DISCUSSION
In this multicentre population of young infants with severe AVB, the failure rate of HFNC was 39%. Failure of this first step of respiratory support occurred early, within 6 hours after HFNC initiation for half of failing patients and within 24 hours for nearly 90% of them. Neither ROX index, nor physiological variables usually collected in infants with acute respiratory failure had early discriminatory capacity to predict failure of management with HFNC.
In a previous study, Kannikeswaran et al. observed an association between ROX index and HFNC failure, in other words the need for positive pressure ventilation, in <2 years infant with bronchiolitis [23]. This result was particularly relevant from the perspective of directing these patients to the most suitable units downstream of the emergency department. In this perspective, our study aimed to identify a threshold for this index, associated with actual predictive capacities. Our AUC results suggested a weak and non-significant relationship between ROX and HFNC failure, which currently does not confirm the interest of this tool in a clinical decision rule. The main difference with studies suggesting that ROX index may be a good marker to predict the risk of HFNC failure probably comes from different patient’s characteristics [22, 23]. Our trial involved much younger patients and, as our results indicate, age is a key risk factor for respiratory failure in this population [12, 13]. In addition, the TRAMONTANE 2 study included patients probably affected by more severe forms of the disease, as mWCAS > 3 was required to be eligible, which signals unambiguous respiratory distress. In Kannikeswaranet al. study, the regression model to estimate the odds ratio of PPV requirement was based on the highest ROX quartile, suggesting marked heterogeneity in the severity of bronchiolitis [23].
The ROX index takes into account only two characteristics of a respiratory distress, namely oxygenation and tachypnea. While these parameters are critical in patients with acute hypoxemic respiratory failure, they do not take into account all of the determinants of HFNC failure in AVB. Indeed, AVB present different phenotypes: sometimes as a restrictive parenchymal disease, but a majority of these infants demonstrate a severe obstructive lung disease, with markedly increased work of breathing and frequent apneas [4, 5, 25]. These two elements, as well as comfort, are not integrated in the ROX index, whereas they often intervene in the clinician’s decision to upgrade respiratory support in a patient with AVB.
No variable, observed or calculated, was able to predict HFNC failure in this work. However, half of failure occurred within 6 hours, which suggests that such delay may be a relevant criterion for triage. Insofar as the volume of patients in the emergency room allows it, it may be consistent to maintain the infant with HFNC and adapted monitoring during this timeframe before deciding admission to pediatric intensive care unit.
This study has several strengths, including a large, multicentric and homogeneous population, in terms of severity and age, of infants with AVB. In addition, predefined HFNC failure criteria had been validated by a panel of experts providing from the 16 participating centres in the trial.
We acknowledge some limitations. Half of the children received a flow rate of 2 L/min/kg, and the other half 3 L/min/kg, but we are uncertain whether this affects the interpretation of our results because the failure rate was exactly the same in the two groups.
The predictive value of the ROX index beyond the first hour of HFNC delivery could have been evaluated, but it seemed more obvious that the relevance of an urgent triage tool can hardly exceed 1 to 2 hours.