High‐flow nasal cannulae for respiratory support in adult intensive care patients

Sharon R Lewis; Philip E Baker; Roses Parker; Andrew F Smith

doi:10.1002/14651858.CD010172.pub2

. 2017 May 30;2017(5):CD010172. doi: 10.1002/14651858.CD010172.pub2

High‐flow nasal cannulae for respiratory support in adult intensive care patients

Sharon R Lewis ^1,^✉, Philip E Baker ², Roses Parker ³, Andrew F Smith ⁴

Editor: Cochrane Emergency and Critical Care Group

PMCID: PMC6481761 PMID: 28555461

Abstract

Background

High‐flow nasal cannulae (HFNC) deliver high flows of blended humidified air and oxygen via wide‐bore nasal cannulae and may be useful in providing respiratory support for adults experiencing acute respiratory failure, or at risk of acute respiratory failure, in the intensive care unit (ICU). This is an update of an earlier version of the review.

Objectives

To assess the effectiveness of HFNC compared to standard oxygen therapy, or non‐invasive ventilation (NIV) or non‐invasive positive pressure ventilation (NIPPV), for respiratory support in adults in the ICU.

Search methods

We searched CENTRAL, MEDLINE, Embase, CINAHL, Web of Science, and the Cochrane COVID‐19 Register (17 April 2020), clinical trial registers (6 April 2020) and conducted forward and backward citation searches.

Selection criteria

We included randomized controlled studies (RCTs) with a parallel‐group or cross‐over design comparing HFNC use versus other types of non‐invasive respiratory support (standard oxygen therapy via nasal cannulae or mask; or NIV or NIPPV which included continuous positive airway pressure and bilevel positive airway pressure) in adults admitted to the ICU.

Data collection and analysis

We used standard methodological procedures as expected by Cochrane.

Main results

We included 31 studies (22 parallel‐group and nine cross‐over designs) with 5136 participants; this update included 20 new studies. Twenty‐one studies compared HFNC with standard oxygen therapy, and 13 compared HFNC with NIV or NIPPV; three studies included both comparisons. We found 51 ongoing studies (estimated 12,807 participants), and 19 studies awaiting classification for which we could not ascertain study eligibility information.

In 18 studies, treatment was initiated after extubation. In the remaining studies, participants were not previously mechanically ventilated.

HFNC versus standard oxygen therapy

HFNC may lead to less treatment failure as indicated by escalation to alternative types of oxygen therapy (risk ratio (RR) 0.62, 95% confidence interval (CI) 0.45 to 0.86; 15 studies, 3044 participants; low‐certainty evidence). HFNC probably makes little or no difference in mortality when compared with standard oxygen therapy (RR 0.96, 95% CI 0.82 to 1.11; 11 studies, 2673 participants; moderate‐certainty evidence). HFNC probably results in little or no difference to cases of pneumonia (RR 0.72, 95% CI 0.48 to 1.09; 4 studies, 1057 participants; moderate‐certainty evidence), and we were uncertain of its effect on nasal mucosa or skin trauma (RR 3.66, 95% CI 0.43 to 31.48; 2 studies, 617 participants; very low‐certainty evidence). We found low‐certainty evidence that HFNC may make little or no difference to the length of ICU stay according to the type of respiratory support used (MD 0.12 days, 95% CI ‐0.03 to 0.27; 7 studies, 1014 participants). We are uncertain whether HFNC made any difference to the ratio of partial pressure of arterial oxygen to the fraction of inspired oxygen (PaO₂/FiO₂) within 24 hours of treatment (MD 10.34 mmHg, 95% CI ‐17.31 to 38; 5 studies, 600 participants; very low‐certainty evidence). We are uncertain whether HFNC made any difference to short‐term comfort (MD 0.31, 95% CI ‐0.60 to 1.22; 4 studies, 662 participants, very low‐certainty evidence), or to long‐term comfort (MD 0.59, 95% CI ‐2.29 to 3.47; 2 studies, 445 participants, very low‐certainty evidence).

HFNC versus NIV or NIPPV

We found no evidence of a difference between groups in treatment failure when HFNC were used post‐extubation or without prior use of mechanical ventilation (RR 0.98, 95% CI 0.78 to 1.22; 5 studies, 1758 participants; low‐certainty evidence), or in‐hospital mortality (RR 0.92, 95% CI 0.64 to 1.31; 5 studies, 1758 participants; low‐certainty evidence). We are very uncertain about the effect of using HFNC on incidence of pneumonia (RR 0.51, 95% CI 0.17 to 1.52; 3 studies, 1750 participants; very low‐certainty evidence), and HFNC may result in little or no difference to barotrauma (RR 1.15, 95% CI 0.42 to 3.14; 1 study, 830 participants; low‐certainty evidence). HFNC may make little or no difference to the length of ICU stay (MD ‐0.72 days, 95% CI ‐2.85 to 1.42; 2 studies, 246 participants; low‐certainty evidence). The ratio of PaO₂/FiO₂ may be lower up to 24 hours with HFNC use (MD ‐58.10 mmHg, 95% CI ‐71.68 to ‐44.51; 3 studies, 1086 participants; low‐certainty evidence). We are uncertain whether HFNC improved short‐term comfort when measured using comfort scores (MD 1.33, 95% CI 0.74 to 1.92; 2 studies, 258 participants) and responses to questionnaires (RR 1.30, 95% CI 1.10 to 1.53; 1 study, 168 participants); evidence for short‐term comfort was very low certainty. No studies reported on nasal mucosa or skin trauma.

Authors' conclusions

HFNC may lead to less treatment failure when compared to standard oxygen therapy, but probably makes little or no difference to treatment failure when compared to NIV or NIPPV. For most other review outcomes, we found no evidence of a difference in effect. However, the evidence was often of low or very low certainty. We found a large number of ongoing studies; including these in future updates could increase the certainty or may alter the direction of these effects.

Plain language summary

High‐flow nasal cannulae for breathing support in adult intensive care patients

Review question

Are high‐flow nasal cannulae (HFNC) a helpful treatment option for adult patients in the intensive care unit (ICU) who need breathing support?

Background

People in the ICU may need support to breathe and HFNC are one option for this. HFNC deliver warm air and oxygen through small plastic tubes that sit inside the nostrils. The airflow is at a higher rate each minute than standard oxygen therapy (which is not always warmed and may be delivered through a plastic face mask or nasal cannulae). Other support options include non‐invasive ventilation (NIV) or non‐invasive positive pressure ventilation (NIPPV). These approaches use mild pressure to push air into the lungs through tightly‐fitting face masks or a helmet covering the entire head. Invasive mechanical ventilation provides the highest level of support, using a ventilator (artificial breathing machine) to push air in and out of the lungs through a plastic tube inserted into the windpipe.

Search date

The evidence is current to April 2020.

Study characteristics

All participants were adults (16 years or older) requiring support to breathe in an ICU. Most participants had respiratory failure (in which the lungs are unable to get enough oxygen into the blood) or had just been taken off a ventilator and needed support to transition to independent breathing.

We searched for randomized controlled trials; these trials give participants an equal chance to be in either trial group and provide the best evidence. We included trials that compared HFNC with standard oxygen therapy or NIV or NIPPV. We included 31 studies with 5136 participants, 51 ongoing studies and 19 studies awaiting classification. Fourteen studies were funded by manufacturers of breathing equipment.

Key results

HFNC compared to standard oxygen therapy

We found that using HFNC may reduce the need for patients to change to another type of breathing support (treatment failure). We found no evidence of a difference between the two interventions for: hospital deaths, length of ICU stay, pneumonia (lung infection), skin damage caused by tubes or masks in contact with the face, comfort while patients received breathing support, or in how well either treatment provided oxygen to the blood.

HFNC compared to NIV or NIPPV

We found no evidence of a difference in treatment failure between using HFNC and NIV or NIPPV. We also found no evidence of a difference for hospital deaths, length of ICU stay, pneumonia, or barotrauma (damage to the body caused by differences in pressure inside and outside the body). NIV or NIPPV may improve how well oxygen gets into the blood. We are uncertain whether HFNC could be more comfortable for patients in the first 24 hours of use. No studies reported skin damage.

Quality of evidence

We used a rating scale to decide the quality of the evidence in these trials. When we rate evidence as very low‐certainty, it means that we are very uncertain about the reliability of the results. High‐certainty means that we are very confident about the results.

We did not always have evidence from enough studies to give us confidence in the key results. Sometimes our findings changed if we removed studies that were less well reported (e.g. regarding how participants were allocated to a treatment). We also found some variation between study results for some outcomes. We are moderately certain in our findings that HFNC did not influence hospital deaths and pneumonia when compared to standard oxygen therapy, but for all other outcomes, we judged the evidence to be of low or very low certainty. This means that our confidence in these results is limited or very limited, and the real effect may be very different.

Conclusion

HFNC may lead to less treatment failure when compared to standard oxygen therapy, but probably makes little or no difference when compared to NIV or NIPPV. For most other review outcomes, we found no reliable evidence of a difference in effect. However, we identified another 51 ongoing trials and we plan to include these in future updates of the review. When these trials are incorporated, we may reach different conclusions about whether HFNC is helpful for breathing support in adult ICU patients.

Summary of findings

Background

Description of the condition

Acute respiratory failure and the subsequent need for respiratory support is a frequent cause of admission of adults to an intensive care unit (ICU) (Behrendt 2000). There are multiple pathological processes which lead to acute respiratory failure, making it difficult to summarise all the possible conditions which would result in a person requiring respiratory support in an ICU environment. In broad terms, respiratory failure in critically unwell people can be considered due to hypoxaemia, ventilatory failure or both (Shelly 1999). These can be a result of conditions such as community or hospital‐acquired pneumonia, sepsis, aspiration or drowning, pneumonia related to being immunocompromised, acute exacerbations of chronic obstructive pulmonary disease (COPD), after surgery (particularly cardiothoracic surgery) and many others. Although drugs may improve some types of respiratory failure (Lewis 2019), respiratory support is the mainstay of treatment. This respiratory support can be provided to the patient in an invasive or non‐invasive manner.

Description of the intervention

Invasive mechanical ventilation involves the insertion of an artificial airway (an endotracheal or tracheostomy tube). Although this is regarded as a life‐saving treatment, it comes with multiple inherent risks to patients. These risks include the development of ventilator‐induced lung injury (Gattinoni 2012), ventilator‐associated pneumonia (Muscadere 2008), neurocognitive sequelae associated with prolonged sedation (Morandi 2011; Nelson 2000), and increased length of ICU and hospital stay (Safdar 2005). Therefore, when possible, invasive mechanical ventilation should be avoided. However, intubation and mechanical ventilation are inevitable if the patient has stopped breathing or is unable to maintain their airway (Nava 2009).

Non‐invasive respiratory support, when possible, is the preferred method of respiratory support and can be delivered via any of the following approaches (O'Driscoll 2008).

Low‐flow nasal cannulae (LFNC).
Simple face mask.
Venturi mask.
Non‐rebreather mask.
Non‐invasive ventilation or non‐invasive positive‐pressure ventilation (NIPPV).
High‐flow nasal cannulae (HFNC).

The type of delivery device chosen depends largely on the severity and the cause of the patient's acute respiratory failure, and each device provides benefits and drawbacks that determine its usefulness in clinical practice.

Physicians use LFNC for patients requiring minimal respiratory support in the form of supplemental oxygen to maintain adequate oxygenation. These cannulae deliver dry oxygen at 1 to 6 litres per minute via small prongs approximately 1.5 cm long, which sit just inside the nostrils (O'Driscoll 2008). Although they are generally well tolerated by patients (Zevola 2001), delivery of higher flows of oxygen through LFNC is not practicable owing to the drying and irritating effects of cold dry gas on the mucosa (Costello 1995; Cuquemelle 2012; Lellouche 2002).

Delivery of oxygen via a face mask is necessary if the patient has higher oxygen requirements than can be achieved with LFNC. Simple face masks can deliver 5 to 10 litres per minute of oxygen. For patients requiring increased oxygen and higher flows to maintain adequate oxygenation, non‐rebreather masks can deliver 10 to 15 litres per minute of oxygen (O'Driscoll 2008). Oxygen may be supplemented with humidification by some devices. Simple face masks and non‐rebreather masks are capable of delivering relatively high oxygen concentrations; therefore, they are generally unsuitable for patients with chronic obstructive pulmonary disease (COPD), who may retain carbon dioxide. For hypercapnoeic patients with COPD, oxygen concentration can be regulated by a Venturi mask, which can deliver between 24% and 60% oxygen at a flow of 2 to 15 litres per minute (O'Driscoll 2008). Although face masks are effective for delivering oxygen to patients with mild to moderate acute respiratory failure, they can be poorly tolerated when compared with nasal cannulae owing to discomfort and feelings of claustrophobia. This may lead to reduced compliance as a result of frequent removal and subsequent treatment interruption (Sasaki 2003).

HFNC, which have been used in the neonatal setting for some years (Wilkinson 2016), are a relatively new method of delivering respiratory support to adults experiencing acute respiratory failure. Cannulae are approximately 1.5 cm long and 0.5 cm in diameter and, as with LFNC, sit just inside the nostrils. A gas flow of up to 60 litres per minute can be delivered because the gas is warmed and humidified, making it less irritating to the nasal mucosa (Papazian 2016). For this review, HFNC will be defined as humidified oxygen delivered via nasal cannulae at a rate greater than 20 litres per minute. Very few adverse reactions have been reported with HFNC use and those reported consist of minor complaints of a runny nose (Price 2008) and some discomfort with heat or flow rate (Roca 2010).

NIPPV can be used in patients who not only require supplemental oxygen but also need support for the mechanical process of ventilation (Mehta 2001). A blend of oxygen and air is delivered at a prescribed fraction of inspired oxygen (FiO₂) via a tight‐fitting mask (nasal mask, oronasal mask, or full face mask). Additionally, continuous positive airway pressure (CPAP) or bilevel positive airway pressure ventilation (BiPAP) is delivered to improve alveolar recruitment, improve gas exchange, and decrease the work of breathing (Mehta 2001). Although CPAP is not a true ventilatory mode, it is often referred to as NIPPV in clinical practice (Nava 2009). Substantial available data show that NIPPV improves outcomes among patients requiring respiratory support owing to cardiogenic pulmonary oedema or acute exacerbations of COPD, and also among patients weaning from invasive mechanical ventilation (Nava 2009). However, its relevance for patients with hypoxaemic acute respiratory failure is less clearly defined (Nava 2009). Despite showing clear benefit for certain conditions, NIPPV inhibits mobilization, is associated with gastric distension, restricts effective communication and oral nutrition, and is poorly tolerated by some patients owing to discomfort (Bello 2016; Gregoretti 2002; Mehta 2001).

Although the conventional non‐invasive delivery devices listed above provide important therapies in the range of respiratory support available to treat patients with acute respiratory failure, it is evident that they have limitations that can impact their usefulness in clinical practice. Failure of these devices to provide adequate respiratory support and to correct acute respiratory failure often results in the need for intubation and mechanical ventilation.

How the intervention might work

HFNC can deliver blended humidified air and oxygen via wide‐bore nasal cannulae at a prescribed FiO₂ at high‐flow rates. HFNC do not need to be removed during oral hygiene care or when patients talk, eat, or drink, resulting in less frequent interruptions to therapy. In the growing body of evidence gathered when effects of HFNC are investigated, improvements in oxygenation (Corley 2011; Parke 2009; Roca 2010; Sztrymf 2011; Sztrymf 2011a), respiratory rate (Corley 2011; Roca 2010; Sztrymf 2011; Sztrymf 2011a), dyspnoea (Corley 2011; Roca 2010; Sztrymf 2011), and patient comfort (Corley 2011; Roca 2010) have been reported in recent observational studies.

Suggested mechanisms of action of HFNC consist of:

flushing of anatomical dead space due to high gas flow, functionally reducing dead space and improving respiratory efficiency (Dysart 2009);
generation of positive airway pressure (Corley 2011; Groves 2007; Parke 2009), which increases functional residual capacity and improves alveolar recruitment;
improved ability to meet high inspiratory flow demands among patients requiring respiratory support and to deliver a more accurate FiO₂ through less dilution by entrainment of room air (Dysart 2009); and
ability to deliver optimal humidification, leading to enhanced mucociliary transport (Salah 1988) and improved patient comfort (Chanques 2009).

We conducted this review to compare the efficacy and safety of HFNC versus other methods of non‐invasive respiratory support in adult patients admitted to the ICU.

Why it is important to do this review

It has been demonstrated that HFNC offer some immediate physiological benefit for patients requiring respiratory support, but it remains to be determined whether they offer any clinically important benefit and improve patient outcomes, such as by preventing progression to invasive mechanical ventilation and reducing mortality. Despite increased popularity as a treatment modality for respiratory support, there is still uncertainty about which patient populations benefit most from HFNC as compared to other therapies (Curley 2015; Demoule 2015; Levy 2016; Nishimura 2015). Individual studies may tend to focus on surrogate outcomes or may be underpowered to detect effects on clinically important outcomes. This is an update of a previously published Cochrane Review (Corley 2017). In the previous version of this review, we found 11 eligible studies but this was insufficient to demonstrate with any certainty whether HFNC is a more effective or safe oxygen delivery device compared with other oxygenation devices in adults in the ICU. Given the COVID‐19 pandemic, in which people are admitted to the ICU requiring respiratory support caused by SARS‐CoV‐2, we believe there is an urgent need to update this review in order to re‐evaluate the effectiveness of HFNC in the adult ICU population and incorporate the most recent evidence.

Objectives

To assess the effectiveness of HFNC compared to standard oxygen therapy, NIV or NIPPV, for respiratory support in adults in the ICU.

Methods

Criteria for considering studies for this review

Types of studies

We included all randomized controlled trials (RCTs) which used either a parallel‐group or cross‐over study design. Owing to the inability of randomized cross‐over studies to detect long‐term patient outcomes, we included this trial design only for the additional outcome measures of positive end‐expiratory pressure, oxygenation, carbon dioxide clearance, respiratory rate, work of breathing, and participant‐reported outcomes.

We did not impose language restrictions.

We excluded cluster‐RCTs, quasi‐RCTs, retrospective studies and prospective cohort or observational studies, as we wanted to focus on evidence of the highest quality from randomized studies.

Types of participants

We included studies that enrolled adults (16 years of age or older) requiring respiratory support and admitted to the ICU.

We excluded participants younger than 16 years of age. Two already published Cochrane Reviews have assessed the effectiveness of HFNC in preterm infants (Wilkinson 2016) and in the paediatric population (Mayfield 2014).

We also excluded participants not admitted to an ICU.

Types of interventions

We included humidified oxygen delivered via the nasal route at a rate greater than 20 litres per minute as the experimental intervention. We referred to this intervention as HFNC.

We compared HFNC to other types of non‐invasive respiratory support which were:

standard oxygen therapy delivered via nasal cannulae or any type of face mask with a gas flow rate of ≤ 15 L/min (with or without humidification and heating)
non‐invasive ventilation (NIV), or non‐invasive positive‐pressure ventilation (NIPPV) which included devices that used bilevel positive airway pressure (BiPAP) or continuous positive airway pressure (CPAP)

We, therefore, included two distinct comparisons in the review:

Comparison 1: HFNC versus standard oxygen therapy
Comparison 2: HFNC versus NIV or NIPPV

Types of outcome measures

The outcome measures in this review are a mix of surrogate, clinical and participant‐reported outcomes. We recognize that, while there may be a correlation between the surrogate and clinical outcomes, it is the clinical outcomes which will provide the strongest evidence regarding the safety and efficacy of HFNC. Similarly, participant‐reported outcomes may help patients to make informed decisions about their care. We considered the short‐term effects and the long‐term effects of treatment, and therefore we collected outcome data up to 24 hours from the initiation of treatment (short‐term) as well as at more than 24 hours (long‐term).

We assessed all outcomes at the time points reported in the included studies. For participant‐reported outcomes, we accepted the study authors’ definitions.

Important outcomes

Treatment failure as indicated by the need for escalation of respiratory therapy (up to 28 days). Escalation of therapy may depend on the initial type of respiratory therapy given to participants and we will be guided by study authors definitions; for example, we will include treatment failure defined as escalation from HFNC or standard oxygen therapy to NIV, NIPPV or invasive mechanical ventilation, as well as escalation from NIV or NIPPV to invasive mechanical ventilation.
In‐hospital mortality (measured up to 90 days).
Adverse events: pneumonia and nasal mucosa or skin trauma. In comparison 2 (HFNC versus NIV or NIPPV); we also included barotrauma (pneumothorax).
Length of ICU stay (in days).
Short‐term oxygenation: partial pressure of arterial oxygen/fraction of inspired oxygen (PaO₂/FiO₂) ratio (mmHg).
Participant‐reported outcomes: short‐ and long‐term comfort.

Additional outcomes

Duration of any type of respiratory support (mechanical ventilation, NIPPV, HFNC, standard oxygen) (in hours);
Long‐term oxygenation: PaO₂/FiO₂ (mmHg);
Short‐term and long‐term other respiratory effects as indicated by any of the following:

- Degree of atelectasis on radiological examination
- Positive end‐expiratory pressure measured at the pharyngeal level (cm H₂O)
- Oxygenation: partial pressure of oxygen in arterial blood (PaO₂; mmHg); oxygen saturation of arterial blood (SaO₂; mmHg); and oxygen saturation (SpO₂; %)
- Carbon dioxide clearance: partial pressure of carbon dioxide in arterial blood (PaCO₂; mmHg)
- Respiratory rate (breaths per minute)
- Work of breathing (joules per litre);

Additional adverse events: tracheobronchitis and abdominal distension;
Length of hospital stay (in days);
Additional short‐term and long‐term participant‐reported outcomes as indicated by any of the following:
- Dyspnoea
- Dry mouth
- Refusal to continue with treatment:
Cost comparison of treatment (in Australian dollars).

Search methods for identification of studies

Electronic searches

We identified RCTs through literature searching with systematic and sensitive search strategies, as outlined in Chapter 4 of the Cochrane Handbook of Systematic Reviews of Interventions (Cochrane Handbook;Higgins 2019). We applied no restrictions on language or publication status. We searched the following databases for relevant trials.

Cochrane Central Register of Controlled Trials (CENTRAL; 2020; Issue 4);
MEDLINE (Ovid SP; 2000 to 17 April 2020);
Embase (Ovid SP; 2000 to 17 April 2020);
Cumulative Index to Nursing and Allied Health Literature (CINAHL; EBSCOhost; 2000 to 17 April 2020);
Web of Science (SCI‐Expanded; 2000 to 17 April 2020);
Cochrane COVID‐19 Study Register (17 April 2020).

We restricted the search start date to 2000, as HFNC have been available for use in the adult population only since the mid‐2000s. For this review update, we amended the search strategies for MEDLINE and Embase, added a search in the Cochrane COVID‐19 Study Register and brought the searches in the other listed databases up to date. We also re‐considered eligibility of the studies included in the previous version of the review (Corley 2017). The search strategy was developed in consultation with the Information Specialist for the Cochrane Emergency and Critical Care Group. Search strategies can be found in: Appendix 1; Appendix 2; Appendix 3; Appendix 4; Appendix 5; Appendix 6.

We searched the following clinical trials registers for ongoing and unpublished trials:

World Health Organization International Clinical Trials Registry Platform (who.int/ictrp; on 6 April 2020);
ClinicalTrials.gov (clinicaltrials.gov; on 6 April 2020).

Searching other resources

We carried out citation searching of identified included studies published since the last review update in Web of Science on 29 April 2020. In addition, we scanned reference lists of relevant systematic reviews which were published since 2018, and we searched Opengrey on 29 April 2020 (www.opengrey.eu).

Data collection and analysis

Two review authors (PB and SL) independently selected studies and extracted data from studies identified in the most recent search. We compared decisions at each stage, and reached consensus through discussion.

Selection of studies

We used reference management software to collate the results of searches and to remove duplicates (Endnote). We used Covidence software to screen results of the search of titles and abstracts and to identify potentially relevant studies (Covidence). We sourced the full texts of all potentially relevant studies and considered whether they met the inclusion criteria (see Criteria for considering studies for this review).

We recorded the number of papers retrieved at each stage and reported this information in the Results.

Data extraction and management

Two review authors (PB and SL) independently extracted information and outcome data from each study using a data extraction template (Appendix 7). We compared collected information and outcome data and reached consensus through discussion.

We collected the following information.

Methods: type of study design; setting; dates of study; funding sources and declarations of interest.
Participants: number randomized to each group; number of losses in each group (with reasons for loss); number analysed in each group; inclusion and exclusion criteria; baseline characteristics (age, gender, body mass index, illness severity score, PaCO₂, PaO₂/FiO₂, respiratory rate).
Interventions: details of intervention and comparison (type of respiratory support, time of initiation, gas flow, duration of support).
Outcomes: data for all reported outcomes to include study author definitions, measurement scales, and time points.

Assessment of risk of bias in included studies

We assessed study quality, study limitations, and the extent of potential bias using the Cochrane 'Risk of bias' tool (Higgins 2011). We considered the following domains:

Sequence generation (selection bias);
Allocation concealment (selection bias);
Blinding of participants, personnel, and outcome assessors (performance and detection bias);
Incomplete outcome data (attrition bias);
Selective reporting (reporting bias);
Other risks of bias.

For each domain, two review authors (PB and SL) judged whether study authors made sufficient attempts to minimize bias in their study design. For performance bias, we accepted that it was not possible to blind participants and personnel to the type or respiratory support used. For detection bias, we separated our judgements for outcomes that we considered to be subjective, which were all the participant‐reported outcomes, and the other outcomes which we considered to be objective. We conducted 'Risk of bias' judgements only for studies in which we reported outcome data.

For each domain, we made judgements using three measures ‐ high, low, or unclear risk of bias. We recorded this in 'Risk of bias' tables and presented a summary risk of bias.

Measures of treatment effect

We collected dichotomous data for the primary outcomes. We collected continuous data for most of the secondary outcomes (duration of respiratory support, length of stay, and respiratory effects). We collected either continuous or dichotomous for participant‐reported outcomes, such as comfort scores, depending on the methods and measurement scales used to report these outcomes in the study reports.

We reported dichotomous outcomes as risk ratios (RR) to compare groups. For continuous data, we reported the mean difference (MD). In the event that studies used different measurement scales, we scaled and inverted the scales to allow calculation of the MD where possible. Where this was not possible, we selected the standardised mean difference (SMD) for measurement. An example of scaling would include dividing the mean and standard deviation by the 10 to convert a 0‐100 scale to a 0‐10 scale. An example of inverting a scale would include subtracting the mean from the highest number on the scale (i.e. 3 out of 10 on a discomfort high scale would be equivalent to 7 out of 10 on a comfort high scale). We reported 95% confidence intervals for RRs and MDs or SMDs. For outcomes for which only one study was available, we used the calculator in RevMan Web to calculate the effect estimates (RevMan Web 2019); we reported the effect estimates from single studies in additional tables. In the event that studies reported data only as median values or used scales that did not easily translate to dichotomous or continuous data, we also reported data separately in additional tables.

Unit of analysis issues

Although we included cross‐over study designs, we only included data in the review if study authors reported findings for the first treatment period.

Included studies measured many of the secondary outcomes (oxygenation (PaO₂, PaO₂/FiO₂ ratio, SpO₂), carbon dioxide clearance, respiratory rate, dyspnoea, mouth dryness, and patient comfort) at multiple time points. To overcome the potential for unit of analysis error, we took a simple approach to analysis of these outcomes on the advice of the statistical editor. We reported outcome data as short‐term and longer‐term effects, with short‐term effects resulting from initiation of therapy up to 24 hours, and longer‐term effects occurring more than 24 hours after initiation of therapy. For short‐term effects, we used the closest data point to 24 hours.

One study reported findings from two separate types of respiratory support (Frat 2015); in the previous version of this review an adjustment was made so that both arms could be included in the same analysis without introducing a unit of analysis issue (Corley 2017). Because we introduced two separate comparison groups in this review, such an adjustment was no longer required.

Dealing with missing data

We updated this review during the Covid‐19 pandemic because the review provides relevant evidence for respiratory support in adults in the ICU. To reduce time to publication, rather than seeking additional information from all study authors, we only attempted to contact study authors to provide clarity when combining data for treatment failure; study authors did not provide additional data. We did not tabulate missing data and perform sensitivity analyses to determine the influence of missing data on effect estimates, as planned in the protocol (Corley 2012).

Assessment of heterogeneity

Using clinical judgement, we assessed participants, interventions, and outcomes for clinical heterogeneity. We assessed methodological heterogeneity during 'Risk of bias' assessments and by visual inspection of forest plots. We assessed statistical heterogeneity by using the I² statistic (on a scale of 0% to 100%) and the Chi² test (Higgins 2019).

Assessment of reporting biases

We attempted to source the published protocols for each our included studies by using the results from our clinical trials register searches. We compared clinical trials register documents, or protocols, with published study results to assess the risk of selective reporting bias for outcomes relevant to this review. We assessed publication bias from the visual inspection of funnel plots for important review outcomes (i.e. those that we included in the 'Summary of findings' tables) if the outcomes included more than ten studies (Egger 1997).

Data synthesis

We conducted meta‐analyses for outcomes for which we had comparable study data and presented a summary statistic for each outcome. We conducted analyses for outcomes using RevMan Web 2019. We performed separate analyses for comparisons of HFNC versus standard oxygen therapy and for HFNC versus NIV or NIPPV. We classified the level of heterogeneity using the I² statistic as: 0% to 40%, not important; 30% to 60%, moderate heterogeneity; 50% to 90%, substantial heterogeneity; and 75% to 100%, considerable heterogeneity (Higgins 2019). We selected a random‐effects model for all meta‐analysis to account for the likely variation in the study population (Borenstein 2010).

Subgroup analysis and investigation of heterogeneity

We performed subgroup analysis for treatment failure, as indicated by the need for escalation to NIV, NIPPV or invasive ventilation, for studies in which respiratory support was given after extubation versus respiratory support that was given without prior mechanical ventilation.

Sensitivity analysis

We explored the potential effect of study limitations in the important review outcomes, as well as the effect of our chosen meta‐analytical effects model. In each sensitivity analysis, we compared the effect estimate with the main analysis. We reported these effect estimates only if they indicated a difference in interpretation of the effect. We performed the following sensitivity analyses on the important review outcomes:

We excluded studies that we judged to have a high or unclear risk of selection bias for either random sequence generation, allocation concealment, or both.
We excluded studies that we judged to have a high risk of bias in any of the other domains (other than selection bias).
We re‐analysed the data using a fixed‐effects model instead of a random‐effects model.
We excluded studies that were funded from commercial sources.

Summary of findings and assessment of the certainty of the evidence

Two review authors (PB and SL) used the GRADE system to assess the certainty of the body of evidence and construct a 'Summary of findings' table associated with the following important outcomes (Guyatt 2008):

Failure of treatment as indicated by the need for escalation of respiratory support;
In‐hospital mortality;
Adverse events (pneumonia, nasal mucosa or skin trauma, barotrauma);
Length of stay in days (ICU);
PaO₂/FiO₂ ratio up to 24 hours after initiation of therapy;
Comfort (short‐term effects);
Comfort (long‐term effects).

The GRADE approach appraises the certainty of a body of evidence based on the extent to which we can be confident that an estimate of effect or association reflects the item being assessed. Evaluation of the certainty of a body of evidence considers within‐study risk of bias, directness of the evidence, heterogeneity of the data, precision of the effect estimates, and risk of publication bias.

We constructed 'Summary of findings' tables using GRADEpro GDT software for the following comparisons in this review (gradepro.org):

HFNC versus standard oxygen therapy;
HFNC versus NIPPV or NIV.

Results

Description of studies

See Characteristics of included studies, Characteristics of excluded studies, and Characteristics of ongoing studies.

Results of the search

After the removal of duplicates from the search results, we screened 4224 titles and abstracts, which included forward and backward citation searches and searches of clinical trials registers. We looked at the full text of 138 reports and selected 31 studies for inclusion, based on review criteria. We identified 51 ongoing studies, found 19 studies for which we could not assess eligibility, and we excluded nine studies (see Figure 1).

Flow diagram. Search conducted in April 2020

Included studies

We included 31 RCTs with 5136 randomized participants. Nine were randomized cross‐over studies (Chanques 2013; Grieco 2020; Lee 2018; Longhini 2019; Mauri 2017a; Mauri 2017b; Rittayamai 2014; Schwabbauer 2014; Vargas 2015); the remaining studies all used a parallel‐group design.

This update included 20 new studies (Azoulay 2018; Brainard 2017; Cong 2019; Fernandez 2017; Futier 2016; Grieco 2020; Hernandez 2016a; Hernandez 2016b; Hu 2020; Jing 2019; Lee 2018; Longhini 2019; Mauri 2017a; Mauri 2017b; Shebl 2018; Song 2017; Vargas 2015; Vourc'h 2020; Yu 2017; Zochios 2018). The remaining studies were previously included in Corley 2017.

During the previous version of the review (Corley 2017), we contacted eight study authors by email to request additional details, including outcome data not available in the published report and information for 'Risk of bias' assessment (Chanques 2013; Corley 2014; Cuquemelle 2012; Maggiore 2014; Parke 2011; Parke 2013a; Rittayamai 2014; Schwabbauer 2014). Chanques 2013, Corley 2014, Parke 2011, Parke 2013a, and Rittayamai 2014 provided participant and outcome data and clarification on methodological issues; Cuquemelle 2012 provided information on methodological issues but was unable to provide data; Schwabbauer 2014 was unable to provide any additional details for the study. Following contact with Maggiore 2014, the full report was published, and we used data from this report, rather than information provided via email communication. In this update, we contacted two study authors by email to request additional details for our important outcomes (Fernandez 2017; Vourc'h 2020).

Study population

We included only studies that examined participants 16 years of age or older requiring respiratory support. Participants in all studies had respiratory failure or were at risk of respiratory failure. Most studies included a heterogeneous study population, with respiratory failure resulting from a variety of causes. Some of the studies included only participants with specific causes of respiratory failure. Three studies specifically included participants requiring support for an exacerbation of chronic obstructive pulmonary disorder (COPD) (Cong 2019; Jing 2019; Longhini 2019), and in one study all participants were immunocompromised (Azoulay 2018). Eight studies required support following cardiothoracic surgery (Brainard 2017; Corley 2014; Futier 2016; Parke 2013a; Stephan 2015; Vourc'h 2020; Yu 2017; Zochios 2018). We included only one study in which all participants had a body mass index (BMI) of at least 30 kg/m² (Corley 2014).

Study setting

All studies were conducted in intensive care units, and 11 of these were multicentre studies (Azoulay 2018; Fernandez 2017; Frat 2015; Futier 2016; Hernandez 2016a; Hernandez 2016b; Lemiale 2015; Longhini 2019; Maggiore 2014; Stephan 2015; Yu 2017).

Interventions and comparisons

All studies randomized a group of participants to receive oxygen via HFNC. We noted differences in flow rates between the studies. Most specified a range of flow rates which was between 30 L/min and 60 L/min (Cong 2019; Cuquemelle 2012; Fernandez 2017; Lee 2018; Maggiore 2014; Parke 2011; Parke 2013a; Rittayamai 2014; Yu 2017; Zochios 2018); in one study, this range had a lower flow rate of 20 to 50 L/min (Futier 2016). Others specified an initial flow rate, with subsequent increases upwards from 10 L/min (Hernandez 2016a), or decreases from 60 L/min (Song 2017). Twelve studies specified a maximum target of up to 50 L/min (Corley 2014; Frat 2015; Grieco 2020; Lemiale 2015; Stephan 2015), or up to 60 L/min (Azoulay 2018; Hu 2020), or a set flow rate of 40 L/min (Brainard 2017; Mauri 2017b), 45 L/min (Vourc'h 2020), 55 L/min (Schwabbauer 2014), or 60 L/min (Vargas 2015). Two multi‐arm studies tested oxygen delivery at different flow rates (Chanques 2013; Mauri 2017a); these were 15, 30, and 45 L/min, and at 30, 45, and 60 L/min, respectively. Flow rates were not specified in four studies (Jing 2019; Lee 2018; Longhini 2019; Shebl 2018).

Three multi‐arm studies included control groups for both of our comparison groups (Chanques 2013; Frat 2015; Schwabbauer 2014). In the standard oxygen therapy comparison group, most studies had a control group in which oxygen was delivered with face masks using a simple face mask, nasal cannulae, a non‐rebreather face mask or Venturi mask; in Chanques 2013, the control used a high‐flow face mask, and in Parke 2011 and Vourc'h 2020, they used a high‐flow face mask with humidifier. Cuquemelle 2012 described the use of 'standard oxygen therapy', with no additional details.

In the NIV and NIPPV comparison group, delivery devices were via Bossignac oxygen therapy (Chanques 2013), and bilevel positive airway pressure (BiPAP) (Cong 2019; Frat 2015; Grieco 2020; Hernandez 2016a; Jing 2019; Schwabbauer 2014; Stephan 2015; Shebl 2018; Vargas 2015).

Fourteen studies initiated the intervention or control after extubation from invasive mechanical ventilation (Chanques 2013; Corley 2014; Fernandez 2017; Futier 2016; Hernandez 2016a; Jing 2019; Maggiore 2014; Parke 2013a; Rittayamai 2014; Song 2017; Stephan 2015; Vourc'h 2020; Yu 2017; Zochios 2018). Participants in Futier 2016, Yu 2017 and Zochios 2018 were at high risk of pulmonary complications. Participants in Hernandez 2016a and Hu 2020 were at high risk of extubation failure, and low risk of extubation failure in Hernandez 2016b. The remaining studies initiated the intervention without previously using mechanical ventilation.

Outcomes

For cross‐over studies, we only included outcome data from the first treatment period. This was reported in only three of the cross‐over studies (Chanques 2013; Cuquemelle 2012; Rittayamai 2014), and we, therefore, did not report data for the remaining cross‐over studies. In Cuquemelle 2012, which included a four‐hour cross‐over period at the end of a 24‐hour parallel‐group assignment period, we included only narrative results from the initial 24‐hour period. For the remaining studies, we reported outcome data as specified in the study reports.

Funding

Whilst some studies were supported by funding that we considered to be independent of the study (for example, university or government health ministries), we noted that 14 studies were supported by grants or by the provision of study equipment by manufacturers (Fisher & Paykel Healthcare (Azoulay 2018; Chanques 2013; Corley 2014; Cuquemelle 2012; Frat 2015; Hernandez 2016b; Lemiale 2015; Maggiore 2014; Mauri 2017b; Parke 2011; Parke 2013a; Schwabbauer 2014; Zochios 2018)) and from Merck Sharp & Dohome (Grieco 2020). Most studies declared that these manufacturers were not involved in the design or conduct of the study, nor in the interpretation of the results or preparation of the final manuscripts for publication. However, in Parke 2011, study authors declared that Fisher & Paykel Healthcare were involved in the study design and data analysis, and provided financial support for the statistical analysis.

Excluded studies

We excluded nine studies during full‐text review (Coudroy 2019; Delorme 2017; Di Mussi 2016; Lemiale 2016; Liu 2019; Pennisi 2019; Sklar 2018; Thille 2018; Thille 2019). We excluded these nine studies owing to the use of HFNC in both the intervention and control arms of the study (Coudroy 2019; Thille 2018; Thille 2019), because participants receiving HFNC or standard oxygen therapy were not randomized to this treatment (Delorme 2017; Di Mussi 2016; Lemiale 2016), and because the study setting was not in an ICU (Pennisi 2019; Sklar 2018). See Characteristics of excluded studies.

This review does not include studies that were previously excluded; details of previous exclusions can be found in the earlier version of the review (Corley 2017).

Studies awaiting classification

We were unable to assess eligibility for 19 studies (Arman 2017; Guoqiang 2018; Gupta 2016; Ischaki 2019; ISRCTN17399068; Lee 2016; Longhini 2017; Macari 2019; Menga 2019; Papachatzakis 2017; Perbet 2014; Saeed 2015; Schreiber 2017; Theerawit 2017; Tseng 2019; Yang 2019; Zhang 2018; Zhao 2019; Zhu 2017); this included two studies that were awaiting classification in a previous version of this review (Perbet 2014; Saeed 2015).

Four of these studies were not published in English and required translation before inclusion in the review (Yang 2019; Zhang 2018; Zhao 2019; Zhu 2017). We identified one completed study in the clinical trial register searches, but because the data in the clinical trials register had not been peer‐reviewed, we did not include it in the review (ISRCTN17399068). The remaining studies were published as abstracts; we are awaiting publication of the full study reports for these studies (and for the study in the clinical trials report) in order to fully assess eligibility and incorporate these results in the review.

Eight of the studies awaiting classification investigated HFNC for post‐extubation respiratory support; of these, two included participants with acute exacerbations of COPD (Guoqiang 2018; Zhang 2018), one included participants post‐surgery (Gupta 2016), and five considered all intubated participants in an ICU setting (Arman 2017; Perbet 2014; Theerawit 2017; Tseng 2019; Zhu 2017). Ten studies investigated HFNC for respiratory support without prior use of invasive mechanical ventilation; of these, five included participants with acute exacerbations of COPD (Ischaki 2019; Lee 2016; Longhini 2017; Saeed 2015; Yang 2019), and five included participants with acute hypoxic respiratory failure (AHRF) (Macari 2019; Menga 2019; Papachatzakis 2017; Schreiber 2017; Zhao 2019).

Ongoing studies

We identified 51 ongoing RCTs with an estimated recruitment of 12,807 participants (see Characteristics of ongoing studies). Three studies marked as ongoing in the previous version of this review had been completed and are now included in this update (Fernandez 2017; Vargas 2015; Vourc'h 2020). Five of the 51 studies were cross‐over RCTs (NCT03811158; NCT03865056; NCT03877172; NCT04036175; NCT04241861), one was a 2 x 2 factorial design RCT (NCT04344730), and the remaining studies were parallel‐group design RCTs.

Six studies, with an estimated 4802 participants, had more than two study arms, and compared HFNC to both standard oxygen therapy and NIV or NIPPV (CTRI/2018/09/015717; ChiCTR‐INR‐17012720; ISRCTN16912075; NCT03171935; NCT03229460; NCT04269681). Twelve studies, with an estimated 1780 randomized participants, compared HFNC to standard oxygen therapy (ACTRN12617000694314; ChiCTR1900021091; NCT01702779; NCT02107183; NCT02290548; NCT03133520; NCT03282552; NCT03361683; NCT03430258; NCT03515031; NCT03811158; NCT03877172). The remaining studies, with an estimated 6225 randomized participants, compared HFNC to NIV or NIPPV.

One study investigated HFNC for both post‐extubation respiratory support and respiratory support without prior use of mechanical ventilation (NCT04269681). Twenty‐three studies investigated HFNC for post‐extubation respiratory support. Of these 23 studies, five included participants with no criteria other than intubation for: at least 24 hours (NCT02107183; NCT04036175); at least 48 hours (NCT01702779; NCT02123940); or with pre‐existing respiratory disease (NCT03632577). Six studies included participants with acute exacerbations of COPD (ChiCTR‐INR‐17011850; ChiCTR‐INR‐17012720; ChiCTR1900025974; NCT02290548; NCT03811158; NCT04156139), five included participants after surgery that had a high risk of extubation failure (ACTRN12617000694314; NCT02713737; NCT03282552; NCT03877172; NCT03928535), two included participants with pneumonia (ChiCTR1900020826; ChiCTR1900021091), four included participants with AHRF (ChiCTR1900023296; NCT02290548; NCT03171935; NCT03361683), one included participants with trauma and acute respiratory distress syndrome (ChiCTR1900023296), one included participants with heart failure (NCT03607357), and one included participants with sepsis (NCT03246893).

Twenty‐seven studies investigated HFNC for respiratory support without prior uses of mechanical ventilation. Of these, 20 included participants with AHRF (CTRI/2018/09/015717; ChiCTR1800017313; ChiCTR1900022241; JPRN‐jRCTs052180236; NCT01166256; NCT02464696; NCT03133520; NCT03229460; NCT03488628; NCT03643939; NCT03788304; NCT03865056; NCT03944525; NCT04035460; NCT04241861; NCT04253405; NCT04293991; NCT04344730; TCTR20171106003; UMIN000008778), five included participants with acute exacerbations of COPD (ChiCTR1800014553; ChiCTR1800018530; Cortegiani 2019; NCT03014869; NCT03643939), two included participants with traumatic injuries (ChiCTR1800017313; NCT03430258), three included participants with pneumonia (ISRCTN16912075; NCT03515031; NCT04344730), of which two focused specifically on patients with COVID‐19 disease (ISRCTN16912075; NCT04344730).

Risk of bias in included studies

We described the risk of bias for each included study in the 'Risk of bias' tables in Characteristics of included studies (see Figure 2). We did not conduct risk of bias assessment for cross‐over studies in which data were not reported for the first period; the 'Risk of bias' figure, therefore, includes blank spaces for seven studies (Grieco 2020; Lee 2018; Longhini 2019; Mauri 2017a; Mauri 2017b; Schwabbauer 2014; Vargas 2015). In addition, we assessed risk of detection bias separately for subjective and objective outcome measures; therefore, some blank spaces in the risk of bias figure indicate that we did not assess risk of bias because the study did not report either objective or subjective outcomes relevant to the review.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study. We only conducted 'Risk of bias' assessments in studies for which we reported outcome data, and for domains that were relevant to reported outcomes (in particular, for detection bias of objective and subjective measures); blank spaces, therefore, indicate that 'Risk of bias' assessment was not conducted for the outcome, or for a particular domain.

Allocation

We found that six studies did not adequately describe a method used to randomize participants to groups (Brainard 2017; Chanques 2013; Cong 2019; Cuquemelle 2012; Rittayamai 2014; Shebl 2018); it was, therefore, unclear whether these studies were at risk of selection bias. We judged the remaining studies to be at low risk of selection bias for random sequence generation because they reported using an appropriate method such as a block system or a computer‐generated sequence.

We judged only 11 studies to be at low risk of selection bias for using a method for concealing allocation (Azoulay 2018; Corley 2014; Fernandez 2017; Futier 2016; Hernandez 2016a; Hernandez 2016b; Jing 2019; Lemiale 2015; Parke 2013a; Vourc'h 2020; Zochios 2018).

Blinding

Owing to the nature of the intervention and comparators, it was not possible to blind participants and their treating clinicians to treatment allocation. Although we believed that knowledge of treatment would not influence performance for the outcomes of interest for this review, we could not be certain of this and we therefore judged all studies to have unclear risk of performance bias.

We judged risk of detection bias according to whether outcomes were objective or subjective. We defined the subjective measures as those being assessed by the participants; these outcomes were dyspnoea, comfort, dry mouth, and refusal to continue with treatment. We defined the remaining outcomes as objective, and we anticipated that knowledge of treatment allocation would not influence the assessment of these outcomes. Therefore, we judged all studies that reported objective measures to have a low risk of detection bias.

Fifteen studies reported subjective measures (Azoulay 2018; Chanques 2013; Cong 2019; Corley 2014; Cuquemelle 2012; Frat 2015; Futier 2016; Jing 2019; Lemiale 2015; Maggiore 2014; Parke 2013a; Rittayamai 2014; Song 2017; Stephan 2015; Vourc'h 2020). However, we believe that the inability to blind participants to treatment allocation would not affect outcome measurements because it would be unlikely that participants would have a particular bias towards one medical intervention over another. We, therefore, judged these studies to have a low risk of detection bias for subjective measures.

Incomplete outcome data

We judged two studies to have a high risk of attrition bias because a large number of participants in these studies were not included in analysis (Brainard 2017; Cuquemelle 2012). In three additional studies, we could not be certain whether participants were lost to follow‐up, or how the data were managed when participants were treated with an alternative therapy (Chanques 2013; Parke 2011; Rittayamai 2014); in these studies, we judged risk of attrition bias to be unclear. The remaining studies reported no or few losses that were sufficiently explained, and we judged these studies to have a low risk of attrition bias.

Selective reporting

Eleven studies reported clinical trials registration which was made prospectively (Azoulay 2018; Corley 2014; Fernandez 2017; Frat 2015; Futier 2016; Hernandez 2016a; Hernandez 2016b; Hu 2020 ;Parke 2011; Parke 2013a; Zochios 2018). However, only five of these prospectively registered studies reported outcomes that were consistent with the clinical trials register documents (Azoulay 2018; Corley 2014; Futier 2016; Hernandez 2016a; Hernandez 2016b). We assessed six of these to be at high risk of bias because they either reported outcomes that were not listed in the clinical trials register documents, or failed to report outcomes as specified in the clinical trials register documents (Fernandez 2017; Frat 2015; Hu 2020; Parke 2011; Parke 2013a; Zochios 2018). We made these risk of bias judgements only according to the outcomes that were relevant to this review.

Six studies reported clinical trials registration which was made retrospectively (Brainard 2017; Chanques 2013; Jing 2019; Lemiale 2015; Maggiore 2014; Stephan 2015). We assessed Maggiore 2014 and Stephan 2015 to be at low risk of selective reporting bias because they were registered only shortly after the study start date, and study authors reported the same outcome data as in the clinical trials register documents. However, we did not think it was feasible to effectively assess risk of selective reporting bias for the remaining retrospectively registered studies, and we judged bias in these studies to be unclear.

The remaining studies did not report clinical trials registration, nor reported a protocol published prior to the completed study report, and we judged risk of selective reporting bias to be unclear (Cong 2019; Cuquemelle 2012; Rittayamai 2014; Shebl 2018; Song 2017; Vourc'h 2020; Yu 2017).

Other potential sources of bias

To date, one of the included studies (Maggiore 2014) has presented three abstracts, and this study is part of a larger ongoing clinical trial (see NCT02107183 in Characteristics of ongoing studies); multiple interim analyses could introduce bias (Bland 1995), and we judged risk of other bias in this study to be unclear.

In Frat 2015, we noted that participants in the NIPPV group could have been exposed to HFNC during breaks in delivery of oxygen, during which choice of oxygen delivery was at the discretion of the attending clinician. Because some participants in the NIPPV group could have received HFNC, we judged this study to have high risk of bias as a result of this methodological decision.

We identified no other sources of bias in the remaining studies.

Effects of interventions

See: Table 1; Table 2

Summary of findings 1. HFNC compared to standard oxygen therapy for respiratory support in adult intensive care patients.

High‐flow nasal cannulae compared to standard oxygen therapy for respiratory support in adult intensive care patients
Population: adults in the ICU, requiring respiratory support Setting: ICUs. In this review, these ICUs were in: Australia; Belgium; China; France; Italy; New Zealand; Spain; Taiwan; Thailand; UK. Intervention: oxygen delivered via HFNC, initiated after extubation from invasive mechanical ventilation or without prior use of invasive mechanical ventilation Comparison: standard oxygen therapy delivered via nasal cannula or face mask
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with standard oxygen therapy	Risk with HFNC
Treatment failure (escalation of respiratory therapy to NIV, NIPPV or invasive ventilation) Measured up to 28 days	Study population		RR 0.62 (0.45 to 0.86)	3044 (15 studies)	⊕⊕⊝⊝ Low^a
	261 per 1000	162 per 1000 (117 to 224)
In‐hospital mortality (up to 90 days; included studies reported in‐hospital mortality, and mortality up to 28 days, up to ICU discharge, and at unspecified time points)	Study population		RR 0.96 (0.82 to 1.11)	2673 (11 studies)	⊕⊕⊕⊝ Moderate^b	‐
	163 per 1000	156 per 1000 (134 to 181)
Adverse events Respiratory infection (pneumonia) Nasal mucosa or skin trauma	Study population for pneumonia		RR 0.72 (0.48 to 1.09)	1057 (4 studies)	⊕⊕⊕⊝ Moderate^c	‐ ‐
	84 per 1000	61 per 1000 (40 to 92)
	Study population for nasal mucosa or skin trauma		RR 3.66 (0.43 to 31.48)	617 (2 studies)	⊕⊝⊝⊝ Verylow^d
	3 per 1000	12 per 1000 (1 to 103)
Length of ICU stay	1.88 days	MD 0.12 days higher (0.03 days lower to 0.27 days higher)	‐	1014 (7 studies)	⊕⊕⊝⊝ Low^e	In addition, 5 studies reported median lengths of ICU stay which we did not combine in analysis; these studies all reported little or no difference in median lengths of ICU stay
Respiratory effects: PaO₂/FiO₂ ratio up to 24 hours after initiation of therapy	188.5 mmHg	MD 10.34 mmHg higher (17.31 mmHg lower to 38 mmHg higher)	‐	600 (5 studies)	⊕⊝⊝⊝ Verylow^f	In addition, 1 study reported median values which we did not combine in analysis; this study reported higher PaO₂/FiO₂ when HFNC was used
Comfort (short‐term effect) Measured up to 24 hours, scales were standardised to allow comparison; higher numbers indicate more comfort	6.81	MD 0.31 higher (0.61 lower to 1.22 higher)	‐	662 (4 studies)	⊕⊝⊝⊝ Verylow^g	In addition, 2 studies reported median values which we did not combine in analysis; 1 of these studies reported little or no difference in comfort according to type of respiratory support used, and 1 study reported improved comfort when HFNC was used
Comfort (long‐term effect) Measured at more than 24 hours, scales were standardized to allow comparison; higher numbers indicate more comfort	7.10	MD 0.59 higher (2.29 lower to 3.47 higher)	‐	445 (2 studies)	⊕⊝⊝⊝ Verylow^g	In addition, 1 study reported data in a figure and we did not combine these data in analysis; this study reported little or no difference in comfort according to the type of respiratory support used
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). For length of stay, PaO₂/FiO₂ and comfort, we present baseline risk values for standard oxygen therapy as the weighted mean values reported in included studies for each outcome. For comfort, these values are scores on a scale from 0 (least comfort) to 10 (most comfort).  CI: confidence interval; HFNC: high‐flow nasal cannulae; ICU: intensive care unit; MD: mean difference; PaO₂/FiO₂: ratio of partial pressure of arterial oxygen to the fraction of inspired oxygen; RR: risk ratio; SMD: standardized mean difference
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: we are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: we have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

High‐flow nasal cannulae compared to NIPPV or NIV for respiratory support in adult intensive care patients
Population: adults in the ICU, requiring respiratory support Setting: ICUs. In this review, these ICUs were in: Belgium, China, France, Saudi Arabia, and Spain Intervention: oxygen delivered via HFNC, initiated after extubation from invasive mechanical ventilation or without prior use of invasive mechanical ventilation Comparison: oxygen delivered via NIV or NIPPV (using BiPAP)
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with NIPPV or NIV	Risk with HFNC
Treatment failure (escalation of respiratory therapy to NIV, NIPPV or invasive ventilation) Measured up to 28 days	Study population		RR 0.98 (0.78 to 1.22)	1758 (5 studies)	⊕⊕⊝⊝ Low^a	We conducted subgroup analysis and found no evidence of a difference in treatment failure when used post‐extubation (RR 1.12, 95% CI 0.89 to 1.41; 3 studies, 1472 participants) and without prior use of mechanical ventilation (RR 0.77, 95% CI 0.58 to 1.03; 2 studies, 286 participants)
	202 per 1000	198 per 1000 (158 to 247)
In‐hospital mortality (up to 90 days; included studies reported in‐hospital mortality, and mortality up to 28 days and up to ICU discharge)	Study population		RR 0.92 (0.64 to 1.31)	1758 (5 studies)	⊕⊕⊝⊝ Low^a	‐
	136 per 1000	126 per 1000 (87 to 179)
Adverse events Respiratory infection (pneumonia)	Study population for pneumonia		RR 0.51 (0.17 to 1.52)	1750 (3 studies)	⊕⊝⊝⊝ Verylow^b	‐
	159 per 1000	81 per 1000 (27 to 241)
Barotrauma (pneumothorax)	Study population for barotrauma		RR 1.15 (0.42 to 3.14)	830 (1 study)	⊕⊝⊝⊝ Low^c	‐
	17 per 1000	19 per 1000 (7 to 53)
Nasal mucosa or skin trauma	Study population for nasal mucosa or skin trauma		‐	‐	‐	No studies reported this outcome
	‐	‐
Length of ICU stay	9.9 days	MD 0.72 days lower (2.85 days lower to 1.42 days higher)	‐	246 (2 studies)	⊕⊕⊝⊝ Low^d	In addition, 2 studies reported median lengths of ICU stay which we did not combine in analysis; these studies reported little or no difference in median lengths of ICU stay
Respiratory effects: PaO₂/FiO₂ ratio up to 24 hours after initiation of therapy	228.9 mmHg	MD 58.1 mmHg lower (71.68 mmHg lower to 44.51 mmHg lower)	‐	1086 (3 studies)	⊕⊕⊝⊝ Low^e	‐
Comfort (short‐term effect) Measured up to 24 hours, scales were standardized to allow comparison; higher numbers indicate more comfort	6.06	MD 1.33 higher (0.74 higher to 1.92 higher)	‐	258 (2 studies)	⊕⊝⊝⊝ Verylow^f	In addition, 1 study reported improved comfort with HFNC (RR 1.30, 95% CI 1.10 to 1.53; 1 study, 168 participants), and 1 study (830 participants) reported little or no difference between types of respiratory support, with comfort rated as 'poor', 'acceptable' or 'good'.
Comfort (long‐term effect) Measured at more than 24 hours	‐	‐	‐	‐	⊕⊝⊝⊝ Verylow^g	1 study (304 participants) reported little or no difference between types of respiratory support, with comfort rated as 'poor', 'acceptable' or 'good'.
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). We present baseline risk values for NIPPV/NIV as the weighted mean values reported in included studies for each outcome. For comfort, these values are a score from 0 (least comfort) to 10 (most comfort).  CI: Confidence interval; HFNC: high‐flow nasal cannulae; ICU: intensive care unit; MD: mean difference; NIPPV: non‐invasive positive pressure ventilation; NIV: non‐invasive ventilation; PaO₂/FiO₂: ratio of partial pressure of arterial oxygen to the fraction of inspired oxygen; RR: risk ratio; SMD: standardized mean difference
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

Important outcomes	HFNC	Standard oxygen therapy	Effect estimate^a	P values^b	Study ID
Length of ICU stay (days)	Median (IQR): 8 (4 to 14)	Median (IQR): 6 (4 to 13)		0.07	Azoulay 2018
Length of ICU stay (days)	Median (IQR): 6 (4 to 16)	Median (IQR): 5 (3 to 13)		0.53	Futier 2016
Length of ICU stay (days)	Median (IQR): 6 (2 to 8)	Median (IQR): 6 (2 to 9)		Not reported	Hernandez 2016b
Length of ICU stay (days)	Median (IQR): 10 (7 to 13)	Median (IQR): 9 (6 to 12)		0.453	Hu 2020
Length of ICU stay (days)	Median (IQR): 1 (1 to 2)	Median (IQR): 1 (1 to 2)		0.949	Zochios 2018
Short‐term oxygenation (PaO₂/FiO₂)	Median (IQR):150 (104 to 230)	Median (IQR):119 (86 to 165)		P value not reported. Study authors described difference as significantly higher in the HFNC group	Azoulay 2018
Short‐term comfort (at 120 minutes) Scale of 0 to 10 (0 = absence of discomfort, 10 = worst possible discomfort)	Median (IQR): 3 (1 to 5)	Median (IQR): 3 (0 to 5)		0.88	Lemiale 2015
Long‐term comfort (at 24 hours) Scale of 0 to 10 (0 = no discomfort, 10 = maximum discomfort)	Median (IQR): 3 (3 to 4.5)	Median (IQR): 7 (6 to 8)		< 0.001	Song 2017
Additional outcomes	HFNC	Standard oxygen therapy	Effect estimate^a	P values^b	Study ID
Duration of respiratory support (hours)	Mean (SD): 59.0 (± 30.8)	Mean (SD): 65.0 (± 41.6)	MD (95% CI) ‐6.00 (‐13.77 to 1.77)	0.13	Parke 2013a
Atelectasis (radiological atelectasis score)	Day 1: median (IQR): 2 (1.5 to 2.5) Day 5: median (IQR): 2 (1.5 to 2.5)	Day 1: median (IQR): 2 (1.5 to 3) Day 5: median (IQR:) 2 (1 to 2.5)		Day 1: 0.70 Day 5: 0.15	Corley 2014
Atelectasis (chest X‐ray)	Day 1: mean (SD): 4.8 (± 1.9) Day 3: mean (SD): 4.8 (± 1.9)	Day 1: mean (SD) 4.9 (± 1.8) Day 3 mean (SD) 4.7 (± 2.1)		Day 1: 0.63 Day 3: 0.69	Parke 2013a
Long‐term PaCO₂ (at 48 hours; mmHg)	Mean (SD): 41.3 (± 7.5)	Mean (SD): 37.2 (± 9.6)	MD 4.10, 95% CI ‐0.43 to 8.63		Hu 2020
Short‐term respiratory rate (at 6 hours; breaths per minute)	Median (IQR): 25 (20 to 30)	Median (IQR): 26 (21 to 31)		Not reported	Azoulay 2018
Long‐term respiratory rate (at 120 minutes; breaths per minute)	Median (IQR): 25 (22 to 29)	Median (IQR) 25 (21 to 31)		Not reported	Lemiale 2015
Length of hospital stay (days)	Median (IQR): 24 (14 to 40)	Median (IQR): 27 (15 to 42)		0.60	Azoulay 2018
Length of hospital stay (days)	Median (IQR): 12 (7 to 20)	Median (IQR): 11 (7 to 18)		0.58	Futier 2016
Length of hospital stay (days)	Median (IQR): 11(6 to 15)	Median (IQR): 12 (6 to 16)		0.76	Hernandez 2016b
Length of hospital stay (days)	Median (IQR): 7 (6 to 9)	Median (IQR): 9 (7 to 6)		0.012	Zochios 2018
Participant‐reported outcomes Dyspnoea Modified Borg scale (0 = no dyspnoea, 10 = maximal dyspnoea	Median (IQR): 1 (0 to 3)	Median (IQR): 0 (0 to 1)		0.008	Corley 2014
Participant‐reported outcomes Dyspnoea Scale of 0 to 10 (0 = absence of dyspnoea, 10 = worst possible dyspnoea)	Median (IQR): 3 (2 to 6)	Median (IQR): 3 (5 to 9)		0.40	Lemiale 2015
Participant‐reported outcomes Dyspnoea Scale of 0 to 10 (0 = no dyspnoea, 10 = maximal dyspnoea). Authors reported proportion of patients with improvement	Mean (SD): 1.6 (1.2)	Mean (SD): 2.9 (1.5)	MD ‐1.3, 95% CI ‐2.60 to 0.00	0.04	Rittayamai 2014
Participant‐reported outcomes Dry mouth Scale of 0 to 10 (0 = no dryness, 10 = maximum dryness)	Mean (SD) 3.6 (2.5)	Mean (SD) 5 (3.1)	MD ‐1.40, 95% CI ‐2.68 to ‐0.12	0.016	Maggiore 2014
Cost comparison of treatment Total hospitalization expenditure, $	Mean (SD): 11522.65 (762.45)	Mean (SD): 12219.73 (1028.66)		0.001	Yu 2017
^acalculated using RevMan Web 2019 ^bas reported by study authors

Additional outcomes	HFNC n/N	Standard oxygen therapy n/N	Effect estimate^a	Study
Atelectasis	2/56	5/54	RR 0.39, 95% CI 0.08 to 1.90	Yu 2017
Adverse events Ventilator‐acquired tracheobronchitis	3/264	7/263	RR 0.43, 95% CI 0.11 to 1.63	Hernandez 2016b
Adverse events Abdominal distension	3/56	0/54	RR 6.75, 95% CI 0.36 to 127.76	Yu 2017
Participant‐reported outcomes Dyspnoea (any improvement; using categorical data reported as marked improvement, slight improvement, no change, slight deterioration, marked deterioration)	65/106	31/94	RR 1.86, 95% CI 1.34 to 2.57	Frat 2015
Participant‐reported outcomes Dry mouth (data included dry mouth, nose, or throat)	18/47	30/43	RR 0.55, 95% CI 0.36 to 0.83	Vourc'h 2020 Vourc'h 2020
Participant‐reported outcomes Throat and nasal pain	1/56	7/54	RR 0.14, 95% CI 0.02 to 1.08	Yu 2017
^acalculated using RevMan Web 2019

Outcome	Study IDs	Effect estimate (short‐term)	Effect estimate (long‐term)	Comment
Duration of respiratory support	Parke 2013a	MD ‐6.00 hours, 95% CI ‐13.77 to 1.77; 1 study, 340 participants; Table 3	‐
Long‐term PaO₂/FiO₂	Maggiore 2014; Vourc'h 2020	‐	MD 27.97, 95% CI 5.60 to 50.33; 2 studies, 195 participants; I² = 81%; Analysis 1.7
Atelectasis	Yu 2017	RR 0.39, 95% CI 0.08 to 1.90; 1 study; 99 participants; Table 4	‐	Additional data available from 2 studies (Corley 2014; Parke 2013a)^a; see Table 3
PaO₂	Frat 2015; Hu 2020; Parke 2011; Maggiore 2014; Song 2017	MD 4.92 mmHg, 95% CI ‐1.24 to 11.07; 4 studies, 415 participants; I² = 47%; Analysis 1.8	MD 12.27 mmHg, 95% CI 7.51 to 17.04; 2 studies, 644 participants; I² = 0%; Analysis 1.8
SpO₂	Maggiore 2014; Parke 2011; Parke 2013a; Rittayamai 2014; Song 2017	MD 0.79 %, 95% CI ‐0.29 to 1.88; 5 studies, 572 participants; I² = 88%; Analysis 1.9	MD 1.28 %, 95% CI 0.02 to 2.55; 2 studies, 445 participants; I² = 81%; Analysis 1.9	Long‐term effect estimate was significant (P = 0.05), however, the high number of comparisons in this review limits our interpretation of this result.
PaCO₂	Frat 2015 Frat 2015; Hernandez 2016b; Hu 2020; Maggiore 2014; Parke 2011; Parke 2013a; Song 2017	MD ‐1.05 mmHg, 95% CI ‐2.24 to ‐0.13; 5 studies, 755 participants; I² = 28%; Analysis 1.10	MD 4.10 mmHg, 95% CI ‐0.43 to 8.63; 1 study, 56 participants; Table 3
Respiratory rate	Corley 2014; Frat 2015; Hu 2020; Maggiore 2014; Parke 2011; Parke 2013a; Rittayamai 2014; Song 2017; Vourc'h 2020	MD ‐2.02 breaths/min, 95% CI ‐3.66 to ‐0.37; 7 studies, 1017 participants; I² = 87%; Analysis 1.11	MD ‐2.01 breaths/min, 95% CI ‐4.39 to 0.37; 4 studies, 591 participants; I² = 92%; Analysis 1.11	Additional data available from 2 studies (Azoulay 2018; Lemiale 2015)^a; see Table 3
Additional adverse events: ventilator‐acquired tracheobronchitis	Hernandez 2016b	RR 0.43, 95% CI 0.11 to 1.63; 1 study, 527 participants; Table 4	‐
Additional adverse events: abdominal distension	Yu 2017	RR 6.75, 95% CI 0.36 to 127.76; 1 study, 110 participants; Table 4	‐
Length of hospital stay	Brainard 2017; Parke 2013a ; Yu 2017	MD ‐0.32 days, 95% CI ‐1.32 to 0.68; 3 studies, 494 participants; I² = 47%; Analysis 1.12	‐	Additional data available from 4 studies (Azoulay 2018; Futier 2016; Hernandez 2016b; Zochios 2018)^ab; see Table 3.
Other participant‐reported outcomes Dyspnoea	Frat 2015; Rittayamai 2014	MD ‐1.30, 95% CI ‐2.60 to 0.00; 1 study, 17 participants; Table 3 RR 1.86, 95% CI 1.34 to 2.57; 1 study, 200 participant; Table 4		Additional data available from 3 studies (Corley 2014; Lemiale 2015; Rittayamai 2014)^a; see Table 3. Azoulay 2018 data reported in figures from which numerical data could not be extracted. Study authors reported no significant difference between groups.
Other participant‐reported outcomes Dry mouth	Maggiore 2014; Vourc'h 2020	RR 0.55, 95% CI 0.36 to 0.83; 1 study, 90 participants; Table 4 MD ‐1.40, 95% CI ‐2.68 to ‐0.12; 1 study, 80 participants; Table 3	‐	Additional data available from Maggiore 2014 reported in Table 3. Additional data from Vourc'h 2020 reported in Table 4. Cuquemelle 2012 effect size was not reported but the authors stated there was no evidence of a difference between groups.
Other participant‐reported outcomes Throat or nasal pain	Yu 2017	RR 0.14, 95% CI 0.02 to 1.08; 1 study, 110 participants; Table 4	‐
Other participant‐reported outcomes Treatment withdrawn due to discomfort	Futier 2016	RR 17.62, 95% CI 1.03 to 301.65; 1 study, 220 participants; Table 4	‐
Other participant‐reported outcomes Refusal to continue treatment	Futier 2016; Parke 2013a	RR 26.89, 95% CI 3.67 to 197.32; 2 studies, 560 participants; Analysis 1.13	‐	Azoulay 2018 reported participant discontinuation in HFNC group due to discomfort, but it was unclear whether any participants in the control group discontinued due to discomfort.
Cost comparison of treatment	Yu 2017	‐	‐	Mean costs reported for HFNC group only. See Table 3
^aWe did not combine data from these studies in analyses because data were reported as median values, or did not include relevant distribution variables for meta‐analysis with other studies ^bFrom visual inspection, we noted that these data were likely to be right skewed due to the comparable magnitudes of the mean and standard deviation. This is expected for outcomes such a length of hospital stay due to most participants being discharged in a short time period with some outliers staying significantly longer. However, right skew introduces artefact into calculation of the effect estimate, limiting the interpretation of the result.

Risk of selection: studies excluded from primary analysis owing to high or unclear risk of selection bias for random sequence generation or allocation concealment
Important outcomes	Excluded studies	Effect of sensitivity analysis
Failure of treatment	Frat 2015; Hu 2020; Lemiale 2015; Maggiore 2014; Song 2017; Yu 2017	Effect estimate no longer indicated improvement with HFNC use (RR 0.85, 95% CI 0.62 to 1.17; 9 studies, 2457 participants; I² = 55%)
In‐hospital mortality	Frat 2015; Hu 2020; Maggiore 2014; Yu 2017	Interpretation of the effect estimate remained the same
Important adverse events: pneumonia	Frat 2015; Yu 2017	Interpretation of the effect estimate remained the same
Important adverse events: nasal mucosa or skin trauma	‐	‐
Length of ICU stay	Brainard 2017; Frat 2015; Maggiore 2014; Yu 2017	Interpretation of the effect estimate remained the same
PaO₂/FiO₂ up to 24 hours	Frat 2015; Maggiore 2014; Parke 2011	Effect estimate indicated higher PaO₂/FiO₂ when standard oxygen therapy was used (MD 25.28 mmHg, 95% CI 7.23 to 43.32; 2 studies, 245 participants; I² = 0%)
Comfort (short‐term)	Frat 2015; Maggiore 2014; Rittayamai 2014	Interpretation of the effect estimate remained the same
Comfort (long‐term)	Maggiore 2014;	Interpretation of the effect estimate remained the same
High risks of other bias: studies excluded from primary analysis owing to high risks of other bias
Outcome	Excluded studies	Effect of sensitivity analysis
Failure of treatment	Fernandez 2017; Hu 2020; Parke 2011; Parke 2013a; Zochios 2018 (selective reporting bias) Frat 2015 (selective reporting bias, and differences in treatment in the HFNC group)	Interpretation of the effect estimate remained the same
In‐hospital mortality	Fernandez 2017; Frat 2015; Hu 2020; Parke 2013a; Zochios 2018 (selective reporting bias) Frat 2015 (differences in treatment in the HFNC group)	Interpretation of the effect estimate remained the same
Important adverse events: pneumonia	Frat 2015 (selective reporting bias, and differences in treatment in the HFNC group)	Interpretation of the effect estimate remained the same
Important adverse events: nasal mucosa or skin trauma	‐	‐
Length of ICU stay	Brainard 2017 (attrition bias) Frat 2015 (selective reporting bias, and differences in treatment in the HFNC group) Parke 2013a (selective reporting bias)	Interpretation of the effect estimate remained the same
PaO₂/FiO₂ up to 24 hours	Frat 2015 (selective reporting bias, and differences in treatment in the HFNC group) Parke 2011 (selective reporting bias)	Effect estimate indicated higher PaO₂/FiO₂ when standard oxygen therapy was used (MD 29.28 mmHg, 95% CI 13.86 to 44.70; 3 studies, 350 participants; I² = 0%)
Comfort (short‐term)	Frat 2015 (selective reporting bias, and differences in treatment in the HFNC group) Parke 2013a (selective reporting bias)	Interpretation of the effect estimate remained the same
Comfort (long‐term)	Parke 2013a (selective reporting bias)	Effect estimate indicated improved comfort when HFNC was used (MD ‐2.10, 95% CI ‐3.16 to ‐1.04; 1 study, 105 participants)
Fixed effect versus random effects: we re‐analysed the data using a fixed‐effect model
Outcomes	Effect of sensitivity analysis
Failure of treatment In‐hospital mortality Important adverse events: pneumonia Important adverse events: nasal mucosa or skin trauma Length of ICU stay PaO₂/FiO₂ up to 24 hours Comfort (short‐term) Comfort (long‐term)	Interpretation of the effect estimate for all outcomes remained the same
Funding: studies excluded from analysis in which funding was from commercial sources
Outcome	Excluded studies	Effect of sensitivity analysis
Failure of treatment	Azoulay 2018; Corley 2014; Frat 2015; Hernandez 2016b; Lemiale 2015; Maggiore 2014; Parke 2011; Parke 2013a; Zochios 2018	Interpretation of the effect estimate remained the same
In‐hospital mortality	Azoulay 2018; Hernandez 2016b; Maggiore 2014; Parke 2013a; Zochios 2018	Interpretation of the effect estimate remained the same
Important adverse events: pneumonia	Frat 2015; Hernandez 2016b	Interpretation of the effect estimate remained the same
Important adverse events: nasal mucosa or skin trauma	Hernandez 2016b	Interpretation of the effect estimate remained the same
Length of ICU stay	Corley 2014; Frat 2015; Maggiore 2014; Parke 2013a	Interpretation of the effect estimate remained the same
PaO₂/FiO₂ up to 24 hours	Corley 2014; Frat 2015; Maggiore 2014; Parke 2013a	Interpretation of the effect estimate remained the same (only one study remaining in analysis)
Comfort (short‐term)	Frat 2015; Maggiore 2014; Parke 2013a	Interpretation of the effect estimate remained the same (only one study remaining in analysis)
Comfort (long‐term)	Maggiore 2014; Parke 2013a	No studies remaining in analysis

Important outcomes	HFNC n/N	NIV or NIPPV n/N	Effect estimate^a	Study ID
Participant‐reported outcomes Comfort	74/84	57/84	RR 1.30, 95% CI 1.10 to 1.53	Cong 2019
Adverse events Pneumothorax	8/414	7/416	RR 1.15, 95% CI 0.42 to 3.14	Stephan 2015
Additional outcomes	HFNC n/N	NIV or NIPPV n/N	Effect estimate^a	Study ID
Adverse events Ventilator‐acquired tracheobronchitis	11/290	18/314	RR 0.66, 95% CI 0.32 to 1.38	Hernandez 2016b
^acalculated using RevMan Web 2019

Important outcomes	HFNC	NIV or NIPPV	Effect estimate^a	P value^b	Study ID
Length of ICU stay (days)	Median (IQR): 9 (4 to 19)	Median (IQR): 10.5 (5 to 19)		Not reported	Hernandez 2016a
Length of ICU stay (days)	Median (IQR) 6 (4 to 10)	Median (IQR) 6 (4 to 10)		0.77	Stephan 2015
Short‐term comfort (1 hour) 5‐point scale of 'poor', 'acceptable', or 'good'	Poor: 16.7% Acceptable: 31.0% Good: 51.0%	Poor: 17.8% Acceptable: 29.3% Good: 53.0%		0.32	Stephan 2015
Long‐term comfort (day 3) 5‐point scale of 'poor', 'acceptable', or 'good'	Poor: 21% Acceptable: 32.4% Good: 47%	Poor: 21% Acceptable: 31% Good: 48.3%		> 0.99	Stephan 2015
Additional outcomes	HFNC	NIV or NIPPV	Effect estimate^a	P value^b	Study ID
Long‐term PaO₂ (mmHg)	Mean (SD): 81.87 (15.27)	Mean (SD): 82.22 (15.64)	MD ‐0.35, 95% CI ‐5.02 to 4.32		Cong 2019
Long‐term SpO₂ (%)	Mean (SD): 87.83 (8.16%)	Mean (SD): 88.65 (7.15)	MD ‐0.82, 95% CI ‐3.14 to 1.50		Cong 2019
Long‐term SpO₂ (%)	Mean (SD): 91.93 (4.35)	Mean (SD): 92.75 (4.07)	MD ‐0.82, 95% CI ‐2.09 to 0.45		Cong 2019
Short‐term PaCO₂ (mmHg) (6 to 12 hours)	Mean (95% CI) 38.2 (37.6 to 38.9)	Mean (95% CI) 39.3 (38.6 to 40.0)		0.19	Stephan 2015
Long‐term PaCO₂ (mmHg)	Mean (SD) 81.87 (15.27)	Mean (SD) 82.22 (15.64)	MD ‐0.35, 95% CI ‐5.02 to 4.32		Cong 2019
Long‐term respiratory rate (breaths/min)	Mean (SD) 22.4 (4.4)	Mean (SD) 21 (4.5)	MD 1.40, 95% CI ‐1.36 to 4.16		Jing 2019
Length of hospital stay (days)	Median (IQR): 23 (14 to 46)	Median (IQR): 26 (16 to 37)		Not reported	Hernandez 2016a
Length of hospital stay (days)	Median (IQR) 13 (9 to 22)	Median (IQR) 14 (9 to 20)		0.59	Stephan 2015
Length of hospital stay (days)	Mean (SD): 18.04 (6.15)	Mean (SD): 18.31	MD ‐0.27 days, 95% CI ‐2.26 to 1.72		Cong 2019
^acalculated using RevMan Web 2019 ^bas reported by study authors

Date	Event	Description
16 June 2020	New citation required and conclusions have changed	The review findings were changed as follows with this update: We found that treatment failure may be reduced with HFNC compared to standard oxygen but there remained no evidence of a difference when HFNC was compared to NIV/NIPPV We increased the certainty of the evidence for mortality to moderate when HFNC was compared to standard oxygen therapy We continued to find low or very low‐certainty evidence for all other important outcomes across both comparisons. Additional evidence is still required to increase certainty in the findings
13 June 2020	New search has been performed	We updated the review and made the following amendments. We added one new author to the review (PB) We re‐ran the searches using new updated search strategies We found 20 new studies and incorporated data from these studies into the review. In addition, we found 19 studies awaiting classification and 51 ongoing studies We separated the findings into two comparison groups (standard oxygen therapy, and NIV or NIPPV). We included a 'Summary of findings' table for each of these comparisons
20 December 2018	Amended	Editorial team changed to Cochrane Emergency and Critical Care

Completed by
Date
Study ID
Methods	Study design: Multicentre or single‐centre: Country: Study aim:
Participants	Total number of participants: Setting (type of ICU; country): Inclusion criteria: Exclusion criteria: Study aim (specify whether intervention is given for respiratory failure, post‐extubation etc.) Baseline characteristics Intervention group Age, (years) mean (SD): BMI (kg/m²) mean (SD): Respiratory rate (breaths/min) mean (SD): PaCO₂ (mmHg) mean (SD): PaO₂/FiO₂ (mmHg) mean (SD): Illness severity score (such as APACHE II) mean (SD): Comparison group Age, (years) mean (SD): BMI (kg/m²) mean (SD): Respiratory rate (breaths/min) mean (SD): PaCO₂ (mmHg) mean (SD): PaO₂/FiO₂ (mmHg) mean (SD): Illness severity score (such as APACHE II) mean (SD):
Interventions	Intervention group Randomized, n = ; losses (with reasons) = ; analysed, n = Details (include type of device; size of nasal cannulae; flow rate; duration; washout period in cross‐over studies): Comparison group Randomized, n = ; losses (with reasons) = ; analysed, n = Details (include type of device; flow rate; duration; washout period in cross‐over studies):
Outcomes	Outcomes measured/reported by study authors: Outcomes relevant to the review:
Outcome data	If study is a cross‐over design, are the data reported separately for the first cross‐over period
Notes	Funding/declarations of interest: Study dates: Additional notes:

Name of outcome:
Time point of measurement:
Intervention group
Number of events	Total number of participants in the group

Control group
Number of events	Total number of participants in the group

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Treatment failure (escalation of respiratory support to NIV, NIPPV or invasive ventilation)	5	1758	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.78, 1.22]
2.1.1 Post‐extubation respiratory support	3	1472	Risk Ratio (M‐H, Random, 95% CI)	1.12 [0.89, 1.41]
2.1.2 Respiratory support without prior use of mechanical ventilation	2	286	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.58, 1.03]
2.2 In‐hospital mortality	5	1758	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.64, 1.31]
2.3 Important adverse events: pneumonia	3	1750	Risk Ratio (M‐H, Random, 95% CI)	0.51 [0.17, 1.52]
2.4 Short‐term respiratory effects: PaO ₂/FiO ₂ (mmHg)	3	1086	Mean Difference (IV, Random, 95% CI)	‐58.10 [‐71.68, ‐44.51]
2.5 Length of ICU stay (days)	2	246	Mean Difference (IV, Random, 95% CI)	‐0.72 [‐2.85, 1.42]
2.6 Short‐term comfort (continuous data)	2	258	Mean Difference (IV, Random, 95% CI)	1.33 [0.74, 1.92]
2.7 Duration of respiratory support (hours)	2	210	Mean Difference (IV, Random, 95% CI)	‐6.12 [‐54.61, 42.37]
2.8 Long‐term respiratory effects: PaO ₂/FiO ₂ (mmHg)	2	344	Mean Difference (IV, Random, 95% CI)	‐31.67 [‐49.37, ‐13.97]
2.9 Short‐term respiratory effects: PaO ₂ (mmHg)	2	384	Mean Difference (IV, Random, 95% CI)	‐9.57 [‐30.25, 11.11]
2.10 Short‐term and long‐term respiratory effects: PaCO ₂ (mmHg)	4		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.10.1 Short‐term effects	4	1254	Mean Difference (IV, Random, 95% CI)	‐0.46 [‐2.08, 1.16]
2.10.2 Long‐term effects	2	208	Mean Difference (IV, Random, 95% CI)	‐1.80 [‐5.57, 1.98]
2.11 Short‐term respiratory effects: breaths/min	4	1090	Mean Difference (IV, Random, 95% CI)	‐1.06 [‐1.80, ‐0.32]
2.12 Dyspnoea (any improvement)	2	1023	Risk Ratio (M‐H, Random, 95% CI)	1.05 [0.74, 1.48]

Name of outcome:
Intervention group
Mean	SD	Total number of participants in the group

Control group
Mean	SD	Total number of participants in the group

Domain	High/Low/ Unclear	Judgement
Random sequence generation (selection bias)
Allocation concealment (selection bias)
Blinding of participants and personnel (performance bias)
Blinding of outcome assessors (detection bias)
Incomplete outcome data (attrition bias)
Selective reporting (reporting bias)
Other bias

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Treatment failure (escalation of respiratory support to NIV, NIPPV or invasive ventilation)	15	3044	Risk Ratio (M‐H, Random, 95% CI)	0.62 [0.45, 0.86]
1.1.1 Post‐extubation respiratory support	11	1912	Risk Ratio (M‐H, Random, 95% CI)	0.50 [0.30, 0.86]
1.1.2 Respiratory support without prior use of mechanical ventilation	4	1132	Risk Ratio (M‐H, Random, 95% CI)	0.85 [0.68, 1.08]
1.2 In‐hospital mortality	11	2673	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.82, 1.11]
1.3 Important adverse events	5		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
1.3.1 Pneumonia	4	1057	Risk Ratio (M‐H, Random, 95% CI)	0.72 [0.48, 1.09]
1.3.2 Nasal mucosa or skin trauma	2	617	Risk Ratio (M‐H, Random, 95% CI)	3.66 [0.43, 31.48]
1.4 Length of ICU stay (days)	6	970	Mean Difference (IV, Random, 95% CI)	0.13 [‐0.02, 0.28]
1.5 Short‐term respiratory effects: PaO ₂/FiO ₂ (mmHg)	5	600	Mean Difference (IV, Random, 95% CI)	10.34 [‐17.31, 38.00]
1.6 Comfort	4		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.6.1 Short‐term effect	4	662	Mean Difference (IV, Random, 95% CI)	0.31 [‐0.60, 1.22]
1.6.2 Long‐term effect	2	445	Mean Difference (IV, Random, 95% CI)	0.59 [‐2.29, 3.47]
1.7 Long‐term respiratory effects: PaO ₂/FiO ₂ (mmHg)	2	195	Mean Difference (IV, Random, 95% CI)	34.28 [‐19.25, 87.80]
1.8 Short‐term and long‐term respiratory effects: PaO ₂ (mmHg)	5		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.8.1 Short‐term effects	4	415	Mean Difference (IV, Random, 95% CI)	4.92 [‐1.24, 11.07]
1.8.2 Long‐term effects	2	644	Mean Difference (IV, Random, 95% CI)	12.27 [7.51, 17.04]
1.9 Short‐term and long‐term respiratory effects: SpO ₂ (%)	5		Mean Difference (IV, Random, 95% CI)	Subtotals only
1.9.1 Short‐term effects	5	572	Mean Difference (IV, Random, 95% CI)	0.79 [‐0.29, 1.88]
1.9.2 Long‐term effects	2	445	Mean Difference (IV, Random, 95% CI)	1.28 [0.02, 2.55]
1.10 Short‐term respiratory effects: PaCO ₂ (mmHg)	5	755	Mean Difference (IV, Random, 95% CI)	‐1.05 [‐2.24, 0.13]
1.11 Short‐term and long‐term respiratory rate (breaths/min)	9	1608	Mean Difference (IV, Random, 95% CI)	‐2.01 [‐3.19, ‐0.83]
1.11.1 Short‐term effects	8	1017	Mean Difference (IV, Random, 95% CI)	‐2.02 [‐3.66, ‐0.37]
1.11.2 Long‐term effects	4	591	Mean Difference (IV, Random, 95% CI)	‐2.01 [‐4.39, 0.37]
1.12 Length of hospital stay (days)	2	450	Mean Difference (IV, Random, 95% CI)	‐0.11 [‐0.43, 0.20]
1.13 Refusal to continue with treatment	2	560	Risk Ratio (M‐H, Random, 95% CI)	26.89 [3.67, 197.32]

*Study characteristics*
Methods	RCT, parallel‐group design. Multicentre study
Participants	Total number of randomized participants: 778 Setting: 32 ICUs; France Inclusion criteria: ICU admission; ≥ 18 years of age; AHRF with PaO₂ < 60 mmHg or SpO₂ < 90% on room air, or tachypnoea > 30 breaths/min or laboured breathing or respiratory distress; need for oxygen flow of ≥ 6 L/min; known immunosuppression; written informed consent Exclusion criteria: people with AIDS; imminent death; refusal to participate in study; anatomical factors precluding use of nasal cannula; hypercapnia indicting NIV; isolated cardiogenic pulmonary oedema indicating NIV; pregnancy or breastfeeding; absence of health insurance coverage; surgery within the last 6 days Baseline characteristics: Intervention group (HFNC): Age, median (IQR): 64 (55 to 70) years Gender, M/F: 270/118 BMI, mean (SD): not reported SOFA, median (IQR) : 6 (4 to 8) SAPS II, median (IQR): 36 (28 to 46) PaCO₂, mean (SD): not reported PaO₂/FiO₂, median (IQR): 136 (96 to 187) mmHg Respiratory rate, median (IQR): 33 (28 to 39) breaths/min Control group (standard oxygen therapy): Age, median (IQR): 63 (56 to 71) years Gender, M/F: 247/141 BMI, mean (SD): not reported SOFA, median (IQR): 6 (4 to 8) SAPS II, median (IQR): 37 (28 to 48) PaCO₂, mean (SD): not reported PaO₂/FiO₂, median (IQR): 128 (92 to 164) mmHg Respiratory rate, median (IQR): 32 (27 to 38) breaths/min
Interventions	Intervention group: Randomized, n = 389; losses, n = 1 (withdrew consent), 13 did not receive intervention as randomized; analysed, n = 388 Details: started within 15 minutes of randomization, and for whole duration of ICU stay. Flow initiated at 50 L/min and 100% FiO₂, then subsequent flow to achieve SpO₂ ≥ 95% up to ≥ 50 L/min within the first 3 days then up to 60 L/min as needed. If participants needed MV, HFNC was used during laryngoscopy and immediately after extubation. Standard oxygen therapy was only used if significant nasal discomfort or skin breakdown. Control group (standard oxygen): Randomized, n = 389 ; losses, n = 1 (withdrew consent), 31 did not receive intervention as randomized, 30 received HFNC; analysed, n = 388 Details: started within 15 minutes of randomization, and for whole duration of ICU stay. Oxygen given by any device or combination of devices (nasal prongs or mask with or without reservoir bag and with or without Venturi system). Flow to achieve target SpO₂ ≥ 95%. HFNC only given if participants had a do‐not‐intubate order or for whom standard oxygen had failed. NIV only used as long as hypercapnia or pulmonary oedema were present.
Outcomes	Mortality within 28 days; hospital and ICU mortality; number needing MV by day 28; respiratory rate (normal values, 12‐20), lowest PaO₂/FiO₂, patient comfort score (range 0 to 10 = severe discomfort to perfect comfort); dyspnoea score (range 0 to 10 = no dyspnoea to severe dyspnoea); ICU and hospital lengths of stay; incidence of ICU‐acquired infections
Notes	Funding/declarations of interest: funded by the French Ministry of Health. Supplies for high‐flow oxygen from Fisher & Paykel Healthcare Ltd. Funders had no role in the design and conduct of the study, nor in preparation of the manuscript etc. Some authors received fees from one or more of: Gilead; Astellas; Baxter; Alexion; Ablynx; Merk Sharp and Dohme, Fisher & Paykel Healthcare, Xenios; Boehringer Ingelheim; Pfizer; Astute; Bristol‐Myers Squibb; Jazz Pharma; Sanofi‐Aventis; Resmed; Philips; Hamilton; Medtronic; French Ministry of Health. One author serves on a data and safety monitoring board for the French Ministry of Health. Study dates: 19 May 2016 to 31 December 2017
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Use of an electronic system
Allocation concealment (selection bias)	Low risk	Quote: "Randomization was achieved using an electronic system incorporated in the electronic case report from to ensure allocation concealment".
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinding was not possible. Although we expected that this would not influence outcome data, we could not be certain of this.
Blinding of outcome assessors (objective outcomes)	Low risk	Quote: "No blinding of adjudication was performed for outcome assessments". The assessors were unblinded; we did not anticipate that this would influence the assessment of objective outcome measures.
Blinding of outcome assessors (subjective measures)	Low risk	Participants were the outcome assessors for comfort and dyspnoea on a standardized scale: we did not anticipate that this would influence the assessment of these outcome measures.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Loss of only two participants
Selective reporting (reporting bias)	Low risk	Study was prospectively registered with a clinical trials register (NCT02739451). Outcomes relevant to the review were reported as described in the clinical trials register.
Other bias	Low risk	We identified no other sources of bias.

*Study characteristics*
Methods	RCT, parallel‐group design. Single‐centre study
Participants	Total number of randomized participants: 51 Setting: ICU, USA Inclusion criteria: ≥ 18 years of age; undergoing thoracic surgery with scheduled admission to the ICU postoperatively Exclusion criteria: < 18 years of age; pregnant or breastfeeding; known diagnosis of obstructive sleep apnoea; current or pervious lung transplantation; previous pneumonectomy; home oxygen > 4 L/min; inability to adhere to assigned treatment for the intended duration Baseline characteristics (for those who continued treatment): Intervention group (HFNC): Age, mean (SD): 57 (± 14) years Gender, M/F: 8/10 BMI, mean (SD): 26 (± 5) kg/m² ASA II/III/IV: 5/12/1 SAPS II, mean (SD): 19 (± 7) PaCO₂, mean (SD): not reported PaO₂/FiO₂, mean (SD): not reported Control group (standard oxygen): Age, mean (SD): 59 (± 16) years Gender, M/F: 14/12 BMI, mean (SD): 25 (± 5) kg/m² ASA II/III/IV: 1/19/1 SAPS II, mean (SD): 23 (± 7) PaCO₂, mean (SD): not reported PaO₂/FiO₂, mean (SD): not reported
Interventions	Intervention group (HFNC): Randomized, n = 25; losses, n = 7 (discontinued due to discomfort); analysed, n = 18 (we included 7 lost participants as outcome data for comfort) Details: HFNC using MaxVenturi, started after transfer to ICU, following surgery, set flow of 40 L/min, FiO₂ titrated by respiratory therapists to maintain SpO₂ ≥ 90%. Therapy continued for 48 hours or until transfer from the ICU to a ward Control group (standard oxygen): Randomized, n = 26; losses, n = 0; analysed, n = 26 Details: standard oxygen given via nasal cannula or face mask titrated by nurses as required to maintain SpO₂ ≥ 90%. Therapy continued for 48 hours or until transfer from the ICU to a ward
Outcomes	Composite of postoperative pulmonary outcomes (severe hypoxaemia, acute respiratory failure, escalation of therapy to non‐invasive ventilation, re‐intubation, occurrence of hospital‐acquired pneumonia); ICU and hospital lengths of stay; postoperative oxygenation Note: study authors did not separately report data for the each event in the primary outcome.
Notes	Funding/declarations of interest: one study author was supported by the NIS/National Institute on Drug Abuse. Study authors declared no conflicts of interest. Study dates: August 2013 to June 2015

*Study characteristics*
Methods	RCT, cross‐over study. Single‐centre study
Participants	Total number of participants: 10 Setting: medical‐surgical ICU; Montpelier, France Inclusion criteria: ≥ 18 years old hospitalized in a medical‐surgical ICU, planned for tracheostomy tube removal which was placed in the ICU for weaning from mechanical ventilation Exclusion criteria: pregnancy, adult under tutelage, contraindications for NIV Baseline characteristics (all patients): Age: 54 to 66 years Respiratory rate, median (IQR): 18 (22 to 20) breaths/min PaCO₂: not reported PaO₂/FiO₂: not reported
Interventions	Flow rates of 15, 30, and 45 litres per minute were tested in a randomized order for each device. High‐flow face mask with a reservoir bag Optiflow high‐flow nasal cannulae Boussignac oxygen therapy system& For each device and flow rate, participants were asked to have their mouth open and mouth closed in a randomized order. Each device was used for 5 minutes, with 15‐minute washout between treatments.
Outcomes	Tracheal pressure, FiO₂ delivered, respiratory discomfort, respiratory rate (at end of each treatment period), noise intensity
Notes	Funding sources/declarations of interest: Study authors disclosed funding of €3000 from Fisher & Paykel Healthcare Ltd, France, which was used to acquire technical equipment and clinical research insurance, and to present results at scientific meetings. Study dates: not reported
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Method of randomization not stated
Allocation concealment (selection bias)	Unclear risk	Method of allocation concealment not stated
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinding was not possible. Although we expected that this would not influence outcome data, we could not be certain of this.
Blinding of outcome assessors (objective outcomes)	Low risk	Investigators were outcome assessors for objective outcomes, but standardized tools were used for measurement, reducing risk of bias.
Blinding of outcome assessors (subjective measures)	Low risk	Participants were outcome assessors for respiratory and auditory discomfort on a standardized scale. We did not think this would influence the subjective outcome data.
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	One participant was excluded owing to major intolerance to the device but possibly should have been regarded as a treatment failure. In such a small study, this is likely to have had an effect. Owing to inability of 4 participants in the Boussignac group to adhere to the protocol, it is likely that data were incomplete; however it was not mentioned how this was handled in the analysis.
Selective reporting (reporting bias)	Unclear risk	ISRCTN15995925. Retrospectively registered in August 2012. Not possible to establish any reporting bias through comparison with the trial register protocol
Other bias	Low risk	We identified no other risks of bias.

*Study characteristics*
Methods	RCT, cross‐over design. Single‐centre study
Participants	Total number of randomized participants: 15 Setting: ICU; Italy Inclusion criteria: adults with AHRF Exclusion criteria: exacerbation of asthma or COPD; clinical evidence of cardiogenic pulmonary oedema; acute respiratory failure occurring within 1 week after surgery; haemodynamic instability and/or shock; metabolic acidosis; GCS < 13; facial anatomy contraindicating helmet or nasal cannula application Baseline characteristics: Overall: Age, mean (SD): 70 (64 to 77) years Gender, M/F: 9/6 BMI, mean (SD): 168 (165 to 175) SAPS II, mean (SD): 28 (24 to 29) SOFA, median (IQR): 45 (36 to 69) PaCO₂, mean (SD): 32 (30 to 34) mmHg PaO₂/FiO₂, mean (SD): 133 (92 to 154) mmHg
Interventions	Cross‐over study, with each phase lasting 60 minutes HFNC: using AIRVO₂ device, Fisher & Paykel Healthcare, with a heated humidifier (MR860). Gas flow set at 50 L/min, and humidifier temperature at 37 ºC. FiO₂ titrated to obtain SpO₂ ≥ 92% and ≤ 98% Helmet NIV: through bi‐tube circuit with no humidification. Initial pressure support was 8 to 10 cm H₂O, adjusted for peak inspiratory flow of 100 to 150 L/min, up to a maximum of 20 cm H₂O; PEEP at 10 to 12 cm H₂O; flow trigger was 2 L/min and increased in presence of auto‐triggering. FiO₂ titrated to obtain SpO₂ ≥ 92% and ≤ 98% Between interventions, there was a 15‐minute washout period with heated and humidified (MR860; Fisher & Paykel Healthcare) oxygen therapy at a flow rate of 50 L/min via a non‐rebreathing face mask (temperature of the humidification chamber set at 37° C, FiO₂ set to achieve a SpO₂ > 92% and < 98%)
Outcomes	PaO₂/FiO₂; PaCO₂; respiratory rate; inspiratory effort; work of breathing; comfort; dyspnoea; end‐inspiratory and end‐expiratory transpulmonary pressure Note: we did not include outcome data in the review because the study authors did not report outcome data from the first study period.
Notes	Funding/declarations of interest: research grant from Società Italiana di Anestesia Analgesia Rianimazione e Terapia Intensiva; and Merck Sharp and Dohme Study dates: May 2017 and December 2018

*Study characteristics*
Methods	RCT, parallel‐group design, multicentre study
Participants	Total number of participants: 102 Setting: 4 ICUs; France Inclusion criteria: consecutive immunocompromised patients admitted to ICU for acute respiratory failure, aged > 18 years Exclusion criteria: hypercapnia (> 45 mmHg), mechanical ventilation before ICU admission, need for immediate NIV or invasive mechanical ventilation, and patient refusal to participate in study Baseline characteristics: Intervention group (HFNC): Age, median (IQR): 59.3 (43 to 70) years Gender, M/F: 38/14 BMI mean (SD): not reported SAPS II, median (IQR): 42 (29.5 to 52) SOFA, median (IQR): 3.5 (2 to 6) PaCO₂: not reported PaO₂/FiO₂, median (IQR): 128 (48 to 178) mmHg Respiratory rate, median (IQR): 26 (21.7 to 31.2) breaths/min Control group (standard oxygen therapy): Age, median (IQR): 64.5 (53.25 to 72) years Gender, M/F: 32/16 BMI mean (SD): not reported SAPS II, median (IQR): 37.5 (31.5 to 46.5) SOFA, median (IQR): 3 (2 to 5) PaCO₂: not reported PaO₂/FiO₂ median (IQR): 100 (40 to 156) mmHg Respiratory rate median (IQR): 27 (22 to 32.2) breaths/min
Interventions	Intervention group: Randomized, n = 53; losses, n = 1 (1 withdrew consent); analysed, n = 52 Details: HFNC; heated, humidified circuit, with initial flow of 40 to 50 L/min; FiO₂ 100%, which was then adjusted to maintain SpO₂ ≥ 95 % Control group Randomized, n = 49; losses, n = 1 (1 withdrew consent); analysed, n = 48 Details: Venturi mask; FiO₂ initially 60%, 15 L/min, then adjusted to maintain SpO₂ ≥ 95% Participants were randomly allocated to oxygen therapy groups for a 2‐hour period.
Outcomes	Need for invasive mechanical ventilation or NIV during or at the end of the 2‐hour study period; VAS scores for comfort, thirst, and dyspnoea (all at 120 minutes); respiratory rate (at 120 minutes); heart rate
Notes	Funding/declarations of interest: Fisher & Paykel Healthcare Ltd provided oxygen delivery devices and funds for study insurance and presentation of results. The sponsors had no role in designing or conducting the study. Study dates: November 2012 to April 2014 Note: we noted some differences in baseline characteristics but these were not clinically significant.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Participants described as randomly allocated, with stratification on study centre by permuted block method
Allocation concealment (selection bias)	Low risk	Use of opaque, sealed envelopes to ensure identity concealment
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Blinding was not possible. Although we expected that this would not influence outcome data, we could not be certain of this.
Blinding of outcome assessors (objective outcomes)	Low risk	Blinding of outcome assessors was not described; we did not anticipate that this would influence the assessment of objective outcome measures.
Blinding of outcome assessors (subjective measures)	Low risk	Standardized approach to measuring subjective outcomes of comfort, thirst and dyspnoea: we did not anticipate that this would influence the assessment of these outcomes.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Loss of two participants after randomization due to withdrawal of consent. Low number, unlikely to influence results
Selective reporting (reporting bias)	Unclear risk	NCT02424773. Retrospective registration in April 2015. Therefore, not feasible to judge if any reporting bias. All outcomes reported from methods section
Other bias	Low risk	We identified no other risks of bias.

*Study characteristics*
Methods	RCT, cross‐over design. Multicentre study
Participants	Total number of randomized participants: 32 Setting: 2 ICUs; Italy Inclusion criteria: COPD, NIV > 24 hours; fully co‐operative; pH ≥ 7.35 during NIV; respiratory rate ≤ 30 breaths/min; improvement of condition (no dyspnoea, no agitation, no fever) Exclusion criteria: diaphragm paralysis; clinical signs of distress or impending respiratory muscle failure; haemodynamic instability; life‐threatening cardiac arrhythmia; ECG signs of ischaemia; impaired renal function; inclusion in other studies; refusal to consent Baseline characteristics (overall, for analysed participants only): Age, mean (SD): 72.5 (± 8.2) years Gender, M/F: 17/13 BMI, mean (SD): not reported SAPS II, mean (SD) : 31.5 (± 6.2) PaCO₂, mean (SD): not reported PaO₂/FiO₂, mean (SD): not reported
Interventions	Cross‐over study design, with 30 minutes for each 5 intervention stages. Three intervention stages, which were all NIV, were not randomized. Two randomized interventions were given as interruptions to the NIV stages: HFNO delivered using Optiflow via nasal cannula connected to a heated humidifier (MR850) Standard oxygen treatment using Venturi mask Randomized participants, n = 32; losses, n = 2 (insufficient ultrasound imaging quality); analysed, n = 30
Outcomes	Evaluation of right hemidiaphragm; respiratory rate; pH; PaO₂; PaCO₂; comfort (using 11‐point NRS) Note: we did not include outcome data in the review because study authors did not report outcome data from the first period.
Notes	Funding/declarations of interest: some authors received fees or institutional funding from one or more of the following: Chiesi; AIM ITALY SRL; Fisher and Paykel Healthcare Ltd; Maquet Critical Care; Draeger; Intersurgical SpA; Orionpharma; Phils; Resmed; Merck Sharp and Dohme; Novartis Study dates: December 2015 to March 2017 Note: we did not complete 'Risk of bias' assessments for this study because we did not report outcome data in the review.

*Study characteristics*
Methods	RCT, cross‐over design. Single‐centre
Participants	Total number of randomized participants: 17 Setting: respiratory ICU; Bangkok, Thailand Inclusion criteria: mechanically ventilated patients who were 18 years of age, successfully weaned by spontaneous breathing, trial with oxygen T‐piece or low level of pressure support for 120 minutes, and ready for endotracheal extubation Exclusion criteria: haemodynamic instability or decreased level of consciousness; patients who lacked co‐operation, tracheotomized patients, and pregnant women Baseline characteristics: Age, mean (SD): 66.8 (± 13.8) years Gender, M/F: 10/7 BMI, mean (SD): not reported SAPS II, mean (SD): 30.9 (± 4.4) Respiratory rate, mean (SD): recorded before each cross‐over period: baseline 1: 20.3 (± 4.5); baseline 2: 21.7 (± 3.8) breaths/min PaCO₂: not reported PaO₂/FiO₂: not reported
Interventions	Intervention group (HFNC): Details: HFNC, Optiflow system, Fisher & Paykel Healthcare; initial inspiratory flow of 35 L/min, and FiO₂ adjusted to achieve SpO₂ ≥ 94% within the first 5 minutes and to maintain this setting for 30 minutes Control group (standard oxygen therapy): Details: Non‐rebreather face mask, 6 to 10 L/min to achieve SpO₂ 94% for another 30 minutes
Outcomes	Dyspnoea, patient comfort, breathing frequency, heart rate blood pressure, SpO₂
Notes	Funding/declarations of interest: study authors did not report funding sources. They disclosed no conflicts of interest. Study dates: August to December 2011
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Methods used to generate group allocation not stated
Allocation concealment (selection bias)	Unclear risk	Methods of allocation concealment not stated
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinding was not possible. Although we expected that this would not influence outcome data, we could not be certain of this.
Blinding of outcome assessors (objective outcomes)	Low risk	Unable to blind outcome assessors owing to nature of the intervention: we did not anticipate that this would influence the assessment of objective outcome measures.
Blinding of outcome assessors (subjective measures)	Low risk	Unable to blind outcome assessors owing to nature of the intervention. Participants were outcome assessors for comfort: We did not expect that this would influence the outcome data.
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	No statement of how many reported. No participant numbers in tables or graphs
Selective reporting (reporting bias)	Unclear risk	Trial registration not reported in paper. Unable to establish whether outcomes were reported according to pre‐published protocol or trial registration documents SpO₂ and mean arterial pressure not reported for all time points set out in methods
Other bias	Low risk	We identified no other risks of bias.

PERMALINK

High‐flow nasal cannulae for respiratory support in adult intensive care patients

Sharon R Lewis

Philip E Baker

Roses Parker

Andrew F Smith

Abstract

Background

Objectives

Search methods

Selection criteria

Data collection and analysis

Main results

Authors' conclusions

Plain language summary

Summary of findings

Background

Description of the condition

Description of the intervention

How the intervention might work

Why it is important to do this review

Objectives

Methods

Criteria for considering studies for this review

Types of studies

Types of participants

Types of interventions

Types of outcome measures

Important outcomes

Additional outcomes

Search methods for identification of studies

Electronic searches

Searching other resources

Data collection and analysis

Selection of studies

Data extraction and management

Assessment of risk of bias in included studies

Measures of treatment effect

Unit of analysis issues

Dealing with missing data

Assessment of heterogeneity

Assessment of reporting biases

Data synthesis

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

Summary of findings and assessment of the certainty of the evidence

Results

Description of studies

Results of the search

1.

Included studies

Study population

Study setting

Interventions and comparisons

Outcomes

Funding

Excluded studies

Studies awaiting classification

Ongoing studies

Risk of bias in included studies

2.

Allocation

Blinding

Incomplete outcome data

Selective reporting

Other potential sources of bias

Effects of interventions

Summary of findings 1. HFNC compared to standard oxygen therapy for respiratory support in adult intensive care patients.

Summary of findings 2. HFNC compared to NIPPV or NIV for respiratory support in adult intensive care patients.

Comparison 1: HFNC versus standard oxygen therapy

Important outcomes

Failure of treatment as indicated by escalation of respiratory therapy to NIV, NIPPV or invasive ventilation

1.1. Analysis.

3.

In‐hospital mortality

1.2. Analysis.

4.

Important adverse events

Pneumonia

1.3. Analysis.

Short‐term oxygenation (PaO₂/FiO₂)

Short‐term oxygenation (PaO₂/FiO₂)

*Study characteristics*
Methods	RCT, parallel‐group design; single‐centre study
Participants	Total number of randomized participants: 60 Setting: ICU; China Inclusion criteria: people with acute respiratory failure, mechanically ventilated in the ICU for at least 48 hours and were ready to be extubated after clinical weaning assessments Exclusion criteria: poor co‐operation; tracheotomy; decreased level of consciousness; < 18 years of age; pregnant; did not sign consent form Baseline characteristics: Intervention group (HFNC): Age, mean (SD): 66 (± 14) years Gender, M/F: 16/14 BMI, mean (SD): not reported APACHE II, mean (SD): 12.87 (± 3) PaCO₂, mean (SD): 41.5 (± 6.7) mmHg PaO₂/FiO₂, mean (SD): not reported Control group (NIV): Age, mean (SD): 71 (± 13) years Gender, M/F: 18/12 BMI, mean (SD): not reported APACHE II, mean (SD): 12.36 (± 3.29) PaCO₂, mean (SD): 42.3 (± 7.1) mmHg PaO₂/FiO₂, mean (SD): not reported
Interventions	Intervention group (HFNC): Randomized, n = 30; losses, n = 0; analysed, n = 30 Details: HFNC via PT101AZ, initial flow rate at 60 L/min with downward adjustments in 5 to 10 L/min decrements; target SpO₂ of 94% to 98% (or 88% to 92% for hypercapnic respiratory failure); FiO₂ set at 40% Control group: Randomized, n = 30; losses, n = 0; analysed, n = 30 Details: oxygen via air entrainment mask, with flow rate at 10 L/min; target SpO₂ of 94% to 98% (or 88% to 92% for hypercapnic respiratory failure); FiO₂ set at 40%
Outcomes	Success of oxygen therapy; needing NIV, or MV, or replacement of oxygen device; respiratory variables (PaO₂; SpO₂; PaCO₂; respiratory rate): haemodynamic variables; discomfort (scale 0 to 10 = no discomfort to maximum discomfort)
Notes	Funding/declarations of interest: funded by grants from the National Natural Science Foundation of China, the Medical and Health Research Program of Zhejiang Province, and the Medical and Health Research Program of Zheijing Province Study dates: January 2013 to December 2014
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated randomization
Allocation concealment (selection bias)	Unclear risk	No details
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinding was not possible. Although we expected that this would not influence outcome data, we could not be certain of this.
Blinding of outcome assessors (objective outcomes)	Low risk	Blinding of outcome assessors was not described; we did not anticipate that this would influence the assessment of objective outcome measures.
Blinding of outcome assessors (subjective measures)	Low risk	Participants were outcome assessors for discomfort using a standardized scale; we did not expect that this would influence the outcome data.
Incomplete outcome data (attrition bias) All outcomes	Low risk	No apparent losses
Selective reporting (reporting bias)	Unclear risk	Study authors did not report clinical trials registration or prepublished protocol. It was not feasible to effectively assess risk of selective reporting bias without these documents.
Other bias	Low risk	We identified no other sources of bias.

Study	Reason for exclusion
Coudroy 2019	This was a multicentre RCT of adults with severe acute hypoxaemic respiratory failure in an ICU setting. Participants were randomized to receive either HFNC alone or HFNC and NIV. We excluded this study as HFNC were used in both study arms.
Delorme 2017	This was a cross‐over RCT of adults with moderate respiratory distress in an ICU setting. Participants were given standard oxygen therapy up to baseline and then received HFNC with 3 different flow rates in a random order. We excluded this study as the control intervention (standard oxygen therapy) was delivered to all participants first, so this is not truly randomized.
Di Mussi 2016	This was a non‐randomized cross‐over study where participants were given HFNC, low‐flow oxygen and HFNC sequentially post‐extubation in an ICU setting. We excluded this study as it was not randomized.
Lemiale 2016	This was a multicentre study of critically‐ill immunocompromised adults receiving treatment for haematological malignancies or solid tumours. Participants were randomized to receive either NIV or oxygen therapy. Within the oxygen therapy group, the decision to use low‐flow oxygen or HFNC was at the discretion of the treating clinician. Therefore, randomization was not at the level required for this review (i.e. HFNC vs low‐flow oxygen therapy).
Liu 2019	This was an RCT of weaning and post‐extubation adults receiving invasive mechanical ventilation in an ICU setting. Participants were randomized to receive T‐tube, NIV or high‐flow oxygen via their endotracheal tube during a 2‐hour spontaneous breathing trial. If they passed the SBT, participants receiving T‐tube or NIV were moved onto low‐flow oxygen facemask whilst the high‐flow oxygen group were moved to HFNC. We excluded this study as the participants received different treatment prior to the initiation of oxygen via HFNC or low‐flow oxygen via facemask.
Pennisi 2019	This was an RCT of adults undergoing elective thoracotomic pulmonary lobar resection. Participants received oxygen either using HFNC or Venturi face mask. We excluded this study because very few participants received therapy in the ICU; most participants started therapy in the PACU before transfer to the surgical ward.
Sklar 2018	This was a cross‐over RCT of adults with cystic fibrosis with a clinical indication for NIV in a respiratory ward. Participants received standard oxygen up to baseline then HFNC and NIV in random order. We excluded this study as participants were treated in a respiratory ward, not an ICU.
Thille 2018	This was a multicentre RCT of adults post‐extubation at high risk of post‐extubation failure in an ICU setting. Participants were randomized to receive either HFNC alone or HFNC and NIV. We excluded this study as HFNC were used in both intervention arms.
Thille 2019	This was a multicentre RCT of adults post‐extubation who were at high risk of post‐extubation failure in an ICU setting. Participants were randomized to receive either HFNC alone or HFNC and NIV. We excluded this study as HFNC were used in both intervention arms.

Methods	RCT, parallel‐group design
Participants	Total number of randomized participants: 36 Inclusion criteria: AECOPD; hypercapnia; ready for extubation Exclusion criteria: none reported
Interventions	Intervention group (HFNC): n = 19 Control group (NIV): n = 17
Outcomes	All outcomes reported: rate of treatment failure; reintubation rate; vital signs; ABG; comfort score; bronchoscopy for secretion management within 48 hours Outcomes relevant to this review: treatment failure; reintubation rate; vital signs; ABG; comfort score
Notes	Contact: Jing Guoqiang, Binzhou Medical University, Binzhou, China Currently published only as an abstract. We are awaiting publication of the full report in order to assess eligibility, collect sufficient study characteristics, and include data in the review.

Methods	RCT. parallel‐group design. Single‐centre study
Participants	Number of participants: 20 Inclusion criteria: postoperative liver transplant; respiratory failure Exclusion criteria: none stated
Interventions	Intervention group (HFNC): n = 10; flow = 60 L/min; flow and FiO₂ titrated to ABG Control group (NIV): BiPAP; n = 10; PEEP = 5 cm H₂O and IPAP = 10 cm H₂O; flow and FiO₂ titrated to ABG
Outcomes	All outcomes measured: ABG, comfort scale, RASS, CAM‐ICU, nutritional deficit; intubation rate Outcomes relevant to this review: intubation rate; ABG; comfort
Notes	Contact: S. Gupta ‐ Medanta ‐ The Medicity, Gurgaon, India Currently published only as an abstract. We are awaiting publication of the full report in order to assess eligibility, collect sufficient study characteristics, and include data in the review.

Methods	RCT, cross‐over design
Participants	Number of participants: 27. Inclusion criteria: acute on chronic respiratory failure; pH > 7.34; respiratory rate ≤ 30 breaths/min Exclusion criteria: none in abstract
Interventions	Intervention group (HFNC): flow = 60 L/min Control group: Venturi mask
Outcomes	All outcomes measured: ultrasound diaphragm displacement; diaphragm thickening fraction; dyspnoea; comfort; arterial blood gases Outcomes relevant to this review: dyspnoea; comfort; arterial blood gases
Notes	Contact: Federico Longhini, Anesthesia and Intensive Care, Sant’Andrea Hospital, ASL VC, Vercelli, Italy Currently published only as an abstract. We noted some similarities with Longhini 2019, but because of some variation in methodology, we have assumed this study to be separate. We are, therefore, awaiting publication of the full report in order to assess eligibility, collect sufficient study characteristics, and include data in the review.

Methods	RCT. parallel‐group design. Multicentre study
Participants	Number of randomized participants: 80 Setting: four ICUs at 2 hospitals, France Inclusion criteria: mechanically ventilated patient ready for extubation Exclusion criteria: none reported
Interventions	Intervention group (HFNC): n = 40 Control group (standard oxygen therapy): n = 40 Both for 48 hours post‐extubation
Outcomes	All outcomes reported: lung ultrasound score, dyspnoea, post‐extubation distress incidence; treatment failure rate, mean time to reintubation; clinical respiratory variables; cardiovascular variable; ICU and hospital mortalities Outcomes relevant to this review: treatment failure rate; clinical respiratory variables; hospital mortality; dyspnoea
Notes	Contact: S. Perbet, University Hospital of Clermont‐Ferrand, ICU, Clermont‐Ferrand, France Currently published only as an abstract. We are awaiting publication of the full report in order to assess eligibility, collect sufficient study characteristics, and include data in the review.

Methods	Not stated if this was an RCT, parallel‐group design. Single‐centre study
Participants	Total number of participants: 85 Setting: respiratory ICU, Egypt Inclusion criteria: COPD; type II respiratory failure; admitted to respiratory ICU Exclusion criteria: none reported
Interventions	Intervention group (HFNC): n = 25 Control group (standard oxygen therapy): Venturi face mask; n = 20
Outcomes	All outcomes reported: ABG variables, successful weaning, treatment failure Outcomes relevant to this review: ABG; treatment failure; successful weaning
Notes	Contact: Adel Saeed, Pulmonary Medicine, Ain Shams University, Abbasia, Cairo Egypt Currently published only as an abstract. We are awaiting publication of the full report in order to assess eligibility, collect sufficient study characteristics, and include data in the review.

Methods	RCT, cross‐over design. Single‐entre study
Participants	Total number of participants: 20 Inclusion criteria: acute respiratory failure; spontaneously breathing patients Exclusion criteria: none in abstract
Interventions	Three 60‐minute trials with the following therapies in random order. Intervention (HFNC): flow = 60 L/min Control 1: (NIV) Control 2: (standard oxygen therapy)
Outcomes	All outcomes reported: lung ultrasound aeration score; diaphragm thickening fraction; diaphragm excursion Outcomes relevant to this review: none
Notes	Contact: Annia Fleur Schreiber, Respiratory Intensive Care Unit and Pulmonary Rehabilitation Unit, Istituti Clinici Scientifici Maugeri IRCCS, Pavia, Italy Currently published only as an abstract. We are awaiting publication of the full report in order to assess eligibility, collect sufficient study characteristics, and include data in the review.

Study name	Prophylactic postoperative high‐flow nasal oxygen therapy versus conventional oxygen therapy in obese patients undergoing bariatric surgery: a randomised controlled pilot study
Methods	RCT, parallel‐group design
Participants	Estimated number of participants: 64 Setting: post‐surgical ICU, Australia Inclusion criteria: age > 18; BMI > 32 kg/m²; Undergoing laparoscopic bariatric procedure for weight reduction Exclusion criteria: refusal of informed consent; contraindication to HFNO therapy; chest circumference too large for EIT belt
Interventions	Extubation in theatre. Hudson face mask with flow = 6 L/min for transfer to ICU. Randomized when RASS ≥ ‐2 Intervention group (HFNC): duration = 6 hrs; flow = 50 L/min. FiO₂ = 0.5 and titrated to achieve target SpO₂ = 95 % Control group (standard oxygen therapy): via Hudson face mask
Outcomes	All outcomes measured: change in end‐expiratory lung impedance as a surrogate for end‐expiratory lung volume, measured by EIT; PaCO₂; change in tidal variance as a surrogate for tidal volume; complication rate; length of hospital stay; PaO₂/FiO₂; patient comfort. Outcomes relevant to this review: PaCO₂; complication rate; length of hospital stay; PaO_2/FiO₂; patient comfort
Starting date	15 May 2017
Contact information	John Fraser, john.fraser@health.qld.gov.au. Rachel Fulton, r.fulton.04@aberdeen.ac.uk
Notes

Study name	Comparative study of nasal high‐flow oxygen therapy and noninvasive positive pressure ventilation for moderate AECOPD: randomized open non‐inferiority trial
Methods	RCT, parallel‐group design
Participants	Total number of participants: 86 Setting: China Inclusion criteria: AECOPD; blood gas analysis pH 7.25‐7.35, PaCO₂ > 50 mmHg Exclusion criteria: age < 18; no informed consent obtained; severe respiratory failure requiring tracheal intubation; NPPV contraindications; patients with short‐term prognosis; other organ failure; tracheotomy
Interventions	Intervention group (HFNC) Control group (NPPV)
Outcomes	All outcomes measured: arterial blood gas, respiratory rate, blood pressure, daily treatment time; parameter setting of NPPV and HFNC; change and time of respiratory support; intubation; time to intubation; dyspnoea score; comfort score; facial skin breakage; number of daily nursing interventions; respiratory and extrapulmonary complications; ICU length of stay; hospital length of stay; discharge outcome (death, improved) Outcomes relevant to this review: arterial blood gas, respiratory rate, intubation; dyspnoea score; comfort score; respiratory and extrapulmonary complications; ICU length of stay; hospital length of stay; mortality
Starting date	21 January 2018
Contact information	Dingyu Tan, 32494845@qq.com. Bingyu Ling
Notes