Introduction
It is generally acknowledged that mechanical ventilation may cause or exacerbate lung damage in critically ill patients with acute lung injury (ALI) or acute respiratory distress syndrome (ARDS). Many studies have examined the effects of different settings of ventilation, such as low vs high tidal volumes, prone positioning and high-frequency oscillation on outcome of intensive care unit (ICU) patients [
1]. Lung-protective mechanical ventilation strategies in patients with ALI/ARDS, applying lower tidal volumes and sufficient levels of positive end expiratory pressure (PEEP) [
2,
3], have been shown to improve outcome.
The mode of mechanical ventilation and the oxygenation targets may influence the outcome for patients. Traditionally, arterial oxygen concentration (measured as partial oxygen pressure, PaO
2) and oxygen saturation by pulse oximetry are used as targets. Common recommendations for oxygenation propose PaO
2 values to be between 7.3 and 10.6 kPa [
2,
4]. The deleterious effects of hypoxia are well known and physicians may be mostly concerned about avoiding hypoxia and give additional oxygen 'to be on the safe side'. Hyperoxia, however, is also to be avoided as oxygen may be toxic. First, it is long known that high fraction of oxygen in inspired air (FiO
2) may be toxic for the lungs. In animals, prolonged hyperoxia causes histopathological changes similar to those seen in ARDS [
5]. Baboons exposed to 100% oxygen demonstrated a progressive reduction in forced vital capacity and functional residual capacity [
6] and proliferative epithelial changes and interstitial fibrosis [
7]. In healthy humans, exposure to 100% oxygen may lead to atelectasis, impaired mucocilliary clearance and tracheobronchitis, alveolar protein leakage and enhanced expression of leukotrienes by alveolar macrophages and increases in alveolar neutrophils [
8]. Apart from its effects on the lungs, oxygen may also lead to systemic toxicity. It has been associated with an increase in vascular resistance and a decrease in cardiac output [
9]. Hyperoxia may result in the generation of central nervous system, hepatic and pulmonary free radicals. Cardiopulmonary resuscitation following cardiac arrest in a canine model is associated with a worsened neurologic outcome when performed in the presence of hyperoxia vs normoxia [
8,
10].
The aim of the present study was to describe the present oxygenation targets applied in ICUs in The Netherlands, and to determine whether outcome of ICU patients was associated with differences in administered oxygen (FiO2) or achieved arterial PaO2.
Discussion
We found that administration of high FiO2 values in ICU patients was associated with increased in-hospital mortality. This association was found for FiO2 values in the first 24 h after admission and also for mean FiO2 during all admission days. The increased risk in patients with high FiO2 remained after correcting for SAPS II, admission type, reduced GCS score and pulmonary dysfunction measured as PaO2/FiO2 ratio. This suggests that the administration of oxygen itself could be deleterious, and that the association between high FiO2 and mortality cannot be explained by the confounding issue that highest FiO2 levels are administered in patients with severe pulmonary dysfunction.
Our observations are in accordance with prior experimental studies showing the potential toxicity of high fractions of inspired oxygen [
5]. Administration of supplemental oxygen can cause lung damage. This risk is especially high in prematurely born infants, probably attributable to inadequate host defences, underdeveloped lungs and immature antioxidant systems [
20]. Exposure to hyperoxia leads to diffuse pulmonary damage characterised by an extensive inflammatory response and destruction of the alveolar-capillary barrier leading to oedema, impaired gas exchange and respiratory failure [
21]. Mouse lungs exposed to > 90% oxygen for 48 h were more susceptible to ventilator-induced lung injury than those exposed to room air [
22]. Hyperoxia also aggravates pulmonary injury following artificial ventilation in rats using high tidal volumes [
23]. Furthermore, hyperoxia impairs the innate immune response by decreased macrophage function, impaired bacterial killing and increased susceptibility to pneumonia in a
Klebsiella pneumoniae model [
24]. Lung injury is likely to be initiated when the rates of generation of reactive oxygen species (ROS) are increased beyond the capacities of the antioxidant defences, such as the enzymes glutathione, superoxide dismutase and catalase. Mitochondrial mediated cell injury by ROS has been identified as a critical event in both apoptotic and necrotic forms of cell death in hyperoxia [
25]. Another organ that may be injured by hyperoxia is the kidney. Hyperoxic reperfusion exacerbates renal dysfunction and histopathologic injury after 30 min of complete normothermic ischaemia in rabbits. This hyperoxia associated dysfunction was prevented by the administration of the radical scavenger allopurinol [
26], suggesting that oxidative injury by ROS plays a role in post-ischaemic renal failure.
Several studies focused on the role of high reperfusion oxygen tensions following cardiac arrest and resuscitation. In a canine model of 10 min of cardiac arrest, resuscitation with 21% vs 100% inspired O
2 resulted in lower levels of oxidised brain lipids and improved neurological outcome [
27]. In another study using the same canine model, it was shown that resuscitation with 100% O
2 resulted in impaired hippocampal neuronal metabolism [
28]. Proposed pathogenetic mechanisms of hyperoxia induced reperfusion injury of the brain include increased production of ROS, a high ratio of oxidised over reduced glutathione [
29] and increased nitric oxide production by endothelium and neuron derived nitric oxide synthase [
30].
Many studies investigated the use of 100% vs 21% oxygen for resuscitation in depressed newborn infants (that is, infants with apnoea or relative bradycardia at birth). A systematic review and meta-analysis of 10 studies reported a significant reduction in the risk of neonatal mortality and a trend towards a reduction in severe encephalopathy in newborns resuscitated with 21% O
2. The reduction in mortality was also found in a subgroup analysis only including strictly randomised controlled trials and in a subgroup of studies enrolled in European countries with a lower risk of mortality than in less developed countries [
31].
Human clinical studies evaluating the effects of hyperoxia in critically ill adult patients are lacking. The effects of hyperoxia in non-ICU settings are not clear. A reduction in surgical site infections by the use of hyperoxia has been reported by one study group [
32], while others reported more surgical site infections in patients treated with hyperoxia [
33].
An alternative explanation for the association between oxygenation and mortality in ICU patients could be that common criteria for weaning from mechanical ventilation are based on FiO2 and PEEP levels. High FiO2 and PEEP, both leading to high PaO2 values, may delay weaning from mechanical ventilation, thus negatively influencing outcome in ICU patients. Also, we cannot exclude that high PaO2 values were achieved by more invasive ventilation strategies, potentially being more injurious to the patients.
Interestingly, apart from FiO2 values, there was also a U-shaped association between achieved arterial oxygen tension (PaO2) during the first 24 h after ICU admission and mortality with higher mortality in patients with either a very low or high PaO2. That mortality is higher in patients with very low PaO2 is not unexpected and possibly related to ischaemia or to selection of the sickest patients. However, mortality was also higher in patients with highest PaO2 values, suggesting the possibility of systemic oxygen toxicity.
In our analysis of mean oxygenation during all admission days, we again found a linear association between mortality and FiO2 values. Low PaO2s were also associated with higher mortality but high PaO2s were not. The shape of the association between PaO2 and mortality was hard to assess. In our data a linear association appeared to best fit the data (data not shown). The number of patients included in this analysis was only 3,322. Only 2% of the patients had a mean PaO2 higher than 20.0 kPa. Thus, the power of our study may have been too low to detect an association between high mean PaO2 values during the ICU stay and increased mortality.
There are limitations to this study. Most importantly, it was a retrospective observational study and the association between mortality and oxygenation is not necessarily causal. Although the association appeared to be independent of a number of potential confounding covariates, we cannot exclude that, despite our efforts, there are still differences in case mix associated with oxygenation that are not taken into account in our multivariate analyses. It is possible that physicians recognised some marker of severity that was not represented in our attempts to adjust for severity, and that they purposefully gave higher concentrations of oxygen to achieve higher levels of PaO2 in these high-risk patients.
The three potential confounders that we corrected for (age, reduced GCS score, and admission type) are part of the SAPS II that was also included as covariate in the multivariate analysis. We have repeated the analyses without these three variables, adjusting for SAPS II only. This yielded similar results for the association between in-hospital mortality and PaO2 and FiO2 respectively.
We corrected for pulmonary dysfunction by including PaO2/FiO2 ratio at admission in the multivariate analysis of data from all admission days. PaO2/FiO2 ratio was not included in the analysis of data from the first 24 h after ICU admission, because including PaO2, FiO2 and PaO2/FiO2 ratio, all from the same arterial blood sample, would introduce problems by colinearity of the data. In this population, however, we performed a separate multivariate analysis substituting PaO2/FiO2 ratio for PaO2 values. Again, FiO2 appeared to be a predictor of mortality, also independent of PaO2/FiO2 ratio (OR 1.15, 95% CI 1.14 to 1.17, model not shown). PaO2/FiO2 ratio is not only influenced by pulmonary dysfunction, but also by ventilator settings, such as PEEP levels. As PEEP was not part of the NICE data collection, we could not include this possible confounder in our analysis. Prospective, controlled trials are necessary to show a causal relationship between high FiO2s and mortality.
As the association between PaO2 and mortality was U-shaped, we categorised PaO2 values for the multivariate analysis using quintiles as categories (as no standard categorisation is available). The boundaries of quintiles are chosen arbitrarily and may not be the optimal cut-off levels to discriminate between patients with low and high risk of mortality. Therefore, we repeated the same multivariate analysis (model not shown) on data from the first 24 h after ICU admission using PaO2 values categorised as deciles and found similar results.
Another finding from our study is the fact that in most patients the achieved PaO
2 values are higher than the targets commonly recommended [
2,
4]. Although oxygen toxicity is a well known entity [
34], FiO
2s up to 0.5 are commonly considered 'safe' by physicians [
5]. It appears that physicians are more concerned about avoiding hypoxia and ischaemia than about the risks of hyperoxia. In The Netherlands, no formal guidelines for oxygenation targets are available. This may be related to the fact that the influence of oxygenation targets has never been studied making it impossible to provide evidence-based recommendations. Based on other observational studies, it may well be that also in other countries actual PaO
2s in ICU patients are higher than recommended [
35,
36]
Competing interests
During the period from 2002 to 2004 LP received an unrestricted educational grant from Eli Lilly Netherlands B.V. The study described in this manuscript was not conducted under the grant, and Eli Lilly Netherlands B.V. has not been involved in any part of the present study. All other authors declare that they have no competing interests.
Authors' contributions
EdJ designed the study and drafted the manuscript. LP and NdK were involved in the set-up of the study, performed the statistical analyses and helped in interpreting the results and writing the manuscript. PK was involved in the set-up of the study, interpreting the results and writing the manuscript. JJ, DdL, PvdV, RB, RdW and RW were involved in interpreting the results and writing the manuscript. All authors read and approved the final manuscript.