Background
Trauma accounts for approximately 10% of deaths worldwide and causes a disproportionately high amount of disability and lost years of life [
1]. The magnitude of this problem may be even greater than recognized, as a significant number of patients survive hospitalization but ultimately succumb to complications of their injuries [
2]. Therefore, improvements in management of patients with severe traumatic injury are needed.
Exposure to supraphysiologic levels of partial pressure of arterial oxygen (hereafter “hyperoxia”) is common in mechanically ventilated, critically ill patients [
3‐
9]. The effect of hyperoxia on outcomes in a non-cardiac arrest population of critically ill patients remains unknown. Patients with severe traumatic injury may be especially susceptible to detrimental neurological effects of hyperoxia due to the prevalence of traumatic brain injury (TBI) in this population, but the biological and clinical effects of hyperoxia are complex and incompletely understood [
3‐
5,
10‐
12]. The observed prevalence and degree of hyperoxia vary widely among centers [
3‐
9] and have been associated with both better [
13,
14] and worse [
3,
6‐
9] outcomes in critically ill adults.
The association of hyperoxia with outcome has been previously studied in patients with critical traumatic injury. One large retrospective cohort analysis of TBI patients showed a U-shaped relationship between in-hospital and 6 month mortality and arterial oxygen tension in the first 48 h of admission, but there was no significant relationship between hyperoxia and mortality on multivariate analysis [
15]. In this study, analysis was limited to arterial blood gases recorded according to APACHE II methodology (i.e., the highest alveolar-arterial gradient for patients on FiO
2 0.5 or greater, or the lowest PaO
2 for patients on FiO
2 < 0.5). This approach would tend to select for the blood gas associated with lowest oxygen tension and thus may not be ideal for measurement of any potential effects of early hyperoxia on outcomes. In another analysis focusing on PaO
2 in the first 24 h, there was an independent association between mortality and hyperoxia (defined as PaO
2 > 300 mmHg) in a TBI cohort [
9]. Thus, data is mixed on this question perhaps due to differences in the time point analyzed and approach to assessing arterial oxygen tension.
Because the peri-contusional brain tissue in TBI is at risk for hypoxic-ischemic injury, strategies have been developed to directly measure the partial pressure of oxygen in brain tissue (PBrO
2) using brain probe monitors to guide oxygen management via adjustments in FiO
2 and/or mean airway pressure. A small phase II RCT suggests improved outcomes with such an approach [
16] and results of a larger RCT studying an aggressive PBrO
2-guided oxygen management are expected soon [
17]. However, these techniques are not yet universally adopted and may not be possible in resource-limited settings. Furthermore, even at institutions using this modality, there is an early period of resuscitation that occurs prior to implantation of PBrO
2 monitors. Consequently, the safe range of PaO
2 in critically ill adults without brain tissue oxygenation monitoring early after traumatic injury is unclear.
We performed a retrospective cohort study in mechanically ventilated patients with severe traumatic injuries to test the hypothesis that the maximum measured partial pressure of arterial oxygen (PaO2) within 24 h of admission would be associated with an increased risk of in-hospital mortality and that this association would be stronger in patients with head injury.
Discussion
Contrary to our hypothesis, we did not find an association between increased PaO2 within 24 h of ICU admission and in-hospital mortality in mechanically ventilated patients with severe traumatic injuries. Additionally, we did not detect such an association in the subgroup of patients with head injury. These findings provide reassuring evidence that hyperoxia early in the course of severe traumatic injury does not have major adverse effects.
It is worth noting that arterial hyperoxia as measured in this study is distinct from cerebral hyperoxia (i.e., elevated oxygen tension in brain tissue). PBrO
2 monitoring in individuals with TBI allows for an individualized approach to oxygen targets in TBI patients [
27‐
29] and its use may obviate concern for brain tissue oxygen toxicity in this population. However, this technology is not available at all centers. Furthermore, in the early period preceding insertion of the PBrO
2 monitor, the management of oxygenation is still empiric. It is tempting to assume that a liberal oxygen target is best in these patients because of the high incidence of cerebral hypo-oxygenation, as is recommended in pre-hospital trauma care guidelines [
29]. However, there may be some patients who would not benefit or would be harmed by such an approach; therefore, this question required further study. Our analysis affirms a prior report that such an approach is indeed safe [
15]. However, because the literature is mixed as to the effects of hyperoxia in TBI and our study was not powered to exclude a small effect in this secondary subgroup analysis, further investigation remains warranted.
The significance of these findings is most apparent when considered in light of the available literature on this topic. Considerable evidence has accrued that hyperoxia can have harmful biochemical and physiological effects. Elevated PaO
2 may increase the formation of reactive oxygen species (ROS) in the neuronal tissue bed, favoring the induction of neuronal cell death, and potentially contributing to poor neurological outcomes [
10,
11]. Supraphysiologic levels of PaO
2 can also cause cerebral vasoconstriction and heterogeneous tissue bed perfusion patterns, potentially resulting in paradoxically lower delivery of oxygen and other important substrates to cerebral tissue [
11,
30,
31].
Conversely, a beneficial effect of increased PaO
2 has been postulated for certain patient groups [
14,
32]. Human studies have associated hyperoxia with improvements in intracranial pressure, tissue bed oxygenation in both peri-contusional and remote neuronal tissue, and more aerobic neural metabolic profiles in patients with TBI [
14]. Early reductions in neurologic deficits and radiographic patterns of injury in patients with stroke who were exposed to hyperoxia have also been reported, supporting a potentially neuroprotective role for hyperoxia in the injured brain [
32].
In previous cohort studies, hyperoxia has been associated with increased mortality in a variety of clinical settings [
3,
5‐
9,
33] but these associations have not been seen in all studies [
4,
15,
34‐
37]. Data in the general mechanically ventilated population have been mixed with the largest retrospective cohort study to date finding no effect [
4,
5]. A recent meta-analysis found arterial hyperoxia to correlate with mortality but noted substantial heterogeneity of effect that could be due to differences in study design [
38].
In light of this heterogeneity, several aspects of the current study enhance the validity of its findings. One strength of this analysis is that all ABGs obtained in the first 24 h after admission were analyzed rather than just those that contributed to APACHE scoring. Furthermore, PaO
2 was analyzed as a continuous variable with the odds ratio of mortality for every fold increase of PaO
2 above 50 mmHg reported. This measurement of oxygen exposure differs from that used in other studies, and may address some of the limitations of prior reports. Past studies of hyperoxia have used an arbitrary PaO
2 cutoff of 200 or 300 mmHg, quintiles of hyperoxia exposure, or in the minority, PaO2 as a continuous variable [
3‐
7,
9]. We chose to analyze PaO
2 as a continuous variable because hyperoxia has no consensus definition, arbitrary partial pressure cut-off values may not accurately reflect biology, and because it increases statistical power. Analysis of the data using an arbitrary cutoff that has been used in some prior studies (300 mmHg) did not change our results (data not shown). The highest PaO
2 during the first 24 h was chosen as a marker of exposure to hyperoxia for this trial. This early time point was chosen because hyperoxia is more common in our center during the early period. Also, vulnerable tissues may be most susceptible to any ill effects of hyperoxia early after trauma. Other publications have examined a varied range of time points and, in some cases, only analyzed the ABG with the lowest alveolar-arterial gradient, lowest PaO
2, or average PaO
2, thus potentially missing many cases of hyperoxia exposure [
8,
15]. This early time point also increases the relevance of our results to the empiric management of oxygen prior to brain tissue monitor placement in centers with PbrO
2 monitoring capability.
Our study is subject to some limitations. The retrospective design limits interpretation of the results. Although about half of our patients were classified as having head injury, this diagnosis was defined as the bedside physician noting head injury in the medical record and likely includes patients with varying degrees of trauma to the neck, face, and cranium. Sufficient data were not collected as part of the VALID study to know whether patients had true TBI. Other explanations for not finding a significant difference in PaO
2 between survivors and non-survivors include a relatively low median PaO
2 observed (142 mmHg) in our entire cohort compared with past studies of oxygen exposure (major studies ranging from 99 to 247 mmHg) [
3‐
9]. Additionally, it is not known if risk accumulates over time with continuing exposure to hyperoxia, and our study was not designed to assess for such an effect. Furthermore, our study may have been underpowered to detect a small effect of hyperoxia. With our sample size of 471, we would have had 80% power to detect an association if the odds ratio corresponding to one-standard-deviation increase in maximum PaO
2 were 1.53. To achieve 80% power to detect our observed OR of 1.27 would require
N = 1973. Finally, there may be outcomes associated with increased PaO
2 other than mortality that the current study did not analyze, including neurologic outcomes, which have previously been shown to worsen in association with exposure to hyperoxia [
6]. However, GCS at time of discharge has been shown to be a reasonable surrogate for longer-term neurologic outcomes [
24] and a lower GCS did not associate with maximum PaO
2 in our cohort.
Acknowledgments
Not applicable.