Identifying predictive patients’ characteristics in case of High Flow Nasal Cannula failure when used for post-extubation failure

Background: High Flow Nasal Cannula (HFNC) is a relatively new but broadly used type of oxygen therapy. Hence, not much is known about HFNC in the setting of post-extubation failure. Aim of this study is to identify patients’ characteristics that predict failure of HFNC when used for post-extubation failure. Methods: This retrospective, observational study was conducted in a Dutch ICU. Between 2008 and 2014 all subjects aged 18 and older who started with HFNC due to respiratory failure within 2 to 72 hours after extubation were included. Primary outcome was patient characteristics predicting failure of HFNC. Failure was defined as reintubation or death following HFNC. Success was defined as the opportunity to stop HFNC after a non-specified time. Secondary outcome was the difference in length of stay (LOS) and mortality between the success and failure group. Results: A total of 246 subjects were included; in 135 (55%) cases HFNC failed. The success and failure group were comparable in terms of age, primary diagnosis and duration of mechanical ventilation prior to extubation. HFNC was started median 14 (min-max: 2-71) after extubation. Significant differences were found for breathing frequency (success: 21 (9-45)/min vs failure: 24(7-45)/min, p=0.009) and the presence of acute kidney injury (51% vs. 79%, p=0.003). There was a non-significant difference in pH pH (7.42(7.27-7.56) vs. 7.41 (7.13-7.58), p=0.08) After logistic breathing frequency (p=0.02) and pH (p=0.01) remained independent predictors of HFNC failure. LOS at the ICU after starting HFNC differed (success: median 131, min-max: 12-1432 hours vs. failure: 250 (23-4726) hours p<0.001. ICU and hospital mortality were not significantly different between groups. Conclusion: In more than half of the patients HFNC failed when used for postextubation failure. Breathing frequency and pH were predictors of HFNC failure.

3 High-flow Nasal Cannula (HFNC) is a relatively new therapy and is being used increasingly as primary therapy or as an alternative for non-invasive respiratory support. HFNC delivers a heated and humidified mixture of air and oxygen at high flow rates. HFNC flushes anatomical dead space and improves respiratory efficiency. 1 2 In addition it generates positive airway pressure and enables to deliver a more accurate FiO₂. HFNC is known to be used in many different settings, including to prevent intubation in patients with respiratory insufficiency 3 4 or as an alternative for patients with respiratory failure who have a limitation regarding invasive mechanical ventilation.
Currently, non-invasive ventilation (NIV) plays an important role in the prevention of respiratory failure post-extubation 5 , but recent studies showed an increased mortality if NIV was used in the setting of post-extubation respiratory failure. 6 As a result a new discussion is emerging on the use of HFNC following extubation to prevent respiratory failure. 7 8 9 HFNC seems to be a reliable alternative for the prevention of re-intubation in high-risk patient by starting HFNC immediately following extubation. 3 8 9 10 However, not much is known about HFNC in the setting of post-extubation failure and thus, HFNC is used for post-extubation failure without solid scientific evidence about the outcomes. As far as we know, only one small study has been published on this subject. 11 As it is used nowadays on large scale in the setting of post-extubation failure without clear evidence, it is important to identify the characteristics that predict success or failure of HFNC, and examine the outcomes of HFNC failure. Especially because HFNC failure in the setting of acute hypoxemic respiratory failure is associated with the increased mortality 12 . Therefore, the aim of this study is to identify patient's characteristics that predict failure of HFNC in the setting of acute respiratory failure after extubation. all subjects aged 18 and older who started with HFNC due to respiratory failure within 2 to 72 hours after extubation were included. Subjects who had respiratory failure within 0 to 2 hours after extubation were excluded to guarantee that only subjects with new respiratory problems were included. Other exclusion criteria were any limitation regarding mechanical ventilation (invasive or non-invasive) at the initiation of HFNC or during the study period.
The study was approved by the regional medical ethical committee, a waiver was given for informed consent due to the retrospective and non-invasive nature of this study.

Study design
Data on subjects' characteristics, length of stay (LOS) and mortality, were obtained from the Patient Data Management System (PDMS). The different variables in the pre-HFNC setting were collected 1 to 4 hours prior to the start of HFNC. Due to the registration method there is no PaO 2 /FiO 2 ratio available, but merely a SpO₂/FiO₂ ratio. In case of conventional oxygen therapy FiO2 was set on 24-40% (O 2 1-6 l/min) for nasal canula, 40-60% (O 2 6-12 l/min) for oxygen mask, the FiO 2 set on the valve of the Venturi-mask and 80% for non-rebreathing mask. Data on reason for admission, primary diagnosis or circumstance in which the respiratory failure occurred, comorbidities and predicted mortality were obtained from the Minimal Data Set (MDS) of the National Intensive Care Evaluation (NICE) in the Netherlands, which is based on the APACHE IV (Acute Physiology And Chronic Health Evaluation) risk model 13 .
Primary outcome of this study were patient characteristics predicting failure of HFNC.
Failure was defined as the need of reintubation or death within 72 hours after the start of HFNC treatment. Success was defined as the opportunity to stop HFNC after a nonspecified time. Secondary outcome was the difference in LOS and mortality between the success and failure group. The necessity of re-intubation was judged by the attending physician. We took death as part of the definition of HFNC failure as well as secondary outcome. The underlying argument to also make death within 72 hours after the start of HFNC treatment part of the definition of HFNC failure was the fact that otherwise subjects who failed HFNC but were not reintubated and died subsequently, had to be considered, by definition, as HFNC success which is not the case.
Subjects were categorized according to the underlying cause or the circumstance in which the respiratory failure occurred, resembling the different categories in the minimal data set: pneumonia, congestive heart failure (CHF), COPD, post-surgery, other respiratory and other non-respiratory. The other respiratory group included subjects with pulmonary embolus, lung cancer, atelectasis and pleural effusion. The other non-respiratory group included subjects with pancreatitis, sickle cell crisis, chest trauma, in the post-CPR setting, gastrointestinal sepsis, acute renal failure, amyotrophic lateral sclerosis and Guillain-Barre Syndrome. All aforementioned diagnosis were not strictly defined, but were found to be present if registered as such in the MDS.

Statistical analysis
Statistical analyses were performed using SPSS, version 22 Command Syntax References (Chicago, IL, USA). Descriptive statistics were calculated for all variables. The relations between variables and endpoints were tested using Mann-Whitney U tests, independent sample t-tests, Chi-square or one-way ANOVA, depending on the origin and distribution of the date. For all analyses, a p-value p< 0.05 was considered statistically significant.
Variables (p<0.1) associated with reintubation after starting HFNC were assessed by means of multivariate logistic-regression analyses with the use of a backward-selection procedure. Variables representing less than 10% of the population were not used. Post hoc subgroup analyses were conducted for variables associated with re-intubation. Were appropriate data is presented as median (minimum-maximum) or mean (±SD).

Results
A total of 246 subjects were eligible for inclusion. Patient characteristics are presented in table 1. Age of this cohort was (median (min-max) 71 (21-90) years, 147 subjects (60%) were male. The main primary reason for admission to the ICU was post-surgery (postcomplex surgery, complications of surgery or non-surgical complications in a surgical patient). Acute kidney injury (AKI) and an immunocompromised status were the main comorbidities according to APACHE IV criteria. The in-hospital mortality was lower than the predicted mortality by APACHE IV. Table 1 Baseline characteristics of subjects treated with HFNC for acute respiratory failure <72h after extubation Length of stay in ICU (hours), median (min-max) 182  Mortality in ICU, n (%) 28 (11) Mortality in hospital, n (%) 43 (18)

Discussion
The present study describes the differences in patients' characteristics between subjects that are being successfully treated with HFNC in the setting of post-extubation acute respiratory failure and subjects in which HFNC fails. HFNC was successful in 45% of the patients and could be stopped after a mean 44 hours. Independent predictors of HFNC failure were breathing frequency and blood gas pH at the start of HFNC. HFNC failure was associated with prolonged stay in ICU and there was a tendency to increased mortality.
To our knowledge our study is the largest study in literature documenting the use of HFNC for acute respiratory failure after extubation. There is one smaller study of Yoo et al., comparing HFNC and NIV therapy in 73 subjects with acute respiratory failure after extubation. 11 They showed HFNC was non-inferior to NIV in the avoidance of reintubation and associated with a shorter stay in the ICU. Reintubation could be avoided in 79% (n=27) of the patients treated with HFNC, which is a much higher rate compared to the reintubation avoidance rate of 45% in our cohort. In terms of age, sex and APACHE II score both cohorts are comparable. In their study the duration of mechanical ventilation prior to extubation was longer (132.7 ± 85.6 hours) and the time between extubation and start of HFNC was shorter (8.6 ± 11.8 hours) compared to our cohort. Based on the data in the article of Yoo there is no clear explanation for the difference in reintubation avoidance rate between the two studies, except for the difference in methodological design.
In contrast to the scarcity of studies on the use of HFNC in the setting of post-extubation acute respiratory failure (ARF), there are numerous studies on the use of HFNC prior to (or the prevention of) intubation. Our success rate of 45% is comparable with the results of Rello who found a success rate of 45% (n=9) in patients treated with HFNC due to severe acute respiratory infection (2009 influenza A/H1N1) 14 , but less successful in comparison with Messika who found a HFNC success rates of 58% (n=26) in the setting of ARDS. 15 Nevertheless, not many studies are available exclusively focussing on the success or failure of HFNC in the setting of ARF. One of the largest randomized trials on HFNC compared the intubation rate of patients with acute hypoxemic respiratory failure treated their patient with either HFNC, conventional oxygen therapy (COT)or NIV and none of these three strategies showed any significant differences in intubation rate (38% vs. 47% vs. 50% respectively, p =0.18 for all comparisons). 3 Primary outcome of our study showed a significant difference between the success and failure group for breathing frequency before the start of HFNC. After logistic regression breathing frequency remained an independent predictor of HFNC failure.
Previous, physical measurements on the use of HFNC revealed a significant reduction in median breathing frequency compared with non-rebreathing mask (NRM). 16 This was confirmed by Sztrymf who not only found a significant reduction in breathing frequency (p= 0.009), but also a reduction in other respiratory parameters and heart rate. 17 Additional analyses showed that 30 and 45 minutes after the start of HFNC a higher breathing frequency, a lower SpO 2 , PaO 2 and PaO 2 /FiO 2 ratio were associated with HFNC failure, with an increasing significance level over time. We also recorded respiratory characteristics at 2, 4, 8 and 24 hour after start of HFNC, but as majority of patients with HFNC failure were re-intubated within 2 hour after start, we could only do limited analysis on these data. From the results of Sztrymf we can conclude that HFNC failure can be expected when no improvement of respiratory parameters occurs after starting HFNC, and, combined with our data, HFNC failure is expected to occur within 2 hours after start of HFNC.
The difference in breathing frequency in our study (success: 22±7/min vs failure: 24±7/min) is statistically significant, but its clinical applicability is more complicated.
The pH of the blood gas obtained (1 to 4 hours) prior to HFNC showed a tendency to significance and after logistic regression it came forward as a predictor for HFNC failure (p=0.01). PaCO₂ did not differ between the two groups. Post hoc analysis showed a higher risk of HFNC failure when HCO 3 < 22 mmol/l. The differential diagnosis for a lowered HCO 3 is broad. However in the setting of post-extubation failure there are some obvious causes such as a lowered circulating volume resulting in lactate acidosis, renal loss of bicarbonate in the setting of renal failure, or more theoretically, an excessive resuscitation with chloride. However we do not have any data on lactate or creatinine levels, diuresis or the amount of chloride that was used during resuscitation, so this differential diagnosis is pure speculative.
Secondary outcome of our study showed a difference in the LOS at the ICU after starting HFNC. No difference in ICU or in-hospital mortality was found between the groups. It is possible that this is due to a relative low sample size, although other authors with smaller simple sizes did find an effect of HFNC on mortality.
Messika did found a positive effect from HFNC success on mortality (ICU survival 96% vs. 50%, p=0.01). 15 In concordance, a significant difference in 90-day mortality was found for HFNC in acute respiratory failure when compared to COT or NIV. 3 Although, this effect was not found in the study of Ni et al. when HFNC was started after extubation to prevent acute respiratory failure. 4 There were several limitations to this study due to the retrospective design of the study, including lack of some data and indication for start of HFNC in some patients. In our centre the moment of extubation was not determined by a spontaneous breathing trial, but determined by the attending physician. As a result one can argue that the subjects in the failure group were extubated not under the right conditions. However, there was no difference between the two groups in the time between extubation and the start of HFNC.
If patients were extubated under suboptimal conditions, one would expect them to be more dependent on respiratory support and as a result a smaller time frame between extubation and the start of HFNC, which is not the case.
Another limitation is that the arterial blood gasses were obtained 1 up to 4 hours before the start of HFNC. With this study we aimed to identify patient characteristics that can predict failure of HFNC.
The need for predictors is evident. As found in a retrospective observational study of patients with respiratory failure in which the HFNC failed, a higher ICU mortality was present in patients intubated >48 hours after the start of HFNC (66.7%) compared with patients intubated <48 hours (39.2%, p=0.001). 18 In our study we did also found a nonsignificantly higher mortality and a significant prolonged ICU-stay.
Roca et al focused in their study on early predictors of HFNC failure and developed a prediction tool that identifies the need for mechanical ventilation in patients treated with HFNC because of pneumonia induced hypoxemic acute respiratory failure. 12 This so called ROX-index is defined as the ratio of pulse oximetry/fraction of inspired oxygen to breathing frequency. A recent subsequent study of the same author found the best prediction accuracy of the ROX index after 12 hours of HFNC treatment. 19 However, due to the fact that subjects in which HFNC failed were quickly reintubated (median 0 (0-45) hours) we could not confirm this prediction accuracy after 12 hours in our study. This

Conclusions
In more than half of the patients HFNC failed when used for post-extubation failure.
Breathing frequency and pH were predictors of HFNC failure. There was a non-significant higher ICU and in-hospital mortality in patients with HFNC failure; length of ICU stay was significant longer in the HFNC failure group.

Ethics approval and consent to participate
The study was approved by the medical ethical committee of the Onze Lieve Vrouwe Gasthuis, Amsterdam. A waiver was given for informed consent due to the retrospective and non-invasive nature of this study. Reference number of this waiver: WO 14.112

Consent for publication
Not applicable

Availability of data and materials
The datasets used and/or analysed during the current study are available from the