High-flow nasal cannula failure in critically ill cancer patients with acute respiratory failure: Moving from avoiding intubation to avoiding delayed intubation

Background High-flow nasal cannula (HFNC) is increasingly used in critically ill cancer patients with acute respiratory failure (ARF) to avoid mechanical ventilation (MV). The objective was to assess prognostic factors associated with mortality in ICU cancer patients requiring MV after HFNC failure, and to identify predictive factors of intubation. Methods We conducted a retrospective study from 2012–2016 in a cancer referral center. All consecutive onco-hematology adult patients admitted to the ICU treated with HFNC were included. HFNC failure was defined by intubation requirement. Results 202 patients were included, 104 successfully treated with HFNC and 98 requiring intubation. ICU and hospital mortality rates were 26.2% (n = 53) and 42.1% (n = 85) respectively, and 53.1% (n = 52) and 68.4% (n = 67) in patients requiring MV. Multivariate analysis identified 4 prognostic factors of hospital mortality after HFNC failure: complete/partial remission (OR = 0.2, 95%CI = 0.04–0.98, p<0.001) compared to patients with refractory/relapse disease (OR = 3.73, 95%CI = 1.08–12.86), intubation after day 3 (OR = 7.78, 95%CI = 1.44–41.96), number of pulmonary quadrants involved on chest X-ray (OR = 1.93, 95%CI = 1.14–3.26, p = 0.01) and SAPSII at ICU admission (OR = 1.06, 95%CI = 1–1.12, p = 0.019). Predictive factors of intubation were the absence of sepsis (sHR = 0.32, 95%CI = 0.12–0.74, p = 0.0087), Sp02<95% 15 minutes after HFNC initiation (sHR = 2.05, 95%CI = 1.32–3.18, p = 0.0014), number of quadrants on X-ray (sHR = 1.73, 95%CI = 1.46–2.06, p<0.001), Fi02>60% at HFNC initiation (sHR = 3.12, 95%CI = 2.06–4.74, p<0.001) and SAPSII at ICU admission (sHR = 1.03, 95%CI = 1.02–1.05, p<0.01). Conclusion Duration of HFNC may be predictive of an excess mortality in ARF cancer patients. Early warning scores to predict HFNC failure are needed to identify patients who would benefit from early intubation.

Introduction Acute respiratory failure (ARF) is a frequent and life-threatening complication in immunocompromised patients, raising major diagnostic and therapeutic challenges. It occurs in up to half of patients with hematological malignancies and 15% of patients with solid tumors and represents the first cause for intensive care (ICU) admission in cancer patients. Mortality can reach 50%, depending on underlying condition, etiology, severity and course of ARF, delayed ICU admission, need for mechanical ventilation (MV), and associated organ dysfunctions at ICU admission [1][2][3][4][5] Initial management of ARF consists of optimizing oxygenation, identifying ARF etiology guided by a standardized diagnostic approach [3] and supporting associated organ dysfunction at the same time [6]. The optimal ventilation strategy in cancer patients with ARF remains controversial [7]. Considering the mortality rates in patients requiring MV, non-invasive ventilation strategies have been recently widely evaluated, priority has been given to avoid intubation [8]. Non-invasive ventilation (NIV) was first investigated with significant reduction in intubation and mortality rates [9], but challenged by larger and multicenter data [2,10].
High-flow nasal cannula (HFNC) oxygen therapy delivers warm and humidified oxygen through a nasal cannula, allowing for airflows as high as 50 liters/minute to achieve inspired oxygen fractions (FiO2) as high as 100% [11,12]. It has been increasingly used recently [13]. In unselected patients with ARF, HFNC was associated with increased ventilator-free days and decreased day-90 mortality [14]. In two recent meta-analyses, HFNC may decrease the need for tracheal intubation without impacting mortality [15,16]. Uncertainty remains about HFNC effects in immunocompromised patients, studies providing conflicting results. Recent data did not find any significant survival or clinical benefit compared with standard oxygen [10,14,17], whereas other publications demonstrated that HFNC may decrease intubation requirement and/or mortality [2,18,19]. The absence of diagnosis and mechanical ventilation requirement are the main prognostic factors [2].
For patients who fail to improve with HFNC, intubation should be strongly considered. HFNC failure in immunocompromised patients has been rarely investigated. The objective of our study was to assess prognostic factors of mortality in ICU cancer patients with ARF requiring MV after HFNC failure, and to identify predictive factors of intubation.

Patients' selection
We conducted a retrospective study from 2012-2016 in our institution (Paoli-Calmettes Institute, Marseille, France), a cancer referral center. All adult patients with solid tumor or hematological malignancy admitted to the ICU for ARF treated with HFNC requiring oxygen �30 L/ minute were included. Exclusion criteria's were acute cardiogenic pulmonary edema, hypercapnic ARF, MV weaning, HFNC during scheduled surgery, treatment-limitation with donot-intubate decision and patients intubated after HFNC weaning at ICU admission. The study was approved by our local Institutional Review Board.

HFNC device
All patients received high-flow oxygen therapy via HFNC device (Optiflow, Fisher & Paykel Healthcare, Auckland, New Zealand). Oxygen was applied continuously by large-bore binasal prongs at a gas flow of 50 L/min with a FiO2 level of 100% initially through a heated humidifier (MR850, Fisher&Paykel Healthcare). Response to treatment was continuously monitored. The aim of oxygenation was to produce peripheral capillary oxygen saturation (SpO2) levels of 92% or more. Patients were treated with complementary non-invasive ventilation according to the intensivist in charge according to local guidelines [19].

Definitions and data collection
Demographical, clinical, biological and outcome data were retrospectively collected from the patient's charts using our ICU management software (MetavisionTM, Dusseldorf, Germany). We recorded the following baseline data at ICU admission and during ICU stay: age, gender, underlying malignancy, allogenic or autologous hematopoietic stem cell transplantation (HSCT), disease status, neutropenia, Charlson comorbidities index [20]. Severity of illness was assessed using Simplified Acute Physiology Score (SAPS 2) [21] and Sepsis-related Organ Failure Assessment (SOFA) score [22].
ARF was defined as a need for oxygen greater than 6L/min to maintain peripheral capillary oxygen saturation >95% or symptoms of respiratory distress (tachypnea >30/min, intercostal recession, labored breathing, and/or dyspnea at rest). Neutropenia was defined by absolute neutrophil count < 0.5 G/L. We quantified the number of involved pathologic quadrants on chest X-ray (0-4) at ICU admission [23]. Fever was classified as clinically documented, microbiologically documented or fever of unknown origin. HFNC settings and respiratory function were collected during ICU stay. HFNC failure was defined by intubation and MV requirement. Criteria for endotracheal intubation included signs of persisting or worsening respiratory failure (defined as two of the following criteria: respiratory rate above 40 breaths per min, lack of improvement in signs of high respiratory-muscle workload, development of copious tracheal secretions, pH <7.35, SpO2 levels <90% for more than 5 min without technical dysfunction, or intolerance to oxygenation techniques), hemodynamic instability (systolic blood pressure <90 mm Hg, mean blood pressure <65 mm Hg, or vasopressors) or neurological deterioration (Glasgow<12) [14]. Outcome endpoints including intubation rate, length of HFNC, time from HFNC initiation to MV initiation (at day 0 (� 24 hours), day 1, day 2-3 and > day 3), ICU, hospital and one-year mortality.

Statistical analysis
All data are presented as rates (percentage) for qualitative variables and medians (25th-75th percentiles) for quantitative variables. Characteristics of patients subsequently intubated were compared across the groups of hospital survivors and non-survivors by using Fisher's exact test and Wilcoxon rank-sum test. We performed logistic regression analyses to identify variables independently associated with hospital mortality, as measured by the estimated odds ratio (OR) with 95% confidence interval (95% CI). Variables yielding p lower than 0.15 in the bivariable analyses were entered into a multivariable logistic regression model with hospital mortality as the outcome and a stepwise forward variable selection.
In the second part, we analyzed all patients treated with HFNC in order to identify predictive factors of intubation. Similarly, we compared characteristics of patients according to HFNC success (no intubation) or failure (intubation) by using bivariable Fine and Gray model which accounts for the competing risk of discharge form ICU without intubation. All variables associated with intubation at a p-value lower than 0.15 were then included in a multivariable Fine and Gray regression model with stepwise forward variable selection. Association between covariate and risk of intubation should have been reported as subdistribution Hazard Ratio (sHR). All tests were two-sided, and p values lower than 0.05 were considered statistically significant. Statistical tests were conducted using the SPSS 13 software package (IBM, Armonk, NY, USA) and R software (R Core Team, 2020).

Results
During the study period, 202 cancer patients with ARF treated with HFNC were included ( Fig  1). 104 patients were successfully treated with HFNC and 98 patients experienced HFNC failure and required subsequent intubation and MV.

Discussion
Our study identified HFNC duration before intubation, disease status and severity of illness as prognostic factors of hospital mortality in ICU cancer patients with ARF requiring intubation. Risk factors of intubation were severity illness at ICU admission, FiO2 at HFNC initiation, SpO2 after HFNC initiation and sepsis. HFNC offers interesting physiological benefits [8,24], improves oxygenation [25], generates low-level positive airway pressure [26], reduces respiration rate [27], attenuates inspiratory resistance and supplies a constant FiO2 [28]. Moreover, it is well tolerated [29,30] and does not increase risks of pneumonia or barotraumas [31]. The impact of HFNC duration on outcome has been poorly assessed. In a general population study, HFNC was safe and well tolerated for long periods [32]. Another retrospective study in unselected patients provided conflicting results, showing that extended use of HFNC before intubation might be harmful [33]. It was conducted on 175 patients unsuccessfully treated with HFNC. Early-intubated (<48 hours) patients had better ICU survival than late-intubated (>48 hours) patients (39.2% vs 66.7%, p = 0.001), extubation success, ventilator weaning and ventilator-free days. Recently, Dumas et al showed in a large cohort of 7736 immunocompromised patients who were intubated that time between ICU admission and intubation is a strong predictor of mortality, suggesting a detrimental effect of late initial oxygenation failure [34]. Although several studies have assessed NIV failure in immunocompromised patients [2,17], our study is, to the best of our knowledge, the first one exploring HFNC failure in this population. Our findings revealed that early MV, before day 2, may be associated with a better outcome [35,36]. There is a need to develop specific early warning scores to predict HFNC failure, in order to identify high-risk patients who would benefit from early intubation. Priority should move from avoiding intubation to avoiding delayed intubation. Severity of presentation, assessed by SAPS II and the number of pulmonary quadrants involved on chest X-ray, were associated with increased hospital mortality, in line with previous publications [1,37].
To avoid late intubation, identifying prognostic factors associated with HFNC failure is important. SAPSII and severity of pulmonary extension were predictive factors of intubation. The number of involved pulmonary fields was previously identified as an early predictor of the severity of acute respiratory distress syndrome (ARDS) in hematology patients [23,38]. Fi02 at HFNC initiation was another significant factor. Similarly, the degree of hypoxemia (PaO2/FiO2 after 1 hour of NIV) was associated with NIV failure [10,18,39]. Predictors of NIV failure have been summarized in an easy to use mnemonic "HACOR" score (Heart rate, Acidosis, Consciousness, Oxygenation, Respiratory rate) to identify patients at high risk for NIV failure [40].
Similarly, further studies are needed to confirm the determinants of HFNC failure. Due to its ease of application, non-evidence-based use of HFNC has spread to non-ICU wards [41]. To consider safely HFNC outside the ICU, the identification of patients at low risk of intubation is essential. Exploratory studies are needed in this context. So far, strategies to improve survival in hypoxemic patients with ARF relied on different oxygenation options [3,7]. More than ventilation strategy, the stronger prognostic features were mechanical ventilation and absence of ARF etiology [2]. This is why we focused in patients requiring intubation in our study, exploring the impact of HFNC in this population. Future researches should focus on optimal timing of ICU admission, personalizing an appropriate oxygenation strategy according to situations [42] and selecting the most relevant diagnostic strategy [2,43]. Collaboration between onco-hematologists and intensivists is crucial [6]. ARF with undetermined etiology impacts outcome, as previously described [3]. A timely and accurate diagnostic strategy that takes into account characteristics of the underlying malignancy, immunosuppression, respiratory symptoms and radiologic pattern [2,3,44,45]. In addition, preventing ICU acquired events from both MV and underlying impairment of immunological functions will also be challenging [1,4,46].
We acknowledge some limitations in our study. First, its retrospective nature is intrinsically susceptible to have selection bias. However all biological and medical settings were prospectively collected. Secondly, this is a monocentric study in a highly specialized cancer center. Generalization of our results may therefore be analyzed with caution. Third, we measured respiratory distress in our study and not dyspnea. Recent data from the GRRROH showed that dyspnea was frequent and intense in patients receiving NIV for ARF and was associated with a higher risk of NIV failure and poorer outcome [47]. Another study showed that the magnitude of inspiratory effort relief as assessed by esophageal pressure variation within the first 2 hours of NIV was an early and accurate predictor of NIV outcome at 24 hours [48]. Dyspnea would be of great interest to assess in future studies. Lastly, our study was unfortunately not designed to explore ROX index, the ratio of oxygen saturation as measured by pulse oximetry/Fi02 to respiratory rate, recently reported in both immunocompetent and immunocompromised patients. Recent data showed that in patients with pneumonia with ARF treated with HFNC, ROX index could help to identify patients with low and high risk for intubation in a 2-year multicenter prospective observational cohort study [49]. Lemiale et al. recently reported the performance of the ROX index to predict intubation in immunocompromised patients receiving HFNC for ARF. A ROX index greater than 4.88 appeared to have a poor ability to predict intubation, although it remained highly associated with the risk of intubation and may be useful to stratify such risk in future studies [50].

Conclusions
Duration of HFNC may be predictive of mortality in ARF cancer patients requiring intubation after unsuccessful HFNC. There is a need to develop specific early warning scores to predict HFNC failure in order to identify high-risk patients who might benefit from early intubation. Whatever the technique of oxygenation used, day-to-day decisions must strive to avoid delayed intubation and identify ARF etiology.