Oxygen supplementation is indispensable for maintaining the appropriate tissue oxygenation of patients with acute respiratory failure. Such patients require a high inspiratory flow that may vary from 30 to 120 l/min [1]. Conventional oxygen delivery methods include nasal cannula, simple mask, Venturi mask, and rebreathing mask. However, these devices can only provide a maximum oxygen flow of 6–15 l/min, which may be inadequate for patients with acute respiratory failure.

The high-flow nasal cannula (HFNC) is a novel oxygen supply device in a high-flow oxygen system that can deliver up to 100 % heated and humidified oxygen via a wide-bore nasal cannula at a maximum flow of 60 l/min [2]. It was recently reported to have beneficial effects on the clinical signs and oxygenation levels of patients with acute respiratory failure in intensive care units (ICUs) and emergency departments [3, 4]. Moreover, HFNC is superior to conventional oxygen therapies in terms of improving the dyspnea of extubated patients or patients with heart failure [5, 6].

In a recent randomized controlled study, HFNC provided to severe respiratory failure patients reduced the intubation rate and lowered ICU mortality [7]. Besides, HFNC might avoid the need for invasive mechanical ventilation [3, 8]. However, failure of HFNC may cause delayed intubation and increased mortality, like noninvasive ventilation (NIV) [9]. The primary objective of this study was ICU mortality according to the intubation timing in respiratory failure patients who received HFNC that failed. Secondary objectives were ventilator weaning, 14- and 28-day mortality, and length of ICU stay. We presented this study at the Congress of the Asian Pacific Society of Respirology [10].


Study design and subjects

We performed a retrospective observation study of critically ill patients older than 18 years who underwent intubation after HFNC failure and were treated in the ICU of Asan Medical Center, Korea, from January 2013 to March 2014. Considering the time to define early and late in NIV study [9], the cutoff value (46 h) analyzed by the receiver-operating characteristic curve, and practicality, patients with HFNC failure were divided into early and late groups based on whether intubation started before or after 48 h HFNC initiation. We compared the outcomes of both groups and adjusted our model using a propensity analysis. The study was approved by the Institutional Review Board of Asan Medical Center (IRB no. 2014-0803). Informed consent was not sought from patients because of the retrospective nature of the study.

HFNC device

In our hospital, patients with respiratory failure receive high-flow oxygen therapy via HFNC device (Optiflow™, Fisher & Paykel Healthcare, Auckland, New Zealand). This device is composed of an air-entrainment device (MaxVenturi™, Maxtec), heated humidification system (MR850 heated humidifier), heated humidification chamber (MR290 autofeed chamber), high-performance breathing circuit (RT202 single-limb adult breathing circuits kit), and a unique wide-bore nasal cannula (OPT 844 Optiflow™ nasal cannula).

Monitoring of patients

All ICU patients usually receive continuous monitoring and multidisciplinary care involving three physicians (one resident, one intensivist, and one professor), nurses, pharmacists, respiratory therapists, and other divisions related to patient treatments. Patients in general wards are also managed by a medical emergency team with an electronic medical record-based screening system and a skilled intervention group (one intensivist, one senior resident, and two nurses) [1113]. The medical emergency team has been carrying out proactive daily rounding of critically ill patients since 2008.

HFNC application and intubation

We routinely consult with an intensivist for HFNC application or intubation in patients with respiratory failure. HFNC is given to the following patients: those who exhibit hypoxia that requires a conventional oxygen device to deliver >9 l/min to achieve an oxygen saturation (SpO2) of >92 %, those who show persistent signs of respiratory distress (respiration rate >24 breaths/min, accessory respiratory muscle use, and thoracoabdominal dyssynchrony) despite adequate oxygen supplementation [3], and those at risk of respiratory failure after extubation [14]. HFNC is not used in patients who require immediate intubation or have hypercapnia.

HFNC failure is defined as a need for endotracheal intubation despite HFNC application. We usually report the complete information to the patients and/or their families, communicating with them before intubation [15, 16]. After consent to the intubation, patients undergo rapid sequence intubation according to the protocol [17] and receive analgesics, sedative agents, and neuromuscular blocking agents based on the pain, agitation, and delirium guideline [8, 18]. The predetermined HFNC failure criteria are: hypoxemic respiratory failure with the patient expected to fail to maintain an SpO2 <90 % despite receiving the maximal fraction of inspired oxygen (FiO2) allowed by the HFNC; hypercapnic respiratory failure accompanied by pH <7.3; respiration rate >35 breaths/min combined with respiratory distress; uncontrolled metabolic acidosis with hypotension (systolic blood pressure <90 mmHg or mean blood pressure <65 mmHg) despite fluid challenge and/or high-dose vasopressors; need for airway protection because of altered mental state or aspiration; cardiopulmonary arrest; or need for an operation to control the underlying disease [19].


Successful extubation was defined as maintenance of spontaneous breathing for >48 h after interruption of mechanical ventilation with extubation. Ventilator weaning was defined as spontaneous breathing without a mechanical ventilator for >48 h in patients with tracheostomy or successful extubation. Use of immunosuppressive agents was defined as treatment with immunosuppressive drugs and/or a cumulative dose of >1,680 mg prednisolone equivalent within 6 months of HFNC [20]. Acute Physiology and Chronic Health Evaluation (APACHE) II and Sequential Organ Failure Assessment (SOFA) scores were calculated by using the worst variable within the first 24 h of ICU admission.

Data collection

One respirologist reviewed all patient data (including baseline characteristics, severity indexes such as APACHE II and SOFA scores, etiology of respiratory failure, extubation and ventilator weaning, length of ICU stay, infectious and noninfectious complications, and cause of death) more than twice, and two professors in the respiratory division also rechecked the findings for accuracy.

The etiologies of respiratory failure before HFNC application were classified into six groups using a system modified from Demoule et al. [21]: acute de novo respiratory failure (i.e., pneumonia and acute respiratory distress syndrome), acute-on-chronic lung disease (i.e., chronic obstructive pulmonary disease and tuberculosis-destroyed lung), cardiogenic pulmonary edema, pulmonary edema due to renal failure, septic shock other than respiratory infection, and after extubation.

Statistical analysis

The data are expressed as median (interquartile range), mean ± standard deviation, or numbers (percentage). Categorical variables were compared using a chi-square test or McNemar test for matched patients. Continuous variables were compared using a Mann-Whitney U test or Wilcoxon signed-rank test for matched patients. P < 0.05 was considered to indicate statistical significance.

To reduce the effect of early and late HFNC therapy bias and potential confounding in this observational study, we conducted careful adjustment for differences in baseline characteristics (Table 1) using a propensity analysis [22]. Propensity scores were estimated by multiple logistic regression analysis (Table E1 in the Electronic Supplementary Material) by means of the stepwise backward elimination methods with P < 0.20.

Table 1 Baseline characteristics and hospital outcomes for patients who were intubated after HFNC therapy failure

The individual propensity score was integrated into each outcome model as a covariable. To compare hospital outcomes between early and late HFNC failure groups, we performed logistic regression (overall ICU mortality, extubation success, ventilator weaning, and 14- and 28-day mortality), negative binomial regression (length of ICU stay), and Cox proportional hazard regression (ventilator-free days) analysis. We also performed propensity score matching. To develop propensity score-matched pairs without replacement (a 1:1 match), the Greedy 5/1 digit match algorithm was used as reported previously [23, 24]. After all of the propensity score matches had been performed, we evaluated the balance in baseline covariables between the two groups. In the propensity score matched cohort, we conducted generalized estimating equations using logistic regression (overall ICU mortality, extubation success, ventilator weaning, and 14- and 28-day mortality) and negative binomial regression (length of ICU stay) to account for the clustering nature of matched pairs. Also, the Cox proportional hazard model (ventilator-free days) with the robust sandwich variance estimator was conducted. Analysis of results was performed using SPSS statistical software (version 21; SPSS, Chicago, IL, USA) and SAS software (version 9.3; SAS Institute, Cary, NC, USA).


During the study period, 616 consecutive patients received HFNC at Asan Medical Center; 177 critically ill patients were intubated after HFNC failure after excluding 439 patients: those who improved and were weaned from HFNC (n = 341), those who deteriorated but were not intubated (n = 85), and those who used another oxygen device before intubation (n = 13). We further excluded two patients who were not transferred to the ICU. Of the remaining 175 patients, 130 underwent intubation before 48 h and 45 underwent intubation after 48 h of HFNC application (Fig. 1).

Fig. 1
figure 1

Distribution of the study population according to HFNC therapy outcome

Patient characteristics

The characteristics of all 175 patients and both groups are shown in Table 1. Many patients had underlying diseases such as immunosuppressive agents use [40.0 % (70/175)], hematological malignancies [31.4 % (55/175)], and solid malignancies [24.6 % (43/175)]. The SOFA score was higher in the early HFNC failure group than in the late HFNC failure group.

Etiology of respiratory failure

The main etiologies of respiratory failure indicating HFNC application are presented in Table 1. The most common etiologies for respiratory failure were acute de novo respiratory failure [33.1 % (43/130)] and acute-on-chronic lung disease [35.6 % (16/45)] in the early and late HFNC failure groups, respectively. However, there was no significant difference between the two groups. The causes of HFNC application and intubation are shown in Table E2 in the Electronic Supplementary Material.

ICU outcomes

The early and late groups did not differ significantly in terms of 14-day mortality, 28-day mortality, or length of ICU stay. However, the early group had better ICU outcomes than the late group (Table 1): they were significantly more likely to have better overall ICU mortality (39.2 vs. 66.7 %; P = 0.001), extubation success (37.7 vs. 15.6 %; P = 0.006), ventilator weaning (55.4 vs. 28.9 %; P = 0.002), and ventilator-free days to day 28 (8.6 ± 10.1 vs. 3.6 ± 7.5; P = 0.001).

Propensity score adjusted and matched ICU outcomes

We confirmed the discrimination and calibration of the model by assessing the c-statistics (c = 0.766) and Hosmer-Lemeshow statistics (P = 0.353), respectively. After propensity score matching, we selected 37 pairs of patients with similar baseline characteristics, severity indexes (PaO2/FiO2, APACHE II score, and SOFA score), and etiology of respiratory failure (Table 2). Propensity score-adjusted and -matched analysis showed that intubation before 48 h of HFNC was associated with better overall ICU mortality [adjusted odds ratio (OR) = 0.317, P = 0.005; matched OR = 0.369, P = 0.046], extubation success (adjusted OR = 3.091, P = 0.020), ventilator weaning (adjusted OR = 3.380, P = 0.004; matched OR = 2.495, P = 0.041), and ventilator-free days [adjusted hazard ratio (HR) = 0.516, P = 0.001; matched HR = 0.639, P = 0.026] (Table 3).

Table 2 Baseline characteristics of the early and late HFNC failure group after matching on propensity score
Table 3 Analysis of hospital outcomes for the early HFNC failure group compared with the late HFNC failure group (as reference) using the propensity score analysis

Complications on ICU admission and during ICU stay and events leading to death

As shown in Table 4, the most frequent presenting complication on ICU admission in both groups was pneumonia (66.9 vs. 80.0 %), and it was the leading cause of death in the ICU [29.6 % (24/81) vs. 21.6 % (11/51)]. During ICU stay, renal failure was the most frequent complication in both groups (25.4 vs. 37.8 %). Most complications during ICU stay were more frequent in the late HFNC failure group than in the early HFNC failure group, but there were no significant differences between the two groups except for cardiac dysfunction (10.8 vs. 31.1 %; P = 0.001).

Table 4 Serious complications presented on ICU admission and occurring during ICU stay and events leading to death in patients with HFNC failure


Our current study findings suggest that early intubation (i.e., within 48 h of starting HFNC) is associated with lower overall ICU mortality than late intubation (i.e., after 48 h of starting HFNC) when patients had received HFNC therapy that failed. In addition, early intubation was found to be associated with better extubation success and ventilator weaning and more ventilator-free days than late intubation. Although several previous studies have assessed the failure of NIV in patients with respiratory failure [9, 25], our present study is, to our knowledge, the first to assess HFNC failure. Moreover, our present study sought to identify the proper HFNC management strategy by analyzing patients with HFNC failure.

HFNC has many advantages as a treatment of acute respiratory failure. First, the heated humidifier system of HFNC improves mucociliary function and secretion expectoration [2] and decreases the metabolic cost of gas conditioning [26]. Second, the high flow achieved by HFNC can improve patient oxygenation [26], generate low-level positive airway pressure [27], reduce the respiration rate of patients [28], attenuate inspiratory resistance [26], and supply a constant FiO2 [29]. Third, HFNC allows the patient to speak and cough and permits oral intake [4]. It is also better tolerated and more comfortable than a face mask [2, 30]. Fourth, HFNC therapy does not increase the risk of complications such as pneumonia, barotraumas, or secondary infections [31].

Despite the advantages of HFNC in patients with respiratory failure, its inappropriate use can be hazardous to patient health. One study has stated that patients who are not intubated tolerate HFNC for long periods and that HFNC is safe for managing patients with respiratory failure for several days [32]. However, our present study findings contradict those conclusions. Thus, extended use of HFNC before intubation in patients with respiratory failure may be harmful.

An earlier study on NIV therapy reported that, of patients who were admitted to the ICU for acute respiratory failure, those who immediately underwent invasive mechanical ventilation had lower in-hospital mortality than those who only underwent invasive mechanical ventilation after NIV failure [33]. A review of Ozyilmaz et al. [25] reported that NIV failure was strongly associated with poor outcomes and that prompt endotracheal intubation should be performed if early or late signs of deterioration are detected. Moreover, Moretti et al. found that late NIV failure (>48 h after starting NIV) was associated with particularly high mortality (67.7 %) and poor prognosis [9]. Similarly, in our present study, the time to define early and late was 48 h, and late HFNC failure was associated with higher overall ICU mortality compared with early HFNC failure. This result may be because prolonged intubation delay in patients with uncontrolled disease can induce respiratory muscle fatigue and cardiac dysfunction, which in turn lead to poor hospital outcomes.

Our data showed that there were no statistical differences in 14- and 28-day mortality, but overall ICU mortality was higher in the late HFNC failure group than early HFNC failure group. This was probably because more patients in the early HFNC failure group died following ICU discharge and more patients in the late HFNC failure group died in the ICU after prolonged ventilation of more than 28 days. Although we analyzed hospital outcomes according to HFNC failure, we could not determine how to anticipate the patients with late HFNC failure. There were no statistical differences in most baseline characteristics between the early and late HFNC failure groups. Diabetes mellitus and the SOFA score were somewhat higher in the early HFNC failure group. Future prospective randomized studies are certainly warranted to address this question.


This study had several limitations. First, it was a retrospective observational study. However, we used propensity score matching and evaluated the influence of HFNC failure in a group of carefully selected patients to identify a possible link between late HFNC failure and poor hospital outcomes. Our retrospective data on early and late HFNC failure may represent a major departure from the data that would be available for such a comparison from a randomized controlled study. Second, the study was conducted in a single tertiary referral center, although we recruited many patients. Selection bias cannot be excluded, and the results should be carefully interpreted. Larger scale multicenter studies are required to confirm the results. Third, the actual delivered FiO2 was not measured; we cannot definitively state that the FiO2 with HFNC was truly higher than the FiO2 with conventional oxygen devices. The PaO2/FiO2 data should be interpreted with caution. Nevertheless, we believe that HFNC does deliver more FiO2 than conventional devices because the oxygenation and respiratory variables of our patients improved when HFNC was applied. Fourth, our study focused on patients with HFNC failure, not patients who had been weaned off HFNC. Therefore, there is a possibility, albeit low, that we might have missed patients with re-respiratory failure after HFNC weaning. Fifth, the patients included in our study exhibited high overall ICU mortality, reflecting the high frequency of poor prognosis factors such as immunosuppressive agent use [34], hematological malignancy [35], and solid malignancy [36]. The patients also frequently had low PaO2/FiO2 ratios and high APACHE II scores compared with patients in previous studies [9, 14, 37]. The study protocol, which stipulated that only patients with treatment failure should be enrolled, is likely to have contributed to the high mortality rate. Finally, we are unable to exclude the possibility of not keeping the predetermined criteria of HFNC failure and need for intubation because of the retrospective study. However, this was less likely because all our patients received management by a skilled medical emergency team, and they underwent continuous monitoring or screening [13].


HFNC is a novel and attractive oxygen device for patients with respiratory failure. However, its inappropriate use may lead to adverse outcomes. In our study, failure of HFNC late (>48 h) after its initiation was associated with significantly higher overall ICU mortality, poorer extubation success and ventilator weaning, and fewer ventilator-free days. Our findings suggest that larger prospective randomized trials on HFNC failure in respiratory failure patients are warranted.