The effect of hyperoxia on mortality in critically ill patients: a systematic review and meta analysis

Oxygen supplement is a life saving treatment commonly used in the critically ill patients [1]. Excess oxygen delivery was reported to be a very common phenomenon, in which about 50% of the patients showed hyperoxemia and 4% in severe hyperoxemia [2].

Animal studies showed that hyperoxia is associated with adverse events such as histopathological injury, interstitial fibrosis, atelectasis, tracheobronchitis, alveolar protein leakage and infiltration by neutrophils [3]. Hyperoxia interacts with mechanical stretch to augment ventilator-induced lung injury [4]. Moreover, hyperoxia could also lead to a decline of cardiac output, [5] coronary blood flow and myocardial oxygen consumption, [6] and generate free radical-mediated damages in various organs [7]. Studies in human demonstrated that hyperoxia could impair the responsiveness of host defense to infections [8]. Hyperoxia may affect a variety of patients’ biological systems, such as antioxidant enzymes [9] and cytokine production [10] through excessive production of reactive oxygen species. Exaggerated apoptosis, in part through the death receptor-mediated signals, accelerates hyperoxia-induced acute lung injury [11]. However, clinical studies testing the relationship between hyperoxia and mortality in critically ill patients have yielded conflicting results. For example, in a study of 36,307 patients admitted to the intensive care units (ICU), no difference in mortality was noted between the patients exposed to hyperoxia and those who did not [12]. In contrast, Page et al. found that there was an association between hyperxia and increased mortality (adjusted odds ratio[aOR] 1.95, 95% confidence interval[CI] 1.34–2.85) [13]. These conflicts were also seen in some specified diseases such as 1) patients with cardiac arrest, in which Elmer et al. showed a decreased survival (aOR 0.83, 95% 0.69–0.99, P = 0.04), [14] while Ihle et al. did not find any difference; [15] and 2) patients after traumatic brain injury(TBI), in which Asher et al. found a reduced mortality by hyperoxia while a contrary result was showed by Davis et al. [16, 17].

Although previous studies have performed the analysis of the relationship between hyperoxia and mortality, no solid conclusion has been drawn. As for several new studies in this topic being published recently, we conducted a systematic review and meta-analysis of all published trials aiming for identifying the roles of hyperoxia in the outcomes of patients in ICU.

Initially 3173 records were identified, of which 3168 were extracted from electronic databases and 5 were from the reference lists review. (Figure 1) By screening the titles and abstracts, 3138 studies were discarded for duplication (n = 679), animal experiments (n = 1134) and non-adult patients (n = 1325). We searched the full-text articles for the remaining 35 studies, and eventually 24 trials were enrolled in our final analysis due to 10 studies not reporting related outcomes, and 1 did not designed as expected.

Hyperoxia defined as partial arterial pressure of oxygen (PaO₂) > 487 mm Hg(mmHg), 341 mmHg, 300 mmHg, 200 mmHg, 156.7 mmHg, 150 mmHg, 120 mmHg and 100 mmHg was reported in 1, 1, 8, 2, 2, 2, 2, and 3 studies, respectively. One study defined hyperoxia as FiO₂ of 1.0, and the other two studies did not specify the definition of hyperoxia. As for the definition of normoxia, 12 studies did not report the details and the details of the other 12 studies have been listed in the Table 1.

In our meta-analysis, we found that hyperoxia would increase the mortality of critically ill patients especially in the ones with cardiac arrest and ELS. In the patients with TBI, stroke and hemorrhage, post cardiac surgery and receiving mechanical ventilation due to mixed primary disease, hyperoxia did not increase mortality.

We found that hyperoxia would lead to adverse outcome in patients with cardiac arrest and ELS. Both of them included a process of reducing blood flow through the cardiac and unstable hemodynamics. The reasons for increased mortality by hyperoxia in patients with these diseases might be as follows. First of all, the exposing to hyperoxia would lead to the increase of reactive oxygen species, which would inhibit the vasodilators such as nitric oxide [37]. Study showed that vasodilator drug-based therapy is a superior solution to reduce mortality in hemodynamically unstable people such as acute myocardial infarction, heart failure or cardiac surgery [38]. Secondly, hyperoxia might increase vascular resistance and reduce cardiac output, coronary blood flow and myocardial O₂ consumption [5, 6]. Reduced intravascular volume would result in cardiogenic shock, which is a life-threatening condition of circulatory failure [39]. Thirdly, pulmonary toxicity such as endothelial cell destruction, interstitial edema, and type I cell injury caused by the hyperoxia was noticed in animal experiment, and the resulting physiology responses included the increasing of pulmonary leukocyte and neutrophil accumulation, extravascular lung water and permeability surface-area [40]. As a result, the total lung capacity decreased and ventilator-associated lung injury augmented [3, 40]. Clinical study also showed that hyperoxia was associated with higher rate of ventilator associated pneumonia [41]. In addition, oxidative stress due to increased oxygen radical formation is one possible mechanism of ICU-required weakness [42]. Forthly, animal studies showed that high oxygen contraction would increase the oxidative stress and decrease antioxidant. These oxidants can react and damage the cellular macromolecules by virtue of the reactivity that leads to cell injury and necrosis. Oxidants are also mediators in damaging various organs and the organ failure is related to increased mortality [43]. It is reported that a high arterial oxygenation leads to myocardial reperfusion damage, with similar inflammatory consequences secondary to O₂ free radical development.

However, in patients with TBI, stroke and hemorrhage, post cardiac surgery and received mechanical ventilation due to mixed primary disease, hyperoxia did not influence the mortality. The most possible explanation for this phenomenon is that the number of patients included in each subgroup was too small to find any significant differences in the mortality between the hyperoxia group and normoxia group. However, there are still some assumptions. Traumatic injury and compartment syndrome may appear to be obvious applications for supplementary oxygen because an increased pressure of oxygen would help overcome the decline in perfusion. Several studies have showed the association between hyperoxia and poor outcome after severe TBI [44]. In TBI patients, oxygen delivery can be affected by a decline in cerebral blood flow. In addition, enhancement of brain edema and endothelial swelling after TBI has been demonstrated to decline the diffusion gradient for oxygen to the mitochondria [45]. There are both experimental and indirect clinical studies suggesting that after TBI aerobic metabolism in the brain goes down, [46] which leads to the mitochondrial dysfunction following TBI. Therefore, it is assumed that hyperoxic therapy could improve the oxygen content and thus raise the partial pressure of oxygen, which is the driving force for oxygen move to the mitochondria. This may promote aerobic metabolism in the brain, and thus improves overall outcomes [47]. The same effect also existed in patients with stroke and hemorrhage. There are increasing evidences illustrating that detrimental ischemia-related processes may be present although cerebral perfusion pressure and intracranial pressure levels are adequate [48]. The most effective way to improve hypoxia in brain tissue is to provide higher fraction of oxygen in inspired gas [49]. Therefore, although hyperoxia plays a role in lung injury and various organ failure, the advantages of hyperoxia might overweight the disadvantages and have no effect on the mortality. As for the patients with post cardiac surgery, although no significant difference was found in the mortality, a slight increase in the ICU and hospital stay may indicate a trend towards harm [32]. In addition, type 2 errors and mixed primary diseases may contribute to the result in patients with mechanical ventilation. In patients with septic shock, hyperoxia could lead to vascular contract and reduce oxygen uptake. Thus, hyperoxia could allow for hemodynamic stabilization during vasodilatory shock and do not increase the mortality [50]. Moreover, hyperoxia contributes to the improvement in tissue bed oxygenation in both peri-contusional and remote neuronal tissue, and more aerobic neural metabolic profiles [51]. This might be the reason that hyperoxia did not increase the mortality in patients with severe traumatic injurie.

We should notice that almost all the included studies in our analysis investigated the oxygenation target in the first 24 h rather than the whole phase in the ICU. According to the previous studies, hyperoxia was most common in the first 24 h in ICU. Meanwhile, the lung would get injuries when exposed to hyperoxia within 24 h [52]. In Kraft’s study [53], they investigated the average oxygenation target during the whole ICU stay, which means the high PaO2 happened in the first 24 h might be offset by the low PaO2 value in the following days.

Moreover, clinical heterogeneity existed in our analysis, which might lead to difference in mortality between included studies. First of all, different definition of hyperoxia and normoxia. For example, in the Helmerhors’ study, the definition of normoxia was 60 < PaO2 < 300 mmHg. However, in Elmer’s study, the definition of hyperoxia was PaO2 > 100 mmHg. Thus, part of patients, who should be included in the hyperoxia group in Elmers’ study, were actually included in the normoxia group in the study of Helmerhors. Thus, the application of our study was limited. Secondly, we should notice that demographic information such as age, gender rate and severity of disease of patients were different. For example, the location of arrest was different in the 7 studies, which only included patients with cardiac arrest. The rate of patients’ cardiac arrest happened outside hospital varied from 57 to 100%, which would significantly influence the mortality of patients after cardiac arrest [54, 55].

There are some limitations of our analysis, which should be demonstrated. First of all, almost all the studies included in our analysis were retrospective studies. In this way, we could not figure out the relationship between the severity of the diseases and the level of arterial oxygen pressure. Second, clinical heterogeneity existed due to the mixed definition of hyperoxia, the oxygenation target in comparison group and the timing for measuring the outcome of patients. Third, in some subgroups, the number of patients was too small. Forth, significant statistical heterogeneity existed both in the overall and subgroup comparisons. Thus, the application of our conclusion is limited.

In conclusion, hyperoxia in patients admitted to the ICU would lead to higher mortality, which has been further confirmed in the patients with cardiac arrest and ELS.