focus not only on “hard” outcomes (eg, mechanical ventilation rates) but also “soft” outcomes (eg, patient comfort), as these “soft” outcomes are often cited to support its use. Second, large administrative database studies are needed to track, understand, and evaluate the use of HFNC in actual clinical practice across a wide range of settings. While these observational studies are limited in their ability to determine efficacy, they serve an essential role in quality measurement and in encouraging high-value care at the population level. Furthermore, accurate assessment of severity of illness will be paramount if we are going to be able to accurately compare outcomes in patients treated with different respiratory support modalities. By attempting to identify HFNC using billing data, Biggerstaff et al provide an admirable first step toward this goal, and in doing so, demonstrate several of the challenges within the current system.¹¹  We hope that others respond to this challenge and continue to build on their important work.