A randomised trial of high-flow nasal cannula in infants with moderate bronchiolitis

Discussion

In this multicentre randomised controlled trial involving infants with moderate bronchiolitis admitted to PEDs or inpatient wards, there was no evidence of a lower rate of failure leading to noninvasive ventilation support in patients receiving high-flow oxygen therapy compared to the control group. There was no significant between-group difference in the rate of ICU admission, while a marginal benefit of HFNC was observed for short-term respiratory parameters or length of oxygen therapy. However, in the HFNC group, three device-associated air leak syndromes were reported.

Our findings are partially supported by the results of two recent randomised trials, which found no difference in ICU admission rates between the two strategies [14, 15].

Regarding the observation that the time to wean off oxygen favoured the HFNC group, this difference may be considered irrelevant, consistent with the negative results reported in the two previously published RCTs (i.e. no significant difference in length of oxygen support between the HFNC and control group).

The first single-centre RCT was designed to demonstrate a reduction in the time to wean off oxygen. No difference was found between the two groups for the primary outcome or in the proportion of patients transferred to the PICU. However, although the percentage of children who experienced treatment failure was lower in the HFNC group (14% compared to 33% in the standard therapy group; p=0.0016), the study was underpowered for this secondary end-point. Finally, the relatively low flow setting of 1 L·kg⁻¹·min⁻¹ in the HFNC group, the low mean m-WCAS score (compared to our data) and the rate of crossover in the standard group raised concerns about generalising these findings to other wards [14]. The crossover rate makes it difficult to draw definitive conclusions regarding the usefulness of high-flow oxygen therapy in very low-severity forms of bronchiolitis. The second study, a large multicentre RCT, aimed to compare HFNC (flow setting of 2 L·kg⁻¹·min⁻¹) to standard therapy with the primary outcome as the rate of escalating therapy, which was defined as a heterogeneous composite failure criterion including meeting an early warning sign-driven protocol, admission to the ICU and/or crossover to HFNC in the control patients. Despite a significantly higher rate of failure-free days in favour of the HFNC group, neither the proportion of patients admitted to the ICU nor the number of oxygen-free days were found to be significantly different. Moreover, the number of patients who underwent noninvasive ventilation in the failure group was unknown [15]. However, the clinical benefit highlighted in these two RCTs (i.e. the proportion of failure-free days in favour of HFNC) is consistent with our observed short-term improvements in respiratory rate, m-WCAS score or mean P_tCO2 at failure in the HFNC group. These findings are consistent with an extensive literature focused on the physiological benefits of HFNC in infants and adults, which stress the benefit of reduced work of breathing.

Finally, several concerns have been noted regarding cost-effectiveness in terms of the high rate of crossover in the control group, as this indicates bias of the attending physicians toward HFNC as a beneficial therapy at the time of crossover, even though the evidence for the benefit of HFNC in “moderate” bronchiolitis (i.e. all patients admitted to the ward and requiring oxygen to target a S_pO2 level of 92–98%) remains to be established [20].

It is worth noting that our ∼10% inclusion rate of total infants admitted to the PED with bronchiolitis as a primary diagnosis is quite similar to that in the study by Franklin et al. [15]. Interestingly, our failure rates in the HFNC and control groups (20% and 14%, respectively) are comparable to their subgroup of patients in recruiting institutions with an on-site PICU. However, neither the inclusion and/or failure criteria nor the primary end-point were similar between the two studies, which makes these comparisons much more difficult, especially because our pragmatic “real-world” trial was not designed to explain these differences.

Indeed, our study included infants aged <6 months during two consecutive winter epidemic outbreaks in order to decrease the risk of including infant asthma patients and to target subgroups at higher risk for admission to the ICU, in contrast to previous RCTs [1, 2, 21, 22]. Furthermore, we chose an escalating respiratory support requirement (mainly with nCPAP) as a pragmatic judgment criterion because it is currently considered standard first-line treatment for severe cases admitted to the PICU in most developed countries, despite a lack of evidence from RCTs [4, 7, 12]. Indeed, we and others suggest that avoiding admission to the PICU or noninvasive support are more relevant end-points, as both are associated with the PICU seasonal burden, substantial complications and higher costs [23]. Finally, some features are likely to reduce interpretation bias, including the fact that the design did not allow crossover between the groups and the absence of a PICU in all but one recruiting centre.

Although the study did not use a crossover design, 10 infants from the control group were escalated to HFNC (nine in the ICU; one remained in the ward) (table 2). Once the patient met the failure criteria in each group, indicating release from the study, physicians were not able to both escalate and remain in the randomisation arm, meaning that the patient had to be admitted to intermediate level care or the ICU. It should be noted that the PICU team caring for infants in the failure group was not involved in the study design. Thus, the choice to escalate support was entirely at the entire discretion of the paediatric intensivist, which is why several patients who failed in the control arm received support with HFNC after their ICU admission.

We chose a 3 L·kg⁻¹·min⁻¹ flow rate in our study given that a previous physiological study suggested that maintaining a pharyngeal pressure-to-flow relationship above 2 L·kg⁻¹·min⁻¹ helps reduce work of breathing [24]. Moreover, a similar flow rate close to or above 3 L·kg⁻¹·min⁻¹ has also been used in two other RCTs that included premature newborns after extubation [25, 26]. We could not rule out worsened work of breathing in some patients due to individual excess inflow rate and/or discomfort, which would mask a potential benefit in the HFNC group. This is consistent with a recent RCT, which indeed suggested that a 3 L·kg⁻¹·min⁻¹ flow rate did not reduce the risk of treatment failure compared to the 2 L·kg⁻¹·min⁻¹ arm in severe bronchiolitis [16]. However, the short-term improvement in respiratory rate or m-WCAS score in the HFNC group is consistent with previous literature evaluating the physiological benefits of HFNC in infants and adults [24, 27–29].

Serious, unexpected adverse events encountered in the HFNC group are a matter of concern, especially because potentially serious air-leak syndromes have previously been reported with high-flow oxygen therapy devices [19, 30]. We propose several hypotheses, including nasal prong sizes that are unable to provide sufficient nostril leakage in some patients (unfortunately, a fixed size apparatus was used for the entire study group), incorrect pacifier use and/or an excessive fixed-flow rate setting at 3 L·kg⁻¹·min⁻¹ in our study.

The limitations and weaknesses of our study include the fact that the median m-WCAS score was slightly but significantly higher at randomisation in the HFNC group, indicating that some inclusion bias cannot be ruled out, though it should be emphasised that this significant difference was not clinically relevant. Given the substantial difference in a physician deciding to escalate to HFNC (especially if the patient is to remain in the ward) versus escalating to NIV, we cannot rule out an evaluation bias regarding the inescapable nonblinded design features, as severity at the time of failure could be lower in the control group despite a lack of evidence (supplementary table E2). In the same way, not knowing the exact time from admission to randomisation make the comparison more challenging. Another potential interpretation bias could be the S_pO2 target chosen for oxygen therapy and failure criteria, given that our chosen value is substantially higher than the 90% threshold listed in AAP guidelines threshold and subsequently recommended by the WHO. The translatability of these results remains, and our findings are likely not generalisable to most centres [31]. However, it could be argued that the number patients who failed in both groups due to hypoxaemia was well balanced (supplementary table E2) and similar to Franklin et al.'s [15] study results using a similar oxygen therapy threshold. The failure rate observed in the control group was lower than expected (20%), and thus the number of patients included in the study did not allow for detection of a minimum difference of 60% with a similar power. This combination of factors puts the study at risk of being underpowered. Attending physicians were not always available 24 h/7 days a week, which may have limited the representativeness of our population by reducing the number of enrolment opportunities.

In conclusion, the results of our study do not support the pre-emptive and routine use of respiratory support by HFNC at a setting of 3 L·kg⁻¹·min⁻¹ in patients admitted to a PED and then onward for moderate viral bronchiolitis. Although HFNC may not be best used as a general practice, the criteria for its use in paediatric wards should be further defined.