Key findings and interpretation
Each of the FIRST-ABC trials in the master protocol recruited 600 children. There was minimal loss to follow-up – the primary analysis in each of the trials included over 92% of randomised children.
Step-up randomised controlled trial
In the step-up RCT, HFNC was shown to be non-inferior to CPAP (i.e. it was not worse than the preset non-inferiority margin) in acutely ill children in whom the treating clinician had decided to start NRS. The median time to liberation from all forms of respiratory support (primary outcome) was 5 hours longer for HFNC compared with CPAP (CIs ranged from 10 hours longer to 17 hours shorter). However, sedation use was lower with HFNC (28% vs. 37%), mean PICU length of stay was shorter (5 days vs. 7.4 days) and mean hospital length of stay was also shorter (13.8 days vs. 19.5 days). The main findings were robust to sensitivity analyses.
In subgroup analyses, HFNC appeared to be non-inferior in all subgroups classified by age (< 12 months vs. ≥ 12 months), diagnosis (bronchiolitis vs. other respiratory illness vs. cardiac vs. other diagnosis), severity of oxygenation defect (SpO2 : FiO2 ratio) and degree of respiratory distress (severe vs. not severe), although these remain exploratory findings since the trial was not designed to detect subgroup effects. In children already receiving NRS at randomisation, however, the non-inferiority of HFNC was unable to be demonstrated (the lower CI of the HR was 0.73, lower than the preset non-inferiority margin). This may reflect the higher severity of illness in children already started on NRS prior to randomisation, suggesting that HFNC may not be non-inferior to CPAP in the subgroup of sicker children.
A third of children started on HFNC required rescue treatment(s), which occurred at a median of 6 hours after starting HFNC. A switch to CPAP was the most common rescue treatment, mainly for clinical deterioration on HFNC.
Step-down randomised controlled trial
In the step-down RCT, in children receiving NRS within 72 hours of extubation, the non-inferiority of HFNC was not demonstrated compared to CPAP (i.e. it was worse than the preset non-inferiority margin). The median time to liberation from all forms of respiratory support (primary outcome) was 8 hours longer for HFNC, with CIs ranging from 20 hours longer to 4 hours shorter. Reintubation rate was similar for HFNC versus CPAP (13.3% vs. 11.5%), as was use of sedation (60.9% vs. 56.4%) and duration of PICU stay (6.6 days vs. 6.9 days). However, there was higher mortality at 180 days in the HFNC group (5.6% vs. 2.4%), which was not explained by differences in patient characteristics. Although this was a secondary outcome, and the trial was not powered to detect differences in secondary outcomes, this finding requires further research. Potential explanations for this finding include the fact that deteriorating children may be supported for longer on HFNC outside the critical care unit, resulting in more physiological derangement and that there might have been a differential impact of previous physiological derangement associated with invasive ventilation. The main findings were robust to sensitivity analyses.
Non-inferiority of HFNC could not be demonstrated in any subgroups classified by age (< 12 months vs. ≥ 12 months), reason for ventilation (non-cardiac vs. cardiac), severity of oxygenation defect (SpO2 : FiO2 ratio) and duration of prior ventilation (0–4 days vs. > 4 days). There was a suggestion of non-inferiority of HFNC in older children (age ≥ 12 months), with the lower confidence limit of a HR of 0.73 close to the preset non-inferiority margin of 0.75 (although a HR of 0.73 still represented at least 16 hours longer for time to liberation from HFNC compared with CPAP); however, these remain exploratory findings since the trial was not designed to detect subgroup effects.
One-third of children started on CPAP required rescue treatment(s) after a median of 8 hours of treatment, mainly a switch to HFNC due to discomfort.
Cost-effectiveness analyses
The main finding from the CEA of both step-up and step-down RCTs is that on average HFNC compared to CPAP reduces costs and improves QALYs by a small magnitude, which leads to positive INB, but there is considerable statistical uncertainty surrounding the cost-effectiveness results. The results of the cost-effectiveness analyses should be interpreted carefully, taking the small QALY gains and uncertainty around the cost-effectiveness analyses into account. The CEA results for the prespecified subgroups are similar to the overall results. The sensitivity analysis finds that this conclusion is robust to alternative assumptions to those made in the base-case analysis.
Findings in context
Other than the FIRST-ABC step-down RCT, there are no other clinical trials comparing modes of post-extubation respiratory support. In acutely ill children, a few small RCTs were published during the time the FIRST-ABC step-up RCT comparing HFNC with CPAP was recruiting – these RCTs were included in a systematic review published in 202161 (prior to the publication of the FIRST-ABC trial findings in 2022). Compared to CPAP, HFNC had a significantly higher risk of treatment failure [relative risk (RR) 1.45, 95% CI 1.06 to 1.99; I2 = 0.0%, n = 6 RCTs]. The need for intubation was similar (RR 1.69, 95% CI 0.97 to 2.94; I2 = 0.0%, n = 5 RCTs). In contrast, we found a higher rate of treatment failure for CPAP, although this was mainly for patient discomfort reasons rather than clinical deterioration. This may also relate to the fact that the systematic review only included children under the age of 2 years, whereas patient discomfort with CPAP is greater in older children.
The baseline characteristics of the children included in the FIRST-ABC step-up RCT were similar to those in other previous RCTs, especially in the subgroup of children with bronchiolitis. In FIRST-ABC, almost 65% of the children had a respiratory rate greater than the 90th centile value for their age; in infants under the age of 1, the 90th centile value ranged from 50 to 55 breaths per minute. In contrast, the mean respiratory rate in the TRAMONTANE trial comparing HFNC with CPAP in infants aged < 6 months with severe bronchiolitis was 53 breaths per minute.62 Similarly, in a large observational data set of infants with bronchiolitis from the USA, 75% of the respiratory rate values at initiation of HFNC or CPAP were > 50 per minute.63 Participants in the FIRST-ABC step-up RCT were also sick even though inclusion criteria were pragmatic and based on clinical decision to start some form of respiratory support: just over 40% of children in both groups had significant hypoxaemia (pulse oximetry oxyhaemoglobin saturation to fractional inspired oxygen ratio < 265, corresponding to the definition of acute lung injury).
In the step-down RCT, the Consolidated Standards of Reporting Trials diagram shows that out of 3121 children extubated in the participating PICUs during the trial, 1421 (45%) were started on some form of NRS. In comparison, in a multicentre cohort of extubated infants with bronchiolitis from the USA, post-extubation NRS was started in 783/1765 infants (43.7%),64 and in a single-centre study of 514 postcardiac surgery infants after extubation, 260 (50%) were started on NRS, reflecting that practice during the FIRST-ABC step-down RCT was comparable to practice in other settings.
Previous systematic reviews had shown that patients receiving HFNC had a lower risk of AEs, mainly nasal trauma.61 These findings were corroborated in the FIRST-ABC step-up RCT, where nasal trauma occurred in 2% of patients started on HFNC and 6.5% of patients started on CPAP.
The main finding when the two trials were compared side by side was that they showed different results. The population of acutely ill children requiring NRS was different from the children requiring post-extubation NRS, and the effect of HFNC versus CPAP treatment was also different. This was shown in the FIRST-ABC pilot RCT,40 which is one of the main reasons why we chose to run the two trials separately within a master protocol. For example, compared with acutely ill children, extubated children were younger (many of whom were post-surgical infants), had received several days of invasive ventilation, and were still weaning off their sedative agents, all factors that might be expected to alter the balance of risks and benefits of HFNC and CPAP.
Reporting equality, diversity and inclusion
The population of patients that are seen in UK PICUs is diverse, and we sought to ensure that this diversity was reflected in the FIRST-ABC trials. The eligibility criteria for the trials were designed to be inclusive and pragmatic, only excluding patients where there was a clear contra-indication or indication for either HFNC or CPAP (e.g. patients with midfacial/craniofacial anomalies).
We monitored the representativeness of the study population by reviewing data on key characteristics, including age, sex, geographic region and ethnicity. With regard to ethnicity, both RCTs had good representation of patients from an ethnic minority background, in line with the ethnic distribution seen in clinical practice. However, ethnicity data were sourced via data linkage to HES and, accordingly, were missing for 26% (step-up RCT) and 32% (step-down RCT) of patients. Future studies might improve on this by collecting ethnicity data via multiple data sources to ensure completeness. Across the FIRST-ABC RCTs, the vast majority of PICUs across the UK were included; however, we note that we had no sites based in Northern Ireland and therefore this region of the UK was not represented.
Language barriers are a potential factor that can limit participation in research at the outset. This barrier was reduced in FIRST-ABC through use of a deferred consent model. This model allowed research teams to aim to enrol all eligible patients, regardless of language spoken, and then follow up at a later appropriate time, with the assistance of locally available translation services. However, it must be noted that availability of translation services varied significantly across sites. Future studies could improve on these efforts by offering a range of translated study documents to participants (noting additional cost and difficulty in predicting which languages will be required). Obtaining consent also appeared more challenging in situations where patients were under the care of local authorities – further guidance on deferred consent for research in these circumstances would be useful to ensure such patients can be included in research.
Strengths and limitations
Strengths
FIRST-line support for Assistance in Breathing in Children was a set of two pragmatic trials, set in a real-world context. They represent the largest RCTs on the topic of paediatric non-invasive breathing support, where the pre-existing evidence base was poor. The master protocol included children receiving NRS in two important clinical scenarios, during step-up treatment in acutely ill children and during step-down treatment in children following extubation from invasive ventilation. This provided an efficient way to answer the research question in two key populations within a common trial infrastructure.
The findings from the trials are generalisable to the NHS, since the vast majority of PICUs in the UK participated (24 critical care units in the step-up RCT and 22 PICUs in the step-down RCT). The age, sex and ethnic distribution of FIRST-ABC trial participants are similar to that of the overall UK PICU population reported by the PICANet national audit report,2 providing confidence that there were no sex/ethnic/racial barriers to being included in the trial. For example, in 2018, 50.5% of patients admitted to UK PICUs receiving respiratory support were infants younger than 1 year (compared to 56% in the step-up trial).
In addition, practice during the FIRST-ABC trial was similar to previous audits of clinical practice, attesting to the generalisability of the trial findings to routine practice. Observational data from UK PICUs in 2015–6 had shown that nearly 20% of admissions received HFNC or CPAP as first-line NRS;65 in the FIRST-ABC step-up trial, out of 18,976 admissions to participating PICUs during the study period, 3825 (20%) met inclusion criteria and were started on some form of NRS.
Patient recruitment was rapid and followed target timelines, especially in the step-down RCT, in which the 600 participants were recruited in < 9 months (as opposed to projected 15-month recruitment). The step-down RCT completed recruitment in May 2020, just after the first COVID-19 lockdown. On the other hand, the step-up RCT recruited through the COVID-19 pandemic, and despite the loss of nearly 6 months of recruitment time on average across sites, managed to complete recruitment in 27 months (as opposed to projected 30 months). This was enabled through good clinical buy-in and the ability of participating sites to embed delivery of the trial into routine clinical practice, allowing randomisation to be carried out 24 hours per day, 7 days per week. The simplicity of the web/telephone randomisation process, with only minimal information needed at the point of randomisation, the use of a research without prior consent model which meant that recruitment could be undertaken in an emergency, and the pragmatic nature of the trial, with trial procedures reflecting clinical practice as far as possible, were other factors contributing to rapid patient recruitment.
Both the protocol and statistical analysis plan were published during the recruitment period to ensure transparency, and all analyses were conducted following the analysis plan. Planned sensitivity analyses were also conducted, which evaluated different definitions of time to liberation, and a post hoc sensitivity analysis to include all randomised patients irrespective of whether they started respiratory support or not was carried out.
From the point of view of the economic evaluation, this study included prospectively designed economic evaluation integrated with well-designed step-up and step-down RCTs, which ensured that detailed resource use and health economic outcome data were collected for each patient randomised. The resource use measurement harnessed information from four linked sources – data from trial CRFs linked to the PICANet database, responses to follow-up health service questionnaires and HES database. The cost measurement utilised the detailed patient-level information from PICANet database on the intensity of resource use for each day in the paediatric critical unit. The CEA also measured HRQoL with the age-appropriate PedsQL instruments.
Paediatric Quality of Life Inventory is a widely used measure of HRQoL among children and adolescents, but the PedsQL scores cannot be directly used to calculate QALYs for the application in economic evaluation because it produces summary scores which are not preference-based. We have mapped the PedsQL to CHU-9D utility score using appropriate mapping algorithms, which provides a preference-based HRQoL score for calculating QALYs that can be used in the application of economic evaluation. The CEA followed the mITT principle and reported results for all patients randomised in the RCT. There were some missing data in the 6 months follow-up of HRQoL and costs, which were imputed using the recommended MICE approach. The study followed a harmonised and prespecified statistical and economic analysis plan, allowing for consistent approach for analysis of clinical and health economic end points. Detailed subgroup analysis was performed, and the cost-effectiveness results were similar across all subgroups. Extensive sensitivity analyses were performed, and the base-case results were not sensitive to alternative assumptions.
Limitations
The main limitation of the trials was the fact the intervention and control treatments could not be blinded. This lack of blinding may have influenced decisions to switch or escalate treatment or to start respiratory support at all. This was highlighted in the step-up RCT in the high rate of switching from CPAP to HFNC for perceived patient discomfort, and a larger than anticipated number of patients in the CPAP group who did not start any respiratory support after randomisation.
The pragmatic inclusion criteria resulted in a heterogeneous population of acutely ill children, mostly younger than 2 years, and although prespecified subgroup analyses based on age, diagnosis and receipt of prior NRS were performed, there may be other unidentified subgroups (e.g. children aged ≥ 10 years) for whom one treatment was more effective over another.
Clinician preference also meant that several children who were eligible for the trials were not randomised, and some children (especially older children randomised to CPAP in the step-up RCT) did not start any respiratory support after randomisation, which created an imbalance between the groups. Changes in clinical practice resulting in greater use of HFNC outside the critical care setting also meant that in the step-up RCT, nearly 1000 children who had received more than 2 hours of prior NRS outside critical care had to be excluded from the trial. Data related to feeding were not collected as part of the trial; therefore, it was not possible to assess the effect of feeding on patient comfort in either of the trials.
In the post-extubation setting, there was a low threshold for starting non-invasive support, with nearly 50% of extubated children receiving some form of support. It was not possible to understand the reasons for this practice; specifically, we did not collect ventilation data, such as mean airway pressure at randomisation, which might have indicated why clinicians chose to start NRS in some patients and not others.
Cost-effectiveness analysis can be challenging in young children who are unable to respond to questionnaires themselves; in addition, the questionnaires used commonly (such as PedsQL) need mapping to CHU-9D. Our CEA presented results for the same prespecified subgroups as for the analysis of clinical effectiveness. The results of these subgroup analyses suggested that the point estimate of the INB was positive for some subgroups and negative for others, but that the CIs around each of these estimates were wide and included zero. In interpreting these findings, it should be recognised that this study was not powered to detect subgroup effect for either clinical effectiveness or cost-effectiveness end points, and hence the subgroup results should be regarded as exploratory.
Implications for practice
Acutely ill children
Other than a handful of small RCTs, mainly from low-middle income settings, there has been a paucity of research to guide clinical practice regarding breathing support in acutely ill children. Previous RCTs have focused on short-term outcomes, such as treatment failure rather than on patient-centred outcomes, such as time to liberation from respiratory support.
Findings from the FIRST-ABC step-up trial indicate that starting an acutely ill child on NRS using HFNC is a reasonable first-line option. It should be expected however that around one in three children will fail HFNC, mainly due to clinical deterioration, and require a switch to CPAP or escalation to other forms of respiratory support, including invasive ventilation (which occurred in one in seven children). Clinicians should expect treatment failure to occur on average 6 hours after starting HFNC treatment.
Following extubation
There have been no previous randomised trials comparing different modes of NRS following extubation.
Results of the step-down trial indicate that in the post-extubation setting, starting CPAP as the first-line mode of NRS is a reasonable option, especially in infants under the age of 1. In children over the age of one, either HFNC or CPAP is a suitable first-line option. Similar to the step-up scenario, clinicians should expect that one in three children started on CPAP will fail, mainly due to patient discomfort with CPAP. Treatment failure in this situation occurred on average 8 hours after starting CPAP.
Recommendations for research
Recommendation 1
Since the trials enrolled a heterogeneous population of children and their trajectory after starting the first-line mode of NRS was variable, further analyses exploring which baseline patient characteristics and patterns of evolving physiological parameters can predict treatment failure, intubation and prolonged time to liberation from respiratory support constitute important areas of future research.
Recommendation 2
The relatively high proportion of extubated children in whom clinicians started NRS following extubation (45%) highlights that developing and testing the utility of protocolised approaches regarding when and which children to start post-extubation respiratory support in future clinical trials is an important area of further research.
Recommendation 3
The results from the prespecified subgroup analyses raise the hypothesis that there may be some subgroups (e.g. cardiac patient subgroup in both step-up and step-down RCTs) for whom HFNC intervention is cost-effective, compared to CPAP. Further research could evaluate clinical and cost-effectiveness of targeted interventions to subgroups who are admitted to paediatric critical care unit for cardiac reasons.
Recommendation 4
Reasons for the unexpected finding of higher mortality at 180 days in the HFNC group in the step-down RCT need to be explored further in future research.