Acute hypoxemic respiratory failure (AHRF) is defined by the presence of hypoxemia without significant hypercapnia. AHRF is usually caused by inadequate ventilation-to-perfusion ratios resulting in poorly oxygenated blood being allowed into the systemic circulation. It can be caused by various conditions including bacterial or viral pneumonia, cardiogenic pulmonary edema, atelectasis, respiratory disease exacerbations—e.g., from chronic obstructive pulmonary disease (COPD)—pleural effusion, or a combination of either of these conditions. AHRF may occur spontaneously or in post-operative, post-extubation, or trauma patients. These different settings involve distinct underlying mechanisms leading to AHRF, present a different prognosis, and are thus usually discussed separately in the clinical practice guidelines on HFNO and non-invasive ventilation (NIV) (1–4). In this present review, we will focus solely on de novo AHRF. We will use the term “AHRF” when not referring to a specific etiology and the term “de novo AHRF” when referring to AHRF occurring without underlying pulmonary disease and excluding cardiogenic pulmonary edema.

Acute respiratory distress syndrome (ARDS) is the most severe form of AHRF. ARDS is defined by the Berlin criteria as follows: new-onset, bilateral pulmonary opacities consistent with pulmonary edema with a PaO₂/FiO₂ of ≤ 300 mmHg under a minimal positive end-expiratory pressure (PEEP) of ≥5 cmH₂O (5). One must first exclude a cardiac origin from pulmonary edema. The severity is determined by the PaO₂/FiO₂ ratio. ARDS carries a poor prognosis with a mortality ranging from 27% for mild ARDS to 45% for severe ARDS (5). The treatment is mainly supportive, focusing on the patient's oxygenation. Historically, conventional oxygen therapy (COT) has been used—nasal cannulas, facemasks, non-rebreather facemasks, or Venturi masks—in association with low-tidal volume invasive mechanical ventilation (IMV) for the most severe patients. However, IMV is associated with several complications, such as ventilator-associated pneumonia, delirium, or ICU-associated polyneuropathy, and is resource-demanding (1). Therefore, non-invasive respiratory support strategies, including continuous positive airway pressure (CPAP), bi-level positive airway pressure (BiPAP), and high-flow nasal oxygen (HFNO), have been explored as possible alternatives to IMV (1–4).

The COVID-19 pandemic resulted in a substantial increase in both AHRF and ARDS overwhelming both acute and intensive care unit capacity, urging the exploration of alternative strategies to avoid the need for IMV and preserve ICU capacity. In this context, HFNO and NIV (CPAP or BiPAP) have been proposed as alternative strategies to COT to avoid IMV. Several observational and randomized controlled trials assessed their impact on mortality and the need for IMV (6–15).

Before reviewing this recent clinical evidence, we will first review the physiological basis and characteristics of these different types of non-invasive respiratory support and then discuss conflicting recommendations in recent international guidelines (3, 4).

The main therapeutic objective in de novo AHRF is to provide sufficient oxygenation and to prevent respiratory exhaustion, which will then require IMV. Various techniques of administration of supplemental oxygen and/or positive end-expiratory pressure serve this purpose. Additional inspiratory pressure may or may not be applied.

A systematic review published in 2020 compiled evidence on HFNO in AHRF before the COVID pandemic (29). This network meta-analysis included 25 trials and 3,804 participants and compared various non-invasive respiratory support strategies. Compared to COT, HFNO was associated with a significant reduction of the need for intubation [risk ratio (RR) 0.76; 95% confidence interval (CI) 0.55–0.99] and a non-significant reduction of 30-day mortality (RR 0.87; 95% CI 0.62–1.15). Similarly, Rochwerg et al. (30) identified nine studies (2,093 patients) comparing HFNO to COT in AHRF of any cause with similar estimates: a significant reduction in the need for intubation (RR of 0.85; 95% CI 0.74–0.99) but no statistically significant reduction in 30-day mortality (RR 0.94; 95% CI 0.67–1.31). The effects of HFNO on key clinical outcomes are summarized in Table 1.

The main RCTs comparing HFNO and COT, including more than 50 patients, and reporting clinically relevant outcomes are detailed in Supplementary Table S1 (31–38). Studies were performed in different settings including ICU (31, 32, 39, 40), emergency unit (35, 36, 38, 41), and acute care ward (37). While pneumonia was the main cause of AHRF, some trials also included patients with cardiogenic edema (34–36), COPD exacerbations (34, 36), and asthma exacerbations (42, 43). Furthermore, three trials focused on patients with immunosuppression from various etiologies (32, 33, 37).

Based on this initial body of evidence, the European Respiratory Society and the European Society of Intensive Care Medicine recommended HFNO over COT in AHRF (2, 3). The American College of Physicians (ACP) recently published a similar guideline (4). While pooling data very similar to the ERS guidelines (3), they concluded that the overall effect of HFNO over COT in AHRF remained uncertain (4). International guidelines regarding HFNO in AHRF are summarized in Table 2. These different interpretations appear to be mainly driven by the use of different methods for the pooled estimates. The ACP guidelines used a Peto odds ratio and Hartung–Knapp–Sidik–Jonkman random effects model, which probably led to an exaggerated weight of two negative studies of a smaller sample size (35, 44). These two studies account for < 1% of the events (3/418) but for 14.4% of the weight of the pooled estimate in the ACP guidelines. Moreover, these two studies differed in their designs; Spoeletini et al. evaluated HFNO during breaks of NIV among patients with hypercapnic respiratory failure, whereas Makdee et al. included patients with acute pulmonary edema, and the study was prognostically imbalanced with patients allocated to COT being more likely to benefit from NIV. For these reasons, the inclusion of these two studies in the meta-analysis and their overweighting appear debatable, hence the different conclusions of the ACP guideline.

Overall, the evidence published before COVID-19 relies on several open-label RCTs, detailed in Supplementary Table S1, and supports the use of HFNO in AHRF to reduce intubation, yet with an uncertain impact on mortality (31–38). Certainty of the available evidence is limited by the imprecision in effect estimates (2, 3, 45) and the risk of bias of included studies, in particular, the absence of blinding due to the nature of the intervention. Moreover, the included populations are heterogeneous in terms of settings and patient characteristics (AHRF etiology, immunosuppression, the inclusion of COPD or cardiac failure, etc.). It remains therefore unclear if the HFNO benefit applies to the majority of included patients or more specifically to some yet unidentified subgroups.

Several RCTs have been conducted during the COVID-19 pandemic to compare different respiratory support strategies. These trials are summarized in Table 5 and detailed in Supplementary Table S3 (7–14).

Ospina-Tascòn et al. (7) randomized 220 COVID-19 patients with AHRF (PaO₂/FiO₂ < 200 mmHg) to receive HFNO or COT. This open-label multicenter trial reported a reduced need for IMV in the HFNO group compared to COT (34.3% in the HFNO group vs. 51% in the COT group, HR 0.62; 95% CI 0.39–0.96). Only one patient crossed over from the COT to the HFNO group. This trial was considered at low risk of bias except for the open-label design, which the authors tried to mitigate using pre-defined intubation criteria (67).

Crimi et al. (11) randomized 364 patients with confirmed SARS-CoV-2 pneumonia with mild AHRF (SpO₂ ≤ 92% or PaO₂/FiO₂ < 300 and the need for oxygen therapy) to receive HFNO or COT (1:1). The primary outcome was a need for escalation to advanced respiratory support (IMV, CPAP, and BiPAP). No statistical difference was observed (RR 0.79; 95% CI 0.59–1.05). This trial included mostly patients with less severe hypoxemia and was underpowered due to a lower number of events than expected.

Frat et al. (14) randomized 782 patients with AHRF to HFNO vs. COT. The trial was initiated before the COVID-19 pandemic and was redesigned during the COVID-19 pandemic. A total of 711 patients with AHRF (PaO₂/FiO₂ < 200) due to COVID-19 were included in the final analysis. No difference in 28-day mortality was observed [10% with HFNO vs. 11% with COT, absolute difference, −1.2% (95% CI −5.8–3.4)]. A significant reduction of IMV was observed in the HFNO group [45% with HFNO vs. 53% with COT, absolute difference −7.7% (95% CI; −14.9 to −0.4%)]. The low crossover rate is a strength of this trial; however, limited by a lack of power to show a difference in mortality because of lower-than-anticipated mortality rates.

Perkins et al. (9) randomized 1,237 patients with confirmed SARS-CoV-2 infection with AHRF (SpO₂ < 94% or less despite receiving 40% FiO₂) to receive HFNO, CPAP, or COT in a pragmatic, open-label, adaptative, and randomized trial. The incidence of the primary composite outcome (requirement of IMV or mortality at 30 days) was significantly lower with CPAP (36.3%) as compared to COT (44.4%) absolute difference −8% (95% CI −15 to −1%) but was not significantly different with HFNO when compared to COT [44.3 vs. 45.1%, absolute difference −1% (95% CI −8 to +6%)]. The difference between CPAP and COT was mainly driven by the requirement of IMV. The lack of benefit of HFNO might be explained by the high number of crossovers with a total of 23.6% of patients crossing from COT to other respiratory support (8.4% received CPAP, 7.6% received HFNO and 7.6% received both CPAP and HFNO), the early termination of the trial and the pragmatic nature of the trial with the lack of predetermined criteria to proceed to IMV. Indeed, while the pragmatic design increases the external validity of the trial, the absence of predefined intubation criteria may increase the risk of bias, especially in the context of the open-label design. Adverse events occurred more often in the CPAP group than in HFNO and COT groups (34.2, 20.6, and 13.9%, respectively). Common adverse events in the CPAP group included the following: interface/therapy intolerance (5.8%), pain (5.5%), cutaneous pressure sore (12.1%), claustrophobia (12.1%), oral dryness (6.6%), and hemodynamic instability (11.3%).

Bouadma et al. (10) randomized 333 patients with confirmed or highly suspected COVID-19-related AHRF to receive HFNO, CPAP, or COT. There were no differences in the incidence of IMV between COT (29.4%), CPAP (31.4%), or HFNO (32.6%). Important limitations of this trial included a high crossover rate (26.6% of patients in COT crossed to HFNO) and lack of power due to a two times lower-than-expected rate of events.

Grieco et al. (8) randomized 109 patients with confirmed SARS-CoV-2 infection and moderate to severe AHRF (PaO₂/FiO₂ ≤ 200) to receive either BiPAP with a helmet interface or HFNO. No significant difference in the number of days free of respiratory support within 28 days was observed. Interestingly, the rate of IMV (secondary outcome) was lower in the helmet group than in the HFNO (30 vs. 51%, p = 0.03). These findings are exploratory, as the primary outcome did not reach a significant effect.

Finally, two smaller trials summarized in Table 5 have been published. One reported a reduction in the rate of escalation toward mechanical ventilation (IMV or NIV) when HFNO was compared to COT and the other one did not find a significant difference in the rate of intubation at 48 h when HFNO was compared to BiPAP (12, 13).

Arabi et al. (15) randomized 320 patients with AHRF (PaO₂/FiO₂ < 200) related to COVID-19 to helmet NIV vs. usual respiratory support (facemask NIV, HFNO, COT, or a combination of these at the discretion of the physician). There was no difference in 28-day mortality (27% in the helmet group and 26.1% in the usual respiratory support group). There was no significant difference in any of the pre-specified secondary outcomes. The main strength includes the low crossover rate in both groups. The main limitations of this trial include the different controls (COT, HFNO, and Facemask NIV) and the lack of power (estimated mortality rates of 40%). More than 36.5% of the patients discontinued helmet NIV because of reported intolerance after a median of 20.5 h of use (IQR 3–48 h).

Overall, the evidence regarding HFNO in AHRF due to COVID-19 seems concordant with the pre-COVID-19 evidence, with a possible decrease in the risk of IMV compared to COT and an uncertain effect on mortality. However, important differences in inclusion criteria, AHRF severity, standardization of intubation criteria, and crossovers lead to some inconsistencies across available trials. Furthermore, different interventions were compared including COT, HFNO, CPAP, BiPAP, and the use of different interfaces and ventilator parameters, with no optimal strategy yet identified.

The COVID-19 pandemic led to an important amount of new evidence regarding the benefits of HFNO in AHRF with more than 3,000 patients included in recent RCTs. Most studies conducted during the COVID-19 period reported positive results in favor of HFNO compared to COT regarding the need for IMV with an uncertain effect on mortality. These recent pieces of evidence are consistent with the pre-COVID evidence regarding HFNO for AHRF of other etiologies. Various phenotypic patterns of COVID-19-related AHRF have been described, some characterized by preserved lung compliance and others with a more restrictive respiratory pattern (68–70). Therefore, it may be questioned if the conclusions of the COVID-19 evidence apply to other forms of AHRF. However, the observed variability in phenotypes among patients with COVID-19 probably also occurs in AHRF of other miscellaneous causes (71). Taken together, these recent RCTs increase the level of evidence regarding the potential benefit of HFNO in AHRF. However, important heterogeneity was present in terms of settings, AHRF severity (as assessed by the variability of the PaO₂/FiO₂ ratios), standardization of intubation criteria, and crossovers leading to some inconsistencies across available trials. In particular, the important rate of crossovers from COT to HFNO in several pragmatic trials may have led to underestimating the benefits of HFNO in these trials. Moreover, the open-label design of available RCTs due to the nature of the intervention, and the absence of strictly pre-defined IMV criteria in a significant proportion of these trials constitute a potential risk of bias and must be acknowledged as an important limitation of the available evidence.

The evidence regarding the comparison between HFNO and other non-invasive respiratory support strategies is still more equivocal. Although pre-COVID pieces of evidence suggested a possible superiority of HFNO over NIV, largely relying on the FLORALI trial, RCTs conducted during the COVID-19 pandemic reported inconsistent results; some studies suggesting a possible superiority of CPAP or BiPAP over HFNO (8, 9), other no difference (10), or even superiority of HFNO over BiPAP (13).

As previously discussed, AHRF encompasses a wide range of phenotypes, etiologies, severity, and contributing factors. HFNO, CPAP, and BiPAP have different physiological effects and individual patient characteristics probably mitigate the clinical effect of the different non-invasive respiratory supports. Gattinoni et al. (68) conceptually described two distinct radiological and physiological patterns of COVID-19 pneumonia: Type L pneumonia characterized by ground-glass densities on Ct-scan, preserved lung compliance, and low recruitability occurring in the early phase of the disease, and Type H pneumonia characterized by extensive non-aerated compartments on imaging, high elastance, and high recruitability. Based on this conceptual model, they suggested that respiratory support strategies should be adapted to these different patterns. Although this would be an interesting hypothesis to explore, these different patterns often co-exist in real-life patients and were therefore not reported in the available randomized data evaluating HFNO or NIV. Moreover, the observed clinical heterogeneity was present not only between RCTs but also within available RCTs, and their subgroup analyses were not systematically reported in particular regarding physiological aspects of lung injury and mechanics. In the recovery trial, no interaction was reported between HFNO effect and age, gender, ethnicity, time from onset to randomization, or body mass index, while patients requiring higher FiO₂ (>60%) tended to benefit more from HFNO than patients requiring lower FiO₂. In the study by Ospina-Tascon et al., younger patients (< 60 years) and patients with lower IL-6 levels (< 100 pg/ml) appeared to benefit more from HFNO than older patients or patients with higher levels of IL-6, while no interaction was detected between HFNO effect and the severity of hypoxemia. In a post-hoc analysis of the HENIVOT trial, patients with low PaCo₂ (< 35 mmHg) or severe hypoxemia (PaO₂/FiO₂ × dyspnea visual analogical scale < 30) tended to benefit from helmet ventilation (72). Finally, Crimi et al. did not report any interaction between HFNO effect, time to randomization, age, co-morbidities, or severity of hypoxemia. In this context of inconsistent results across secondary analyses, it remains difficult based on the current evidence to identify a subgroup of patients likely to benefit specifically from HFNO or other non-invasive respiratory support, until more studies specifically designed to explore these subgroup hypotheses are reported.

Delayed intubation is a matter of concern as NIV failure in AHRF is frequent and associated with increased mortality (73). A correlation between the duration of respiratory distress (time with hypoxemia and RR >25/min) and reduced lung compliance (driving pressure >14 cmH₂O) has been reported (74). However, the optimal timing for intubation remains uncertain (75). For these reasons, early recognition of patients failing to respond to non-invasive respiratory support is crucial, so that the use of HFNO and NIV should, to pour opinion, be restricted to settings such as intermediate/respiratory care units or intensive care units, where the response to treatment can be closely monitored and treatment strategies quickly adapted (76). The ROX index [(SpO₂/FiO₂)/respiratory rate] has been validated as a simple predictor of failure in patients with AHRF receiving HFNO and might be helpful to evaluate patient response, guiding therapy adaptation and to avoid potential risks associated with delayed intubation (77, 78). Given its ease of use, excellent tolerance, and the potential benefits compared to COT, HFNO could arguably be the first option for patients with AHRF and may constitute an adequate comparator to evaluate the benefits of alternative or additional strategies such as CPAP or BiPAP in future trials.

The COVID-19 pandemic was associated with an unprecedented increase in hospital admissions for AHRF and was an opportunity to enrich the available evidence regarding non-invasive respiratory support strategies in this condition. HFNO is a type of well-tolerated respiratory support that is relatively easy to provide and probably reduces the need for IMV in patients with AHRF compared with COT. This may contribute to reducing IMV-associated complications and preserving ICU capacity. Several studies suggested possible additional benefits of NIV in COVID-19 patients. However, no optimal supportive respiratory strategy has yet been defined due to heterogeneity in interfaces, ventilator settings, AHRF definitions. and results. Given important phenotypic differences among patients with AHRF, the use of non-invasive respiratory supports should probably be restricted to monitored units and tailored to patient tolerance and response to therapy.

LG, TA, and CM designed and wrote the manuscript. CE and AK revised the work and contributed to critical appraisal. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.