Trials on oxygen targets in the critically ill patients: do they change our knowledge and practice?

In the last decades, the description of the U-shaped association between PaO₂ and mortality [1] highlighted the potential dangers of high PaO₂ levels (hyperoxaemia) in critically ill patients. Subsequent randomised controlled trials (RCTs) have generated conflicting results. Earlier trials favoured conservative oxygen therapy; two moderate-sized RCTs both reported an 8% absolute decrease in mortality compared to either a liberal oxygenation group [2] or to septic patients exposed to an FiO₂ of 100% for 24 h (HYPERS2S) [3]. A meta-analysis of 25 RCTs performed in acutely ill patients found a 1.21 (95% CI 1.03–1.43) increase in relative risk for hospital mortality with liberal oxygen therapy [4]. Two later and larger RCTs (ICU-ROX and HOT-ICU), however, failed to show any mortality difference [5, 6], while two other moderate-sized RCTs (O2-ICU and LOCO2) even trended towards a better clinical outcome in the liberal oxygenation groups [7, 8]. Important limitations of these trials include heterogeneous populations and the use of different arterial oxygenation targets that generally compared mild hypoxaemic against normoxaemic targets. The only conclusion that can be currently drawn is that extremes of oxygenation are disadvantageous. A fairly broad range of less extreme values appear to be safe and any clinical impact is likely to be minor.

Several important questions remain. Are there specific subpopulations which benefit from a higher or lower arterial oxygenation target? Since the nadir of the U-shaped PaO₂-mortality association was in the mild hyperoxaemic range at around 130 mmHg [1], is the optimal target higher than currently investigated? What upper limit of hyperoxaemia can be considered safe?

Recent trials

Three large RCTs assessing PaO₂ targets in different critically ill populations of patients are summarised in Table 1. The BOX trial [9] enrolled 789 adult patients comatose after out-of-hospital cardiac arrest (OHCA) admitted to two Danish hospitals. After admission, mechanically ventilated patients were randomised to a higher (97–105 mmHg) or lower (67–75 mmHg) PaO₂ target (median duration 60 h). Both primary (death or hospital discharge with coma) and secondary endpoints, including adverse events, did not differ between the two PaO₂ groups. The limited data on PaO₂ levels showed a minimal difference (~ 6–11 mmHg) between the two groups over the first 48 h as median PaO₂ levels exceeded 75 mmHg in the conservative group.

Table 1 Characteristics of patients included, setting, oxygenation protocols and outcomes of the three recent trials on oxygen targets in critically ill patients

The EXACT trial [10] assessed two different SpO₂ targets, 90–94% vs 98–100%, in the immediate management of patients with return of spontaneous circulation (ROSC) after OHCA in two Australian emergency medical services. The protocol was initiated within 40 min after ROSC and terminated after the first blood gas analysis taken in the intensive care unit (ICU). Hospital survival, the primary endpoint, was higher (47.9 vs 38.3%; p = 0.05) in patients randomised to a higher SpO₂ target. This group also suffered fewer hypoxaemic episodes (16.1 vs 31.3% p < 0.001) pre-ICU admission. Unfortunately, the study was stopped prematurely due to the coronavirus disease 2019 (COVID-19) pandemic after enrolling a third of the planned sample size. The authors also acknowledged difficulties in providing an accurate FiO₂ and following the protocol in an out-of-hospital setting. As with the BOX trial, differences in SpO₂ between the two groups on arrival in the Emergency Department (97% vs 99%) and at protocol end (98% vs 99%) were minimal.

The cluster crossover PILOT trial [11] evaluated three different SpO₂ targets [90% (88–92%) vs 94% (92–96%) vs 98% (96–100%)] in mechanically ventilated adult patients in one US hospital. More liberal targets were allowed during transport or procedures. The protocol was applied to approximately half the patients for at least 72 h. The three groups did not differ, either for primary outcome (i.e. ventilation-free days through day 28), secondary outcomes or adverse events. Unfortunately, between-group differences in SpO₂ were lower than intended, ranging between 1 and 3% among groups rather than a 4% separation. This was particularly relevant to the lowest SpO₂ group where the median SpO₂ value was 94%. Surprisingly, PaO₂ data were available for only 20% of included patients on day 1 and even less over the following days.

As with any other drug, benefit and harm from oxygen therapy depend on total dose, i.e. the percentage of O₂ within the inspired gas mixture and exposure duration, and patient’s factors that may increase susceptibilities to oxygen toxicity, such as severe brain injuries [12]. Beyond the substantial overlap between groups in SpO₂ and/or PaO₂ achieved during the study, it is relevant to note that the time of exposure to different O₂ levels is short ranging from few to 48–72 h for most of the patients included in the trials, resulting in a minimal difference in total O₂ exposure among groups.

Take-home messages

In virtually all trials to date in critically ill patients, liberal and conservative oxygenation targets have not been extreme and, accordingly, little effect has been seen on clinical outcomes. The impact of greater degrees of hyperoxaemia or hypoxaemia remains uncertain, as does any differential effect in pre-specified patient subsets, such as those surviving cardiac arrest.

Many more studies on oxygenation targets are in progress, of which Mega-ROX is the largest, targeting a sample size of 40,000 patients [13]. The hypothesis being tested here is that conservative oxygen therapy (91–95% SpO₂) in patients requiring unplanned mechanical ventilation reduces in-hospital all-cause mortality by at least 1.5% when compared with liberal oxygen therapy (lower SpO₂ limit of at least 91%, no specified upper limit, and a minimum use of 0.3 FiO₂ while the patient remains intubated). Within the overall trial there will be three nested RCTs in pre-specified patient subgroups: suspected hypoxic ischaemic encephalopathy (HIE), sepsis, and acute brain injuries other than HIE. However, we express the same concerns here as with previous studies in terms of achieving adequate separation between groups to demonstrate any clear outcome difference. Furthermore, whether a small clinical impact, even if statistically significant, is sufficient to change routine clinical practice is questionable, especially considering that practice outside a trial protocol is likely to be even less rigorous.

A study we would like to see performed is one that incorporates a mild hyperoxaemia target, e.g. 120–130 mmHg, corresponding to the nadir of the U-shaped-relationship with mortality. Alternatively, perhaps preferably, and as is being recognised after multiple negative RCTs in sepsis, a more directed and biological approach to study design may yield greater advances. Patients could be stratified into liberal and conservative targets by a biomarker identifying, for example, brain injury, endothelial activation, or excessive reactive oxygen species production where either a high or low oxygen target may be postulated as beneficial or detrimental. Such an approach does, however, require prior studies to identify suitable biomarkers with repeated blood sampling to delineate the baseline biological signature of enrolled patients and to assess the impact of oxygenation targets, given the heterogeneity of other management practices. So-called pragmatic trials, where no attempt is made to appreciate the underlying biology, have been uniformly disappointing to date. This repeating pattern is likely to continue unless a different trial strategy is adopted. In the waiting period for the new trials, we suggest considering oxygen as a powerful drug that should be carefully titrated for maintaining the patient in the nadir part of the U-Shaped relationship, that is, for most of the critically ill patients included in the normoxia-mild hyperoxaemia range.