Methods

Some text in this chapter is reproduced from Richards-Belle et al.1 and Orzechowska et al.30 These are Open Access articles distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon the work, including for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. Includes additions and changes to the originals.

Also reproduced with permission from Ramnarayan et al.31 and Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved.

Trial design

FIRST-line support for Assistance in Breathing in Children was a master protocol comprising two pragmatic, multicentre, parallel-group, non-inferiority RCTs (step-up RCT and step-down RCT) with shared infrastructure, including an internal pilot stage and integrated health economic evaluation.1

The master protocol design allowed the research question to be addressed in each of the two important populations (step-up and step-down NRS) in an efficient way by minimising time and infrastructure costs as compared with conducting two sequential RCTs.33

The non-inferiority design was chosen based on previous RCTs on this topic as well as feedback from Paediatric Intensive Care Society Study Group in July 2017 which indicated that the potential benefits of HFNC (in terms of patient comfort and ease of use) would mean that it would likely be preferred in usual practice even if it was shown not to be superior to CPAP.

The pragmatic design ensures that research findings can be more easily generalised to real-world practice.

Setting

The trial was set in NHS paediatric critical care units [PICUs and/or high-dependency units (HDUs)] across England, Wales and Scotland.

Trial sites

The trial aimed to recruit eligible patients from a representation sample of 25 paediatric critical care units (PICUs and/or HDUs). Trial sites could be either general (medical-surgical), cardiac or mixed (general-cardiac) units.

Sites were required to commit to the following criteria to take part in FIRST-ABC:

• identify a principal investigator (PI) to lead FIRST-ABC locally, supported by a research nurse with responsibility for day-to-day local trial co-ordination
• able to provide both treatments (HFNC and CPAP) to trial participants
• confirm collective equipoise regarding the choice of first-line NRS
• agree to incorporate FIRST-ABC into routine paediatric critical care clinical practice, highlighting the importance of systematic screening for potential eligible patients and prompt randomisation
• active participation in the PICANet for the UK and Ireland audit or able to collect detailed data on patient interventions and outcomes
• compliance with all responsibilities and requirements as stated in the trial protocol and FIRST-ABC Clinical Trial Site Agreement
• compliance with the UK Policy Framework for Health and Social Care Research and International Conference on Harmonization Guidelines on Good Clinical Practice.

Site identification, initiation and activation

Site identification commenced in February 2019 with an e-mail from the chief investigator (PR) to potential site investigators. A FIRST-ABC Collaborators’ Meeting was also held in London in June 2019 with representatives from the clinical and research teams from potential sites to raise awareness of the trial and to gather feedback on the proposed trial procedures, including the algorithms for the delivery of CPAP and HFNC.

A staggered opening of sites was used to allow for in-person site initiation visits to be conducted at each participating site. Site initiation visits were facilitated by the chief investigator (PR) and/or trial manager (ARB) and attended by research and clinical team members at the site. During each visit, the background/rationale for the trial was presented and training was provided in the trial procedures for screening, randomisation, delivery of HFNC and CPAP, consent, data collection and safety monitoring. An investigator site file, containing all essential trial documents [e.g. trial protocol, standard operating procedures, participant information sheets (PISs) and consent forms, case report forms (CRFs) and relevant approvals], was also provided.

A ‘green light’ e-mail was issued to the local PI and research team by the Intensive Care National Audit & Research Centre (ICNARC) Clinical Trials Unit (CTU), authorising the commencement of screening and recruitment, once the following were in place:

• completed site initiation visit
• local confirmation of capacity and capability (e.g. local research and development department approval)
• fully signed FIRST-ABC Clinical Trial Site Agreement
• copy of the signed off delegation log submitted to the ICNARC CTU.

Patients

Eligibility

The target population was critically ill children requiring NRS for (1) an acute illness (step-up RCT) or (2) following extubation after a period of invasive ventilation (step-down RCT). Patients were considered eligible if they met all inclusion criteria and none of the exclusion criteria.

Inclusion criteria

1. Admitted/accepted for admission to PICU/HDU.
2. Age > 36 weeks corrected gestational age and < 16 years.
3. Assessed by the treating clinician to require NRS, EITHER
   1. for an acute illness (step-up RCT) OR
   2. within 72 hours of extubation following a period of invasive ventilation (step-down RCT).

Exclusion criteria

1. Assessed by the treating clinician to require immediate intubation and invasive ventilation due to severe hypoxia, acidosis and/or respiratory distress, upper airway obstruction, inability to manage airway secretions or recurrent apnoeas.
2. Tracheostomy in place.
3. Received HFNC/CPAP for > 2 hours in the prior 24 hours.
4. On home non-invasive ventilation prior to PICU/HDU admission.
5. Presence of untreated air leak (pneumothorax/pneumomediastinum).
6. Midfacial/craniofacial anomalies (unrepaired cleft palate, choanal atresia) or recent craniofacial surgery.
7. Agreed ‘not for intubation’ or other limitation of critical care treatment plan in place.
8. Previously recruited to the FIRST-ABC trial.
9. Clinician decision to start other form of NRS (i.e. not HFNC or CPAP).

Exclusion criterion 9 was added following an early amendment to the protocol (see Health Research Authority and research ethics application for details).

Screening

Potentially eligible patients admitted/accepted for admission to the participating unit were to be screened against the inclusion/exclusion criteria by the local clinical team, supported by the site research team. In the step-up RCT, all admissions to the critical care unit were to be screened. In the step-down RCT, all patients extubated during critical care admission were to be screened. In each RCT, screening and enrolment logs recorded screened patients, detailing which patients were randomised, reasons for exclusion and reasons for non-enrolment of eligible patients.

Randomisation

Randomisation of patients was performed after confirming eligibility and as soon as possible to the anticipated start of the randomised treatment.

In each RCT, eligible patients were randomised on a 1 : 1 basis to either CPAP or HFNC using a dedicated, centralised 24 hours/7 days per week telephone/web-based randomisation service hosted by Sealed Envelope Ltd (https://sealedenvelope.com/). The randomisation sequence was computer generated by Sealed Envelope and used variable block sizes of 4 and 6 to strengthen allocation concealment. Randomisation was stratified by site and age (< 12 months vs. ≥ 12 months) to minimise imbalance arising from unit practices and interface selection.

Following randomisation, the allocated treatment was commenced as soon as practically possible, each participant was assigned a unique FIRST-ABC trial number, and the local site research team were notified of the randomisation by e-mail.

Treatment groups

High-flow nasal cannula

Any approved medical device capable of delivering heated, humidified, high-flow through nasal cannulae was used to provide HFNC at prescribed gas flow rates (based on patient weight) during the trial period. To standardise treatment, clinical criteria and guidance for the initiation, maintenance and weaning of HFNC (and CPAP) were provided in a trial algorithm (Figure 1). The trial algorithms were developed iteratively in consultation with paediatric critical care clinicians across the UK (both via e-mail and in person at a Collaborators’ Meeting held prior to the start of the trial). The trial recommended that patients were assessed for response to the treatment, readiness to wean and for stopping HFNC, as per the HFNC algorithm, at least twice per day (e.g. at ward rounds).

FIGURE 1 Trial algorithm for the delivery of HFNC. Reproduced from Richards-Belle et al.1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. Includes additions and changes to the original. (Figure not reproduced here; see the original article.)

Continuous positive airway pressure

Continuous positive airway pressure was started using an approved medical device at a set expiratory pressure of 7–8 cm H₂O. The trial did not specify any particular device or patient interface for the provision of CPAP. To standardise treatment, clinical criteria and guidance for the initiation, maintenance and weaning of CPAP were provided in a trial algorithm (Figure 2). It was recommended that patients were assessed for response to the treatment, readiness to wean and for stopping CPAP, as per the CPAP algorithm, at least twice per day (e.g. at ward rounds).

FIGURE 2 Trial algorithm for the delivery of CPAP. Reproduced from Richards-Belle et al.1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. Includes additions and changes to the original. (Figure not reproduced here; see the original article.)

Clinical practice during the trial

Since staff in participating sites already used HFNC and CPAP, no additional central training related to the use of HFNC or CPAP was provided for the trial, but resources for training in the trial algorithms were provided. As the medical devices and interfaces that deliver HFNC and CPAP are easily distinguishable from each other, it was not possible to blind the patient, parents/guardians or clinical staff.

The trial algorithms were to be followed until the patient has been liberated from all forms of respiratory support for at least 48 continuous hours. As per current clinical practice, clinicians were able to stop HFNC/CPAP and switch to the other treatment or escalate to other forms of respiratory support, if clinically deemed necessary. Prespecified objective criteria to identify non-responders to HFNC/CPAP were provided in the algorithms as a guide for clinicians considering switching or escalating respiratory support. Reasons for switches or escalations were recorded on the CRF. Patients who switched or escalated treatments remained in the trial and continued to be monitored until liberation from respiratory support.

Co-interventions

All other usual care (e.g. sedation, feeding, etc.) was provided at the discretion of the treating clinical team, as per local practice. Note that respiratory support, as defined for the purposes of FIRST-ABC, did not include supplemental/low-flow oxygen.

Consent procedures

Consent was sought for the child (patient) from their parent/legal guardian. Children became eligible for FIRST-ABC when critically ill, a profoundly stressful time for parents/guardians, during which there are ethical concerns both about the burden of trying to understand the trial and the ability to provide informed consent. Initiation of NRS typically occurs during a time-sensitive situation, where delays could be detrimental to the child and to the trial’s scientific validity. Moreover, both CPAP and HFNC are already widely used in standard practice across the NHS. Considering these reasons, FIRST-ABC was granted ethical approval by the East of England – Cambridge South Research Ethics Committee (REC) to use a model of research without prior consent. Once a patient was screened and confirmed as eligible for the trial, they were randomised, and the allocated treatment (CPAP or HFNC) was commenced as soon as possible. This model, developed in line with the CONseNt methods in paediatric Emergency and urgent Care Trials (CONNECT) guidance,34 has been found acceptable to parents/guardians and clinicians in several recent RCTs in the PICU setting35–40 and is informed by experience/feedback from the pilot RCT.40

Following randomisation, a trained, delegated member of the local research team approached the child’s parents/guardians as soon as appropriate and practically possible to discuss the trial (usually within 24–48 hours of randomisation). Before approaching the parent/legal guardian, the research team member would check with the relevant clinical staff that the participant was stable and that the timing was appropriate. If the participant’s condition had not stabilised, additional time was given before approaching the parent/guardian. Once approached, a PIS was provided (see Report Supplementary Material 1), covering information about the purpose of the trial; the consequences of participating or not; confidentiality; use of personal data; data security; and the future availability of the trial results. A consent form (see Report Supplementary Material 2) was provided, indicating that: the information given had been read and understood; participation was voluntary and consent could be withdrawn at any time without consequence; and consent was given for access to medical records to continue data collection, to receive a follow-up questionnaire and for anonymised data to be shared in the future. Parents/guardians were given time to read the information sheet, invited to ask any questions that they may have had about their child’s participation, and the opportunity to discuss with other family members or friends before confirming their decision. Due to age and severity of illness, it was not possible to involve the patient in the consenting process. Instead, assent was to be obtained prior to hospital discharge if their condition allowed (e.g. they regained mental capacity).

A modification of the consent procedure would be utilised for two rarer situations where either the patient: (1) was discharged from hospital prior to obtaining consent or (2) died prior to consent being sought.38,41 In the former, the local research team followed up with the parent/guardian, initially by phone and then by post, for consent. Postal contact was made again if there was no response after 4 weeks. If no consent form was received within 4 weeks of the second letter, the participant was to be included in the trial unless they notified the research team otherwise. In the latter situation, the local research team obtained information from colleagues and bereavement counsellors to establish the most appropriate clinical/research team member to notify the parents/guardians of involvement in the trial. If approach for consent was deemed not appropriate prior to the parent/guardian’s departure from hospital, then they were approached by post 4 weeks post randomisation. The letter explained how to opt out of the trial. Postal contact was made again if there was no response after 4 weeks. If no consent form was received within 4 weeks of the second letter, the participant’s data were included in the trial.

If informed consent was refused or withdrawn, this decision was respected and abided by, and no further contact made. All data occurring up to the point of this decision would be retained in the trial, unless parents/guardians requested otherwise.

Safety monitoring

Adverse event (AE) reporting followed the Health Research Authority guidelines on safety reporting in studies which do not use Investigational Medicinal Products (non-CTIMPs). The following events were prespecified as potential AEs that could be related to CPAP and/or HFNC and observed in participants from the date and time of randomisation until 48 hours of liberation from all forms of respiratory support:

• nasal trauma
• facial/neck trauma
• abdominal distension
• pneumothorax
• pneumomediastinum
• subcutaneous emphysema
• facial thermal injury
• respiratory arrest
• cardiac arrest
• aspiration.

Occurrences of the above specified, expected AEs were recorded for all randomised patients. Considering that eligible patients were critically ill and at increased risk of experiencing AEs, occurrences of other, non-specified, AEs were only reported if considered to be related to either CPAP or HFNC (i.e. ‘possibly’, ‘probably’ or ‘definitely’ related). The following events were not reported as AEs or serious adverse events (SAEs) as they were instead collected as study outcomes:

• intubation or reintubation
• sedation
• death (note that death itself was not reported as a SAE, but the suspected cause of death was assessed for severity, relatedness and expectedness).

Each event was assessed for its severity, according to the below definitions:

• None: indicates no event or complication.
• Mild: complication results in only temporary harm and does not require clinical treatment.
• Moderate: complication requires clinical treatment but does not result in significant prolongation of hospital stay. Does not usually result in permanent harm and where this does occur the harm does not cause functional limitation to the participant.
• Severe: complication requires clinical treatment and results in significant prolongation of hospital stay or permanent functional limitation.
• Life-threatening: complication that may lead to death or where the participant died as a direct result of the complication/AE.

Any reportable event classified as ‘severe’ or ‘life-threatening’ in severity was considered a SAE and was reported to the ICNARC CTU. If a SAE was evaluated by a clinical member of the Trial Management Group (TMG) as a related and unexpected SAE, then the ICNARC CTU submitted a report to the REC within 15 calendar days.

Questionnaire follow-up

At 6 months, after assessing the child’s survival status, each consenting parent was sent a questionnaire (via e-mail or post) by the ICNARC CTU to assess health-related quality of life (HRQoL) and health service/resource use. If a parent requested a questionnaire to be sent via post, then a pen and self-addressed stamped envelope was provided for ease of return. Non-responders were followed up via telephone by a trained member of the FIRST-ABC team from the ICNARC CTU 3 weeks later.

If a patient was an inpatient at a participating site at the 6-month time point, then the site research team approached the parent/guardian to complete the questionnaire.

Clinical outcomes

Primary clinical outcome

Time to liberation from respiratory support

The primary outcome was the time from randomisation to liberation from respiratory support, defined as the start of a 48-hour period during which the child was free of all forms of respiratory support. In this definition, respiratory support included HFNC, CPAP, other forms of NRS (e.g. bilevel positive airway pressure, pressure support, etc.) and invasive ventilation. It did not include the administration of supplemental oxygen alone.

Secondary clinical outcomes

Rate of reintubation at 48 hours

Reintubation at 48 hours is defined as occurring if the child has started invasive ventilation at any time up to and including 48 hours and 0 minutes after the time of randomisation. Patients are included in the denominator if they have received invasive ventilation by 48 hours or are known not to have received any invasive ventilation from randomisation to 48 hours following randomisation. Patients discharged from PICU/HDU before 48 hours are assumed not to have been invasively ventilated post-discharge.

Duration of paediatric intensive care unit/high-dependency unit and acute hospital stay

Duration of PICU/HDU stay was calculated as the sum of the duration (in days and fractions of days) from the date and time of randomisation to the date and time of first discharge from a critical care unit (or ultimate discharge from critical care if transferred directly to another critical care unit) or to death in the critical care unit.

Patient comfort, during randomised treatment and during non-invasive respiratory support (i.e. High-flow nasal cannula and/or continuous positive airway pressure), measured using the COMFORT Behavior score

Patient comfort was measured during HFNC or CPAP using the COMFORT Behavior (COMFORT-B)42 score and summarised at the patient level using the median of all recorded scores. Patient comfort was reported in all patients with at least one recorded COMFORT-B score in the first 6 hours of support following randomisation, and, while respiratory support continues, at least one COMFORT-B score per day during at least the first 48 hours of respiratory support.

Proportion of patients in whom sedation is used during non-invasive respiratory support

Sedation was defined as any medication given with the intention of improving patient comfort (analgesics/sedatives) while on NRS. These included (but were not limited to): chloral hydrate, alimemazine, opiates (e.g. morphine, fentanyl), benzodiazepines (e.g. midazolam, lorazepam), clonidine and dexmedetomidine. Examples of analgesics which would not be considered a sedative were ibuprofen and paracetamol.

Need for sedation was reported as the proportion of patients in whom sedation was used during NRS at any point from randomisation until liberation from respiratory support. Patients were be included in the denominator if they had a minimum of three non-missing observations in the first 6 hours of respiratory support.

Parental stress, in hospital at/around the time of consent, measured using the Parental Stressor Scale: paediatric intensive care unit

Parental stress was measured using the validated Parental Stressor Scale: PICU (PSS:PICU)43 in hospital at/around the time of consent (anticipated to be within 24–48 hours post randomisation). This scale consists of 37 items, each scored in whole numbers from 1 (not stressful) to 5 (extremely stressful). A total score was calculated as the mean of all completed items.

Mortality at paediatric intensive care unit/high-dependency unit discharge, day 60 and day 180

Mortality at discharge from the PICU/HDU was defined as death due to any cause before discharge to any location providing a level of care less than Level 2 (high dependency care). Mortality at days 60 and 180 was calculated as binary end points using all patients with known survival status at those times and additionally using time-to-event methods with surviving patients censored at the date last known to be alive (to a maximum of day 180).

Data collection and management

Case report forms

Case report forms were developed to capture important data fields for the RCTs relating to: confirming eligibility, consent and patient details (to enable data linkage and follow-up), baseline observations and comorbidities, delivery of respiratory support (including switches and escalations, weaning from HFNC or CPAP, and liberation), critical care and hospital discharge details, and safety monitoring. The trial data collection schedule is shown in Table 1.

A dedicated, secure, electronic CRF was developed to enable site staff to enter and submit the trial data. Access to the electronic CRF was centrally managed, with access granted only to authorised site staff (as per the delegation log).

Data management

Data validation checks were built into the electronic CRF, such that only values that were physically possible could be saved as well as checks to identify and query unusual or missing data items. Sites were contacted throughout the trial period with potential data queries in an attempt for data to be validated and cleaned as early as possible.

Data linkage: The Paediatric Intensive Care Audit Network and National Health Service Digital

To reduce the burden of data collection on sites, FIRST-ABC data were linked to PICANet and NHS Digital. Linkage with PICANet provided further data on interventions delivered during critical care admissions within the first 6 months following randomisation (a separate CRF was developed for sites not participating in PICANet). Following the signing of a data-sharing agreement with NHS Digital, FIRST-ABC data were linked to Hospital Episode Statistics (HES; admitted patient care, outpatients and emergency care data sets), to provide data on resource use, and to civil registrations, to identify any deaths, both within the 6 months following randomisation.

Governance, management and oversight

Health Research Authority and research ethics application

Following submission of an application for Health Research Authority approval and REC favourable opinion on 26 April 2019, a research ethics committee meeting was scheduled for 23 May 2019 and attended by ARB and PM. Provisional opinion was issued on 12 June 2019 and full Health Research Authority approval and REC favourable opinion given on 26 July 2019.

Amendments to the protocol

There was one amendment to the protocol (substantial amendment 1). The main purpose of this amendment was to:

• add an exclusion criterion (‘clinical decision to start other forms of NRS i.e. not HFNC or CPAP’)
  • As FIRST-ABC investigated HFNC and CPAP, patients deemed to require other forms of NRS (such as bilevel support) were not eligible. This message was reinforced in study training materials; however, some participating sites requested this be added as an explicit exclusion criterion for clarity among their teams.
• remove the step-down RCT interim analysis
  • Considering a higher-than-expected recruitment rate in the step-down RCT, a decision was agreed with the DMEC that no formal interim analysis would be performed given that the trial would have almost completed recruitment by the time the first 300 patients had been followed up to 60 days (the prespecified time point for the interim analysis). Instead, safety data (counts and percentages of AEs by arm, and a line listing of SAEs) was made available for scrutiny by the DMEC by the end of the internal pilot stage. There were no changes to the planned interim analysis of the step-up RCT.

This amendment was submitted on 6 February 2020 and received Health Research Authority approval and Research Ethics Committee favourable opinion on 27 February 2020.

Local governance

Prior to the commencement of patient recruitment, each participating NHS trust/health board signed a Clinical Trial Site Agreement, based on the model agreement for non-commercial research in the NHS, with the sponsor and issued confirmation of capacity and capability to deliver the trial.

Trial registration

FIRST-line support for Assistance in Breathing in Children was prospectively registered with the ISRCTN Registry on 19 June 2019 (reference: ISRCTN60048867).

Patient and public involvement

Two parent representatives with experience of paediatric intensive care contributed to the design of FIRST-ABC and were co-investigators on the grant. One parent representative continued their involvement through membership of the TMG, and contributed to reviewing and advising on participant documents, consent procedures and interpretation of the results. The TSC included two patient and public involvement (PPI) members.

Trial monitoring

The trial team members at the ICNARC CTU had regular communication with sites via e-mail, telephone, teleconferences and regular newsletters. The local PI was responsible for ensuring all queries were addressed and for overall quality of their site data. Adherence to the protocol was paramount in the central monitoring plan, including a review of eligibility data and adherence to the HFNC and CPAP algorithms.

The on-site monitoring plan followed a risk-based strategy. The timing and frequency of visits to sites were based on risk assessment, with a view to visiting ~ 25% of sites at least once during the recruitment period. In total, 6 of 26 (23%) were visited for a monitoring visit. Additional visits were not possible due to restrictions introduced following the start of the COVID-19 pandemic. During these visits, the investigator site file was checked for completeness, consent forms were reviewed, and source data verification was carried out on a random sample of patient CRFs. The visits were also used to discuss progress on delivering the trial at the site, including identifying any potential barriers or difficulties and suggesting potential solutions. Following each visit, a report was sent to the site by the trial monitor summarising the findings and any outstanding actions.

Trial Management Group

The TMG, led by the chief investigator (PR), was responsible for the management of the trial. The chief investigator (PR) took overall responsibility for the delivery of the trial and oversaw progress against timelines. The TMG included paediatric critical care clinicians, trialists and PPI representation. The trial manager (ARB) was responsible for the day-to-day management of the trial, supported by the trial co-ordinator (LD), data manager (MS) and trial statisticians (KT, IO).

Trial Steering Committee

An independently chaired and majority independent Trial Steering Committee was convened by the National Institute for Health and Care Research (NIHR) to provide overall supervision of the trial on behalf of the funder and sponsor. The committee was chaired by Professor Carrol Gamble (University of Liverpool).

Data Monitoring and Ethics Committee

An independent Data Monitoring and Ethics Committee was convened by the NIHR in order to monitor recruitment, protocol adherence and patient safety. The committee was chaired by Professor Neal Thomas (Penn State University).

Sponsorship

FIRST-line support for Assistance in Breathing in Children was sponsored by Great Ormond Street Hospital for Children NHS Foundation Trust (reference: 17IA05).

Network support

FIRST-line support for Assistance in Breathing in Children was adopted onto the NIHR Central Portfolio Management System on 17 May 2019 (reference: 42112).

Statistical analysis

Analysis principles

The primary clinical outcomes were tested for non-inferiority. Other secondary outcomes were tested for superiority, where testing was specified, or analysed using descriptive statistics only if no testing was specified in the statistical analysis plan. All analyses were performed separately for each of the two trials, and any results were not combined.

Statistical tests will be two-sided with significance set at p < 0.05 unless otherwise specified. Effect estimates will be reported with 95% CIs. There will be no adjustment for multiple testing. The results of subgroup analyses were interpreted taking into account accepted criteria for credible subgroup effects.44,45

Analysis population

All randomised patients will be included in the intention-to-treat (ITT) population. A modified ITT (mITT) population will be used for analysis of the primary end point (‘primary analysis set’), consisting of the ITT populations excluding those with no recorded respiratory support post randomisation.

The per-protocol population will consist of all randomised patients who met the eligibility criteria and started on the randomised respiratory support, as the first respiratory support post randomisation.

Sample size calculation

The sample size was calculated as follows: to achieve 90% power with a one-sided type I error rate of 2.5% to exclude the prespecified non-inferiority margin of HR = 0.75 (corresponding to approximately a 16-hour increase in median time to liberation) requires 508 events to be observed. Based on data from the FIRST-ABC pilot RCT,40 we anticipate 5% censoring due to death or transfer, leading to a required sample size of 268 patients per group in each of the two RCTs. To allow for withdrawal/refusal of deferred consent, and for exclusion due to non-adherence in the per-protocol population, we aimed to recruit a total sample size of 600 patients in each of the two RCTs.

Internal pilot

The internal pilot phase was evaluated 6 months after the first site opened to recruitment. At this point, the following key progression criteria were assessed and classified as green, amber or red (Table 2).

All progression criteria in both RCTs were classified as green and so both proceeded to the full sample size as planned.

Interim analyses

A single interim analysis was planned for each RCT, after recruitment and follow-up to day 60 of 300 patients. At this point, the following end points were to be analysed in the intention-to-treat (mITT) population only:

• Time to liberation from respiratory support, which will be tested using an unadjusted log-rank test, with early termination of the trial recommended if any one arm is shown to be superior with p < 0.001 (Peto–Haybittle stopping rule).
• Mortality to day 60, which will be tested using a log-rank test, with early termination of the trial recommended if any one arm was shown to be superior with p < 0.05.

For this interim analysis, patients discharged from hospital alive with no further death after discharge recorded are assumed to be alive on the day of data extract. Patients who have withdrawn or refused consent for access to medical records will be censored on the date of withdrawal or refusal of consent.

Due to a higher than anticipated recruitment rate, the step-down RCT interim analysis was not performed (see Amendments to the protocol).

Clinical effectiveness analysis

Screening

Screening logs will be used to record all patients who are admitted or accepted for admittance to critical care (step-up RCT), and all patients extubated during critical care unit stay (step-down RCT). The following summaries will be presented:

• Number and percentage of patients who did not meet inclusion criteria, overall and by criteria.
• Of the patients who met the inclusion criteria, number and percentage who met exclusion criteria, overall and by criteria.
• Of the eligible patients (i.e. met inclusion criteria and did not meet exclusion criteria), number and percentage not randomised, overall and by reason (if known).

Recruitment

Consolidated Standards of Reporting Trials flow diagrams for the ITT and per-protocol populations will be completed for each trial.

Consent

The parent/legal guardian of trial participants will be asked to consent to the study as soon appropriate and practical after randomisation (usually within 24–48 hours of randomisation but the timing will vary according to the child’s clinical situation). They may consent to any one or more of the following aspects: trial continuation (i.e. treatment); access to medical records for ongoing data collection; completion of the parental stress questionnaire (at/around the time of consent); to receive a follow-up questionnaire at 6 months post randomisation; sharing of anonymised data to support future research; to be contacted regarding future research participation. When consent is refused for access to medical records (regardless of whether or not consent has been given for trial continuation), all trial data collection should cease and no data linkage to PICANet or NHS Digital should be performed. Data collected by site staff directly to the trial CRF up to the point of consent refusal will be retained and used for analysis, but no events after this point will be recorded or reported on. If any data have already been obtained via linkage from PICANet or NHS Digital, these data will be deleted.

Where consent has been refused for trial continuation, but granted for access to medical records, data collection and linkage may continue, and the patient may be included in the analysis as appropriate for each end point.

If consent is refused for access to medical records and/or trial continuation, the parental stress questionnaire may still be completed and reported on if this has been consented to.

Exposure

Exposure to the intervention will be assessed by the following parameters, which will be calculated for each treatment group and summarised using descriptive statistics [mean, standard deviation (SD), median and interquartile range (IQR), or counts and percentages for binary and categorical variables] unless otherwise specified:

• In patients randomised to CPAP, pressure (in cm H₂O as a continuous variable, and grouped as < 7 cm, 7–8 cm, > 8 cm), by hour during the first 6 hours from randomisation.
• In patients randomised to HNFC, flow rate (as percentage of recommended starting rate, and grouped as ≤ 50%, 51–75%, 76–85%, 86–95%, ≥ 95% of recommended starting rate), by hour during the first 6 hours from randomisation.
• Time from first recorded observation meeting weaning/failure/stopping criteria to time of weaning/switch or escalation/treatment stop.

Further treatment patterns across each group and time from first meeting weaning criteria to start of weaning attempt will be explored using summary statistics and graphic methods only, no formal statistical testing will be performed.

Protocol adherence

The number and percentage of patients affected will be reported for each of the following potential protocol deviations:

• Did not start randomised treatment (i.e. first recorded respiratory support post randomisation is not the randomised treatment).
• Switched or escalated from randomised treatment without meeting treatment failure criteria.
• Weaning attempt made, when weaning criteria are not met in last recorded observation prior to weaning.
• Respiratory support is discontinued while FiO₂ ≥ 0.3 and moderate or severe respiratory distress is present.

Safety

Adverse events (nasal trauma, facial/neck trauma, abdominal distension, pneumothorax, pneumomediastinum, subcutaneous emphysema, facial thermal injury, respiratory arrest, cardiac arrest), and any other possibly related AE, are recorded only in patients who commenced respiratory support post randomisation, and are recorded from randomisation up to 48 hours after date/time of liberation of respiratory support.

The percentage of patients experiencing one or more AEs in patients who commenced respiratory support post randomisation will be compared between groups using Fisher’s exact test. Counts and percentages of AEs, and SAEs, overall and by type, will be presented by allocated treatment group.

Withdrawal/follow-up

Once given, consent can be withdrawn at any time up to the end of the study. Data collected up to the point of non-consent or withdrawal of consent to data collection will be retained.

Timing of outcome assessments

Following randomisation, details of respiratory support (type of support, flow rate/pressure), physiological parameters (respiratory rate, heart rate, SpO₂, FiO₂) and measures of patients' comfort (respiratory distress scored as none/mild/moderate/severe, sedation delivered yes/no, and COMFORT-B scores) are recorded hourly for the first 6 hours, and 6 hourly thereafter until the end of respiratory support (or to at least 48 hours following randomisation, if patients are transferred to another unit or ward).

Survival status is recorded at unit discharge, at ultimate discharge from critical care (if the patient has been transferred to another critical care unit) and at discharge from acute hospital. Where consent is given for access to medical records, longer-term survival is collected from linked NHS Digital records.

Parental stress is measured using the PSS at the time of consent, which is expected to be within 24–48 hours of randomisation. Pediatric Quality of Life Inventory (PedsQL) and the Child Health Utility 9 Dimension (CHU-9D) and health services/resource use is assessed at 6 months post randomisation.

Analysis methods

Baseline patient characteristics

Baseline data are collected at critical care admission via data linkage to PICANet, and directly via trial CRF for physiology at randomisation. The following baseline demographic and clinical data will be summarised in the mITT and per-protocol populations, by allocated treatment group (using mean, SD, median and IQR, or counts and percentages for binary and categorical variables) but not subjected to statistical testing:

In both RCTs

• Age (years) – median and IQR, and number and percentage by age group (≤ 28 days, 29–180 days, 181–364 days, 1 year, 2 years, 3 years, 4 years, 5–10 years, 11–15 years).
• Sex (male, female) – number and percentage.
• Respiratory distress at randomisation – number and percentage by category.
• Heart rate at randomisation (both as absolute values, and converted to centile for age) – median and IQR, mean and SD.
• SpO₂ at randomisation – median and IQR, mean and SD.
• FiO₂ at randomisation – median and IQR, mean and SD.
• Ratio of SpO₂ : FiO₂ at randomisation – median and IQR, mean and SD.
• COMFORT-B score at randomisation (last available) – mean and IQR, number and percentage with COMFORT-B score ≥ 23 (representing possible distress1).
• Comorbidities – number and percentageby type of comorbidities (as specified on the CRF).

Step-up RCT only

• Main reason for admission to critical care – number and percentage.
• Any respiratory support received in 24 hours prior to randomisation (overall, and by type and duration of support) – number and percentage.
• Whether on respiratory support at time of randomisation – number and percentage.
• Received general anaesthesia for surgery/procedure in the 6 hours preceding randomisation – number and percentage.

Step-down RCT only

• Main reason for invasive ventilation.
• Duration of invasive ventilation – median and IQR, and number and percentage with duration < 5 days, number and percentage with duration ≥ 5 days.

Primary outcome

The median (with 95% CI) time to liberation from respiratory support will be reported for each arm using Kaplan–Meier estimates, and compared between groups using Cox regression, unadjusted and adjusted for important baseline characteristics (including shared frailty at the site level). The covariates for inclusion in the regression models are the following, which have been selected a priori based on an established relationship with outcome for critically ill children:

In both RCTs

• Age (< 12 months vs. ≥ 12 months).
• Severity of respiratory distress at randomisation (severe vs. mild/moderate).
• SpO₂ : FiO₂ ratio at randomisation (linear).
• Comorbidities (none vs. neurological/neuromuscular vs. other).

Step-up RCT only

• Reason for admission [bronchiolitis vs. other respiratory (airway problem, asthma/wheeze or any other respiratory) vs. cardiac vs. other (neurological, sepsis/infection, any other)].
• Whether the patient was on NRS at randomisation (yes/no).

Step-down RCT only

• Length of prior IMV (< 5 days vs. ≥ 5 days).
• Reason for IMV (cardiac vs. other).

The primary effect estimate will be the adjusted HR, reported with a 95% CI. HFNC will be considered non-inferior to CPAP if the lower bound of the 95% CI is above 0.75 in both the mITT and per-protocol populations. Patients without a recorded time of liberation will be censored at date and time of death (for patients who died while on treatment) or at date and time of last recorded respiratory support. The assumption of proportional hazards will be explored by fitting a Cox model with time dependent covariates.

Subgroup analyses will be performed to test for interactions between the effect of allocated treatment group and the following baseline covariates, with groupings defined as for the adjusted model specification above:

In both RCTs

• Age.
• Severity of respiratory distress at randomisation.
• SF ratio at randomisation.
• Comorbidities.

Step-up RCT only

• Reason for admission.
• Whether the patient was on NRS at randomisation.

Step-down RCT only

• Length of prior IMV.
• Reason for respiratory support post extubation, categorised as planned (randomisation followed by extubation), indeterminate (extubation followed by randomisation within 60 minutes of extubation) versus rescue (extubation followed by randomisation more than 60 minutes post-extubation) breathing support.
• Reason for IMV.

The interaction effect for linear covariates (SF ratio) will be illustrated by calculating the adjusted HR within five categories at quintiles of the continuous variable.

Planned sensitivity analyses included a repeat of the primary analysis using alternative durations: from start of respiratory support to liberation from respiratory support; from randomisation to start of weaning; and from randomisation to meeting weaning criteria. A post hoc analysis was performed to assess the effect of patients who did not start any respiratory support by including them in a sensitivity analysis of the primary end point that assigned them to a nominal 2 hours of respiratory support.

Secondary outcomes

Binary outcomes [mortality at discharge from critical care, at 60 and 90 days post randomisation, (re-)intubation at 48 hours, sedation use during randomised treatment, sedation use during HFNC or CPAP] will be reported in each treatment group, in the per-protocol and mITT populations. Absolute risk reduction and unadjusted odds ratios (ORs) will be reported with 95% CIs. Multilevel logistic regression (adjusted for the same baseline variables as the adjusted analysis of the primary outcome) will be used to calculate adjusted ORs with 95% CIs.

Continuous outcomes (duration of PICU and hospital stays) will be summarised by treatment groups, stratified by survival status, in the per-protocol and mITT populations. Mean difference between groups will be calculated, with 95% CI using bootstrapping to account for anticipated non-normality in the distribution.

Duration of survival to day 180 will be plotted as Kaplan–Meier survival curves, in the per-protocol and mITT populations, and unadjusted and adjusted HRs with 95% CIs will be calculated using Cox regression models.

Parent-/patient-reported outcomes (PSS:PICU score, PedsQL score) will be summarised by treatment groups, in the per-protocol and mITT populations. Mean difference between groups will be calculated, with 95% CI using bootstrapping to account for anticipated non-normality in the distribution. Linear regression will be used to calculate adjusted mean differences.

For each patient, their median COMFORT-B score while on randomised treatment, and their median COMFORT-B score while on either HFNC or CPAP, will be calculated. These median scores will be summarised by treatment groups, using median (IQR) and mean (SD). The number and percentage of patients with any recorded COMFORT-B score ≥ 23 while on randomised treatment, and the number and percentage of patients with any recorded COMFORT-B score ≥ 23 while on either HFNC or CPAP will be reported. Mean difference between groups will be calculated, with 95% CI using bootstrapping to account for anticipated non-normality in the distribution. Linear regression will be used to calculate adjusted mean differences.

Health economic analysis

Methods overview

A full cost-effectiveness analysis (CEA) was undertaken to assess the relative cost-effectiveness of HFNC versus CPAP in acutely ill children admitted to a critical care unit clinically assessed to require NRS. The CEA was undertaken for both step-up and step-down RCTs and compared the costs and health economic outcomes of treatment comparators within each trial over 6 months after randomisation. The CEA used patient-level healthcare resource use and outcome data collected as part of the trial databases linked to routine data from PICANet (the national clinical audit of PICUs) and NHS Digital’s HES data sets and through completion of the follow-up Health Services Questionnaire (HSQ) to report cost-effectiveness at 6 months. The cost analysis adopted a health and personal health services perspective.46

Health-related quality of life was measured using the age-appropriate Pediatric Quality of Life Generic Core Scales (PedsQL) questionnaire, which was then mapped onto preference-based CHU-9D score to estimate HRQoL. HRQoL data were combined with survival data to report preference weighted quality-adjusted life-years (QALYs), which is the preferred outcome measure for CEA. The base-case CEA followed the mITT principle and reported the mean (95% CI) incremental costs, QALYs and incremental net monetary benefit (INB) at 6 months of HFNC versus CPAP, overall and for the same prespecified subgroups as for the evaluation of clinical effectiveness. Patients who did not consent to being sent the questionnaire or to their data being used in the CEA were not included in this part of the study. The main assumptions of the CEA were subjected to extensive sensitivity analyses. The study used similar CEA methods for the step-up and step-down RCTs, which are described below.

Resource use and costs at 6 months

The resource use associated with hospital stays (index admission and re-admissions up to 6 months), and visits to outpatients and community healthcare services were chosen a priori as possible drivers of incremental costs. The interventions, CPAP and HFNC, are provided in the usual care practice and their costs are assumed to be included within the hospital bed-day costs. No additional intervention costs, such as cost of device, consumables or staff time were deemed required for delivering the interventions in PICU/HDU. Total costs were calculated by combining the resource use with unit costs at 2020–1 prices (£ GBP).

The length of stay within PICU/HDU and general medical ward from index and re-admission were extracted from FIRST-ABC CRFs linked to PICANet database and follow-up HSQ. The duration and location of the index hospital admission, that is, the combined duration of any stay in PICU/HDU and in the general medical ward following randomisation, were recorded for each patient on the CRF. The length of stay in the initial PICU/HDU admission during the index admission was calculated as the total duration in days (including fractions of days), from the date and time of randomisation until the time of discharge to the medical ward or death, including duration of transfers to other critical care units. Subsequent re-admissions into PICU/HDU units following initial discharge to the ward were also measured. Each activity day within paediatric critical care was assigned a Healthcare Resource Group (HRG) applying the 2019–20 HRG4+ Grouper algorithm.47 The cost of paediatric critical care bed-days was valued according to realistic scenarios reflecting workload. Patients receiving HFNC or CPAP in a PICU or HDU were considered to receive the same level of care. Intermediate paediatric critical care (XB06Z HRG code) was considered the appropriate level of critical care associated with providing the interventions. The cost of HFNC use in the paediatric ward following discharge from HDU/PICU was costed as basic critical care (XB07Z HRG code).

A hospital re-admission was defined as a further hospital admission following ultimate discharge from the index hospital admission. Information on re-admissions was collected from two sources. Firstly, data on re-admissions to critical care were accessed from the trial CRFs linked to the PICANet database.48 From these databases, information was accessed on the duration of stay within the critical care unit as well as the total hospital stay including subsequent transfers to other care areas (e.g. general medical wards) within the same hospital and to other hospitals. Secondly, information on re-admissions that did not include a further stay in critical care wards was collated from responses to the HSQ administered to patients surviving to 6 months post randomisation.

The resource use associated with re-admissions to hospital, hospital outpatient visits and community services use following discharge from the index hospital admission but before 6 months post randomisation were collected via HSQ. In sensitivity analysis, use of health services following discharge from the index hospital admission was derived from HES.

Unit cost

The unit costs required for valuing the resource use data were taken from national unit cost databases and are listed in Appendix 1. The costs per critical care bed-day by HRG and general medical bed-day were taken from the NHS benchmark prices (personal communication). Unit costs for hospital outpatient visits and community service use were obtained from a recommended published source for health and social care costs.49 All unit costs were reported in 2020–1 prices.

Health-related quality of life, life-years and quality-adjusted life-years up to 6 months

Health-related quality of life data were collected via age-appropriate Pediatric Quality of Life Generic Core Scales (PedsQL™) questionnaires sent to patients at 6 months post randomisation.50 The responses to PedsQL questionnaires from eligible surviving patients at 6 months were mapped onto the CHU-9D index score.51,52

FIRST-line support for Assistance in Breathing in Children trial data were linked with national death registrations using the Medical Research Information Service Database Administrative System held by NHS Digital. Information on the date and time of deaths was used to calculate the survival time and life-years up to 6 months for each randomised patient. QALYs at 6 months post randomisation were calculated by valuing each patient’s survival time by their HRQoL (CHU-9D utility score) at 6 months according to the ‘area under the curve’ approach.53 For 6-month survivors, QALYs was calculated using the CHU-9D scores at 6 months, assuming a CHU-9D score of zero at randomisation, and a linear interpolation between randomisation and 6 months. An alternative published algorithm for mapping PedsQL to the CHU-9D index score was considered in a sensitivity analysis. For decedents between randomisation and 6 months, a zero QALY was assumed for the 6-month period.

Incremental net monetary benefit at 6 months

The INB of HFNC compared to CPAP at 6 months was calculated by multiplying the mean gain or loss in QALYs by a recommended cost-effectiveness threshold in the UK (£20,000 per QALY gained) and subtracting the difference in costs.46

Statistical analysis of cost-effectiveness analysis at 6 months

Cost-effectiveness analyses for both step-up and step-down RCTs followed the mITT principle, and reported incremental costs, QALYs and cost-effectiveness up to 6 months, according to randomised group. Missing data in baseline covariates, resource use and outcomes (see Appendix 2 and Appendix 3) were handled with multivariate imputation via chained equations (MICE).54 Under this approach, each variable was imputed conditional on fully observed baseline variables, resource use, outcomes and all other imputed variables. Patients who did not return or fully completed the PedsQL questionnaire administered at 6 months had their HRQoL scores imputed from those survivors who did fully complete the questionnaire. Similarly, for those eligible patients who did not return the HSQ, information on the use of outpatient services up to 6 months post randomisation was imputed from those patients who completed and returned the HSQ.

Analysis of uncertainty and sensitivity in cost-effectiveness at 6 months

We estimated the incremental costs and QALYs from the imputed data sets with a single-level bivariate seemingly unrelated regression model to allow for correlation between costs and QALYs. The incremental results from multiply imputed data sets were summarised using Rubin’s rule.55 The economic analysis was adjusted for same baseline covariates as for the clinical analysis to adjust for baseline imbalances between the randomised arms (see Clinical effectiveness analysis section for further details). To express the uncertainty in the estimation of the incremental costs and QALYs, we used the estimates of the means, and variances from the single-level regression model, to generate 800 estimates of incremental costs and QALYs from the joint distribution of these end points, assuming asymptotic normality. The uncertainty around the differences in average costs and QALYs at 6 months between the treatment groups was illustrated on the cost-effectiveness plane.56

Subgroup analyses

We undertook prespecified subgroup analysis as per analysis of clinical effectiveness, for both step-up and step-down RCTs and reported INBs for each subgroup.

Sensitivity analyses

The main assumptions made in the base-case analysis and how they were relaxed in sensitivity analyses are described below (see also Table 3).

• Analysis principle: The base-case analysis followed the mITT population as per the primary clinical analysis. The sensitivity analysis was performed according to the ITT principle.
• Intervention costs irrespective of location: The base-case analysis considered intervention costs of HFNC or CPAP to be the same across locations (PICU or HDU) and equivalent to the cost of intermediate paediatric critical care (XB06Z HRG code). In the sensitivity analysis, we examined whether the cost-effectiveness results are sensitive to intervention base costs irrespective of location, by assuming the cost of HFNC is equivalent to basic critical care (XB07Z HRG code) and CPAP is equivalent to intermediate critical care (XB06Z HRG code).
• Follow-up costs from HES database: The base-case analysis included follow-up costs calculated from responses to the HSQ. In the sensitivity analysis, follow-up costs derived from resource use information from HES databases were used.
• Unit costs of resources: The unit costs of resources in the base-case analysis were taken from NHS benchmark prices and Personal Social Services Research Unit (PSSRU) costs in 2020–1. In the sensitivity analysis, unit costs were increased/decreased by 10% to allow for possible changes in unit costs in recent years.
• CHU-9D mapping: The HRQoL in the base-case analysis was derived using a mapping algorithm of PedsQL responses to the CHU-9D utility score derived in the UK context.51 In the sensitivity analysis, we consider an alternative mapping algorithm derived using responses to both questionnaires from a paediatric population from Australia.57
• Distributional assumptions for costs and QALYs: The base-case analysis assumed that costs and QALYs were normally distributed when reporting the 95% CIs around incremental costs, QALYs and INB. In a sensitivity analysis, we assessed the robustness of CEA results to alternative distributional assumptions about both outcomes. Following methodological guidance,58,59 the sensitivity analysis considered a gamma distribution for costs given the observed right-skewed distribution.60 The sensitivity analysis also considered a gamma distribution for QALYs given the large proportion of decedents with zero QALYs and the observed right-skewed distribution of QALYs for patients who were alive at 6 months.
• Modelling assumption: The base-case analysis model followed single-level bivariate regression model to allow for correlation between costs and QALY. The sensitivity analysis followed a multilevel bivariate regression model to allow for clustering of patients at sites.
• Analysis model and missing data: The base-case analysis was based on full eligible trial participants in which missing data were imputed using MI methods. In the sensitivity analysis, complete case analysis was performed excluding patients whose costs/outcome data were missing.

The results of the sensitivity analyses were reported as mean INBs with corresponding 95% CIs.

Handling of missing data

As the primary end point will be analysed using time-to-event methods, patients with missing data will be included in the analysis as censored at the point of last recorded NRS. Time to censoring will be compared between arms using Kaplan–Meier curves to explore the assumption of censoring at random.

Multiple imputation will be used to complete missing data in secondary outcomes, costs and HRQoL, under the assumption that the responses are missing at random conditional on the observed data. Multiple imputation will be undertaken using the MICE algorithm, with the model including all baseline variables included in the adjusted models and all outcome variables. The number of imputations will be determined according to level of missingness in the outcome variables. Models will be fitted in each imputed data set and results combined using Rubin’s rules.55

Statistical software

All analyses were conducted in Stata/SE Version 14.2 64-bit x86-64 (StataCorp LP, College Station, TX, USA). Some additional CEAs were carried out in R (The R Foundation for Statistical Computing, Vienna, Austria) as required.