Introduction
Application of noninvasive ventilation (NIV) remains debated in patients admitted to intensive care unit (ICU) for de novo acute hypoxemic respiratory failure [
1‐
13], and clinical practice guidelines remain unable to offer a recommendation given the uncertainty of evidence [
14]. By contrast, the guidelines suggest NIV as first-line therapy for oxygenation in immunocompromised patients [
14]. Indeed, by pooling all randomized controlled trials, NIV has been associated with lower intubation and mortality rates than standard oxygen therapy [
15‐
18].
Mortality of immunocompromised patients requiring intubation and invasive mechanical ventilation is particularly high, ranging from 50 to 70% [
19,
20]. Therefore, oxygenation strategies aiming at preventing intubation of these patients deserve consideration. Although recommendations for the management of acute respiratory failure differ between immunocompromised and nonimmunocompromised patients, it is not clear whether or not immunocompromised patients are actually at increased risk of intubation. A recent study including a large sample of patients with acute respiratory distress syndrome (ARDS) reported higher intubation rates in immunocompromised patients than in the others [
21]. However, immunocompromised patients had more severe organ failure than nonimmunocompromised patients, and based on unadjusted analysis, this result may simply indicate greater disease severity.
The aim of our study was to assess whether immunosuppression is an actual risk factor for NIV failure by performing post hoc analysis of two studies including immunocompromised and nonimmunocompromised patients admitted to ICU for de novo acute hypoxemic respiratory failure and treated with NIV. The primary aim was to compare adjusted intubation rates according to immunosuppression status in the matched cohort. The secondary aim was to compare in-ICU mortality rates according to immunosuppression status in the matched cohort.
Discussion
The direct influence of immunosuppression status on outcomes of patients with acute hypoxemic respiratory failure treated with NIV remains poorly known. Our main results are that immunosuppression status was strongly associated with mortality but not with intubation after adjustment, nor after matching on markers of respiratory disease severity. Indeed, PaO2/FiO2 and large tidal volumes 1 h after NIV initiation were stronger predictors of NIV failure than immunosuppression status per se. By contrast, once intubated after NIV failure, mortality was markedly higher in immunocompromised patients than in the others.
In our cohort, immunocompromised patients treated with NIV required intubation in 61% of cases, knowing that the vast majority (89%) met the criteria for ARDS [
26]. Our findings are in keeping with these from Azoulay and colleagues, who reported NIV failure rates of 71% in a pooled analysis including 387 immunocompromised patients with ARDS [
28], and with the 63% failure rate reported in a sub-analysis of a large international cohort study on ARDS [
21]. Similarly, mortality rates in these two cohorts were 44 and 43% [
21,
29], i.e., very close to the 38% we report, thereby reinforcing the external validity of our results.
Immunosuppression status was associated with a 2.6-fold risk for mortality after adjustment. It has been extensively reported to be an independent risk factor for mortality in patients treated with NIV for ARDS or influenza [
30‐
32]. By contrast, whether immunosuppression is a risk for NIV failure has been poorly studied. Despite higher intubation rates in immunocompromised than in nonimmunocompromised patients, immunosuppression status did not remain associated with intubation after adjustment on variables of respiratory disease severity such as hypoxemia and large tidal volumes generated under NIV. These results are in line with previous studies reporting that immunosuppression was associated with NIV failure by univariate analysis, but not after adjustment [
33,
34]. However, in the first study, adjustment was performed on SAPS II, which shares collinearity with immunosuppression [
33]. Moreover, in the second study, most of the patients had cardiogenic pulmonary edema or acute-on-chronic respiratory failure [
34], in whom strong benefits of NIV have been reported [
35,
36]. Due to the above-mentioned limitations, the impact of immunosuppression status on NIV failure rate could have been mitigated. However, the absence of association between immunosuppression status and NIV failure after adjustment in both our overall cohort and in our propensity score-matched cohort suggests that immunosuppression status is not per se a risk factor for NIV failure, and that immunocompromised patients might have been more severe than nonimmunocompromised at inclusion despite similar vital signs, oxygen requirements, and arterial blood gas at baseline. Whether similar results would have been observed with less severe patients (such as patients with less oxygen requirement or less hypoxemia) is unclear. Indeed, although oxygen requirement at ICU admission is independently associated with mortality in immunocompromised patients, its effect on intubation has not been tested yet [
37].
After adjustment, mortality rate was higher in immunocompromised than in nonimmunocompromised patients despite similar NIV failure rates. As patients with a do-not-intubate order were excluded from the analysis, this striking difference pertains exclusively to different mortality rates of intubated patients. This difference could be related to differences in respiratory mechanics under invasive ventilation between the two groups, such as driving pressure, an important determinant of ARDS prognosis [
38]. Unfortunately, driving pressure of the respiratory system was not collected in the two studies. Likewise, the cause of death was not collected and it is possible that the proportion of patients with limitation of life-sustaining measures was different between the two groups [
21]. Last, we cannot rule out some differences in the management of invasive ventilation between the two groups given it was not protocolized in the two studies. However, we believe that patients studied were managed according to the best practice given that all participating centers are experienced in the field of invasive ventilation, and have been involved in several trials on invasive ventilation. Our findings highlight the urge for standardization of invasive ventilation settings, collection of respiratory mechanics, and the cause of death, even in studies comparing two first-line, early, noninvasive ventilation strategies to better understand the influence of one strategy over the other (NCT02978300).
PaO
2/FiO
2 1 h after NIV initiation was independently associated with NIV failure. Hypoxemia is a well-known risk factor for NIV failure. In line with our findings, two prospective cohort studies have reported that PaO
2/FiO
2 lower than 146 mmHg and 175 mmHg after 1 h of NIV was independently associated with NIV failure [
39,
40]. Likewise, the HACOR score proposed by Duan and colleagues to predict NIV failure allocated a stepwise increasing score for PaO
2/FiO
2 drop 1 h after NIV initiation [
41]. A similar association between hypoxemia and NIV failure had previously been reported [
30,
31]. However, the timing of hypoxemia assessment was unclear, and we believe that early evaluation (1 h after NIV initiation, for instance) is important to avoid the harmful effects of late intubation [
42]. All in all, PaO
2/FiO
2 1 h after NIV initiation seems an interesting early predictor of NIV failure and mortality. As a consequence, patients remaining hypoxemic after 1 h of NIV should be monitored closely to avoid the deleterious effects of spontaneous breathing [
43].
Large tidal volumes 1 h after NIV initiation were also independently associated with NIV failure. Although tidal volumes were not normalized to the patient’s predicted body weight due to missing data, our results are in line with a recent study reporting that large expired tidal volumes were independent predictors of NIV failure in de novo acute hypoxemic respiratory failure [
33]. Interestingly, large tidal volumes 1 h after NIV initiation were also associated with mortality. This finding may illustrate the concept of patient self-inflicted lung injury [
43]. Spontaneously breathing patients with acute respiratory failure have a high respiratory drive, which generates large transpulmonary driving pressures and subsequent large tidal volumes, which stretch the lungs and may worsen lung injury [
44]. Our results suggest that large tidal volumes and low PaO
2/FiO
2 are more important risk factors for NIV failure than immunosuppression status by itself, and that particular attention should be paid to large tidal volumes and blood gas analysis as early as 1 h after NIV initiation in patients with de novo acute hypoxemic respiratory failure.
Some limitations have to be acknowledged. First, this is an unplanned post hoc analysis of two studies with inherent possible bias. Although in the study from Thille and colleagues, clinical data were collected retrospectively based on computerized medical charts, NIV settings were collected prospectively, thereby reducing potential selection and information bias. Moreover, NIV failure and mortality rates are in line with previous studies suggesting that our results might be generalizable. Second, the sample size is relatively small and our results deserve confirmation in larger multicenter prospective cohorts. Third, it is worth noting that patients from the Frat study received high-flow oxygen therapy, whereas those from the Thille study received standard oxygen therapy during NIV breaks [
11,
22]. This difference was not considered in our analysis because we did not collect data between NIV sessions (data were exclusively collected under standard oxygen therapy before ICU admission or under NIV after ICU admission). Moreover, in a previous study on immunocompromised patients, intubation and mortality rates of patients treated with high-flow or standard oxygen therapy during NIV breaks were not different [
45]. Fourth, the choice of variables included in multivariate analysis could be argued. We chose early easy-to-assess variables, rather than variables to be calculated 24 h after ICU admission such as SAPS II [
24]. Indeed, given the harm of delayed NIV failure [
42], it seems important to identify as early as possible the factors associated with poor outcomes. Fifth, although our results suggest that immunosuppression status is independently associated with mortality in patients with de novo acute hypoxemic respiratory failure treated with NIV, whether a different protocol to carry out NIV aiming at more protective ventilation (high PEEP, low pressure support) would lead to different results is unknown. Indeed, a recent randomized trial including patients with ARDS, half of whom were immunocompromised, reported better outcomes when NIV was applied with a helmet and high PEEP levels than with a face mask and lower PEEP levels [
46].
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.