Background
High-flow nasal cannula oxygen (HFNC) is a relatively new and increasingly used therapy in adults with acute respiratory failure [
1]. HFNC can deliver heated humidified oxygen through nasal prongs and provide much higher and more predictable rates of gas flow (maximum flow of 60 L/min) and fraction of inspired oxygen (FiO
2) (up to 1.0) [
2‐
4].
HFNC has several advantages when compared to conventional oxygen therapy (COT): (1) the high-flow rates match the patient’s inspiratory flow rates, which creates a positive pressure effect [
5,
6] and reduces the anatomic dead space [
7]; (2) HFNC can deliver a predictable and constant FiO
2 [
8]; (3) HFNC can increase the partial arterial pressure of oxygen (PaO
2)/FIO
2 ratio, which reduces the entrainment of room air and the dilution of oxygen [
1,
2]; (4) the heated and humidified gas that is inhaled can improve mucociliary motion and sputum clearance [
3,
9]; and (5) there is reduced upper airway resistance, reduced work of breathing [
10] and improvement in thoraco-abdominal synchrony [
11,
12]. Based on the above advantages, several studies found that HFNC could improve comfort level [
13‐
15], increase oxygenation [
12,
14‐
17] and decrease the dyspnoea score in adult patients [
3,
12,
14,
17]. Nevertheless, there has been no clear consensus on treatment outcomes (such as intubation rate, escalated respiratory support rate and mortality) [
13,
18‐
22].
Compared to HFNC, noninvasive mechanical ventilation (NIV) can create a much higher gas flow rate and positive airway pressure but is not comfortable and has many complications [
23,
24]. Recently, well-designed randomized controlled trials (RCTs) have not differentiated the effect of HFNC and NIV on intubation rates and mortality [
18,
25]. To decrease complexity, a recently published meta-analysis showed no significant difference in mortality or intubation rates when HFNC was compared to COT and NIV [
26]. The conclusion of this study might be misleading: first, COT and NIV are very different in terms of the mechanism of action and clinical application; second, the combination of COT and NIV not only increased patient heterogeneity but also increased the power of statistical bias. A separate comparison of HFNC with COT and NIV should result in more reasonable conclusions. Maitra et al. included only six RCTs in their meta-analysis that compared only the prognosis of higher respiratory support, and found no significant difference between HFNC and either standard oxygen therapy or NIV [
27]. The small number of studies included in the meta-analysis makes its application limited.
Therefore, we conducted a meta-analysis to investigate the effect of HFNC on the rates of intubation, mechanical ventilation, the escalation of respiratory support and its effect on mortality versus COT or NIV in adult patients with respiratory failure.
Methods
We conducted a systematic review in accordance with the methods recommended in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [
28].
Search strategy
Using electronic and manual searching, a literature review was performed from the electronic databases inception up to 20 October 2016 in the Cochrane Library, Embase and Ovid Medline without language restrictions. The following search terms were used: ‘hfnc’, ‘hfnp’, ‘hhfnox’, ‘hfno’, ‘(high flow) adj5 nasal’, ‘(high flow) adj5 oxygen’, ‘high-flow adj5 nasal’ and ‘nasal adj5 (high flow)’. The search strategy is in the Appendix (Additional file
1). In addition, we manually searched clinical trials.gov and the bibliographies of randomized controlled trials, meta-analyses, and systematic reviews to identify other potentially relevant studies. The references of all included articles were also checked manually to identify additional eligible studies. The literature review was conducted independently by two authors (HYZ and HXW). Disparities in the literature review were resolved by a consensus of all authors.
Inclusion and exclusion criteria
The following criteria were used for inclusion in our meta-analysis:
1.
Type of study: randomized controlled trials;
2.
Population: adult patients with respiratory failure who received oxygen therapy;
3.
Intervention: HFNC treatment compared either COT or NIV;
4.
Predefined outcomes: intubation, mechanical ventilation (includes noninvasive mechanical ventilation and invasive mechanical ventilation), escalation of respiratory support (HFNC, NIV or intubation) and mortality.
If there was more than one eligible trial from one team, the study with the most recent publication date was used in the analysis.
Exclusion criteria: the types of articles excluded from the analysis were reviews, retrospective studies, observational studies, case reports, animal studies, studies conducted on children, studies examining only psychological mechanisms, unrelated studies (e.g., HFNC not used in patients), duplicate reports, studies involving repeated experiments (commentary articles on specific studies or secondary analyses of experimental data), studies not in either the English or the Chinese language and non-randomized trials.
The primary outcome of our study was to investigate whether HFNC versus COT or HFNC versus NIV resulted in a similar intubation rate. The secondary outcomes were whether HFNC versus COT had similar rates of mechanical ventilation and rates of escalation of respiratory support and mortality, and whether HFNC versus NIV had similar rates of escalation of respiratory support and mortality.
Two reviewers (HYZ and HXW) extracted the data independently using a predefined data extraction form. Disagreements were resolved through discussion or consensus with a third reviewer (FS). The data extracted included the study ID (together with the first author’s name and publication year), country, study design, setting, control therapy, duration of therapy, primary and secondary outcomes and clinical outcomes. We also checked the supplementary files and contacted the authors for more detailed information if necessary.
Quality assessment and publication bias
Two independent reviewers assessed the methodological quality of the included trials using the Cochrane Collaboration Risk of Bias tool [
29] within the RevMan5.3 software, which considers seven different domains: adequacy of sequence generation; allocation sequence concealment; blinding of participants and caregivers; blinding for outcome assessment; incomplete outcome data; selective outcome reporting; and the presence of other potential sources of bias not accounted for in the other six domains (the current research primarily refers to receiver sponsorship or oxygen therapy devices from Fisher and Paykel Healthcare or from other organizations). Based on the method of the trials, each was graded as “yes”, “no” or “unclear”, to reflect a high, low risk or uncertain risk of bias, respectively. Two reviewers (HYZ and HXW) made judgments independently. In cases of disagreement, resolution was first attempted by discussion and then by consulting a third author (FS) for arbitration.
Funnel plots were used to assess the possibility of publication bias and were implemented in RevMan 5.3 software. The Egger regression test was used to measure funnel plot asymmetry [
30] and was implemented using Stata 12.0 (StataCorp LP, TX, USA).
Grading the quality of the evidence
We used the methodology of the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) Working Group to assess the overall quality of the evidence for the primary and secondary outcomes in the following domains: risk of bias, inconsistency, indirection, imprecision and publication bias. These were classified as very low, low, medium and high [
31].
Statistical analysis
Statistical analyses were performed using RevMan 5.3 software (Cochrane Library, London, UK). We combined data from all trials to estimate the pooled odds ratio (OR) with 95% confidence intervals (CIs) for the rates of intubation, mechanical ventilation, escalation of respiratory support and mortality. Pooled ORs with 95% CI were estimated by the Mantel-Haenszel method. ORs were undefined and excluded for studies with no event in either arm. The analysis was performed using a random effects model. Heterogeneity was tested using a weighted Mantel-Haenszel
χ
2
test and was quantified using the
I
2
statistic as implemented in RevMan
I
2
values, which describes the percentage of the variability in effect and computes estimates that are due to heterogeneity rather than sampling error.
I
2
values of 25–50% indicated low, 50–75% indicated moderate, and >75% indicated high heterogeneity. A value >50% may be considered substantial heterogeneity [
32]. A
P value of 0.1 was used to denote the statistical significance of heterogeneity. Differences between subgroups were analysed using the test of subgroup differences described by Deeks et al., and the results were expressed using the
P values.
P < 0.05 was considered statistically significant.
Subgroup analysis
We utilized subgroup analysis to assess possible influences of the oxygen therapy system on clinical outcomes, which allowed us to explore the possible causes of the heterogeneity. In the comparison of HFNC to COT, we first explored whether there was a different treatment effect of the oxygen therapy system in patients with post-extubation acute respiration failure (ARF) and patients with ARF that occurred for other reasons. Second, we assessed the effect of trials that allowed COT to escalate to HFNC versus those studies that did not. Moreover, we also compared the results of RCTs with patients from single-centre studies versus patients from multi-centre studies.
Discussion
The main finding of our study was that HFNC significantly reduced the rate of intubation, mechanical ventilation and escalation of respiratory support compared with COT in adult patients with respiratory failure, but there was no difference in mortality. On the other hand, when compared to NIV, no significant difference in intubation rate, escalation of respiratory support rate or mortality was detected.
Our results were partially different from two recent meta-analyses of HFNC in adult patients [
26,
27]. Our meta-analysis was registered on the PROSPERO website before the other authors published, and we included more RCTs (11 RCTs recruiting 3459 patients); Maitra et al. only included 6 RCTs in their meta-analysis and only compared the prognosis of higher respiratory support, while also finding no significant difference between HFNC and standard oxygen therapy or NIV [
27]. We included more outcomes for HFNC, including intubation rate, mechanical ventilation rate, the rate of escalation of respiratory support and mortality. Furthermore, we focused on various prognostic indicators of HFNC compared with COT or NIV, to decrease the population heterogeneity and population bias of the sample.
Our meta-analysis found that HFNC also significantly reduced the rate of intubation, mechanical ventilation and the escalation of respiratory support compared to COT. HFNC is superior to COT due to the delivery of heated humidified oxygen, higher and more predictable gas flow rates and rates of FiO
2, a positive pressure effect and improvement in sputum clearance [
3,
4]. Nevertheless, HFNC did not reduce mortality when compared to COT, which could be explained by the complex causes of respiratory failure. Frat et al. [
18] enrolled patients in the ICU with acute respiratory failure, most of which had community-acquired pneumonia. Jones et al. [
20] included patients in the ED with respiratory failure with the main cause indicated as chronic obstructive pulmonary disease (COPD). The other three RCTs included patients with respiratory failure post extubation [
19,
21,
22], and patients in the study of Parke et al. [
22] were included after undergoing cardiac surgery. In addition, in three RCTs [
13,
22,
44] COT was allowed to escalate to HFNC, and the subgroup analysis found no significant differences in the intubation rate between the two groups, which suggests early conversion to HFNC after failure of COT. Care should also be taken since some patients could not tolerate HFNC: in the study of Jones [
20], 11 of 172 patients did not tolerate HFNC and were placed on COT. Finally, the prolonged use of HFNC could result in delayed intubation and increased mortality [
48].
NIV has been proven to improve gas exchange and reduce the need for intubation and mortality in patients with respiratory failure, especially in patients with an exacerbation of COPD and those with acute cardiogenic pulmonary oedema [
49,
50]. Compared with NIV, HFNC has some advantages, such as greater patient comfort, lower costs, increased ease in clearing secretions [
34] and a lower incidence of adverse effects that could lead to a poorer outcome [
18]. We included three RCTs that showed no difference between HFNC and NIV in terms of intubation rate and mortality, which suggests that HFNC was not inferior to NIV pertaining to the patients’ prognoses, at least in patients with hypoxaemic respiratory failure. Of the three RCTs, Frat et al. [
18] enrolled patients in the ICU with acute hypoxaemic respiratory failure and excluded chronic respiratory failure and cardiogenic pulmonary oedema. A study by Stephan [
47] enrolled hypoxemic patients after cardiothoracic surgery. Hernandez et al. [
25] included patients with post-extubation respiratory failure and excluded those with hypercapnia. The exacerbation of COPD and acute cardiogenic pulmonary oedema may require further study.
A few RCTs [
19,
22,
25] showed that the length of ICU stay was not influenced by choice. Parke et al. [
22] reported no difference in HFNC compared with COT (HFNC 33.4 h versus COT 28.9 h,
P = 0.08 [
22], and Hernandez et al. [
19,
25] reported that the patients’ length of stay in the ICU was similar between HFNC and COT (6 days (IQR 2–8) with HFNC versus 6 days (IQR 2–9) with COT,
P = 0.29), and between HFNC and NIV (9 days (IQR 4–19) in HFNC versus 10.5 days (IQR 5–19) in NIV,
P = 0.23). Thus, more RCTs are necessary in the future.
There are several limitations to our meta-analysis. First, the causes of the respiratory distress were heterogeneous in the patients recruited for the studies included. The studies included in the meta-analysis had been performed by researchers who operated independently, and the subjects or interventions in these studies would have differed in ways that could have impacted the results; therefore, we used a random-effects model to make the outcomes conservative and easy to justify. The subgroup analysis showed that HFNC may be more suitable for patients with post-extubation ARF. Second, the duration of treatment and the modes of HFNC and NIV were also variable, which further increased the heterogeneity. Third, all the RCTs included had a high risk of performance bias due to the dramatic differences between HFNC, COT and NIV, which made blinding impossible. Finally, all the studies included in this meta-analysis received sponsorship from Fisher and Paykel Healthcare, although most of the articles [
13,
18‐
20,
44‐
46] state that the study design and data analysis were independent of commercial interference.