Results: step-down randomised controlled trial

Sites and patients

Site selection and set-up

Expressions of interest and completed site feasibility questionnaires were received from 22 potential sites across England, Wales and Scotland. All sites were invited to take part in the trial, of which two were ultimately not able to progress with set-up (one due to research capacity issues and one due to equipoise issues).

In total, 20 sites, covering 22 PICUs, recruited patients into the FIRST-ABC step-down RCT. The first sites opened in August 2019, on schedule, and the final sites opened in January 2020. By the end of the internal pilot, 20 sites were open to recruitment. Characteristics of the participating PICUs are compared with non-participating PICUs in Table 15.

In relation to research governance, the median time from provision of the final local information pack to the issuing of local confirmation of capacity and capability was 39 (IQR 19–98) days. The median time from local confirmation of capacity and capability to the start of patient screening at sites was 18 (IQR 6–59) days. The median time from the start of patient screening to the first patient recruited at sites was 5 (IQR 2–13) days. In total, the process from provision of the final local information pack to first patient recruited took a median of 101 (IQR 64–130) days.

Of the 20 sites, 15 formally paused recruitment at least once due to the impact of COVID-19 on clinical and research activity. However, 91.2% of the sample size were already recruited prior to March 2020, limiting the overall impact on trial recruitment. Overall, each site participated in the step-down RCT for a median (IQR) of 5.6 (4.0–6.6) months.

Patient screening, randomisation and consent

Between August 2019 and May 2020, 3121 patients were extubated following a period of IMV at participating PICU/HDUs and screened for inclusion in the step-down RCT, from which 1397 were deemed to meet inclusion criteria (Figure 14). Of those not meeting inclusion criteria (n = 1724), most did not require NRS within 72 hours of extubation (n = 1700, 98.6%).

Of patients meeting inclusion criteria, 346 additionally met exclusion criteria and were not eligible for the trial. The most common reasons for exclusion were: a clinical decision to commence a form of NRS other than HFNC or CPAP (n = 153, 44.2%) and receipt of home non-invasive ventilation prior to admission (n = 64, 18.5%).

Of the remaining 1051 eligible patients, 600 (57.1%) were randomised into the trial and 451 (42.9%) were not. The most common reasons for eligible patients not being enrolled in the trial were that patients were either missed or identified too late for recruitment (n = 157, 34.8%) or a clinical decision was made not to enrol the patient (n = 263, 58.3%). Regarding the latter, a greater number of patients were not randomised due to the clinician preferring HFNC rather than CPAP. The full sample were recruited 5 months ahead of the schedule (Figure 15).

Among the 600 enrolled, 299 were randomised to the HFNC group and 301 to the CPAP group. After accounting for complete withdrawals, 291 in the HFNC group and 296 in the CPAP group formed the ITT population. In the HFNC group, 281 started respiratory support following randomisation compared with 272 in the CPAP group, forming the primary analysis set. Of those randomised to HFNC, 272 started HFNC, while 252 in the CPAP group started CPAP as randomised, forming the per-protocol population (see Report Supplementary Material 5).

In the primary analysis set, the primary outcome was observed in 260 in the CPAP group and in 258 in the HFNC group, with 12 and 23 censored in each group, respectively. In the per-protocol population, the primary outcome was observed in 244 in the CPAP group and in 248 in the HFNC group.

Baseline characteristics

The groups were well matched at baseline (Table 16), except for a higher proportion of children receiving ventilation for cardiac reasons in the HFNC group (28.8% vs. 20.2%). The median age in months was 3, and just over three quarters were aged < 1 year, in both groups. The median (IQR) duration of prior IMV was 89 (56–145) hours in the HFNC group and 87 (51–140) hours in the CPAP group, with most patients randomised prior to extubation (63.3% vs. 61.8%, respectively), indicating a pre-planned use of NRS post-extubation. Baseline characteristics of the per-protocol population were similar to those of the primary analysis set (Table 17). The baseline characteristics of the full ITT population (i.e. including the small number of patients who did not start any respiratory support following randomisation) are shown in Report Supplementary Material 6.

Clinical management

In both groups, the randomised treatment was started in the majority of children who started respiratory support [272/281 (96.8%) in the HFNC group, 252/272 (92.6%) in the CPAP group]. A variety of devices and interfaces were used to delivery HFNC and CPAP (Report Supplementary Material 7). Adherence to the trial algorithms was good, with most patients receiving HFNC at the prescribed gas flow rates (Figure 16) and CPAP at the specified pressure level (Figure 17).

A flow of the clinical management in each randomised group is shown in Figure 18. More patients (101/272, 37.1%) in the HFNC group experienced treatment failure compared to the CPAP group (85/252, 33.7%) (Report Supplementary Material 8). Almost a quarter (23.5%) of those experiencing treatment failure in the HFNC group switched to CPAP, whereas only 12.3% in the CPAP group switched to HFNC. Patients in the HFNC group were most often switched due to severe respiratory distress, whereas patients in the CPAP group were most often switched due to patient discomfort reasons (Report Supplementary Material 8).

Clinical effectiveness

Primary outcome

The median time from randomisation to liberation from respiratory support was 50.5 hours (95% CI 43.0 to 67.9) in the HFNC group and 42.9 hours (95% CI 30.5 to 48.2) in the CPAP group (adjusted HR 0.83; one-sided 97.5% CI 0.70 to ∞). The bound of the one-sided 97.5% CI (adjusted HR 0.70) was below the prespecified non-inferiority margin (HR 0.75) (Figure 19; seeTable 18). Results were similar in the per-protocol analysis (adjusted HR 0.82, 95% CI 0.68 to 0.98) (Figure 20; see Table 18) (see Report Supplementary Material 9). Proportional hazards assumptions were checked by examining plots of −ln(−ln(survival)) over time and scaled Schoenfeld residuals from each Cox proportional hazards model. No evidence of departures from proportional hazards was observed. A breakdown of the duration of invasive and NRS prior to liberation in the primary analysis set is shown in Table 19 and the time to liberation according to whether treatment failure occurred is shown in Report Supplementary Material 9.

Subgroup analyses

There were no significant differences in treatment effects across prespecified subgroups (Figure 21).

Secondary outcomes

The rate of reintubation within 48 hours was not significantly different between the groups (HFNC, 13.3%; CPAP, 11.5%; adjusted OR, 1.11, 95% CI 0.7 to 1.9) (see Table 18). The time to reintubation was a median of 25 hours (IQR, 8–79) after randomisation for HFNC compared with 11 hours (IQR, 2.5–49) for CPAP (see Table 18 and Report Supplementary Material 10).

There was no significant difference in the duration of critical care unit stay or acute hospital stay between the groups. By day 180, there was a higher rate of mortality in the HFNC group (15/268, 5.6%) compared to the CPAP group (6/253, 2.4%) (adjusted OR 2.44, 0.9–6.4) (see Table 18). Kaplan–Meier survival plots to day 180 are the shown for primary analysis set in Figure 22 and for the per-protocol population in Figure 23. Characteristics of the patients who died by day 180, by randomised group, are shown in Report Supplementary Material 11.

Adverse events

The number of participants with one or more AEs was low in both groups, occurring in 25 out of 281 (8.9%) in the HFNC group and in 28 out of 272 (10.3%) in the CPAP group (Table 20). The proportion of patients who experienced one or more SAEs was the same in both groups (1.8%).

Sensitivity analyses

Planned and post hoc sensitivity analyses did not alter interpretation of the primary outcome (Report Supplementary Material 12). When repeating the primary analysis in the full ITT population (assigning a nominal 2 hours of respiratory support to patients who did not start any respiratory support following randomisation), the adjusted HR was 0.80 (95% CI 0.68 to 0.96) (Figure 24).

Cost-effectiveness analysis

Resource use and costs up to 6 months

Table 21 shows the resource use up to 6 months post randomisation. Patients in the HFNC compared to the CPAP group had, on average, shorter stays in paediatric critical care units (7.43 vs. 7.88 days) but longer stays in general medical ward (11.63 vs. 10.72 days) during the index hospital admission. Up to 6 months post randomisation, re-admissions to critical care units and general medical wards were higher in the HFNC compared to the CPAP group, but the mean lengths of stay in both critical care unit and general medical wards from re-admissions were lower in the HFNC group. The mean total lengths of stay up to 6 months in HFNC and CPAP group were 26.82 days and 28.49, days respectively. Overall, total length of stay up to 6 months was lower in the HFNC group than the CPAP group.

Resource use reported from responses to the HSQ is summarised in Report Supplementary Material 13. Patients in the HFNC group had on average more outpatient visits and contacts with GP and nurse than the CPAP group. Patients in the CPAP group had higher average number of visits to occupational therapist, physiotherapist and social worker. All other community care contacts up to 6 months were low and similar between the randomised groups.

Total costs (GB£)

Table 22 reports total costs per patient up to 6 months post randomisation. For both randomised groups, costs of the index hospital stay account for a major share of total costs associated with the provision of HFNC and CPAP in this population, accounting for up to 75% and 71%, respectively, of the total cost at 6 months. Within the index admission, costs in the critical care unit were lower in the HFNC compared to CPAP group (£11,755 vs. £12,592), and higher that than those of general medical ward costs. The re-admission costs were lower in the HFNC group than in the CPAP group. The re-admission costs of stays in general medical wards were higher than the costs of stays in critical care units for both randomised groups. The costs of outpatient visits and community healthcare services were lower in the HFNC group than in the CPAP group (£984 vs. £1013). The mean total costs per patient were lower in the HFNC group (£28,275) compared to the CPAP group (£30,303).

Health-related quality of life

The health status profiles reported from responses to the PedsQL questionnaires administered at 6 months post randomisation are summarised by randomised group in Table 23. The mean overall scores across the four PedsQL dimensions were similar between HFNC and CPAP patients (79 vs. 81 in the physical dimension, 71 vs. 73 in the emotional dimension, 89 vs. 89 in the social dimension, and 79 vs. 77 in the school functioning dimension). The mean total PedsQL scores at 6 months were high in both randomised groups (77 and 79 in HFNC and CPAP group, respectively).

The mean mapped CHU-9D utility scores at 6 months were similar between the randomised groups (Table 24). At 6 months, HFNC compared to CPAP group had lower life-years. The resultant mean QALYs at 6 months were slightly lower in the HFNC (0.218) compared to CPAP (0.227) group.

Cost-effectiveness

The unadjusted incremental cost of HFNC compared to CPAC was −£2028, and this estimate was surrounded by a wide 95% CI that included zero (Table 25). After adjustment for baseline characteristics, the estimated incremental cost of HFNC compared to CPAP was −£4565, but the 95% CI around it was wide. HFNC compared to CPAP results in fewer life-years and QALYs at 6 months, but the magnitude of the effect on both end points was small. The cost-effectiveness plane shows that the majority of points representing mean incremental costs and mean incremental QALYs are in the third (south-west) quadrant of the cost-effectiveness plane, indicating that HFNC had lower mean QALYs and lower costs (Figure 25). The INB for HFNC versus CPAP was positive, but there is considerable statistical uncertainty around INB; at £20,000 per QALY, the INB from adjusted analysis was £4388 (95% CI −£2551 to £11,307).

FIGURE 15 Step-down RCT – actual vs. expected patient randomisation. Comparison of actual vs. expected cumulative randomisation of patients into the FIRST-ABC step-down RCT. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 16 Step-down RCT – HFNC flow rates during the first 6 hours of treatment. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 17 Step-down RCT – CPAP pressures during the first 6 hours of treatment. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 18 Step-down RCT – clinical management of trial patients. a, Time to liberation was censored in five patients started on HFNC and three patients started on CPAP. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 19 Step-down RCT – time to liberation from respiratory support in the primary analysis set. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 20 Step-down RCT – time to liberation from respiratory support in the per-protocol population. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 21 Step-down RCT – subgroup analyses of the primary outcome in the primary analysis set. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 22 Step-down RCT – Kaplan–Meier survival curve in the primary analysis set. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 23 Step-down RCT – Kaplan–Meier survival curve in the per-protocol population. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 24 Step-down RCT – time to liberation from respiratory support – post hoc sensitivity analysis in all randomised children including those who were not started on respiratory support. a, To enable the inclusion of all randomised patients we assigned a minimal time to liberation of 2 hours in those who did not start any respiratory support following randomiation, and repeated the primary analysis. Reproduced with permission from Ramnarayan et al.32 Copyright © 2022 JAMA. All rights reserved. (Figure not reproduced here; see the original article.)

FIGURE 25 Step-down RCT – uncertainty in the mean costs (GB£) and QALY differences and their distribution for HFNC vs. CPAP (within 6 months post randomisation). (Figure not reproduced here; see the original article.)

The results of the subgroup analyses presented in Report Supplementary Material 14 show that the INBs were similar across all subgroups. For the neurological/neuromuscular, cardiac and SF ratio (quintile 2) subgroups, mean INB associated with HFNC versus CPAP was positive. For all subgroups, as for the overall results, the 95% CIs around the INB included zero.

Report Supplementary Material 15 reports the mean INB and 95% CI around it (at £20,000 per QALY gain) according to alternative assumptions, compared with the base case (first row). The INB estimates remain around the base-case INB (£4388), whether analysis sample is ITT, or alternative assumption for costs (intervention costs irrespective of location, follow-up costs from HES database, 10% increase or decrease in unit costs), alternative mapping algorithm for CHU-9D, and alternative analysis model is considered (complete case analysis, gamma distribution for costs and QALYs, and multilevel model). These sensitivity analyses showed that the results were robust to alternative scenarios.