The clinical effectiveness and cost-effectiveness of heated humidified high-flow nasal cannula compared with usual care for preterm infants: systematic review and economic evaluation

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 14/151/03. The contractual start date was in January 2015. The draft report began editorial review in May 2015 and was accepted for publication in December 2015. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

none

Copyright statement

© Queen’s Printer and Controller of HMSO 2016. This work was produced by Fleeman et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Description of health problem

Respiratory problems are one of the most common causes of morbidity in preterm infants,1 that is infants born before 37 completed weeks of gestation. Respiratory distress syndrome (RDS), also known as hyaline membrane disease, is a serious medical condition in which the lungs of a newborn baby lack surfactant and are not functioning at a level that is able to provide their body with enough oxygen. 2–4 It is a particular problem for preterm infants, as surfactant is usually produced between weeks 24 and 28 of pregnancy. European data for 2010 show an incidence of RDS of 92% at 24–25 weeks’ gestation, 88% at 26–27 weeks’ gestation, 76% at 28–29 weeks’ gestation and 57% at 30–31 weeks’ gestation. 4 The proportion of infants with RDS has been reported to fall to around one-tenth of those born at 34 weeks’ gestation2 (although the proportion with ‘respiratory problems’ at weeks 34 to 36 may be around three times higher).

Clinically, RDS presents with early respiratory distress comprising cyanosis, grunting, inter- and subcostal retractions and tachypnoea, and if left untreated it may result in death from progressive hypoxia and respiratory failure. 4 Consequences of RDS include:3

• hypoxia, acidosis, hypothermia and hypotension

• bronchopulmonary dysplasia (BPD) also commonly known as chronic lung disease

• pulmonary haemorrhage

• apnoea of prematurity/bradycardia

• intraventricular haemorrhage (IVH).

Advances in care over the years have, however, resulted in significant decreases in mortality from RDS. 4,5 Although data on RDS mortality are not routinely collected in the UK, data from the USA show it has fallen from 2.89 per 1000 live births between 1969 and 19736 (or 2.6 per 1000 live births in 19707) to 0.37 per 1000 live births between 1987 and 19958 (or 0.4 per 1000 live births in 19947). This decrease in RDS is also reflected by a decrease in mortality from all causes as reported by a number of worldwide studies. 9

Epidemiology

According to the UK Office for National Statistics,10 there were 729,312 live births in England and Wales in 2012 and the gestational age was known and verified for 726,572 infants. Of these, 52,909 (7.3%) were born preterm, prior to 37 weeks. The majority (43,993, 83.1%) were born between 32 and 36 weeks, with 5693 (10.8%) born between 28 and 31 weeks, 2474 (4.7%) born between 24 and 27 weeks and 749 (1.4%) born before 24 weeks.

Birthweight is associated with gestational age. In England and Wales in 2012,10 the vast majority of infants born before 24 weeks or those born between 24 and 27 weeks weighed under 1500 g (99.5% and 96.2%, respectively). At between 28 and 31 weeks, 85.6% weighed 1000–2499 g and between 32 and 36 weeks 96.7% of those born weighed 1500–3999 g. 10

Infant mortality is associated with gestational age and birthweight, decreasing with advanced gestational age and increasing birthweight (Table 1). 10

Current treatment options for preterm infants

Over the years, several modalities for respiratory support have been developed. The treatments which have arguably had the largest impact in reducing mortality are the administration of surfactant5,7 and antenatal corticosteroids. 11 Improved methods of mechanical ventilation, regionalised perinatal care and continuous improvement in general neonatal care have also been highlighted as having an important impact, particularly in the period between 1970 and 1985, prior to the use of surfactant therapy in the 1990s. 5,7 Recently updated European Consensus Guidelines for the management of RDS in preterm infants4 highlight that, in many instances, the risk of a preterm birth is known and this should enable preterm infants at risk of RDS to be born in centres where appropriate facilities are available for stabilisation and ongoing respiratory support, including intubation and mechanical ventilation, following birth.

Once born, preterm infants require stabilisation. In practice, preterm infants who present with early respiratory distress may receive any one of the following interventions (described in more detail in the following sections):

1. mechanical endotracheal ventilation

2. nasal continuous positive airway pressure (NCPAP)

3. oxygen

4. nasal intermittent positive-pressure ventilation (NIPPV)

5. heated humidified high-flow nasal cannula (HHHFNC).

The technology: heated humidified high-flow nasal cannula

A number of differently branded HHHFNC devices exist including the Vapotherm 2000i (Vapotherm Inc., Stevensville, MD, USA) and the Fisher & Paykel Healthcare (Auckland, New Zealand, and Irvine, CA, USA) devices. Three main features are common to any HHHFNC device:15

1. a respiratory circuit with a means to maintain the temperature and, by extension, the humidity of the delivered gas until the distal end of the circuit

2. a humidifier to effectively warm and humidify respiratory gases

3. a nasal cannula with adapter that connects to the delivery circuit and which should allow little or no excess tubing between the end of the delivery circuit and the actual nasal prongs, thereby minimising further any potential for gas cooling and precipitation.

In addition to HHHFNC, variations of this technology exist in which gas flow is provided at a high rate but not heated [high-flow nasal cannula (HFNC)]. Unheated gas cannot be adequately humidified even if it passes through a humidifier. 26

With regard to gas flow rate, no optimal level exists. 15 One early study reported that the flow rate should vary from infant to infant depending on weight. 27 It has also been stated that gas flow rate should be adjusted according to clinical response, generally being increased for increasing respiratory distress or oxygen requirement and decreased for improving respiratory distress or decreasing oxygen requirement. 15 Unlike the nasal prongs for NCPAP (which fit tightly in the nares), the nasal cannulae for HHHFNC are smaller and looser fitting. Nasal cannulae size varies from infant to infant, this being dictated by the size of the infant’s nares. 18,20

The HHHFNC is gaining popularity and is increasingly used in clinical practice in many units in the UK and other countries, particularly in North America and Australasia. 28 This is largely because of the perceived greater ease of use of such devices compared with NCPAP, allowing both practitioners and family members to handle and care for infants more easily. 15,20,29 In addition, it is considered that HHHFNC should improve patient tolerance and outcomes: heat and humidity should prevent airway water loss, airway cooling, thickened secretions and nasal irritation, allowing high-flow rates without nasal drying or bleeding while the comparably lighter and easier-to-apply interface may lessen nasal septal damage. 15,20 Other perceived advantages compared with NCPAP include a reduction in the number of ventilator days, an improvement in weight gain and being able to introduce oral feeding earlier. 18,20

However, there are concerns about the unpredictability of the positive airway pressures generated by HHHFNC and the potential for infection. Unless the infant’s mouth is closed and the leak around the nares minimised, it is unlikely that nasal cannulae deliver a clinically relevant level of positive airway pressure,15 while in the absence of an effective way of controlling distending pressure there is also the theoretical risk of lung overdistension and pneumothoraces;18 pressure appears to be related to gas flow, prong size and patient size. 15 The potential for infection was discovered in 2005 when instances of Gram-negative bacteria known as Ralstonia spp. were reported from Vapotherm devices in the USA. 30 This led to the recall of all devices in January 2006 but the product returned to the market with US Food and Drug Administration approval in January 2007, with new instructions for use including the recommendation to utilise only sterile water in the system. 15,30

Evidence for the effectiveness of heated humidified high-flow nasal cannula from previous reviews

In 2011, a Cochrane review related to heated and non-heated HFNC by Wilkinson et al. 31 concluded that there was ‘insufficient evidence to establish the safety or effectiveness of HFNC as a form of respiratory support in preterm infants’. Evidence was derived from two randomised controlled trials (RCTs)32,33 comparing HHHFNC with NCPAP (including one RCT that was unpublished and halted early when the equipment was recalled33), a RCT comparing two types of HHHFNC equipment (Vapotherm vs. Fischer & Paykel)34 and a crossover trial comparing HHHFNC with a non-humidified high-flow device. 35 A whole range of efficacy and safety outcomes were considered by this review, none of which could be pooled for a meta-analysis. More recently, a meta-analysis by Daish and Badurdeen,36 which included three RCTs37–39 that were published after the Cochrane review, examined the effects of HHHFNC on extubation failure (i.e. need for reintubation) and BPD. No significant differences were found between HHHFNC and NCPAP for either outcome. It is worth noting that one of the trials included in the meta-analysis (Yoder et al. 39) included both preterm and term infants.

Rationale for the current review

The wide variety of indications reported in studies included in systematic reviews,31,36 surveys28,29,40,41 and guidelines20,42 support the need for updated evidence of the effectiveness of HHHFNC for a variety of indications, not simply following ventilation. Although a recent meta-analysis has been published examining extubation failure and the incidence of BPD for HHHFNC compared with NCPAP,36 there is also the need for a review of the evidence for other relevant outcomes and comparators.

Clarification of research question and scope

The aim of this project was to answer the question: what is the clinical effectiveness and cost-effectiveness of HHHFNC compared with usual care for preterm infants? This was carried out by a systematic review of the available evidence and the subsequent assessment of the cost implications. We conducted a primary analysis of HHHFNC compared with usual care for preterm infants following ventilation and a secondary analysis of HHHFNC with usual care for preterm infants with no prior ventilation.

Search strategy

The following databases were searched for eligible studies:

• MEDLINE and MEDLINE In-Process & Other Non-Indexed Citations (via OvidSP)

• EMBASE (via OvidSP)

• Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, Database of Abstracts of Reviews of Effectiveness, Health Technology Assessment database

• ISI Web of Science – Science Citation Index Expanded and ISI Web of Science – Proceedings (Index to Scientific and Technical Proceedings)

• PubMed (limited to the last 6 months).

Search terms included a combination of index terms (for the study population) and free-text words (for the technologies involved). No study design filters were applied. All databases were searched from 2000 to 8 September 2014, apart from PubMed which was searched from 1 March to 9 September 2014. The searches were then updated on 12 January 2015.

Details of the search strategies can be found in Appendix 1.

Trial and research registers were searched for ongoing trials and reviews including:

• Clinicaltrials.gov

• metaRegister of Controlled Trials and International Standard Randomised Controlled Trial Number Register

• World Health Organization International Clinical Trials Registry Platform

• PROSPERO systematic review register

• National Institute for Health Research Clinical Research Network Co-ordinating Centre Portfolio Database

• Turning Research into Practice Database Plus

• US Food and Drug Administration.

Bibliographies of previous reviews and retrieved articles were searched for further studies.

Study selection

A decision was made by the review authors to trial the new and freely available web-based software platform developed for the production of systematic reviews, including Cochrane Reviews. The citations identified were independently assessed for inclusion through two stages by two reviewers (YD and RD). Initially the reviewers independently scanned all the titles and abstracts identified (and de-duplicated) through the searching exercise to identify the potentially relevant articles to be retrieved. Full-text copies of the selected studies were then subsequently obtained and assessed again for inclusion using the inclusion and exclusion criteria outlined in Table 2. Disagreements were resolved by discussion at each stage. There was no need to consult a third reviewer.

Data extraction strategy

Data relating to study design and findings were extracted by one reviewer (VB) and independently checked for accuracy by a second reviewer (RD). Study details were extracted on pre-tested data extraction forms. Data from studies presented in multiple publications were extracted and reported as a single study with all other relevant publications listed. When studies included preterm and non-preterm infants, only data for preterm infants were extracted and study authors were contacted for missing data as necessary.

Assessing the risk of bias

The plan for the conduct of risk of bias of the individual studies was originally based on the Cochrane risk-of-bias criteria44 because the intention was to use the new and freely available web-based software platform developed for the production of systematic reviews, including Cochrane Reviews, for the entire review. However, it became clear that the data extraction tool used in this software did not allow us to easily produce tables for the review. We therefore opted to quality assess the included studies using criteria adapted from the Centre for Reviews and Dissemination at the University of York. 43 Criteria were assessed independently by one reviewer (VB) and then crosschecked by a second reviewer (YD). Disagreements were resolved through consensus and there was no need to consult a third reviewer.

Methods of analysis/synthesis

The results of the data extraction and quality assessment for each study were presented in structured tables and as a narrative summary for the primary analysis (preterm infants treated following ventilation) and secondary analysis (preterm infants with no prior ventilation). When data permitted, we conducted a meta-analysis of primary and secondary outcomes using an appropriate software package (RevMan; The Nordic Cochrane Centre, Copenhagen, Denmark). We also conducted subgroup analyses based on gestational age. We planned to use the categories < 30 weeks and ≥ 30 weeks (but the data did not permit us to use these specific thresholds once we had extracted the data). For dichotomous outcomes, we planned to use risk ratio (RR) and the corresponding 95% confidence intervals (CIs) to summarise results from each trial and for continuous outcomes, we planned to use the mean difference (or standardised mean difference when different scales are used). It was only possible to pool data for dichotomous outcomes.

The decision to conduct a meta-analysis depended on there being sufficient data (at least two studies with the same interventions and comparators measuring the same outcome in the same way) and an assessment of heterogeneity. Heterogeneity was explored through consideration of the study populations (e.g. differences in gestational age), interventions (e.g. starting flow rate for HHHFNC or starting pressure for NCPAP), outcome definitions (e.g. different definitions for reintubation) and, in statistical terms, by the chi-squared test for homogeneity and the I² statistic. 45 The I² statistic, with a level of > 50%, was considered to indicate moderate levels of heterogeneity, and the chi-squared test of < 0.10 to indicate statistically significant heterogeneity. Based on these assessments, a decision was made on whether to combine the results using a fixed-effects model (in the case of minimal heterogeneity) or a random-effects model (in the case of substantial levels of heterogeneity).

If data had allowed, we would have conducted sensitivity analyses excluding trials deemed to be of low quality to assess the robustness of the findings. Had we included ≥ 10 studies in a meta-analysis an assessment of the risk of publication bias would have been conducted by constructing a funnel point and conducting a simple test of asymmetry to test for possible bias. 46

Modelling clinical pathway and outcomes

The definition of the patient pathway was determined through consultation of one of the authors who was a clinician (BS) and the economic modeller (JM). The pathway that was developed is shown in Figure 1. Data required to populate this patient pathway were taken from the studies included in the review (see Chapter 4, Included studies).

FIGURE 1.

Treatment pathway.

It was determined that the pathway was best modelled as a decision tree, as there is no long-term progression of disease over time. It is assumed that any loss in utility from the primary outcome is once and for all and that any short-term loss in utility from, for example, nasal injury, is a one-off utility decrement before a return to the long-term prior health state.

The model time horizon could, in theory, be lifetime provided, and there was evidence from the clinical review that the difference in outcomes between technologies had lifetime consequences.

Analysis of uncertainty

If a formal economic model could be constructed, appropriate sensitivity analyses were planned in order to assess the robustness of model results to realistic variations in the levels of the underlying data. When the overall results are sensitive to a particular variable, the sensitivity analysis would analyse the exact nature of the impact of variations.

Imprecision in the principal model cost-effectiveness results with respect to key parameter values was to be assessed by use of techniques compatible with the modelling methodology deemed appropriate to the research question and available evidence. This would include multiway sensitivity analysis and cost-effectiveness acceptability curves.

Initial searches and application of inclusion criteria

The results of the application of the study inclusion criteria are presented in Figure 2.

FIGURE 2.

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram. a, Ten papers report on seven separate studies.

Included studies

In total, 10 records33,37–39,49–54 were included. These report on seven separate studies summarised in Table 3. For the remainder of this report, only the primary paper for each of the studies will be referred to. In one instance, this was only an abstract. 33

Three studies37,38,49 reported on preterm infants that had been previously ventilated and were applicable for the primary analysis. An additional study,39 in which the majority of infants received prior ventilation (see Study characteristics), was also included in this primary analysis. The remaining three33,51,52 studies reported on preterm infants requiring respiratory support following no prior ventilation and were applicable for the secondary analysis.

Study quality assessment

A summary of the quality assessment conducted is presented in Table 4 and a more detailed assessment is presented in the sections following. Overall, the RCTs included in the review were of reasonable methodological quality, although it was not possible to blind administrators or participants in any study. Studies included in the primary analysis of HHHFNC compared with usual care for preterm infants following ventilation were generally of better quality than those in the secondary analysis of HHHFNC compared with usual care for preterm infants with no prior ventilation. One of the studies included in this latter analysis, by Nair and Karna,33 was not published but only presented as an abstract.

Study characteristics

Study characteristics are presented in Tables 5 and 6. A total of 859 infants were involved in the seven trials and the trial sizes ranged from 2051 to 303 participants. 38

Study characteristics of studies included in primary analysis

As per the inclusion criteria, the four included studies37–39,49 of infants who had been treated following ventilation were RCTs. A total of 735 infants were involved in the trials and the trial sizes ranged from 13237 to 303. 38

Three studies38,39,49 were multicentred; no study was carried out internationally, with two studies conducted in Australia,37,38 one in the USA39 and one in China. 49 The earliest study started enrolling participants in December 200739 and the most recent in 2012. 49 HHHNFC was compared with NCPAP in all four studies. 37–39,49

The length of the study follow-up was only explicitly stated by Collins et al. 37 in which it is stated that 132 infants were followed up for 7 days and 121 infants were followed up until their discharge home; reasons for loss to follow-up after 7 days are provided. Yoder et al. 39 also appear to have followed up infants until discharge, as they present a study flow chart presenting numbers of patients until discharge. It can be assumed that in the other two studies37,49 infants were followed up for a minimum of 7 days (as the primary outcome in each study required follow-up for 7 days).

Study participants were generally similar across the studies (in terms of inclusion and exclusion criteria), although the two Australian studies37,38 limited participation to infants with a gestational age of < 32 weeks and the US study39 to ≥ 28 weeks. The US and Chinese studies39,49 included preterm, term and post-term infants, but only data for preterm infants has been synthesised in the remainder of this report (56.6% of participants in the Chinese study49 and 32.4% with a gestational age of < 32 weeks in the US study39). In addition, the US study39 also included infants who had not received prior ventilation (32.4% of all participants, including term and post-term babies, the proportion of preterm infants being unknown). The type of HHHFNC device and flow rate varied across studies, as did the NCPAP devices and starting flow rates.

Characteristics of the preterm infants included in the studies

Characteristics of the preterm infants that participated in the trials are presented in Table 7 (primary analysis of preterm infants treated following ventilation) and Table 8 (secondary analysis of preterm infants with no prior ventilation). There is a lack of data for two studies39,49 reporting on preterm infants treated following ventilation because both of these studies also included term and post-term infants and did not present baseline data for only preterm infants.

When data on birthweight were provided, birthweight was generally lower in those studies relevant to the primary analysis (mean < 1150 g) than those relevant to the secondary analysis (mean > 1490 g). Similarly, when data on mean gestational age were provided, this was generally lower in those studies relevant to the primary analysis (mean < 28 weeks) than those in the secondary analysis (mean ≥ 30 weeks). Prior steroid use was only reported in three studies37,38,52 and this was notably higher (≥ 88%) in the two studies relevant to the primary analysis37,38 than in the study52 included in the secondary analysis (50%). These differences in baseline findings suggest that infants in the primary analysis are heavier and have a shorter gestational age than those in the secondary analysis, which is not unexpected as these are the infants who tend to most need mechanical ventilation as soon as they are born.

Efficacy findings from primary analysis

For preterm infants treated following ventilation, it was possible to pool data in a meta-analysis for three outcomes: need for reintubation < 7 days, BPD and death. The primary outcome for our review was treatment failure as defined by the need for reintubation at < 72 hours, < 7 days or ever. For the primary analysis, three studies37,38,49 measured the need for reintubation within the first 7 days. The data for these three studies37,38,49 were pooled into a meta-analysis (Figure 3). Data were also pooled for BPD and death from three studies37–39 (Figures 4 and 5). For all analyses, a fixed-effects model was employed, as there was no evidence of statistical heterogeneity (or indeed clinical heterogeneity based on the data presented in Tables 5 and 7). The forest plots show that all the findings are in the direction of favouring HHHFNC. However, no statistically significant differences were reported between arms for any of the outcomes. No significant statistical heterogeneity between studies was noted in any of the three meta-analyses (I² = 0% and chi-squared test, p ≥ 0.10).

FIGURE 3.

Meta-analysis for need for reintubation < 7 days. df, degrees of freedom; M–H, Mantel–Haenszel.

FIGURE 4.

Meta-analysis for BPD. df, degrees of freedom; M–H, Mantel–Haenszel.

FIGURE 5.

Meta-analysis for death. df, degrees of freedom; M–H, Mantel–Haenszel.

One trial39 only reported reintubation within 72 hours for preterm infants. As reported in Table 9, marginally fewer preterm infants required reintubation in the HHHFNC arm than in the NCPAP arm. However, the proportions were small and no statistically significant difference between arms was reported.

All other outcomes reported in the trials are also presented in Table 9. No statistically significant differences were reported between arms in any of these studies37,38 and so no study reported the superiority of HHHFNC over usual care. However, it should be noted that the Manley et al. 38 study was a non-inferiority trial and so the aim of the trial was not to demonstrate superiority.

It should also be noted that the definition of extubation failure/treatment failure in two studies37,38 differed to that used for our review; both studies37,38 based failure on a composite outcome including apnoea, acidosis and increase in the fraction of inspired oxygen. In addition, Manley et al. 38 also included these three outcomes plus an urgent need for intubation in their composite outcome. Using these study definitions, it is noted that Manley et al. 38 reported a numerically higher rate of treatment failure with HHHFNC than NCPAP (but the opposite was the case with regard to need for reintubation). In contrast, Collins et al. 37 reported a numerically lower rate of extubation failure with HHHFNC (and reintubation rates were also numerically lower in the HHHFNC arm).

Hours on mechanical ventilation, days on oxygen support and length of hospital stay were reduced with HHHFNC than with NCPAP in the study by Manley et al. ;38 however, the differences were not statistically significant. In the same study,38 median weight gain also appeared to be higher in the HHHFNC arm than in the NCPAP arm, but again the difference was not statistically significant. Days to full feeds was reported only by Collins et al. 37 This was marginally higher in the HHHFNC arm by around half a day; the between-arm difference was not statistically significant.

A number of other secondary outcomes that we had planned to measure were not reported by any study, namely BPD/death (composite outcome), duration of respiratory support on NCPAP or HHHFNC, length of stay in NICU or measures of quality of care.

Exploratory subgroup analyses

Extubation failure/treatment failure (as defined, differentially, in the individual studies37,38) was considered by gestational age in two trials. 37,38 In Manley et al. ,38 it was considered in those infants born before 26 weeks of completed gestation and in those born from 26 weeks onwards, and in Collins et al. 37 it was considered in those born before/from 28 weeks. Unsurprisingly, the extubation failure/treatment failure rate was higher in infants with gestational ages below 26/28 weeks (extremely low gestational age) than in infants born later. As shown in Figure 6, and as noted above for the whole-trial population in Efficacy findings from primary analysis, the treatment effect was in opposite directions in the two included studies. 37,38

FIGURE 6.

Extubation failure/treatment failure by subgroup. Note that the extremely low gestational age subgroup was defined as < 28 weeks’ gestational age in Collins et al. 37 and < 26 weeks’ gestational age in Manley et al. 38 The higher gestational age subgroup was defined as ≥ 28 weeks’ gestational age in Collins et al. 37 and ≥ 26 weeks’ gestational age in Manley et al. 38 Collins et al. 37 defined extubation failure by composite criteria based on apnoea, acidosis and increase in fraction of inspired oxygen, whereas Manley et al. 38 defined treatment failure by composite criteria based on apnoea, acidosis, increase in fraction of inspired oxygen and urgent need for intubation. M–H, Mantel–Haenszel.

Reintubation rates were only presented by subgroup in one study. 37 As reported in Table 10, reintubation rates appeared to be higher in those treated with NCPAP than in those treated with HHHFNC, regardless of gestational age.

TABLE 10 - Subgroup analysis of reintubation rate by gestational age

Study	Arm	Gestational age
		< 28 weeks, n (%)	≥ 28 weeks, n (%)
Collins et al., 201337	HHHNFC	5 (16.7)	4 (10.8)
	NCPAP	7 (24.1)	7 (19.4)

Adverse events reported for primary analysis

A summary of the adverse events reported in the included trials is presented in Table 11. Adverse event data were reported for preterm, term and post-term infants combined by the Collaborative Group49 and Yoder et al. ;39 therefore, these data are not presented here.

Data were pooled into a meta-analysis for pneumothorax (Figure 7), nasal trauma leading to change of treatment (Figure 8), IVH (grade ≥ 3) (Figure 9), necrotising enterocolitis (NEC) (Figure 10), apnoea (Figure 11) and acidosis (Figure 12). With the exception of apnoea and acidosis, the forest plots show the findings are in the direction of favouring HHHFNC. Statistically significant differences were reported for nasal trauma leading to change of treatment, with fewer preterm infants changing treatment with HHHFNC than with NCPAP. No statistically significant differences between arms were reported for any other adverse events, although for IVH (grade ≥ 3) and NEC events were noticeably numerically fewer in the HHHFNC arm.

FIGURE 7.

Meta-analysis for pneumothorax. df, degrees of freedom; M–H, Mantel–Haenszel.

FIGURE 8.

Meta-analysis for nasal trauma leading to change of treatment. df, degrees of freedom; M–H, Mantel–Haenszel.

FIGURE 9.

Meta-analysis for IVH (grade ≥ 3). df, degrees of freedom; M–H, Mantel–Haenszel.

FIGURE 10.

Meta-analysis for NEC. df, degrees of freedom; M–H, Mantel–Haenszel.

FIGURE 11.

Meta-analysis for apnoea. df, degrees of freedom; M–H, Mantel–Haenszel.

FIGURE 12.

Meta-analysis for acidosis. df, degrees of freedom; M–H, Mantel–Haenszel.

In addition to data that could be pooled, differences in the nasal trauma score were statistically different favouring HHHFNC in Collins et al. 37 In Manley et al. 38 the difference in the incidence of nasal trauma was statistically significant whether reported as any documented nasal trauma, nasal trauma leading to a change of treatment or nasal trauma caused by the assigned treatment. Manley et al. 38 was the only study to report on nosocomial sepsis and gastrointestinal perforation, both of which were numerically fewer in the HHHFNC arm than in the NCPAP arm (17.1% vs. 19.9% and 0.7% vs. 1.3%, respectively).

Efficacy findings from secondary analysis

Findings for infants who had not received prior ventilation are summarised in Table 12. The primary outcome of our review, treatment failure as defined by the need for intubation, was reported in one study. 52 Respiratory failure, defined by a composite outcome incorporating blood gas and another outcome such as fraction of inspired oxygen > 70% or frequent apnoea or bradycardia, was reported by one other. 33 Neither study33,52 reported a statistically significant difference between arms for treatment failure/respiratory failure for either HHHFNC compared with NIPPV52 or HHHFNC compared with NCPAP. 33

In the study by Kugelman et al. ,52 compared with NIPPV, time on mechanical ventilation and length of hospital stay were reduced with HHHFNC and days on oxygen support were increased; however, the differences between trial arms were not statistically significant. In the same study,52 days to full feeds also appeared to be greater in the HHHFNC arm than in the NIPPV arm, again the difference was not statistically significant. None of these outcomes was reported by either of the two other studies33,51 comparing HHHFNC with NCPAP. A number of the other secondary outcomes that we had planned to measure were not reported by any study at all, namely BPD/death (composite outcome), duration of respiratory support on NCPAP or HHHFNC or length of stay in NICU.

Adverse events reported for secondary analysis

The authors of the study by Kugelman et al. 52 reported adverse events for infants who had not received prior ventilation. These were numerically higher with HHHNFC than with NIPPV (except for apnoea), but no statistically significant differences were reported. The following adverse events were reported for HHHFNC compared with NIPPV: air leak (5.3% vs. 0.0%), nosocomial sepsis (10.5% vs. 7.8%), IVH (5.3% vs. 2.6%), NEC (5.3% vs. 0.0%) and apnoea (10.5% vs. 13.1%). There were no incidences of nasal trauma in either arm.

Quality of care

Klingenberg et al. 51 reported the results of a crossover study comparing HHHFNC with NCPAP for preterm infants who had received no prior ventilation (secondary analysis); this was the only study to report on outcomes relevant to quality of care (within two 24-hour periods). The primary outcome of the study was patient comfort, defined as a state free of prolonged pain by a validated neonatal pain and discomfort scale [the Echelle Douleur Inconfort Nouveau-Né (Neonatal Pain and Discomfort) scale]. 55 No statistically significant differences between arms were reported for this outcome or for noise of equipment (measured by a handheld audiometer). There were, however, statistically significant differences for all parental assessment measures (from a visual analogue scale rated 1–10) with parents preferring HHHFNC to NCPAP (Table 13). In addition, it was noted by the study authors that infants had significantly lower respiratory rates in the HHHFNC arm than in the NCPAP arm in this study51 but that all other respiratory parameters were similar.

Treatment resource use and costs

Resource use of treatment included capital equipment, consumable costs and clinician time taken to establish a preterm infant onto either HHHFNC or NCPAP. All prices are in 2015 GBP unless otherwise stated. Given the time horizon is the period up to discontinuation of NCPAP or HHHFNC, no discounting needed to be applied to costs.

Analysis of uncertainty

Ordinarily in an economic evaluation, scenario analysis and deterministic and probabilistic sensitivity analysis would be used to explore parameters when there was uncertainty in the economic model.

As we carried out a cost-minimisation analysis, this analysis has focused on the resources and costs associated with two treatments that the clinical evidence suggests are equally efficacious for the primary outcome of interest. The only notable difference between the treatments was in nasal injury as an adverse event and this has a very low cost per patient.

No distributions on any of the costs or resource use are available and so any probabilistic analysis of uncertainty is not possible. However, assumptions were made on the life expectancy of NCPAP and HHHFNC machines. As the cost saving for HHHFNC is driven by the greater capital cost of NCPAP, these assumptions were explored with sensitivity analysis.

Table 15 shows the two-way sensitivity analysis of the cost differential with HHHFNC compared with NCPAP as the utilisation rates vary between 20% and 100% and the lifespan varies between 2 and 10 years.

The threshold analysis shows that if the lifespan of the machines reaches 6.8 years then HHHFNC would no longer be cost saving compared with NCPAP. A machine lifespan above 6.8 years means that NCPAP becomes the less costly option.

Changes in machine lifespan and utilisation rate are positively related to the number of infants that can be used by each machine and therefore negatively related to the machine unit cost per infant (i.e. lower utilisation rates/machine lifespans lead to a lower number of infants that can use a machine over its lifespan, and therefore higher unit costs of the machine per infant). Although these changes in unit cost will be proportionally the same for each technology, the machine cost of NCPAP is higher than with HHHFNC. As such, the change in the absolute difference in unit cost per infant between the technologies is negatively related to the utilisation rate and machine lifespan (i.e. higher utilisation rates/machine lifespans lead to a smaller absolute difference in the machine unit costs per infant between NCPAP and HHHFNC).

It is also possible that different neonatal units pay different costs for consumables depending on the NCPAP and HHHFNC systems employed. However, what is important for our economic analysis is the size of the cost differential in consumables rather than the consumable costs per se. As costs can vary between units, a two-way sensitivity analysis was also undertaken to show how the differential in consumable costs together with the lifespan of the different machines change. The difference in consumable costs ± £24 (200%) is shown in Table 16; in the initial analysis there is a cost difference of –£12 per week (consumable cost with NCPAP is £55 and with HHHFNC is £67).

The results presented in Table 16 demonstrate that the main finding of the economic analysis, that is HHHFNC is cost saving compared with NCPAP, is relatively sensitive to changes in the difference in weekly consumable costs of the two technologies. Assuming a 5-year lifespan for equipment as in the initial analysis, if the difference in consumable prices rises approximately by 35% from £12 to £16.24 then HHHFNC will no longer be cost saving compared with NCPAP.

Principal findings

We have conducted a systematic review of the literature to summarise the clinical effectiveness of HHHFNC compared with usual care for preterm infants. Usual care was considered to consist of NCPAP, oxygen or NIPPV with five RCTs33,37–39,49,51 comparing HHHFNC with NCPAP and one RCT52 with NIPPV. Evidence was derived from four RCTs37–39,49 for effectiveness of treatment following ventilation (primary analysis) and three RCTs33,51,52 for clinical effectiveness following no prior ventilation (secondary analysis) including a crossover trial by Klingenberg et al. 51 The quality of the studies included in the primary analysis of treatment following ventilation could be considered to be superior to that of the studies included in the secondary analysis of treatment with no prior ventilation.

In the primary analysis, the primary outcome for our systematic review was treatment failure; we defined treatment failure to be the need for reintubation within 72 hours, within 7 days or ever (i.e. time period not specified). There were proportionally fewer cases of reintubation of preterm infants treated following ventilation in the HHHFNC arm than in the NCPAP arm in all four RCTs,37–39,49 although no statistically significant difference was found between treatment arms, either as reported in the individual studies37,38,49 or in the meta-analysis of these three trials reporting reintubation within 7 days. Two RCTs37,38 used composite outcomes to define extubation failure/treatment failure rather than simply defining it as the need for reintubation. Interestingly, despite the reintubation rate being lower for those treated with HHHFNC than those treated with NCPAP, the largest RCT by Manley et al. 38 reported a higher rate of treatment failure for HHHFNC compared with NCPAP.

Extubation failure/treatment failure was the only outcome that was considered in a subgroup analysis in which two trials37,38 considered this outcome by gestational age. In our review protocol, we had proposed conducting subgroup analyses of gestational age prior to and from 30 weeks but the included studies reported these prior to and from 26 weeks38 and prior to and from 28 weeks. 37 Unsurprisingly, infants with extremely low gestational age appeared to have higher rates of treatment failure in the individual studies,37,38 although the difference between subgroups was not statistically significant. The subgroup findings must be treated with extreme caution and can only be considered exploratory because different gestational age thresholds were used to define subgroups in the two studies and because extubation failure/treatment failure was also defined differently in the two studies;37,38 as discussed previously, these studies37,38 used composite outcomes, as opposed to our definition that was simply the need for reintubation.

In the secondary analysis, with regard to preterm infants who had received no prior ventilation, treatment failure was defined by the need for endotracheal ventilation52 or by a composite outcome. 33 Neither study33,52 reported a statistically significant difference in treatment failure rates for HHHFNC compared with NCPAP33 or HHHFNC compared with NIPPV. 52

Secondary efficacy outcomes for the comparison of HHHFNC compared with NCPAP were only reported in three studies;37–39 all three studies were included in the primary analysis of treatment following ventilation. Meta-analyses found that the findings for both outcomes are in the direction of favouring HHHFNC but no statistically significant differences were found. The majority of other relevant secondary outcome data (e.g. days on mechanical support and length of hospital stay) also suggested an improvement for HHHFNC over NCPAP but these were not reported by two or more trials and no statistically significant differences were reported between arms. 37,38

The authors of the study by Kugelman et al. 52 reported relevant secondary outcomes for HHHFNC compared with NIPPV. In this study, no preterm infant had received prior ventilation (secondary analysis). Although the findings from this small study52 appeared to marginally favour HHHFNC over NIPPV in terms of days on mechanical support and length of hospital stay and marginally favour NIPPV over HHHFNC in terms of days on oxygen support and days to full feeds, none of the between-arm differences was statistically significant.

Adverse event data for the comparison of HHHFNC compared with NCPAP were only available from two studies37,38 that were included in the primary analysis (preterm infants treated following ventilation). Importantly, nasal trauma was statistically significantly lower in the HHHFNC arm in the largest study by Manley et al. 38 Meta-analysis of nasal trauma leading to change of treatment also showed statistically significantly fewer infants changing treatment from HHHFNC than with NCPAP. With the exception of apnoea and acidosis, where mixed results were reported in the individual studies, pneumothorax, IVH (grade ≥ 3) and NEC appeared to be less common with HHHFNC than NCPAP but differences were not statistically significant.

Adverse event data for the comparison of HHHFNC with NIPPV were only available from one study52 that was included in the secondary analysis. Generally the adverse event profile appeared to marginally favour NIPPV over HHHFNC, but there were no between-arm statistically significant differences.

Klingenberg et al. 51 reported outcomes from the smallest RCT included in our review (n = 20) and was the only study to report quality-of-care outcomes. Although there were no statistically significant differences between arms in terms of noise or neonatal pain and discomfort, there were statistically significant differences between study arms in terms of parental preferences for HHHFNC over NCPAP. Parental preferences were based on the belief that (1) child satisfaction, (2) contact and interaction and (3) opportunities to take part in care were all improved with HHHFNC compared with NCPAP. In this study,51 preterm infants were not supposed to have been treated following ventilation, although a minority of infants had, in fact, received prior ventilation (n = 7, 30%).

In summary, therefore, following ventilation (primary analysis), there is a lack of convincing evidence for a difference in the need for reintubation, BPD or death between HHHFNC and NCPAP; there is, however, some evidence for a decrease in nasal trauma. For preterm infants with no prior ventilation (secondary analysis), there is some suggestive evidence for parental preferences for HHHFNC over NCPAP but overall an absence of any consistent evidence to suggest that HHHFNC is superior or inferior to usual care.

The lack of evidence supporting the clinical effectiveness of HHHFNC compared with usual care, or vice versa, precluded us from being able to conduct a cost–utility or cost-effectiveness analysis for either HHHFNC compared with usual care following ventilation or HHHFNC compared with usual care with no prior ventilation. Instead, we were only able to conduct a cost-minimisation analysis. Given the absence of evidence for infants who had no prior ventilation, a cost-minimisation analysis was only performed for infants who had been treated following ventilation.

The results of our cost-minimisation analysis suggest that HHHFNC would be cost saving over NCPAP for infants who have been treated following ventilation. However, the results of our economic analysis are sensitive to both the size of the machine lifespan and utilisation of equipment. When estimating and valuing resources for these two items in the analysis, it was necessary to make assumptions and so there is a degree of uncertainty associated with the results. If the HHHFNC and NCPAP machines last, on average, longer than 6.8 years and assuming an 80% utilisation rate for equipment, NCPAP is likely to become the less costly of the two technologies. Although the cost differential of consumables has a higher degree of certainty than the lifespan of the machines, as costs have been derived from an individual neonatal unit, it is not known how representative this difference might be across units in the UK. If HHHFNC consumables cost £16.24 or more than NCPAP consumables per week, then NCPAP will become the less costly of the two technologies. Hence, while the best estimate from the economic analysis is that HHHFNC will cost just over £26 less per infant than NCPAP, the cost saving could be as high as £326 per infant with HHHFNC over NCPAP or NCPAP could save £173 compared with HHHFNC, depending on differences in the lifespan of machines, utilisation rates and cost differences in consumables.

In reality, the actual total cost differential between infants on either technology is relatively insignificant compared with the cost per day in a neonatal intensive care ward, regardless of the assumptions employed in the analysis. The NHS Reference Cost47 for a day in a neonatal high-dependency unit in 2013/14 was £839 per day or just under £36,500 for a 43.5 day stay. The cost of either treatment with HHHFNC or NCPAP during this period therefore costs less than 2% of the total care while the infant requires oxygen. The economic analysis therefore shows that cost does not seem to be a paramount consideration when deciding between the two technologies.

Similarities and differences with previous systematic reviews and meta-analyses

We are aware of two other published meta-analyses of HHHFNC compared with NCPAP; one published alongside a systematic review by Daish and Badurdeen36 and another which accompanies a review of the literature by DeMauro et al. 58 Both of these meta-analyses include the same three trials. 37–39 In addition, we are aware of an unpublished ‘pooled analysis’ of HFNC compared with NCPAP which has only been presented as an abstract by Rotta et al. 59 and which includes four trials; as data have only been presented in abstract form for this unpublished analysis59 it is unclear which trials were included and if the HFNC described in all four trials is heated.

All analyses reported no statistically significant differences between arms for extubation failure, although the RR exceeded one, suggesting that the treatment effect may be in favour of NCPAP (RR 1.12, 95% CI 0.85 to 1.4736 and RR 1.05, 95% CI 0.79 to 1.3958 in the published meta-analyses; RR 1.17, 95% CI 0.90 to 1.5159 in the unpublished analysis59). Although our meta-analysis also reported no statistically significant differences, the RR was less than one suggesting that the treatment effect may be in favour of HHHFNC (RR 0.76, 95% CI 0.54 to 1.09).

The reasons for marginal differences in results arise from including different studies in the meta-analyses and from differences in how data were pooled in each of the meta-analyses. It is unclear from the unpublished abstract which four trials were included in the pooled analysis by Rhotta et al. 59 However, both the published meta-analyses36,58 and our meta-analysis reported on three trials, including the same two Australian RCTs. 37,38 Whereas Daish and Badurdeen36 and DeMauro et al. 58 also included the US study by Yoder et al. ,39 we excluded this study from our meta-analysis as extubation failure reported for the subgroup of preterm infants was for reintubation within 72 hours. Data from a subgroup analysis of preterm infants from the Chinese study49 were also included in our meta-analysis but not in the other studies.

Crucially, we also used a standard definition of treatment failure across all studies included in our meta-analysis (reintubation rates within 7 days). The other two published meta-analyses,36,58 however, used the original study definitions of treatment failure/extubation failure which differed across all three studies37–39 and, importantly, were measured over different time points (within 7 days in the two Australian studies37,38 and within 72 hours in the study by Yoder et al. ). 39 Finally, our meta-analysis included data describing only preterm infants who had received treatment following ventilation; the inclusion of the study by Yoder et al. 39 in the other two published meta-analyses36,58 resulted in a mixed population of infants, some of whom had received treatment following ventilation and some of whom had received no prior ventilation.

Differences in the choice of studies that were included in the meta-analyses and differences in how the data were pooled and analysed are reflected in the measures of statistical heterogeneity reported. A moderate level of statistical heterogeneity was identified for the meta-analysis of treatment failure (I² = 56% and χ² = 4.5; p = 0.11) by Daish and Badurdeen. 36 Greater and statistically significant (p < 0.10) levels of heterogeneity (I² = 59.5%; p = 0.085) were reported in the meta-analysis by DeMauro et al. 58 Our meta-analysis reported no statistical heterogeneity at all (I² = 0% and χ² = 0.34; p = 0.84).

As per our meta-analysis, Daish and Badurdeen36 also pooled data for BPD and found no statistically significant differences between treatment arms. Data were pooled from the same three RCTs37–39 in both our meta-analysis and in the analysis conducted by Daish and Badurdeen. 36 Hence, the findings of the meta-analyses were identical (RR 0.92, 95% CI 0.72 to 1.17) with no statistical heterogeneity evident (I² = 0% and χ² = 0.87; p = 0.65). In this instance, it should be noted that the inclusion of the study by Yoder et al. 39 did result in a mixed population of infants in our meta-analysis, some of whom had received treatment following ventilation and some of whom had received no prior ventilation.

No previous meta-analysis of death has been previously published. The meta-analysis we conducted again found no statistically significant differences between arms. However, as perhaps expected from a meta-analysis of only two trials with few events, CIs were wide (RR 0.56, 95% CI 0.22 to 1.44). No significant statistical heterogeneity between studies was reported (I² = 0% and χ² = 1.21; p = 0.23).

Prior to the publication of the meta-analyses by Daish and Badudeen36 and DeMauro et al. 58 a Cochrane review31 included narrative results from a systematic review of the effectiveness of HFNC (as opposed to HHHFNC) from four RCTs. 32–35 Four different analyses were presented, with one study included in each analysis: HFNC compared with NCPAP for preterm infants who had received prior ventilation,32 HFNC compared with NCPAP with no prior ventilation,33 HHHFNC compared with ‘standard’ HFNC35 and a comparison of two different brands of equipment for HHHFNC. 34 It was not possible to conduct meta-analyses given each analysis only included one study. The review authors concluded that there was insufficient evidence to establish the safety or effectiveness of HFNC and that HFNC may be associated with a higher rate of reintubation than NCPAP when used after ventilation. It should be noted that this latter conclusion was drawn from one study32 comparing HFNC (not HHHFNC) to NCPAP and reporting a significantly worse outcome for HFNC (RR 4.0, 95% CI 1.33 to 12.05). No statistical differences were reported for HHHFNC compared with HFNC in the study by Woodhead et al. 35 which examined reintubation rates within the first 24 hours. However, the study was small (n = 30) and only two infants who received standard HFNC as opposed to no infants who received HHHFNC required reintubation; the data therefore suggest that HHHFNC may be superior to HFNC. Furthermore, infants were statistically significantly more likely to have a normal appearance of their nasal mucosa in the HHHFNC arm than in the HFNC arm in this study35 (p < 0.0005). This arguably highlights the importance of distinguishing between HHHFNC and HFNC.

Strengths and limitations

One of the strengths of our systematic review is that we have limited the inclusion of our evidence to RCTs in which evidence has been presented for only preterm infants, as opposed to a mixed population of preterm, term and post-term infants. We have also limited our review to only include studies when it was clear that the intervention was HHHFNC; HFNC that is neither heated or humidified is now considered by many review authors36,58 to be inconsistent with clinical practice. Finally, we have considered the clinical effectiveness of HHHFNC compared with usual care both following ventilation and in preterm infants who have received no prior ventilation. This distinction is of importance given that the European Consensus Guidelines4 recommend that NCPAP should be the preferred option for the stabilisation of preterm infants when possible, ventilation being preferred for less mature infants.

While we consider that limiting the inclusion of studies to only those in which it was clear HFNC was heated to be a strength, this approach may also be considered to be a limitation; study authors do not always explicitly state that the interventions they are studying are heated. Therefore, it is possible we have excluded some studies that we should have included. Certainly, we have excluded three abstracts by Collins60 and Collins et al. 61,62 that report on the same study that we have included. 37 This is because it was not stated in these three abstracts60–62 that the intervention was heated. Excluding these abstracts was, however, of no importance because we did include the fully published study with the relevant results. It does, however, suggest that there is a need for common and consistent terminology when describing whether HFNC is heated or humidified. Of the other six papers that we excluded for not being heated,27,32,63–66 three27,32,63 explicitly stated that they were unheated, meaning that three other papers64–66 (of two studies, one64 reported only as an abstract) may actually have been studies of HHHFNC.

An advantage of the data available for our primary analysis of infants who have received treatment following ventilation is that there is an element of consistency in how outcomes have so far been reported. This is particularly true for reintubation, which has been reported within 7 days, enabling comparisons across trials, and also for BPD and death. However, it still remains unclear if reintubation within 7 days is the optimal outcome and, arguably, reintubation should be reported at three different time points, within 72 hours, within 7 days and ever. The only study we are aware of that has reported reintubation at different time points is the study by Yoder et al. 39 Unfortunately, the findings at these two time periods are for a mixed population of preterm, term and post-term infants. This study39 did, however, provide a subgroup analysis for some, but not all, preterm infants [gestational age < 32 weeks (34.7% of the study population) as opposed to < 37 weeks] but only for reintubation within 72 hours. We contacted the principal author of the Yoder et al. 39 study to request further information about all preterm infants but we have not received a reply. It should also be noted that the study by Yoder et al. 39 also included infants who had received no prior ventilation alongside those who had been ventilated. However, it is unclear how many of the preterm infants had received prior ventilation.

A limitation of our review is the lack of evidence regarding the quality of care delivered in the clinical studies. It is often cited that HHHFNC is preferred over NCPAP by staff and parents of preterm infants, as it enables infants to be more easily handled and cared for than does NCPAP. 15,20,29 Only one of the RCTs51 we identified examined outcomes relating to quality of care, and data were available from only 20 participants and hence the generalisability of the findings should be treated with caution. Nevertheless, this study51 did report that parents preferred HHHFNC to NCPAP. In terms of neonatal pain and discomfort and noise, there were no statistically significant differences between HHHFNC and NCPAP. However, RCTs are not necessarily the best types of study to evaluate such outcomes, with qualitative studies and surveys probably being better suited to studying such outcomes. For example, it would be illustrative to know whether or not improved parental contact which was reported with HHHFNC over NCPAP by Klingenberg et al. 51 included an increase in the amount of time spent in ‘Kangaroo Care’. ‘Kangaroo Care’ entails skin-to-skin care between mother and infant. Previous studies have reported this practice to be beneficial to the development of infants67,68 and to reduce mortality. 69,70 Nonetheless, the inclusion of outcomes such as those measuring parental preferences as secondary outcomes in RCTs is informative.

Another limitation of the evidence base is that it was not possible for investigators to blind health-care staff or study participants to the treatment that they delivered or received in any of the RCTs. This is commonly cited as a major weakness of clinical trials44 but when comparing an intervention such as HHHFNC to an intervention such as NCPAP, such blinding would be impossible to employ; realistically, only those responsible for the analysis of the results could be blinded. Only one study (included in the primary analysis) reported that assessors were blinded to treatment allocation. 37

Arguably the largest limitation of our review, however, is the lack of published RCT data from relatively large-sized populations in which HHHFNC is compared with usual care. The lack of evidence is perhaps most stark when we present the secondary analysis of our review, assessing the clinical effectiveness of interventions in preterm infants with no prior ventilation. As discussed, there were only 124 preterm infants from the three relevant trials33,51,52 in the secondary analysis (although as also highlighted, seven of the participants in one trial51 had in fact received treatment following ventilation); this figure (n = 124) is smaller number than the number of participants in the smallest trial (n = 132)37 of preterm infants in the primary analysis. However, even for the primary analysis of those who received treatment following ventilation, more RCT evidence is required.

Finally, the lack of evidence describing treatment failure across trials and from our meta-analysis has also precluded us from being able to conduct a cost-effectiveness or cost–utility analysis, another limitation of our research. Instead we have only been able to conduct a cost-minimisation analysis which has levels of uncertainties around the costs and lifespan of different HHHFNC and NCPAP devices and associated consumables.

Uncertainty in the evidence base is evident from comparing the (statistically non-significant) findings for treatment failure from our meta-analysis to those of other authors;36,58,59 the results of our meta-analysis suggest that the treatment effect may be in favour of HHHFNC over NCPAP, whereas other authors36,58,59 suggest the opposite effect. However, as discussed, other authors use different definitions of treatment failure and include mixed populations, whereas we have limited the data in our meta-analysis to reintubation within 7 days in a population limited to preterm infants who have previously been ventilated.

Recommendations for future research

Based on the available evidence, the following research recommendations are made:

1. There is a need for more RCT evidence comparing HHHFNC with usual care including, but not limited to, a comparison with NCPAP. End points should include (re)intubation, BPD, death and adverse events. In particular, there is a need for research into the need for (re)intubation at both 72 hours and 7 days, both outcomes which should ideally be measured in individual trials. This is because trials have utilised both outcome measures, and results with respect to efficacy may differ at different follow-up times (as preterm infants may remain extubated for the first 72 hours but then be reintubated at 7 days).

2. Ideally, studies should only include preterm infants, and when infants may have received either previous ventilation or no prior ventilation RCTs should be stratified for these factors and subgroup analyses conducted.

3. Given the evidence has not shown HHHFNC to be statistically superior to NCPAP but the direction of the treatment effect appears to favour HHHFNC over NCPAP, a non-inferiority trial may be of particular value. As the primary outcome, BPD may be particularly clinically important and meaningful, as it has been shown to be associated with long-term disability and morbidities. The sample size for such a trial would then depend on the significance level and desired statistical power as well as the rate of BPD and preferred non-inferior margin, as detailed in Appendix 4, Table 23.

4. There is also a need for more research on quality of care in terms of staff and parental preferences, and infant comfort. Although these outcomes are arguably best investigated via qualitative studies and surveys, including such outcomes in future RCTs will be informative.

Contributions of authors

Nigel Fleeman (Review Co-ordinator, LRiG) drafted the review protocol, managed process of study selection and provided input to the final report.

James Mahon (External Economic Consultant, Coldingham Economics) commented on the protocol, conducted assessment economic costs and implications, and provided input to the final report.

Vickie Bates (Reviewer, LRiG) commented on the protocol, conducted data extraction and quality assessment, and provided input to the final report.

Rumona Dickson (Director, LRiG) commented on the protocol, conducted study selection, checked data for accuracy and provided input to the final report.

Yenal Dundar (Reviewer, LRiG) contributed to the initial search strategy, conducted study selection and quality assessment, and provided input to the final report.

Kerry Dwan (Statistician, LRiG) contributed to the proposed statistical approach.

Laura Ellis (parent of premature infants) commented on the review protocol and draft of final report.

Eleanor Kotas (Information Specialist, LRiG) conducted literature searches for studies and costs, and gave input to the final report.

Marty Richardson (Statistician, LRiG) conducted the statistical analyses.

Prakesh Shah (Clinical Consultant, Mount Sinai Hospital and Departments of Paediatrics and of Health Policy, Management and Evaluation, University of Toronto) reviewed the protocol and provided input to the final report.

Ben NJ Shaw (Clinical Consultant, Consultant in Neonatal and Respiratory Paediatrics Neonatal Unit, Liverpool Women’s Hospital) reviewed the protocol and provided input to final report.

Data sharing statement

The RevMan file used for conducting the data analyses can be obtained from the corresponding author on request.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.