Oxygenation derangements after subarachnoid hemorrhage: incidence and impact on the outcome. Data from a large multicenter, retrospective study.

Background Hypoxemia and hyperoxemia are frequent after acute subarachnoid hemorrhage (aSAH) and are associated with an increase in morbidity and mortality. Among the pulmonary complications causing oxygen derangements, acute respiratory distress syndrome (ARDS) seems to be crucial, with a reported incidence ranging from 11 to 50%. Methods We designed a multicentric, retrospective cohort study in tree intensive care in Europe. We collected data between January 2009 and December 2017. We included adult patients (≥18 years) with a diagnosis of aSAH. Hypoxemia was defined as PaO 2 <60mmHg, mild hyperoxemia as PaO 2 >120mmHg, and severe hyperoxemia as PaO 2 >200mmHg. The primary aim of this study was to assess the incidence of episodes of hypoxemia, hyperoxemia, and the oxygen variability values (calculated as the daily difference between the highest and the lowest arterial partial pressure of oxygen (PaO 2 )) during the first week after the intensive care unit (ICU) admission. Secondary aims included the evaluation of the incidence of ARDS according to the Berlin criteria, and the assessment of the effect of oxygen derangements on patients’ outcomes.

and duration of MV are associated with patients' outcomes Introduction Over the last years, attention has risen towards the levels of oxygenation as an essential determinant of secondary brain injury [1][2][3][4]. The deleterious consequences of hypoxemia, (the presence of a partial pressure of oxygen (PaO 2 ) < 60 mmHg) on the patients' outcome, are well known [5]. More recently, research has also focused on the role of hyperoxemia. In fact, PaO 2 levels > 100 mmHg seem to have deleterious effects on outcome in critically ill patients overall [6,7] and the threshold changes only slightly in specific subpopulations such as post-cardiac arrest (PaO 2 levels > 120 mmHg) [8,9] or aneurysmal subarachnoid hemorrhage (aSAH) (PaO 2 > 200 mmHg) patients [10]. Moreover, the concept of PaO 2 variability, defined as the daily difference between the highest and the lowest PaO 2 values, has recently been identified as a potential surrogate marker for relative ischemiareperfusion injury [11,12].
Pulmonary complications causing oxygen derangements after aSAH are frequent and associated with an increase in morbidity and mortality [13], with the worst clinical manifestation being acute respiratory distress syndrome (ARDS) [14]. The reported incidence of ARDS in aSAH patients ranges from 11 to 50% [4,13,15,16]. This heterogeneity could be explained by the absence of standardized treatment protocols, by differences in center policies, or by the use of different ARDS definitions. The Berlin Criteria [17], an updated definition for ARDS aimed to simplify the diagnosis and better prognosticate outcome, has never been applied to the aSAH population.
Altogether, the optimal oxygenation strategies (i.e., both oxygen dose administration and PaO 2 target) remain unclear in aSAH patients.
We, therefore, conducted a multicenter, retrospective, international study, including a large cohort of patients with aSAH, to assess the incidence of oxygen derangements during the first week after intensive care unit (ICU) admission.

Study cohort
This multicenter retrospective study includes patients with aSAH admitted to the Neuro-intensive care unit (NICU) of the "San Gerardo" hospital in Monza, the NICU of the "Medical University Hospital" of Innsbruck and the medical-surgical ICU of the "Erasme Hospital (ULB)" in Bruxelles between January 2009 and December 2017.

Inclusion criteria were:
age ≥18 years; a diagnosis of aSAH confirmed by a Computed Tomography (CT) scan and by an angio-CT or angiography; admission to an ICU involved in the study.

Aims of the study
The aims of this study were: to assess the incidence of episodes of hypoxemia, hyperoxemia, and PaO 2 variability during the first week from ICU admission; to assess the incidence of ARDS, according to the Berlin Criteria; to evaluate the effect of oxygen derangements on patients' outcomes.

Patients' characteristics
We used the electronic data systems of each center to collect data regarding the patients' previous medical history, including age, smoking habits, and pre-injury comorbidities (e.g., hypertension and chronic obstructive pulmonary disease history). We assessed the Glasgow Coma Scale score (GCS) [20] on admission and the World Federation of Neurosurgical Societies grading system for aSAH (WFNS) [21] for each patient according to their medical records. We recorded the modified Fisher score (mFisher) [22], the arteriography or angio-CT findings (such as aneurysm location), the information regarding the aneurysm treatment (if surgical or endovascular procedure), and we noted whether and how the invasive intracranial pressure was monitored (through an external ventricular drain (EVD) or an intraparenchymal catheter). We assessed the neurological outcome through the Glasgow Outcome Score (GOS) [23] and the modified Rankin scale (mRS) [24] at ICU discharge. We defined an unfavorable outcome as a modified Rankin score ≥3 or GOS ≤3. We also recorded the length of mechanical ventilation in hours and ICU length of stay (LOS) in days.

Ventilation and oxygenation data
For each patient, we collected data regarding oxygenation, ventilator settings (such as tidal volume (TV) and positive end-expiratory pressure (PEEP)), and arterial blood gases analysis (ABG; pH, PaO 2 , and partial pressure of carbon dioxide (PaCO 2 )) at day one, three, and seven from ICU admission. For each variable, we collected the lowest and highest value for each daily time point. At the corresponding time interval, we calculated the lowest values of the ratio of the PaO 2 over the inhaled fraction of oxygen (PaO 2 /FiO 2 ).
Two trained investigators for each center reviewed all chest X-rays independently, checking for the presence of bilateral infiltrates. In case of disagreement, a third investigator was involved in interpreting the radiological findings.
We defined the episodes of hypoxemia as the detection of at least one measurement of PaO 2 <60mmHg on any ABG of each day of observation (one, three, and seven). We analogously defined the hyperoxemia episodes as mild if a PaO 2 >120mmHg and as severe if a PaO 2 >200mmHg was found at least once on an ABG for each day [25]. We assessed the oxygen variability as the difference between the highest and the lowest PaO 2 values in mmHg during the day.
To evaluate the ARDS incidence in our cohort, we applied the Berlin definition of ARDS [6]  When we reached a diagnosis of ARDS, we also assessed whether the same episode lasted for multiple days during the first week of ICU admission, thus avoiding the risk of a repeated diagnosis.

Statistical analysis
We described quantitative variables as the median and interquartile range (IQR), and categorical variables as number and percentage. We used the Fisher's exact test for categorical variables and Wilcoxon tests for continuous variables to compare the characteristics of patients with or without ARDS at ICU admission.
The association between the admission characteristics and the primary outcome, dichotomized as favorable (GOS 4-5, mRS 0-2) or unfavorable (GOS 1-3, mRS 3-5), was first analyzed using Fisher's exact test for categorical values and Wilcoxon test for continuous variables. We then conducted a multivariable logistic regression to adjust on predefined variables according to our hypothesis and previously published models. We included the following factors in the multivariate logistic regression: age, GCS at admission, WFNS score, mFisher scale, ARDS occurrence, mild hyperoxemia episodes, hypoxemia episodes, high PaO 2 variability, and duration of mechanical ventilation. Aspiration pneumonia, the presence of pathological secretions at admission, PEEP values at admission, chest Xray results, and the administration of neuromuscular blocking. The WFNS score includes the GCS and hence was discarded from the multivariate analysis. We compared the c index of models with and without PaO 2 variability to evaluate its additive value in outcome prediction. Results are presented as Odds ratios with 95% confidence interval.
All tests were two-sided at a 0.05 significance level; we performed the statistical analysis using R, software version 3.5.2 [26].

Results
To the best of our knowledge, this is the most extensive multicenter study describing oxygen derangements after aSAH. For the first time, we assessed the role of both hypoxemia and hyperoxemia in aSAH patients, including the concept of oxygen variability that could account for unfavorable outcome. Moreover, this is the first study evaluating the occurrence of ARDS using the Berlin definition in this population. The large size of our cohort and the inclusion of three different centers make this study well representative for the evaluation of the incidence and effects on patients' outcomes of respiratory derangements after aSAH.
According to our results: episodes of oxygenation derangements, i.e., hyperoxemia and hypoxemia, are frequent in aSAH patients, whereas the incidence of ARDS after aSAH reaches only 3.6% within the first week of ICU admission. the duration of mechanical ventilation, neurological status at admission (GCS), and PaO 2 variability are all independent factors for unfavorable outcome. We also confirmed that hypoxemia and ARDS are associated with unfavorable outcome after aSAH.
The main goal of mechanical ventilation in aSAH patients is to provide adequate oxygenation and to prevent the risk of hypoxemia, which is a well-known cause of secondary brain injury [30]. The recommended target of oxygenation as per PaO 2 values in brain-injured patients with healthy lungs is > 75 mmHg, whereas in those suffering from ARDS, lower PaO 2 targets (55-80 mmHg) are suggested [19]. Also, oxygenation derangements are frequent, with an incidence ranging from 43% to 92% in aSAH patients [27].
Supplemental oxygen is a fundamental part of the acute management of these patients; however, aggressive treatment of hypoxemia might lead to the patients' exposure to unnecessarily high levels of FiO 2 , even higher than the values strictly necessary to maintain physiologic values of PaO 2 , sometimes for a prolonged amount of time.
Although moderate hyperoxemia can increase cerebral oxygenation as measured by near-infrared spectrometry (NIRS), its clinical implications remain unclear [28,29]. Indeed, hyperoxemia can have detrimental effects on the patients' outcome by several pathophysiological mechanisms, including oxidative damage and an increase in reactive oxygen species production, deriving from high tissue oxygen tension with consequent mitochondrial damage and lipid peroxidation, leading to neuronal cells apoptosis. Several cohort studies support the hypothesis that hyperoxemia could be detrimental in the emergency room [30] and after cardiac arrest [9]. Although the effects are particularly pronounced following long-term exposition to high PaO 2 levels, several studies suggest that even short term hyperoxemia can lead to increased mortality and morbidity, especially in brain-injured patients [31, 32].
A randomized trial [18] which compared two targets of PaO 2 (70 vs. 100 mm Hg or arterial oxygen saturation (SaO 2 ) between 94% and 98% vs. SaO 2 values between 97% and 100%) found that a conservative target for oxygen therapy is associated with reduced ICU mortality (absolute risk reduction of 0.086 [95% CI, 0.017-0.150]). More recently, a retrospective analysis of an extensive database confirmed these results for severe but not for mild hyperoxemia after cardiac arrest [28].
We also found that PaO 2 variability was independently associated with poor outcome. PaO 2 variability could cause the so-called "reperfusion injury," defined as the tissue damage resulting from the restoring of appropriate oxygen and blood supply after a period of hypoxemia [39]. Low oxygen arterial pressure is responsible for tissue hypoxemia or ischemia, which can be followed by reperfusion upon PaO 2 increase. Such a reperfusion mechanism could increase neuro-inflammation and brain damage [33]. Thus, PaO 2 variability could be a surrogate marker for relative ischemiareperfusion injury in the brain, which could cause secondary brain injury. Indeed, in pediatric intensive care, oxygen variability measured on the plethysmography has recently been shown to predict chronic lung disease [11,12] or the need for hospitalization [12].
ARDS after aSAH can be consequent to several factors, including neurogenic pulmonary edema (NPE), aspiration pneumonia, or ventilator-associated pneumonia. In our study, we did not investigate the etiology of ARDS, but the incidence of aspiration pneumonia and the presence of pathological pulmonary secretions was similar in patients with and without ARDS. The peak of incidence of ARDS was in the first 24 hours, corresponding to that of NPE, thus suggesting that probably NPE was the leading cause of ARDS in our cohort.
A retrospective study reported an incidence of PaO 2 /FiO 2 ratio <300 in nearly one-third of patients with aSAH (27%) [13]. However, other authors suggested that the prevalence of ARDS is much lower, ranging from 4 to 18% [34,35]. The results of our study suggest that the incidence of ARDS is even lower when compared to previous literature [1,[3][4][5]. The reason could be related to the increasing application of protective ventilation strategies [20] in our institutions over the last years. The beneficial effect of protective lung ventilation and respiratory strategies are well established in the general ICU population and the operating room [36, 37]. However, the application of these techniques in neuro-critical care patients is contrasting because of this group's specific therapeutic needs and ventilator targets, and for the potential risks of lung-protective ventilation strategies on intracranial pressure and cerebral perfusion pressure. However, recent studies support the use of protective strategies and, in particular, of low driving pressure and low tidal volumes even in brain-injured patients [38].
Of note, we chose to evaluate the ARDS incidence according to the daily worst PaO 2 /FiO 2 value. Thus, we might have overestimated the incidence of ARDS, as daily lowest PaO 2 /FiO2 could be the result of temporary episodes of hypoxemia such as those occurring during an atelectasis episode, during the transports to the radiologic suite, or during an angiographic procedure.
We also showed that patients with ARDS have more frequent episodes of hyperoxemia, a trend towards more frequent episodes of hypoxemia, and a subsequent increased PaO 2 variability. Several causes could explain the high PaO 2 variability found in ARDS patients. Aggressive treatment of hypoxemia might have led to higher levels of FiO 2 and, therefore, high levels of PaO 2 . Also, patients with ARDS are more susceptible to episodes of derecruitment, atelectasis, and desaturation, promptly treated with high FiO 2 and recruitment maneuvers by the attending physician, causing rapid changes in oxygenation values.
One of the limits of this hypothesis is that the PaO 2 variability might represent the combined predictive value of both hypoxemia and hyperoxemia and therefore present a better predictive value than each variable alone. Nevertheless, in the multivariate model, hypoxemia had a modest predictive value, and PaO 2 variability presented a predictive value independent from hypo-and hyperoxemia, supporting its independent impact on the outcome.

Limitations
Several limitations to this study need to be mentioned. First, we evaluated ARDS secondary to subarachnoid hemorrhage within the first seven days after the ICU admission, evaluating its occurrence on days one, three, and seven. The median ventilation time in our cohort was 2.00 [0.00, 11.00] days; therefore, patients could have developed ARDS afterward, which could result in an underestimation of the real incidence of ARDS in this group of patients.
Second, we focused on the early outcome after aSAH, at ICU discharge. However, the evaluation of long-term neurological outcome in aSAH would have provided a much stronger message to our study. Third, we have reported associations but not causality, due to the retrospective observational design of this study; moreover, a retrospective data collection resulted in the lack of precise and complete data regarding ventilation and other confounders that might have influenced the patients' outcome.
Finally, we evaluated the daily PaO 2 through discontinuous arterial blood gases analysis that might represent less accurately the PaO 2 variability when compared to continuous measures.

Discussion
To the best of our knowledge, this is the most extensive multicenter study describing oxygen derangements after aSAH. For the first time, we assessed the role of both hypoxemia and hyperoxemia in aSAH patients, including the concept of oxygen variability that could account for unfavorable outcome. Moreover, this is the first study evaluating the occurrence of ARDS using the Berlin definition in this population. The large size of our cohort and the inclusion of three different centers make this study well representative for the evaluation of the incidence and effects on patients' outcomes of respiratory derangements after aSAH.
According to our results: episodes of oxygenation derangements, i.e., hyperoxemia and hypoxemia, are frequent in aSAH patients, whereas the incidence of ARDS after aSAH reaches only 3.6% within the first week of ICU admission. the duration of mechanical ventilation, neurological status at admission (GCS), and PaO 2 variability are all independent factors for unfavorable outcome. We also confirmed that hypoxemia and ARDS are associated with unfavorable outcome after aSAH.
The main goal of mechanical ventilation in aSAH patients is to provide adequate oxygenation and to prevent the risk of hypoxemia, which is a well-known cause of secondary brain injury [30]. The recommended target of oxygenation as per PaO 2 values in brain-injured patients with healthy lungs is > 75 mmHg, whereas in those suffering from ARDS, lower PaO 2 targets (55-80 mmHg) are suggested [19]. Also, oxygenation derangements are frequent, with an incidence ranging from 43% to 92% in aSAH patients [27].
Supplemental oxygen is a fundamental part of the acute management of these patients; however, aggressive treatment of hypoxemia might lead to the patients' exposure to unnecessarily high levels of FiO 2 , even higher than the values strictly necessary to maintain physiologic values of PaO 2 , sometimes for a prolonged amount of time.
Although moderate hyperoxemia can increase cerebral oxygenation as measured by near-infrared spectrometry (NIRS), its clinical implications remain unclear [28,29]. Indeed, hyperoxemia can have detrimental effects on the patients' outcome by several pathophysiological mechanisms, including oxidative damage and an increase in reactive oxygen species production, deriving from high tissue oxygen tension with consequent mitochondrial damage and lipid peroxidation, leading to neuronal cells apoptosis. Several cohort studies support the hypothesis that hyperoxemia could be detrimental in the emergency room [30] and after cardiac arrest [9]. Although the effects are particularly pronounced following long-term exposition to high PaO 2 levels, several studies suggest that even short term hyperoxemia can lead to increased mortality and morbidity, especially in brain-injured We also found that PaO 2 variability was independently associated with poor outcome. PaO 2 variability could cause the so-called "reperfusion injury," defined as the tissue damage resulting from the restoring of appropriate oxygen and blood supply after a period of hypoxemia [39]. Low oxygen arterial pressure is responsible for tissue hypoxemia or ischemia, which can be followed by reperfusion upon PaO 2 increase. Such a reperfusion mechanism could increase neuro-inflammation and brain damage [33]. Thus, PaO 2 variability could be a surrogate marker for relative ischemiareperfusion injury in the brain, which could cause secondary brain injury. Indeed, in pediatric intensive care, oxygen variability measured on the plethysmography has recently been shown to predict chronic lung disease [11,12] or the need for hospitalization [12].
ARDS after aSAH can be consequent to several factors, including neurogenic pulmonary edema (NPE), aspiration pneumonia, or ventilator-associated pneumonia. In our study, we did not investigate the etiology of ARDS, but the incidence of aspiration pneumonia and the presence of pathological pulmonary secretions was similar in patients with and without ARDS. The peak of incidence of ARDS was in the first 24 hours, corresponding to that of NPE, thus suggesting that probably NPE was the leading cause of ARDS in our cohort.
A retrospective study reported an incidence of PaO 2 /FiO 2 ratio <300 in nearly one-third of patients with aSAH (27%) [13]. However, other authors suggested that the prevalence of ARDS is much lower, ranging from 4 to 18% [34,35]. The results of our study suggest that the incidence of ARDS is even lower when compared to previous literature [1,[3][4][5]. Of note, we chose to evaluate the ARDS incidence according to the daily worst PaO 2 /FiO 2 value. Thus, we might have overestimated the incidence of ARDS, as daily lowest PaO 2 /FiO2 could be the result of temporary episodes of hypoxemia such as those occurring during an atelectasis episode, during the transports to the radiologic suite, or during an angiographic procedure.
We also showed that patients with ARDS have more frequent episodes of hyperoxemia, a trend towards more frequent episodes of hypoxemia, and a subsequent increased PaO 2 variability. Several causes could explain the high PaO 2 variability found in ARDS patients. Aggressive treatment of hypoxemia might have led to higher levels of FiO 2 and, therefore, high levels of PaO 2 . Also, patients with ARDS are more susceptible to episodes of derecruitment, atelectasis, and desaturation, promptly treated with high FiO 2 and recruitment maneuvers by the attending physician, causing rapid changes in oxygenation values.
One of the limits of this hypothesis is that the PaO 2 variability might represent the combined predictive value of both hypoxemia and hyperoxemia and therefore present a better predictive value than each variable alone. Nevertheless, in the multivariate model, hypoxemia had a modest predictive value, and PaO 2 variability presented a predictive value independent from hypo-and hyperoxemia, supporting its independent impact on the outcome.

Limitations
Several limitations to this study need to be mentioned. First, we evaluated ARDS secondary to subarachnoid hemorrhage within the first seven days after the ICU admission, evaluating its occurrence on days one, three, and seven. The median ventilation time in our cohort was 2.00 [0.00, 11.00] days; therefore, patients could have developed ARDS afterward, which could result in an underestimation of the real incidence of ARDS in this group of patients.
Second, we focused on the early outcome after aSAH, at ICU discharge. However, the evaluation of long-term neurological outcome in aSAH would have provided a much stronger message to our study.
Third, we have reported associations but not causality, due to the retrospective observational design of this study; moreover, a retrospective data collection resulted in the lack of precise and complete data regarding ventilation and other confounders that might have influenced the patients' outcome.
Finally, we evaluated the daily PaO 2 through discontinuous arterial blood gases analysis that might represent less accurately the PaO 2 variability when compared to continuous measures.

Conclusion
Our results show that episodes of hypo-and hyperoxemia are frequent in aSAH patients. Also, ARDS has a low incidence in the first seven-day period of ICU admission after aSAH, and this could be due to the systematic use of protective ventilation strategies in our institutions. The most important predictors of outcome at ICU discharge are the severity of aSAH, the duration of mechanical ventilation, the ARDS occurrence, and the oxygenation parameters. The precise impact of hypoxemia, hyperoxemia, and oxygen variability needs to be further evaluated in interventional trials.

Ethics approval and consent to participate
The local ethics committee of each center approved the conduct of the study and waived the need for informed consent due to the retrospective nature of the study. In Monza: Comitato Etico Brianza, protocol number 3115; in Innsbruck: The Medical University of Innsbruck, protocol number AM4091-292/4.6; in Bruxelles: Erasme Hospital, protocol number P2018/128.

Consent for publication
Not applicable.

Availability of data and materials
The anonymized datasets analyzed during the current study are available from the corresponding author upon reasonable request.

Competing interests
The authors declare no conflict of interest for this manuscript.

Funding
No external funding has been obtained for this research.   Table 2 Characteristics of patients with a diagnosis of subarachnoid hemorrhage according to the presence or absence of a concomitant acute respiratory distress syndrome. An adverse outcome is defined as the presence of a modified Rankin score > 2 or of a Glasgow outcome score <4. ICU: Intensive Care Unit; IQR: interquartile range; SD: standard deviation; PaO 2 /FiO 2: the ratio between the arterial partial pressure in oxygen (PaO 2 ) in mmHg and the inhaled fraction of oxygen (FiO 2 ).   Figure 1 The course of patients admitted to the intensive care unit on days 1, 3, 7, and at discharge.

Figures
The outcome was evaluated at discharge from the intensive care unit; an adverse outcome was defined as a modified Rankin score >2 or a Glasgow outcome score <4. Data are presented as percentages of the whole population (n=855). ARDS: acute respiratory distress syndrome; w/o: without.

Figure 2
The ratio of arterial partial pressure in oxygen in mmHg and inhaled fraction of oxygen values at days 1, 3, and 7 in patients with and without acute respiratory distress syndrome.
X-axis: the ratio between arterial partial pressure in oxygen in mmHg and the inhaled fraction of oxygen. The blue bar represents the patients without acute respiratory distress syndrome and the red bar the patients with a diagnosis of acute respiratory distress syndrome; the white bar represents missing values. Data are presented as relative frequency among the whole population. NA: not available.

Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download. Additional_materials_oxygenation_derangements_after_SAH.docx