Hyperoxia during extracorporeal cardiopulmonary resuscitation for refractory cardiac arrest is associated with severe circulatory failure and increased mortality

Background High levels of arterial oxygen pressures (PaO2) have been associated with increased mortality in extracorporeal cardiopulmonary resuscitation (ECPR), but there is limited information regarding possible mechanisms linking hyperoxia and death in this setting, notably with respect to its hemodynamic consequences. We aimed therefore at evaluating a possible association between PaO2, circulatory failure and death during ECPR. Methods We retrospectively analyzed 44 consecutive cardiac arrest (CA) patients treated with ECPR to determine the association between the mean PaO2 over the first 24 h, arterial blood pressure, vasopressor and intravenous fluid therapies, mortality, and cause of deaths. Results Eleven patients (25%) survived to hospital discharge. The main causes of death were refractory circulatory shock (46%) and neurological damage (24%). Compared to survivors, non survivors had significantly higher mean 24 h PaO2 (306 ± 121 mmHg vs 164 ± 53 mmHg, p < 0.001), lower mean blood pressure and higher requirements in vasopressors and fluids, but displayed similar pulse pressure during the first 24 h (an index of native cardiac recovery). The mean 24 h PaO2 was significantly and positively correlated with the severity of hypotension and the intensity of vasoactive therapies. Patients dying from circulatory failure died after a median of 17 h, compared to a median of 58 h for patients dying from a neurological cause. Patients dying from neurological cause had better preserved blood pressure and lower vasopressor requirements. Conclusion In conclusion, hyperoxia is associated with increased mortality during ECPR, possibly by promoting circulatory collapse or delayed neurological damage. Supplementary Information The online version contains supplementary material available at 10.1186/s12872-021-02361-3.

treated with ECPR [3,4], implying that this strategy should be limited to highly selected patients [4][5][6]. Besides patient selection, improving the quality of postcardiac arrest care might also be key to improve ECPR outcome. The restoration of systemic oxygenated blood flow after cardiac arrest may trigger reperfusion injury, largely mediated by the generation of oxidants and free radicals [7], whose magnitude may be influenced by the arterial partial pressure of oxygen (PaO 2 ) during reperfusion [8,9]. In conventional CPR, an association between hyperoxia (defined as supranormal PaO 2 caused by high fractions of inspired oxygen, with a cut-off value of 300 mmHg frequently used in clinical studies) after return of spontaneous circulation (ROSC) and outcome has been indeed reported in large observational studies [10,11].
Patients undergoing ECPR may be exposed to significant hyperoxia, given the ease to oxygenate blood through the oxygenator [12], and recent retrospective analyses indicated that hyperoxia during the first 24 h of ECPR was associated with reduced survival [13,14]. One of these studies reported a significant association between circulatory shock during the first 24 h and mortality [14], whereas another study showed that most deaths (88%) occurred during the first 48 h and were related to multiple organ failure [13]. These findings suggest the hypothesis that early hyperoxia during ECPR may precipitate cardiovascular dysfunction leading to organ failure and death, by exacerbating post-resuscitation disease related to reperfusion injury. To address this hypothesis, we sought to determine the potential association between hyperoxia during the first 24 h of ECPR and the severity of hemodynamic failure, as well as its possible association with mortality in a cohort of 44 patients treated with ECPR for refractory IHCA and OHCA. In addition, since hyperoxia-mediated toxicity is largely related to the generation of oxygen free radicals, one can hypothesize that a longer duration of hyperoxia might be associated with a cumulative exposure to such species, hence greater toxicity, as recently shown by Roberts et al. in a prospective study on hyperoxia in cardiac arrest [11]. Therefore, instead of relying on a single value of PaO 2 , we averaged the values of several PaO 2 measurements during the first 24 h of ECPR, in order to take into account the duration of exposure to elevated PaO 2 levels during this period of time. We then evaluated a possible association between PaO 2 levels, circulatory failure and death during ECPR.

Study setting
This study was approved by our ethical committee (Commission Cantonale d'Ethique de la Recherche sur l'Etre Humain/CER-VD-Nr: 2017-01,184), as a retrospective use of clinical data with waiver of consent (CER-VD-Nr: 2017-011,184), and conforms to the STROBE statement for the report of observational studies. All methods were performed in accordance with the relevant guidelines and regulations. The cohort included 44 consecutive patients hospitalized in our 35-bed multidisciplinary ICU for refractory, non-traumatic and non-hypothermic, OHCA and IHCA treated with ECPR from June 2017 to June 2019. During this period, our internal recommendations for starting ECPR were: refractory OHCA or IHCA; no flow < 5 min; low flow < 80 min; age < 75 years; absence of major co-morbidities; absence of Do Not Resuscitate order. A subset of patients of the current cohort were also included in an unrelated study evaluating automated pupillometry during Veno-Arterial ECMO (VA-ECMO) for various indications [15].

Treatment strategy
The insertion of venous and arterial femoral cannulas was performed in the hospital by a cardiac surgeon. Initial VA-ECMO settings included a sweep fraction of oxygen (F S O2) of 100% and a pump setting for a target blood flow of 40-60 ml/kg, using the Maquet Cardiohelp ECLS system ® . Sweep gas flow was adapted to maintain normal PaCO 2 . Systemic anticoagulation (intravenous heparin) initiated at the end of surgery was maintained to achieve an anti-Xa activity of 0.3-0.5. A coronary angiography was performed at the discretion of the in charge physicians, when an ischemic origin of cardiac arrest was strongly suspected, according to internal recommendations. Noradrenaline (NA) and Adrenaline (Adre) were given to maintain mean blood pressure (BP) ≥ 65 mmHg. Intravenous (IV) fluids were administered to maintain the target blood flow. Mechanical ventilation was performed with a tidal volume of 6 ml/kg, rate of 10-15/ min and PEEP of 6-10 cm H 2 O. Both FiO 2 and FsO 2 were set by default at 100% on ECPR initiation. Following these initial settings, decisions to reduce FiO 2 or/and FsO 2 were entirely at the discretion of the treating physicians, without specific recommendation. Sedation was maintained with Propofol (2-4 mg/kg/h) or Midazolam (0.05-0.15 mg/h). Targeted temperature management at 35-36 °C for 24 h was implemented in all patients. Criteria for discontinuing VA-ECMO were the absence of cardiac function recovery, intractable circulatory shock, or evidence of severe neurological injury (major stroke on cerebral imaging or evidence of severe anoxic brain injury according to a multimodal outcome prediction, combining neurological examination, electrophysiological data, and plasma levels of serum neuron-specific enolase (NSE), as described in [16]

Data collection
Demographic variables included age, sex, location of CA (OHCA, IHCA), initial rhythm, duration of no flow and low flow (total duration of CPR before ECMO initiation), SAPS 2 score, ICU and hospital length of stay, duration of ECMO treatment, proportion of patients undergoing coronary angiography and angioplasty (PTCA), as well as the causes of death. Hemodynamic variables included: (A) Mean arterial blood pressure (mean BP, obtained via an arterial catheter in all but 3 patients), from which we computed the average from all measurements performed each 2 h during the first 24 h (or until death if it occurred before 24 h). (b) Pulse pressure (systolic minus diastolic blood pressure), determined during the first 24 h as an indirect indicator of native cardiac output recovery [17], which was averaged from values obtained at 2, 6, 12 and 24 h (or until death if it occurred before 24 h). (C) The amount of catecholamines, fluids, packed red blood cells and fresh frozen plasma administered (first 72 h).
Blood gas data were obtained from an indwelling intraarterial catheter, whose position was recorded (right radial, left radial or femoral). The values of PaO 2 measured at 5 different times of ECPR (first 15 min, 2 h, 6 h, 12 h and 24 h) were averaged as the mean 24 h PaO 2 (in patients dying before 24 h, mean PaO 2 was computed from values obtained during the time spent under ECPR until death). Arterial blood lactate was determined in the first arterial blood sampling (first 15 min of ECPR).

Data analysis
Continuous variables are expressed as means ± SD, or medians and interquartile ranges (IQR), and categorical data as absolute numbers and percentages. All comparisons were done using the Wilcoxon's rank sum test for continuous variables and the chi-square test for categorical variables. We determined which variables were associated with survival using univariate logistic regression analysis for continuous variables and contingency analysis with Pearson's test for categorical variables. To evaluate the association of hyperoxia with mortality, multivariable logistic regression was applied. We first considered a model including only co-variables significantly associated with mortality in univariate analysis (low flow duration, first arterial lactate, mean BP and mean PaO 2 ). We next applied a model including co-variables commonly considered as important clinical predictors of poor outcome, including age, low flow and no flow duration, shockable rhythm, the duration of ECMO, as well as mean PaO 2 . In both models, hospital mortality was the response binomial variable. Odds ratios (ORs) with 95% CI were calculated (for continuous variables, ORs were calculated per unit change of each variable), and Wald statistics was used to assess the significance of each variable. Furthermore, due to the relatively low number of events in our cohort and the inherent risk of overestimating regression coefficients [18], we complemented this analysis by running several logistic regressions using only mean PaO 2 and a second co-variable at a time (including co-variable with a p value < 0.2 in the multivariable logistic regression: no flow, low flow, and shockable rhythm). To control for possible type I error, we introduced a Bonferroni adjustment for assessing significance in these 3 models (thus, a p value of 0.05/3 = 0.016 was used as the significance limit in these analyses). To determine a possible association between PaO 2 and circulatory failure, we performed bivariate analyses and simple linear regressions between mean PaO 2 and mean blood pressure and between mean PaO 2 and the amount of vasopressive catecholamines administered, with calculation of the Pearson r coefficient. To determine the influence of the arterial catheter position on PaO 2 , PaO 2 values obtained in the different catheter positions were compared using Wilcoxon rank test. The JMP software, version 15, was used for all the analyses, and a p value < 0.05 was considered statistically significant.

Results
The characteristics of patients and differences between survivors and non survivors are shown in Table 1. No patient in our cohort displayed ROSC before ECMO initiation. For the whole cohort, in-hospital survival was 25% (11pts), including OHCA: 7/27 = 26% and IHCA: 4/17 = 24% (p = 0.86). Survivors had lower SAPS 2 scores, longer LOS and ECPR duration. There were no differences in terms of location of CA, initial rhythm, no flow duration and PaCO 2 levels, whereas low flow duration, first arterial lactate and intravenous fluids administration were significantly higher in non survivors.
In two patients dying early after ECPR initiation, arterial blood samples were not obtained. Non-survivors had significantly higher mean PaO 2 during the first 24 h (Table 1 and (Table 1 and Fig. 2A, B). Pulse pressure during the first 24 h was not different between non-survivors and survivors (Table 1 and Fig. 2C). The mean PaO 2 value during the first 24 h displayed a significant negative correlation with mean blood pressure (Fig. 2D) and positive correlation with total amount of vasopressors (NA + Adre, in μg/kg/min during the first 3 days, Fig. 2E).
Variables associated with in-hospital mortality at the 5% level in univariate analysis (Table 2) included the duration of low flow, mean PaO 2 , the first arterial lactate concentration, the amounts of catecholamines and IV fluids administered, as well as mean blood pressure during the first 24 h. Table 3 shows the results of the two models of multivariable logistic regression analyses, which indicate that mean PaO 2 was the only co-variate showing a p value < 0.05 for the association with mortality in both models. The results of the logistic regressions including only 2 co-variates (mean PaO 2 and low flow; mean PaO 2 and no flow; mean PaO 2 and initial rhythm), also show an association between mean PaO 2 and mortality following Bonferroni correction. The causes of refractory CA included acute myocardial ischemia (n = 24, 55% of patients), acute hypoxic CA (n = 4, 9%), primary arrhythmia (n = 3, 7%), acute pulmonary embolism (n = 3, 7%), toxic/metabolic origin (n = 3, 7%, including one cocaine intoxication, one aluminium phosphorus intoxication and one severe metabolic acidosis), heart transplant rejection (n = 2; 4%), undetermined cause (n = 2, 4%), and other causes (n = 3, 7%, including one intraoperative CA, one pericardial tamponade and one subarachnoid hemorrhage). The two main causes of death under ECPR were profound circulatory shock (n = 15, 46% of patients) and neurological injury (severe anoxic encephalopathy or major stroke, n = 8, 24% of patients). Other causes of death included hemorrhagic shock (n = 3). Multiple organ failure (n = 4), septic shock (n = 2) and technical issue (n = 1, related to an intractable failure to obtain an adequate venous inflow, possibly due to extensive thrombosis in the inferior vena cava). Comparisons between patients dying from circulatory or neurological causes are shown in Fig. 3. Low flow was not statistically different in the two groups (Fig. 3A). Patients dying from circulatory cause had significantly higher initial lactate (Fig. 3B) and shorter ICU LOS (Fig. 3C). The mean PaO 2 during the first 24 h was not different (p = 0.1 between both groups, Fig. 3D). Patients with neurological death had significantly higher mean BP during the first 24 h (Fig. 3E) and received significantly less vasopressive catecholamines (Fig. 3F).
Mean PaO 2 measured via a right arterial catheter was not different from that measured from a left radial catheter, but was lower than that measured via a femoral catheter, and mean PaO 2 measured from a left radial catheter  was not different from that measured via a femoral catheter (Fig. 4A). When comparing mean PaO 2 between survivors and non survivors according to the catheter position, values were significantly lower in survivors for each position of the arterial catheter (Fig. 4B).

Background and previous work
The restoration of systemic oxygenated blood flow after a prolonged period of hypoxia may trigger widespread reperfusion injury. A key mechanism of reperfusion injury is the generation of oxidants and free radicals [19], whose flux increases in proportion with the local PO 2 [20][21][22], implying an increased risk of oxidant-mediated damage at higher PO 2 during reperfusion [23]. In the setting of ECPR, the risk of hyperoxia is particularly elevated, due to the ease of oxygenating blood through the membrane oxygenator [5,12], and a few retrospective studies have indeed reported a negative impact of hyperoxia on survival after ECPR [13,14,24]. In a retrospective cohort of 291 ECPR patients, Chang et al. reported that a first PaO 2 value during the first 24 h after ECPR initiation between 77 and 220 mmHg, was significantly associated with neurologically intact survival in comparison to higher values [14]. In a retrospective cohort of 66 patients undergoing ECPR for refractory OHCA, Halter et al. reported a significant association between PaO 2 measured at 30 min after ECPR initiation and 28 days mortality. Using a threshold value of PaO 2 of 300 mmHg to define hyperoxia, the odds ratio for 28 days mortality was 4.07 [13]. In our study, non survivors displayed a higher mean 24 h PaO 2 , lower mean blood pressure, higher needs in vasopressors, and profound circulatory shock was the primary cause of death in a majority of non survivors. Taken together, these findings suggest that severe vascular failure with refractory circulatory shock may represent an important mechanism of hyperoxia toxicity during ECPR.

Hyperoxia toxicity in ECPR
Hyperoxia results in an increased vascular generation of superoxide (O 2 .− ), which reacts rapidly with nitric oxide (NO . ) to form peroxynitrite, that can trigger significant vascular contractile failure through a number of processes [25][26][27]. In addition, oxidants such as peroxynitrite promote the expression of multiple inflammatory cytokines and mediators [28][29][30][31], which also reduce   vascular tone and may precipitate hypotension. These effects may contribute to foster a sepsis-like state, characterized by an irreversible loss of vascular contractility and refractory hypotension with negative prognostic impact after prolonged resuscitation and ECPR, as recently reported by Jouffroy et al. [32]. Obviously, the formation of peroxynitrite and other oxidants, together with the generation of inflammatory mediators, at different levels of PaO 2 during ECPR, should be evaluated in future investigations to explore these mechanisms.

Importance of the localization of arterial blood sampling in ECPR
A critical aspect in the interpretation of arterial blood gases during peripheral VA-ECMO is the localization of arterial blood sampling and right radial artery sampling is recommended [33], owing to the risk of upper body hypoxia in case of cardiac recovery and impaired pulmonary gas exchange [34]. Accordingly, we found that PaO 2 was lower when measured from a right radial artery than from a femoral artery, while it did not differ from values obtained from the left radial artery. One could therefore argue that the lower PaO 2 in survivors might reflect an earlier recovery of native cardiac function, but this appears unlikely in view of the similar values of pulse pressure, an indirect, real-time measure of native cardiac output during extracorporeal support [35], in survivors and non survivors.

Main causes of deaths during ECPR
The two main causes of death were early circulatory failure and delayed neurological damage. Patients dying from either cause had similar durations of low flow and mean PaO 2 , but initial lactate levels were lower in patients dying from neurological damage, pointing to less profound systemic anoxia. These patients also had higher mean blood pressure and required significantly less vasopressors. Overall, these findings suggest that patients with more severe anoxia (higher lactate levels) might develop more severe reperfusion injury under hyperoxic conditions, leading to predominantly vascular failure and early deaths, whereas those with less severe anoxia would survive the early stage and develop delayed hyperoxic neurological damage. This hypothesis should require validation in larger cohorts of patients treated with ECPR for refractory cardiac arrest.

PaO 2 is the main variable associated with mortality in ECPR
We did not find significant association between the initial rhythm and survival, which differs from the notion that survival under ECPR is better in patients with an initial shockable rhythm [36]. This discrepancy most likely reflects the small sample size in our study. Only 52% of patients had an initial shockable rhythm, and among those with a non-shockable rhythm, all patients with asystole died, whereas 3 out of 14 patients (21%) with pulseless electrical activity (PEA) survived to hospital discharge. Interestingly, these 3 patients had the lowest mean PaO 2 (152, 162 and 184 mmHg, respectively) among the 14 PEA patient, which may suggest that the avoidance of hyperoxia during ECPR could be critical to determine outcome in PEA patients undergoing ECPR. Besides PaO 2 , survival in our cohort was significantly associated with the duration of low flow in univariate analysis, in line with previous investigations [37][38][39]. However, this association was not observed in a multivariable analysis evaluating several co-variates commonly associated with poor outcome after CA and ECPR, such as no flow, low flow, initial rhythm, localization of CA (IHCA vs OHCA) and age, together with mean 24 h PaO 2 . We also included the duration of ECMO as a co-variate, given that the initial sweep gas oxygen fraction was set at 100%, which could have favored persisting hyperoxia in these patients. With the exception of mean PaO 2 , none of these variables displayed a significant association with mortality. Although these findings may suggest a particularly negative prognostic implication of hyperoxia in ECPR, they warrant cautious interpretation. Owing to the relatively small sample size in our study, the results of multivariate analysis may be subject to some bias related to an overestimation of regression coefficients [18]. For this reason, we performed additional analyses using only one co-variate with mean PaO 2 , including no flow, low flow and initial rhythm, which confirmed the association of mean PaO 2 with mortality. In addition, due to the retrospective nature of our study, the association of PaO 2 with mortality could represent a surrogate for the sickness of patients. Physicians would indeed be inclined to maintain high levels of administered O 2 to the most severely affected patients, and hyperoxia would therefore be a side effect of such management.

Study limitations
Our study has several limitations. First, we must acknowledge the usual limitations related to the retrospective design of our study [40] and to its relatively small sample size. Second, although we established an association between hyperoxia and mortality, as well with refractory circulatory failure, such associations do not necessarily mean a causal relationship. Third, the absence of internal recommendations for the management of arterial oxygen levels during ECPR may have favored the development of hyperoxia, which could have been avoided with a dedicated clinical protocol. Fourth, we relied on pulse pressure as an indirect evaluation of native cardiac output, which may be of limited reliability under conditions of varying cardiac loading conditions, especially afterload [41]. Direct assessment of native cardiac function using echocardiography would have been more appropriate in this setting, but we do not routinely perform echocardiography in the first 24 h after ECPR initiation. Fifth, we did not measure circulating mediators related to oxidative stress and inflammation, which could have given important insights on the effects of hyperoxic reperfusion during ECPR. Such measurements will be the matter of additional future investigations.

Conclusion
In conclusion, although we acknowledge several important limitations, our study shows that, in patients undergoing ECPR for refractory cardiac arrest, a high PaO 2 during the first 24 h of support is associated with worse outcome, possibly by promoting severe vascular failure and refractory circulatory shock. PaO 2 should therefore be strictly controlled during the first 24 h of ECPR, via blood sampling from the right radial artery. Future studies should evaluate the impact of different PaO 2 levels during ECPR on biomarkers of oxidative stress and inflammation, to provide further insights into possible mechanisms of hemodynamic collapse in this setting.