How much oxygen in adult cardiac arrest?

Sudden cardiac arrest (CA) is the leading cause of death among adults worldwide [1],[2]. In most patients, attempts at cardiopulmonary resuscitation (CPR) remain ineffective and spontaneous cardiac activity cannot be restored [3]. Among those patients who do achieve return of spontaneous circulation (ROSC), there are two key periods when death may occur: early (during the first three days), usually because of recurrent CA or severe cardiovascular failure resulting in multiple organ failure (MOF), and late (beyond day 3), usually secondary to withdrawal of life-sustaining therapies in the absence of neurological recovery [4]. Although several interventions, including target temperature management (TTM), have been introduced into the post-CA care of these patients [5],[6], conflicting results have been obtained [7], and these approaches are not sufficient to prevent the deleterious consequences of brain ischemia in all patients. During the post-CA care, secondary brain insult must be avoided [8] and optimization of brain oxygenation is likely to be an important component of brain recovery. The restoration of adequate systemic hemodynamics is a prerequisite to provide adequate cerebral blood flow in CA patients [9],[10], but brain oxygenation is also determined by the arterial oxygen content. Arterial oxygen pressure (PaO₂) itself may influence brain cellular oxygen supply; if hypoxemia (that is, PaO₂ of less than 60 mm Hg) is associated with poor outcomes after CA [11], a high PaO₂ may also be detrimental in a vulnerable brain, as suggested in patients with traumatic brain injury or stroke [12],[13]. The aims of this article are to review the potential mechanisms of oxygen toxicity after CA and to discuss the clinical impact of oxygen management on post-CA care.

Post-cardiac arrest syndrome (PCAS) is a complex phenomenon, which shares several features with septic shock [7],[14]. In particular, PCAS includes a systemic inflammatory response that can be triggered by the ischemia-reperfusion injury and also specific precipitating events, such as concomitant infections or heart disease. Moreover, PCAS can contribute to brain injury and myocardial dysfunction and can rapidly lead to MOF. The primary ischemia-reperfusion injury [15] activates various intracellular pathways, promoting ion concentration disequilibrium with increased intracellular levels of inorganic phosphate, lactate, and H⁺, and resulting in an influx of calcium into the cell [16], which aggravates mitochondrial dysfunction and eventually leads to programmed cellular death (apoptosis). After reperfusion has occurred, other mediators, including superoxide (O₂^-), peroxynitrite (NO₂^-), hydrogen peroxide (H₂O₂), and hydroxyl radicals (OH^-), contribute to worsen cellular function by oxidizing and damaging numerous cellular components [17] (Figure 1). These reactive oxygen species (ROS) then have a central role in initiating and enhancing the post-ischemic damage [15]. Indeed, supra-normal oxygen concentrations in this context may further stimulate ROS production and contribute to worsen cellular function in a setting of impaired mitochondrial function and impaired oxygen utilization. Moreover, some other systemic detrimental effects of hyperoxemia have been known for many years [18]-[20]. Hyperoxemia causes systemic and coronary vasoconstriction, which can decrease cardiac output and induce myocardial ischemia. In some experimental models of global cerebral ischemia, hyperoxemia has been shown to be detrimental to the brain, probably also because of its vasoconstrictor effects [21],[22]. Hyperoxemia may also provoke or exacerbate seizures, which could aggravate brain injury [23],[24].

Several studies have assessed the effects of administering high oxygen concentrations in experimental models of CA. Pilcher and colleagues [25] recently reviewed studies that evaluated the role of different oxygenation strategies - that is, one with 100% inspired oxygen fraction (FiO₂) and the other with lower FiO₂ as a control - after ROSC. Six studies including 95 animals of different species were included in their final meta-analysis; in general, administration of high FiO₂ (100%) for 1 hour after ROSC resulted in a worse neurological outcome, as assessed by a neurological deficit score, than other FiO₂ values. Four of the five studies that assessed histological damage reported a significantly higher neuronal injury with high FiO₂; cerebral metabolic function was also more altered in the high FiO₂ group. The extrapolation of such findings to humans, however, may be misleading. First, in all of these experimental studies, the animals were already mechanically ventilated before CA, and this is not usually the case in humans; also, control animals did not all receive the same FiO₂ (ranging from 21% to higher levels based on PaO₂ values), and the intervention period was rather short (that is, 1 hour). Moreover, different models of CA were employed (that is, asphyxia versus electrical-induced ventricular fibrillation), with shorter durations of CPR than in humans, and in some articles clear PaO₂ targets were aimed at, whereas in others only FiO₂ was modified without looking at PaO₂ values. In addition, no potential neuroprotective therapy, such as TTM, was provided, so that its influence cannot be ascertained. In summary, animal data have highlighted a clear correlation between the application of high FiO₂ after CA and poor neurological recovery. Nevertheless, experimental models are quite remote from the human setting, so that the observations cannot be readily translated to humans in the post-CA management.

Two studies reported conflicting results regarding oxygen management during the early phase of CA (Table 1). In 145 out-of-hospital CA patients, Spindelboeck and colleagues [26] observed that high PaO₂ (more than 300 mm Hg) levels during CPR were associated with higher rates of ROSC and intact neurological survival when compared with low (PaO₂ of less than 60 mm Hg) and normal PaO₂. In contrast, Kuisma and colleagues [27] randomly assigned 28 patients to receive either 30% or 100% FiO₂ for 1 hour after ROSC; five patients (36%) in the group with lower FiO₂ required an increase in FiO₂ to 40% because oxygen saturation fell below 95%, according to a standardized protocol. The primary endpoints were biomarker levels of brain injury - that is, neuron-specific enolase (NSE) and S100B - at 24 and 48 hours after ROSC, with no differences in absolute values between the two groups overall. However, in the subgroup of patients who did not undergo TTM, those receiving lower FiO₂ had lower NSE levels at 24 hours than the other patients (7.6 ± 4.2 versus 13.5 ± 9.6 μg/mL, P = 0.049). Thus, one may argue that high PaO₂ during CPR reflects better lung function or better quality CPR or both, with higher blood flow and better tissue oxygenation (indeed all patients received FiO₂ of 100%), whereas administration of high FiO₂ immediately after ROSC could enhance brain injury. Importantly, the retrospective nature of the study by Spindelboeck and colleagues [26] may have limited the analysis of all factors possibly affecting outcome (that is, quality of CPR, comorbidities, and so on). Also, blood gas analysis samples were drawn within 60 minutes from CPR initiation, and it is difficult to compare patients analyzed in the early CPR with those included in the later phase, who may have developed pulmonary injury with reduced PaO₂ due to prolonged resuscitation itself. It is difficult to determine the role of oxygen levels during CPR per se on neurological outcome of CA patients because they may be a surrogate marker of resuscitation performance or better cardiorespiratory status or both. Moreover, brain injury is a continuous process, and the time course of PaO₂ may more significantly influence neurological outcome in these patients.

Few clinical studies have evaluated the role of hyperoxemia during the ICU stay on outcome of CA patients. Kilgannon and colleagues published two different analyses on the same large database from a cohort of patients from 131 US hospitals (Increase Minority Participation and Awareness of Clinical Trials (IMPACT) Database). In the first study [28], including 6,326 patients, hospital mortality was higher in patients with hypoxemia (defined as a PaO₂ of less than 60 mm Hg or altered gas exchange with a PaO₂/FiO₂ ratio of less than 300) or hyperoxemia (PaO₂ of more than 300 mm Hg) detected in the first arterial blood gas (ABG) available within 24 hours after admission, compared with patients with normal oxygen levels (hypoxemia 53% versus hyperoxemia 67% versus normoxemia 45%, P < 0.001). In a multivariable logistic regression analysis, hyperoxemia was independently associated with in-hospital mortality: odds ratio (OR) 1.8, 95% confidence interval (CI) 1.5 to 2.2. In the second study, after exclusion of patients with hypoxemia, the investigators wanted to evaluate the association between PaO₂, considered as a continuous variable, and in-hospital mortality [29]. In the 4,459 patients studied, the median PaO₂ was 231 (interquartile range 149 to 349) mm Hg and in-hospital mortality was 54%. Multivariable analysis yielded a significant 6% increase in mortality for any PaO₂ increase by 25 mm Hg (OR 1.06, 95% CI 1.05 to 1.07). Although these two studies support the hypothesis that hyperoxemia is associated with worse outcome after CA, some limitations must be acknowledged. First, only a small proportion of patients were treated with hypothermia. Second, the use of the PaO₂/FiO₂ ratio does not really reflect hypoxemia, because it is a surrogate of lung failure rather than of low arterial oxygen content. Third, data were not collected following the Utstein style [30] or corrected for severity of disease (for example, Acute Physiology and Chronic Health Evaluation III, or APACHE III). Finally, and even more importantly, only the first ABG was taken into account in the final analysis instead of a mean value over the first 24 hours, which might be more representative of real oxygen exposure during this crucial period.

In a large Australian database (Australian and New Zealand Intensive Care Society-Adult Patient Database, or ANZICS-APD), Bellomo and colleagues [31] evaluated the correlation between different PaO₂ levels and hospital mortality in 12,108 patients. Although these authors used the same definitions of hypoxemia and hyperoxemia used in previous studies [28],[29], they reported a lower percentage of patients with hyperoxemia (11% [31] versus 18% [28]) but still a higher mortality rate for patients with hyperoxemia (59%) and hypoxemia (60%) when compared with normoxic patients (47%). In a multivariable model including some major confounding factors, hyperoxemia was significantly associated with mortality (OR 1.2, 95% CI 1.1 to 1.6); however, in a Cox proportional hazards regression model, after adjustment for other relevant covariates (year of admission, treatment limitations, patient’s lowest glucose level in the first 24 hours, patient’s indigenous status, and hospital source from home), PaO₂ was no longer statistically associated with poor outcome (hazard ratio 1.1, 95% CI 1.0 to 1.2; P = 0.20). Despite the limitation of a retrospective analysis, this study highlights that some confounders not taken into account in previous studies may have influenced the statistical association between high PaO₂ and poor outcome. Importantly, other issues may further limit the power of these findings. The PaO₂ from the ABG analysis associated with the worst alveolo-arterial oxygen difference was used, and this PaO₂ correlated only fairly with the mean PaO₂ (as shown in the analysis of a subgroup of patients) and may not have adequately reflected the exposure of patients to high oxygen concentrations. This observation was also suggested by the lower PaO₂ values recorded in this study when compared with previous publications (112 versus 231 mm Hg) [29]. Also, no data on neurological outcomes were reported. Finally, as in previous studies, only a minority of patients underwent hypothermia, but cooling is considered to mitigate ROS production and possibly influence hyperoxemia-mediated brain injury [8].

The first article to include patients treated with hypothermia [32] showed that those with a poor outcome had higher PaO₂ values than others (254 versus 198 mm Hg; P = 0.022). A multivariable regression model, including factors known to be associated with poor outcome after CA, confirmed an independent correlation of PaO₂ with mortality (OR for a PaO₂ increment of every 100 mm Hg above 54 mm Hg 1.48, 95% CI 1.03 to 2.01) and worse neurological outcome (OR 1.48, 95% CI 1.03 to 2.14) at hospital discharge; however, neither FiO₂ nor APACHE III score was included in the final analysis. Unfortunately, no specific cutoff of PaO₂ was identified to predict poor outcome, although a PaO₂ of more than 228 mm Hg was associated with a lower likelihood of neurological recovery. More recently, Ihle and colleagues [33] reviewed 584 patients selected from the ANZICS-APD database, for whom variables could be found following the Utstein model [30]. Unadjusted in-hospital mortality did not differ across different PaO₂ ranges (51% hypoxia, 41% normoxia, and 47% hyperoxia; P = 0.28). In a multivariable model including CA characteristics, neither hypoxemia nor hyperoxemia was an independent predictor of in-hospital mortality. In a retrospective cohort of 213 adult patients with CA, Lee and colleagues [34] reported that PaO₂ obtained during the first 24 hours was not related to hospital mortality or neurological outcome. In a multivariable model adjusted for established confounding factors after CA, the authors showed a V-shaped relationship between PaO₂ and neurological outcome, with the highest probability of good neurologic outcome at PaO₂ around 130 mm Hg. Finally, in a prospective observational study (n = 409), Vaahersalo and colleagues [35] calculated the proportion of time spent in different oxygen categories (less than 75 mm Hg, 75 to 150 mm Hg, 150 to 225 mm Hg, and more than 225 mm Hg) during the first 24 hours after CA. The proportion of time spent with a PaO₂ of more than 225 mm Hg was similar between groups, as was the association of time within different PaO₂ categories and outcome. Mean and highest PaO₂ values were higher in patients with good neurological outcome than in those with poor neurological outcome (120 versus 113 mm Hg and 173 versus 150 mm Hg, respectively). Interestingly, the proportion of patients with good neurological outcome was higher in patients with the combination of high mean PaO₂ and PaCO₂ values.

The International Liaison Committee on Resuscitation states that oxygen administration should be titrated to obtain an oxygen saturation of 94% to 96% after ROSC [36]. Thus, routine administration of an FiO₂ of 100% is no longer recommended after CPR [36],[37]; however, it still seems prudent to use 100% oxygen during CPR, although the impact of high PaO₂ on survival needs to be further evaluated. Although some studies have suggested that hyperoxemia after CA should be considered a cost-free, potentially modifiable risk factor for poor outcome, many potential confounders have been identified and strongly challenge this concept. Moreover, it is very difficult to identify a specific threshold of toxicity, as most of the studies used a PaO₂ of more than 300 mm Hg to define hyperoxemia, while some brain injury may also potentially occur at lower values. Also, when a single ABG value was used, the proportion of patients with hyperoxemia ranged from 3% to 25% in the different studies [27]-[29],[31]-[34]. Nevertheless, when all ABG values are considered, the incidence of hyperoxemia may exceed 40% [38], so that the real impact of exposure to high oxygen concentrations has probably been underestimated. Other parameters obtained from the ABG, such as acidemia or hypocapnia, may also be important determinants of poor outcome after CA [39],[40]; however, these variables were not considered in these studies. Importantly, it is worthwhile to prospectively evaluate the mechanisms of oxygen toxicity after CA, such as seizure development, ROS production, impaired microcirculation, or the development of organ dysfunction. Finally, further studies are needed to help understand whether an absolute PaO₂ value (that is, `peak’) could be more detrimental than a continuous exposure above a specific threshold and to assess the impact of the timing of hyperoxemia occurrence (early versus late phase after arrest) or of variability in oxygen levels (that is, from hypoxemia to hyperoxemia) in this setting.