High‐Flow Nasal Cannula Versus Noninvasive Ventilation in the Treatment of Hypercapnic Respiratory Failure: Meta‐Analysis of Randomized Controlled Trials

ABSTRACT

Aim

To evaluate efficacy and safety of high‐flow nasal cannula (HFNC) versus noninvasive ventilation (NIV) in adults with hypercapnic respiratory failure (HRF).

Methods

We conducted a meta‐analysis of randomized controlled trials from PubMed, Embase, The Cochrane Library, Wanfang, CNKI, and Weipu comparing HFNC with NIV in HRF patients. Primary outcomes were mortality and endotracheal intubation. Secondary outcomes included blood gas values, hospital/ intensive care unit (ICU) length of stay, nasal skin breakdown, and comfort. Pooled risk ratios (RR) and mean differences (MD) with 95% confidence intervals (CI) were calculated.

Results

Twenty trials involving 1835 patients were included. No significant differences were observed in mortality (RR, 0.94; 95% CI, 0.69 to 1.28), endotracheal intubation (RR, 0.87; 95% CI, 0.71 to 1.06), PaCO₂ (MD, –0.87 mmHg; 95% CI, –2.98 to 1.25) or PaO₂ (MD, 2.67 mmHg; 95% CI, –0.66 to 6.01). However, HFNC significantly reduced hospital stay (MD, –0.69 days; 95% CI, –1.08 to –0.30) and ICU stay (MD, –0.98 days; 95% CI, –1.50 to –0.45), lowered nasal skin breakdown risk (RR, 0.20; 95% CI, 0.09 to 0.45) and improved comfort (MD, –1.19; 95% CI, –1.98 to –0.41). Subgroup and sensitivity analyses confirmed the principal findings.

Conclusions

HFNC was not significantly different from NIV for mortality and endotracheal intubation in HRF but was associated with shorter hospital and ICU stays, reduced nasal facial skin breakdown and improved comfort. Larger trials in diverse populations are needed.

1. Introduction

Chronic obstructive pulmonary disease (COPD) is the third leading cause of mortality globally, resulting in 3.3 million deaths and 74.4 million disability‐adjusted life years (DALYs) [1]. Hypercapnic respiratory failure (HRF) predominantly occurs in COPD patients and is associated with increased mortality [2]. High‐quality randomized controlled trials (RCTs) indicated that combining noninvasive mechanical ventilation (NIV) with usual care could reduce the risk of mortality by 46% and the risk of needing endotracheal intubation by 65% in acute exacerbations of COPD (AECOPD) patients complicated by HRF [3]. Therefore, NIV is recommended as the primary therapy for patients who require augmentation of ventilation to manage acute and chronic respiratory failure [4, 5]. However, prolonged inhalation of dry gas may result in complications such as gastric inflation and impaired mucus clearance. Additionally, tightly fitted masks or helmets may increase the risk of facial skin breakdown and claustrophobia, often resulting in poor tolerance and treatment failure [6].

High‐flow nasal cannula (HFNC) has emerged as a promising alternative, delivering heated and humidified gas mixtures at high flow rates through wide‐bore nasal prongs, thereby enhancing comfort, humidification, and tolerance [7]. Several meta‐analyses have evaluated whether HFNC is superior, inferior, or equivalent to NIV in acute HRF patients, but their findings are largely limited by small sample sizes [8, 9]. Due to imprecision and heterogeneity, evidence remains insufficient to determine their relative efficacy and safety. Several recently published large‐scale RCTs on this topic have provided critical supplementary data, greatly increasing the sample size. Accordingly, a new analysis incorporating these trials is needed to evaluate the efficacy and safety of HFNC versus NIV in adults with HRF.

2. Methods

2.1. Protocol and Guidance

This systematic review is presented according to Preferred Reporting Items for Systematic Reviews and Meta‐analyses (PRISMA) guidelines. This protocol was preregistered on the PROSPERO international prospective register of systematic reviews (registration number: CRD42025649318).

2.2. Selection Criteria

Studies were included if they (1) enrolled adults (≥18 years) with HRF, defined as PaCO₂ greater than 45 mmHg with concomitant pH less than 7.35 (acute) or pH 7.35 or greater (chronic) [10]; (2) compared HFNC with NIV; (3) reported outcomes of mortality or endotracheal intubation rate; and (4) included only RCTs. Cross‐over randomized trials were excluded.

2.3. Outcomes

The primary outcomes were mortality and endotracheal intubation. Mortality was defined as all‐cause mortality at the longest follow‐up reported in each study, with follow‐up periods ranging from 72 h, during hospitalization, 28 days, 30 days, or 90 days to maximize data inclusion across trials. Secondary outcomes included partial pressure of carbon dioxide (PaCO₂), partial pressure of oxygen (PaO₂), length of stay in hospital and intensive care unit (ICU), nasal facial skin breakdown, and comfort score (ranging from 1 to 10, where 1 indicated very comfortable and 10 indicated very uncomfortable, with higher scores indicating greater discomfort). For details for the definitions of these outcomes was in Table S1.

2.4. Search Strategy

We conducted a systematic search of PubMed, Embase, the Cochrane Library, Wanfang, CNKI, and Weipu (VIP) from database inception through March 2026, using the key search terms “HFNC,” “NIV,” and “HRF.” No language restrictions were applied. In addition, supplementary searches were performed by screening the reference lists of relevant studies and previously published systematic reviews and meta‐analyses to identify potentially eligible studies.

2.5. Study Selection and Extraction

Two investigators (LY and SJF) independently screened titles and abstracts identified through the systematic search. Discrepancies in eligibility assessments were resolved through discussion with, or adjudication by, a third reviewer (CKC). Full texts of potentially eligible studies were independently reviewed by the same pair of investigators, with disagreements resolved through discussion.

Data were extracted independently by the same two reviewers (LY and SJF) using a standardized form. Extracted data included study characteristics (including study region and sample size), patient characteristics (age, sex, baseline PaCO₂, PaO₂/FiO₂, and pH), ventilatory settings for HFNC and NIV, and prespecified outcome measures (including the primary and secondary outcomes). When the data were reported as medians, ranges, and/or interquartile ranges (IQRs), the means and standard deviations (SDs) were estimated using the method outlined by Wan et al. [11].

2.6. Assessment of Risk of Bias

The risk of bias for all included studies was assessed using the Cochrane Risk of Bias (RoB 2) tool [12]. Two investigators independently evaluated across the following domains: the randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Each study was rated as having low risk, high risk, or some concerns in each domain. Discrepancies between reviewers were resolved through discussion.

2.7. Data Analysis

Statistical analyses were performed using RevMan (version 5.4) and R (version 4.3.3). Heterogeneity across studies was assessed using the I ² statistic, a fixed‐effects model was used if heterogeneity was low (I ² < 50%). Otherwise, the random‐effects model was applied to perform the pooled analysis. For dichotomous outcomes, we calculated the relative risk (RR) with 95% confidence intervals (CIs). For continuous outcomes, we estimated the mean difference (MD) with 95% CIs.

Several prespecified subgroup analyses were conducted by HRF type (acute HRF vs. chronic HRF), COPD status (all patients were AECOPD vs. not all patients were AECOPD), and post‐extubation (yes vs. no). Sensitivity analyses were conducted using a random‐effect model, excluding the largest trial, and using mortality data within 30 days. Publication bias was assessed using funnel plots. A two‐sided p‐value < 0.05 was considered statistically significant.

2.8. Confidence of Evidence

Two authors (SJF and CKC) used the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach to assess the quality of evidence for primary and secondary outcomes [13]. Discrepancies between reviewers were resolved through consensus, or, when necessary, by consultation with a third reviewer (LY).

3. Results

3.1. Characteristics of Included Studies and Assessment of Risk of Bias

A total of 5744 records were initially identified. After removing duplicates, 5023 records underwent title and abstract screening. Of these, 4558 were excluded as irrelevant, leaving 465 articles for comprehensive assessment of full texts. Finally, 20 studies met the inclusion criteria and were included in the systematic review [14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33]. The study selection process is detailed in Figure 1.

FIGURE 1.

PRISMA flow diagram of the study selection and identification.

The characteristics of the RCTs included in this study are summarized in Tables 1 and S2. These trials, published between 2018 and 2025, enrolled a total of 1835 patients, with sample sizes ranging from 40 to 225 participants. The mean age of participants across studies was 52.60 to 77.0 years, and the proportion of females ranged from 22.7% to 54.5%. The RCTs were conducted in China (n = 13) [15, 16, 17, 19, 21, 23, 24, 25, 26, 28, 29, 30, 33], Greece (n = 2) [22, 27], Brazil (n = 1) [31], Italy (n = 1) [18], South Korea (n = 1) [14], Turkey (n = 1) [32], and the United States (n = 1) [20]. Sixteen studies focused on the treatment of acute HRF [14, 15, 17, 18, 19, 20, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33], while four addressed chronic HRF [16, 21, 22, 23]. Seventeen studies evaluated the efficacy and safety of HFNC compared to NIV in AECOPD patients [14, 15, 16, 17, 18, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33]. Two studies included mixed populations (not all AECOPD) [20, 22], and another did not specify the COPD proportion [19].

TABLE 1.

Characteristics of included studies.

		Sample size, n					PaCO2 baseline, mmHg		PaO2/FiO2 baseline		PH baseline
Study	Study region	Total	HFNC	NIV	Age, years	Female, n (%)	HFNC	NIV	HFNC	NIV	HFNC	NIV
Lee et al. 2018	South Korea	88	44	44	73 (66.5–79)	31 (35.2)	56.4 ± 10.1	52.6 ± 8.8	134.8 ± 7.3	134.4 ± 7.5	7.32 ± 0.28	7.31 ± 0.29
Wang et al. 2019	China	63	32	31	73.40 ± 6.56	31 (49.2)	66.84 ± 3.92	68.35 ± 3.14	NA	NA	7.24 ± 0.02	7.24 ± 0.02
Jing et al. 2019	China	42	22	20	75.73 ± 7.07	NA	53.2 ± 6.7	53.7 ± 8.6	239.2 ± 80.8	250.8 ± 75.8	7.46 ± 0.04	7.44 ± 0.06
Chai et al. 2020	China	66	33	33	58.69 ± 11.54	15 (22.7)	NA	NA	247.10 ± 46.05	250.05 ± 43.03	7.26 ± 0.20	7.30 ± 0.11
Cortegiani et al. 2020	Italy	79	40	39	75.48 ± 12.56	39 (49.4)	73.7 ± 12.8	72.0 ± 13.0	203.2 ± 45.5	222.4 ± 71.0	7.30 ± 0.03	7.29 ± 0.03
Deng et al. 2020	China	80	40	40	52.60 ± 7.29	27 (33.8)	73.28 ± 9.48	73.19 ± 8.49	271.07 ± 10.82	275.92 ± 17.83	NA	NA
Doshi et al. 2020	United States	65	34	31	NA	34 (52.3)	56 (48–72)	64.6 (48–91)	NA	NA	7.33 ± 0.12	7.32 ± 0.10
Lu et al. 2020	China	120	60	60	62.19 ± 10.31	51 (42.5)	70.97 ± 14.25	67.22 ± 16.05	159.81 ± 36.98	151.34 ± 36.34	7.39 ± 0.39	7.39 ± 0.46
Papachatzakis et al. 2020	Greece	40	20	20	77.0 ± 11.0	21 (52.5)	60.4 (9.9)	62.1 (10.3)	NA	NA	7.4 ± 0.1	7.4 ± 0.1
Tan et al. 2020	China	86	44	42	69.87 ± 8.73	36 (41.9)	50.5 (48–57.8)	53 (48.8–61.3)	239.2 ± 47.0	229.3 ± 42.0	7.48 ± 0.07	7.45 ± 0.07
Wang et al. 2021	China	105	50	55	70.97 ± 16.00	36 (34.3)	74.93 ± 12.81	77.71 ± 14.28	115.53 ± 78.60	123.03 ± 69.95	7.25 ± 0.06	7.26 ± 0.05
Liu et al. 2023	China	72	36	36	NA	30 (41.7)	63.97  ±  4.79	63.68  ±  4.81	158.21 ± 20.89	158.38 ± 21.46	7.21 ± 0.03	7.23 ± 0.03
Lu et al. 2024	China	97	49	48	68.09 ± 9.95	39 (40.2)	54.42 ± 3.95	54.53 ± 3.82	NA	NA	7.32 ± 0.07	7.31 ± 0.09
Pantazopoulos et al. 2024	Greece	105	51	54	72.83 ± 9.28	30 (28.6)	60.8 ± 10	65 ± 13	237.78 ± 58.27	238.10 ± 43.19	7.30 ± 0.03	7.29 ± 0.05
Sun et al. 2024	China	112	56	56	57.53 ± 3.43	50 (44.6)	NA	NA	NA	NA	NA	NA
Tan et al. 2024	China	225	113	112	NA	92 (40.9)	63 (59–68)	61 (58–65)	175 ± 23.72	184 ± 25.94	7.31 ± 0.03	7.30 ± 0.03
Wang et al. 2024	China	96	49	47	73.93 ± 7.70	48 (50.0)	76.40 ± 1.65	76.17 ± 2.49	NA	NA	7.28 ± 0.03	7.27 ± 0.02
Maia et al. 2024	Brazil	77	35	42	72.36 ± 10.55	42 (54.5)	61 (52–93)	80 (60–114)	189 ± 108.97	209 ± 107.49	7.32 ± 0.04	7.30 ± 0.09
Haciosman et al. 2025	Turkey	137	90	47	68.1 ± 10.5	47 (34.3)	61.91 ± 9.48	64.69 ± 10.81	NA	NA	7.31 ± 0.04	7.30 ± 0.04
Yang et al. 2025	China	80	40	40	65.45 ± 8.05	30 (37.5)	75.41 ± 6.89	75.21 ± 6.91	NA	NA	7.11 ± 0.06	7.12 ± 0.06

Abbreviations: BiPAP, bi‐level positive airway pressure; FiO₂, fraction of inspired oxygen; HFNC, high‐flow nasal cannula; NA, not applicable; NIV, noninvasive ventilation; PaCO₂, arterial partial pressure of carbon dioxide; PaO₂, arterial partial pressure of oxygen; PEEP, positive end‐expiratory pressure; PSV, pressure support ventilation.

Risk‐of‐bias assessments indicated an overall moderate risk of bias across the included studies. According to the GRADE assessment, the certainty of evidence was moderate for the length of hospital and ICU stay and nasal skin breakdown, low for mortality, endotracheal intubation, and comfort, and very low for PaO₂ and PaCO₂ (Table S3).

3.2. Mortality and Endotracheal Intubation

A total of 13 studies [14, 15, 16, 17, 18, 22, 23, 26, 27, 29, 30, 31, 32], including 1181 patients (613 receiving HFNC treatment and 568 receiving NIV treatment), reported data of mortality. The time points for reported mortality varied across the studies: 72 h in one study [27], in‐hospital mortality in two studies [18, 22], 28 days in seven studies [15, 16, 23, 26, 29, 30, 32], 30 days in one study [14], and 90 days in two studies [17, 31]. There was no significant difference in mortality between HFNC and NIV group (RR, 0.94; 95% CI, 0.69 to 1.28; p = 0.70; I ² = 0%) (Figure 2A).

FIGURE 2.

Forest plot of mortality and endotracheal intubation comparing HFNC and NIV. CI, confidence interval; HFNC, high‐flow nasal cannula; NIV, noninvasive ventilation; RR, risk ratio.

Subgroup analyses by acute or chronic HRF, COPD status, and post‐extubation yielded consistent results (all p for interaction > 0.05, Table 2). Sensitivity analyses using a random‐effects model and excluding the largest trial yielded results consistent with the main analysis. Importantly, restricting mortality to follow‐up within 30 days, thereby excluding longer‐term data, produced similar estimates, supporting the robustness of the findings despite variability in mortality time windows across trials (Table S4).

TABLE 2.

Subgroup analysis of mortality and endotracheal intubation.

Subgroup	Number of studies	Number of patients	RR (95% CI)	I 2 (%)	p for overall effect	p for interaction
Mortality
HRF type
Acute HRF	10	1013	0.90 (0.64, 1.28)	0	0.57	0.63
Chronic HRF	3	168	1.09 (0.56, 2.15)	0	0.80
All patients were COPD
Yes	12	1141	0.94 (0.68, 1.29)	0	0.70	0.93
No	1	40	1.00 (0.23, 4.37)	/	/
Post‐extubation
Yes	2	128	1.12 (0.52, 2.39)	0	0.77	0.62
No	11	1053	0.91 (0.65, 1.28)	0	0.59
Endotracheal intubation rate
HRF type
Acute HRF	16	1527	0.79 (0.57, 1.11)	39	0.18	0.21
Chronic HRF	3	248	1.14 (0.72, 1.80)	0	0.57
All patients were COPD
Yes	18	1710	0.89 (0.67, 1.17)	31	0.40	0.27
No	1	65	0.36 (0.08, 1.75)	/	/
Post‐extubation
Yes	2	128	1.06 (0.41, 2.77)	0	0.90	0.64
No	17	1647	0.84 (0.62, 1.14)	38	0.26

Abbreviations: CI, confidence interval; COPD, chronic obstructive pulmonary disease; I ², heterogeneity index; RR, risk ratio.

Nineteen studies reported the endotracheal intubation rate [14, 15, 16, 17, 18, 19, 20, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33]. We found no statistically significant difference in the rate of endotracheal intubation rate between HFNC and NIV (RR, 0.87; 95% CI, 0.71 to 1.06; p = 0.17; I ² = 31%; Figure 2B) and the finding remained robust in subgroup (Table 2) and sensitivity analyses (Table S4).

3.3. PaCO₂ and PaO₂

PaCO₂ data were reported in fourteen studies [14, 16, 18, 19, 20, 21, 22, 23, 24, 25, 26, 29, 30, 33] and PaO₂ data were reported in eleven studies [14, 17, 19, 20, 21, 22, 24, 25, 26, 30, 33]. No significant differences were observed between groups in PaCO₂ (MD, –0.87 mmHg; 95% CI, –2.98 to 1.25; p = 0.42; I ² = 89%; Figures  3A ) or PaO₂ (MD, 2.67 mmHg; 95% CI, –0.66 to 6.01; p = 0.12; I ² = 95%; Figures  3B ). Subgroup analyses showed no significant differences in PaCO₂ across HRF type or post‐extubation subgroups (Table S5). In 2 studies not limited to AECOPD (n = 92), HFNC was associated with greater PaCO₂ reduction than NIV (MD, –7.69 mmHg; 95% CI, –13.62 to –1.76; p = 0.01; I ² = 32%). For PaO₂, subgroup analyses by HRF type and COPD status showed no significant differences between HFNC and NIV (Table S5). In patients with acute HRF, HFNC was associated with a modest but statistically significant improvement in PaO₂ compared with NIV (MD, 3.79 mmHg; 95% CI, 0.08 to 7.49; p = 0.05; I ² = 96%). No significant differences were observed in other subgroups.

FIGURE 3.

Forest plots of PaCO₂ and PaO₂ comparing HFNC and NIV. (A) PaCO₂; (B) PaO₂. CI, confidence interval; HFNC, high‐flow nasal cannula; NIV, noninvasive ventilation; PaCO₂, arterial partial pressure of carbon dioxide; PaO₂, arterial partial pressure of oxygen; SD, standard deviation.

3.4. Length of Stay in Hospital and ICU

Eight studies reported the length of stay in hospital [15, 18, 20, 22, 23, 26, 29, 32] and five studies reported the length of stay in ICU [15, 16, 20, 23, 29]. The pooled analysis found that compared with receiving NIV, patients receiving HFNC had a significantly shorter hospital stay (MD, –0.69 days; 95% CI, –1.08 to –0.30; p < 0.001; I ² = 0%; Figure 4A) and ICU stay (MD, –0.98 days; 95% CI, –1.50 to –0.45; p < 0.001; I ² = 23%; Figure 4B).

FIGURE 4.

Forest plots of arterial blood gas parameters comparing HFNC and NIV. (A) length of stay in hospital; (B) length of stay in ICU; (C) nasal facial skin breakdown; (D) comfort. CI, confidence interval; HFNC, high‐flow nasal cannula; NIV, noninvasive ventilation; SD, standard deviation.

3.5. Nasal Facial Skin Breakdown and Comfort

Four studies reported the differences in nasal facial skin breakdown between the two groups [15, 23, 26, 29]. HFNC group significantly reduced the incidence of nasal facial skin breakdown in HRF patients compared with NIV (RR, 0.20; 95% CI, 0.09 to 0.45; p < 0.001; I ² = 0%; Figure 4C). Additionally, three studies reported comfort scores [16, 23, 29]. One study utilized a 1–10 point scale, where one indicated very comfortable and ten indicated very uncomfortable [16]. The other two studies used a modified 10‐cm visual analog scale, where one indicated very uncomfortable and 10 indicated very comfortable [23, 29]. To ensure consistency, we adjusted the scores from the latter two studies by subtracting them from 11, so that higher values indicated greater discomfort. Compared with NIV, HFNC treatment significantly improved patient comfort (MD, –1.19; 95% CI, –1.98 to –0.41; p = 0.003; I ² = 70%; Figure 4D).

3.6. Publication Bias

Funnel plots showed no evidence of asymmetry for mortality, endotracheal intubation, the length of hospital and ICU stay, and nasal facial skin breakdown. However, asymmetry was observed for PaCO₂, PaO₂, and patient comfort.

4. Discussion

In this meta‐analysis of twenty randomized controlled trials involving 1835 patients with HRF, there was no significant difference in mortality (RR, 0.94; 95% CI, 0.69 to 1.28, low certainty), endotracheal intubation rate (RR, 0.87; 95% CI, 0.71 to 1.06, low certainty), PaCO₂ (MD, –0.87 mmHg; 95% CI, –2.98 to 1.25, very low certainty), or PaO₂ (MD, 2.67 mmHg; 95% CI, –0.66 to 6.01, very low certainty) between the HFNC group and the NIV group. These results are consistent with accumulating evidence showing no statistically significant differences between HFNC and NIV across key clinical and physiological outcomes in the management of HRF. Evidence is currently insufficient to establish noninferiority or equivalence. Notably, this is the first meta‐analysis to demonstrate that HFNC was associated with significantly shorter length of stay in hospital (MD, –0.69 days; 95% CI, –1.08 to –0.30, moderate certainty) and ICU (MD, –0.98 days; 95% CI, –1.50 to –0.45, moderate certainty), a lower incidence of nasal facial skin breakdown (RR, 0.20; 95% CI, 0.09 to 0.45, moderate certainty), and improved patient comfort (MD, –1.19; 95% CI, –1.98 to –0.41, low certainty) compared with NIV.

Among patients with de novo acute hypoxemic respiratory failure, HFNC may be preferred over NIV due to its potential association with lower mortality, greater ease of use, and better tolerability [34, 35]. In contrast, for acute HRF secondary to COPD exacerbations, NIV remains the first‐line intervention [3, 4], although a substantial proportion of patients experience treatment failure due to poor tolerance [6]. HFNC could be used either between NIV sessions or in patients with mild‐to‐moderate hypercapnia who are unable to tolerate NIV [36]. Several previous meta‐analyses have evaluated the comparative effectiveness of HFNC compared with NIV. Huang et al. [37] reported no significant differences between HFNC and NIV in outcomes such as endotracheal intubation, mortality, oxygenation, PaCO₂ and pH improvement, respiratory distress relief, ICU length of stay. Similar findings were observed in studies by Ovtcharenko et al. [8] and Fahey et al. [9].

In the current meta‐analysis, we included newly published RCTs and expanded our search to Chinese databases to minimize language bias and ensure comprehensive coverage. We applied more stringent inclusion and exclusion criteria, limiting the analysis to parallel‐group RCTs to enhance methodological rigor. The recent trials accounted for 81.7% (1500 of 1835) of the total study population and the outcome measures demonstrated narrow CIs.

For physiological mechanisms, HFNC could generate a low level of positive airway pressure [38], which may improve ventilation‐perfusion matching and counterbalance intrinsic positive end‐expiratory pressure (PEEP) [39]. Additionally, HFNC delivered a constant, adjustable flow of blended air‐oxygen at high velocity, flushing nasopharyngeal anatomical dead space, thereby improving alveolar ventilation [40, 41]. The current meta‐analysis showed no statistically significant difference between HFNC and NIV in mortality, endotracheal intubation rate. Subgroup analysis did not find a statistically significant difference in the effect of HRF type, COPD status, and post‐extubation status. For PaCO₂ and PaO₂, substantial heterogeneity was observed. This variability likely reflects differences in baseline respiratory severity, ventilatory settings, patient populations, and timing of arterial blood gas measurement across studies. Therefore, pooled estimates for these outcomes should be interpreted cautiously and considered exploratory rather than definitive evidence of comparable effects.

A further important finding was HFNC significantly decrease hospital and ICU length of stay. This reduction may be attributable to several factors. First, the warmed and humidified gas in HFNC enhanced secretion clearance and reduced upper airway resistance and work of breathing [41, 42]. Second, in contrast to NIV, which is often associated with claustrophobia, patient‐ventilator dyssynchrony [43], and nasal facial skin breakdown, HFNC is better tolerated and significantly improves treatment adherence [44]. These advantages collectively enhance patient comfort, a finding robustly supported by this meta‐analysis.

The external validity of these findings should also be considered. The majority of included trials enrolled patients with acute exacerbations of COPD and predominantly mild‐to‐moderate hypercapnic respiratory failure. Patients with severe acidosis, marked hypercapnia, or advanced respiratory failure were generally excluded. Consequently, extrapolation to patients with severe disease requires caution. Furthermore, evidence remains limited for HRF due to non‐COPD etiologies, such as obesity hypoventilation syndrome, neuromuscular disorders, chest wall deformities, or central hypoventilation. Therefore, extrapolation of these findings to broader HRF populations should be undertaken cautiously.

This meta‐analysis has several limitations. First, heterogeneous mortality time points required using the longest follow‐up in each study, which may obscure time‐dependent effects, limiting our pooled estimate to an overall rather than time‐specific treatment effect. Second, most participants in this analysis were patients with acute exacerbation of COPD, and evidence for non‐COPD causes of HRF remains limited, which may partially limit the generalizability of the findings.

In conclusion, in patients with HRF, HFNC showed no statistically significant difference from NIV in mortality or endotracheal intubation, but was associated with shorter hospital and ICU stays, and improved patient comfort. Further large, high‐quality trials in diverse HRF populations are warranted to evaluate the noninferiority of HFNC compared with NIV.

Funding

This work was supported by the Noncommunicable Chronic Diseases‐National Science and Technology Major Project (2023ZD0506104/2023ZD0506100), Sichuan Science and Technology Program (2024NSFSC1528), and National Natural Science Foundation of China (92159302).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File 1: jebm70139‐sup‐0001‐SupMat.doc