Editorial August 31, 2021

Titrating Oxygen Therapy in Critically Ill Patients

Martin Urner, Carolyn S. Calfee, Eddy Fan

JAMA. Published online August 31, 2021. doi:10.1001/jama.2021.9843

The development of oxygen therapy is one of the major advances of modern medicine. Oxygen therapy laid the groundwork for intensive care medicine as a specialty and has saved millions of lives. During the COVID-19 pandemic, millions of patients survived due to provision of supplemental oxygen, with or without mechanical ventilation. In some regions of the world, where oxygen supply is limited, however, there were many preventable deaths from acute hypoxemic respiratory failure.

Oxygen therapy is not without potential risks. In the late 18th century, animal models demonstrated that severe lung injury resulted from hyperoxia, ie, breathing a high fraction of inspired oxygen (Fio₂) for prolonged periods. Research in the early 20th century demonstrated that sustained exposure to Fio₂ of0.7 or greater was toxic across numerous species, and that hyperoxic and ventilator-induced lung injuries may be synergistic. Hyperoxemia (increased blood oxygen content), which may result from hyperoxia, may also be harmful. A large observational study involving 36 307 critically ill patients from 50 intensive care units (ICUs) reported a u-shaped relationship between Pao₂ and in-hospital mortality.¹

A key lesson in critical care over the past 20 years is that correcting abnormal physiology to normal levels may often lead to harm, and a “less is more” approach to targeting physiological goals might result in a more favorable risk/benefit trade-off for many interventions. Oxygen therapy, likewise, has been the subject of many clinical trials, in which patients were randomized to a specific Pao₂ or peripheral or arterial oxyhemoglobin saturation (oxygen saturation as measured by pulse oximetry [Spo₂] or arterial oxygen saturation [Sao₂]) target.

The potentially harmful effects of hyperoxemia and hyperoxia for critically ill patients have been supported by the results of the Oxygen-ICU trial (434 critically ill adults) and the HyperS2S trial (442 adults with septic shock who received mechanical ventilation).^2,3 Both trials had limitations (ie, small sample size, potential bias in the modified intention-to-treat analysis), but both demonstrated higher mortality with higher oxygen targets or higher oxygen delivery (20.2% vs 11.6% in the Oxygen-ICU trial and 43% vs 35% in the HyperS2S trial). The results of the IOTA meta-analysis that investigated a heterogeneous group of 16 037 critically ill patients from 25 trials further supported potentially adverse effects of hyperoxemia.⁴

On the other hand, the LOCO₂ trial (205 mechanically ventilated adults with acute respiratory distress syndrome) indicated that aiming for lower oxygen targets could lead to harm: while there was no significant difference in 28-day mortality (34.3% vs 26.5%), 5 mesenteric ischemic events occurred in the conservative oxygen therapy group and none in the liberal oxygen group.⁵ More recently, the ICU-ROX trial (965 mechanically ventilated adults) reported no significant differences in the primary outcome of ventilator-free days during the first 28 days,⁶ and the HOT-ICU trial (2928 patients with acute hypoxic respiratory failure) reported no difference in terms of 90-day mortality or adverse events,⁷ in direct contrast to the results of the LOCO₂ study. Thus, uncertainty remains about the optimal approach to oxygen targets in critically ill patients.

It is also uncertain which physiological variable (ie, Pao₂, Spo₂, or Sao₂) should be the target for oxygen treatment. Pao₂ is sensitive at detecting hyperoxemia and tissue hypoxia but requires placement of an arterial catheter for frequent blood gas sampling. The continuous measurement of Spo₂ is less invasive and widely available but may be unreliable in severe hypoxemia (due to the shape of the hemoglobin-oxygen dissociation curve), tissue hypoperfusion, or with dyshemoglobinemia (ie, carboxyhemoglobinemia or methemoglobinemia). The use of Spo₂ for medical decision-making may also be influenced by racial bias and systematically places Black patients at a higher risk for hypoxemia.⁸

There are individual differences in the adaptive response to hypoxemia,⁹ and there is heterogeneity in the treatment effect of oxygen therapy among different categories of critically ill patients. For example, hyperoxemia is important for the treatment of carbon monoxide poisoning or necrotizing skin infections, while the exposure to higher Pao₂ levels in ischemia/reperfusion injury (eg, cardiac arrest, myocardial infarction, or stroke) is associated with worse outcomes.⁴

In this issue of JAMA, Gelissen and colleagues¹⁰ report on the results of a randomized clinical trial (RCT) comparing 2 different oxygenation targets in 400 critically ill patients with systemic inflammatory response syndrome treated in 4 ICUs in the Netherlands. Patients were treated with supplemental oxygen targeting either a Pao₂range from 60 to 90 mm Hg (8 to 12 kPa; low normal; n = 205 patients) or a Pao₂ range from 105 to 135 mm Hg (14 to 18 kPa; high normal; n = 195 patients). The selection of patients with systemic inflammatory response syndrome enriches for a population that might particularly benefit from the hypothesized vasoconstrictive and antimicrobial effects of supplemental oxygen.^11,12 The oxygenation targets were chosen within a normal range to avoid detrimental effects from lower targets found in prior work.⁵ The authors used a protocolized approach for the adjustment of supplemental oxygen that sought to prevent Fio₂ levels greater than 0.6 whenever possible, in order to separate the effects of hyperoxia from hyperoxemia.

The primary outcome of the trial was the SOFA_RANK, a measure of nonrespiratory organ failure quantified by the nonrespiratory components of the Sequential Organ Failure Assessment (SOFA) score, summed over the first 14 study days. Patients were ranked from those with the most rapid organ failure improvement (lowest scores) to those with worsening organ failure or death (highest scores). The median SOFA_RANK score was −35 points in the low-normal group vs −40 in the high-normal group after 14 days of follow-up (median difference, 10 [95% CI, 0 to 21]; P = .06), suggesting a low-normal Pao₂ target compared with a high-normal target might not significantly reduce organ dysfunction at 14 days.

The current trial does not provide the definitive answer on the optimal targets for oxygen therapy and will not change clinical practice. However, there are several important implications for future research. Importantly, the results confirm that definitive knowledge is lacking regarding how to optimally titrate oxygen in critically ill patients. In addition, the novel primary outcome used in this trial opens a discussion on advantages and disadvantages of different outcomes for trials in critical care. Unlike other clinical trials of oxygen targets, the investigators used SOFA_RANK score to reflect the severity and temporal evolution of organ failure, death, and discharge. They posit this approach overcomes the challenges of achieving the required sample sizes to detect a difference in mortality with adequate statistical power. Mortality is often selected as the primary outcome in trials of critically ill patients because it is considered a patient-centered and “hard” end point with minimal risk of biased assessment. However, there is considerable heterogeneity among critically ill patients, and the clinically meaningful effect sizes for mortality are often relatively small, making the conduct of traditional RCTs in critically ill patients challenging, costly, and time-consuming. This tension has prompted consideration about the use of surrogate outcomes. A good surrogate end point does not need to be a mediator between the exposure and outcome, but must be easily measurable, treatment responsive, and consistently and strongly associated with a patient-important outcome (eg, mortality or long-term functional outcomes).¹³

The SOFA_RANK score has several limitations. First, differences in this primary outcome are not consistently related to changes in the exposure, nor are they intuitively interpretable. Second, in a parallel-group RCT, the outcome of interest is the difference between groups and not the changes within groups in response to the intervention. The SOFA_RANK score is a change score relative to baseline and summed over the first 14 study days, which may be problematic for a number of reasons (eg, floor and ceiling effects of the SOFA score, assumptions about linearity).¹⁴ Third, it is unclear whether the SOFA_RANK score is sensitive enough to detect differences arising from targeting different Pao₂ levels in critically ill patients (treatment responsiveness of the surrogate). In patients with COVID-19–related hypoxemic respiratory failure, the discriminative performance of the SOFA score was poor and significantly inferior to a variable such as age.¹⁵ Fourth, the authors assumed that the intervention influenced all components of the SOFA_RANK score equally. Changes in the severity of organ failure relative to baseline (eg, an increase in platelet count) were incorporated on the same scale with vastly different events, such as death and being discharged, resulting in a complex composite outcome that is difficult to interpret.

Several ongoing and planned trials of oxygen therapy may provide additional insight. The UK-ROX trial (ISRCTN13384956) will compare the effect of conservative oxygen therapy vs usual care on 90-day mortality and plans to enroll 16 500 patients from 100 ICUs across the United Kingdom. The multifactorial, adaptive Mega-ROX trial (ACTRN12620000391976) is aimed to provide more definitive evidence regarding the role of conservative oxygen therapy, powered to detect a 1.5% absolute risk difference in mortality among 40 000 patients. The enormous sample sizes required for these studies underline the issues that Gelissen et al¹⁰ sought to overcome with a surrogate outcome. While SOFA_RANK score is not an ideal surrogate outcome, further research on the development of novel end points for clinical trials is needed to provide researchers with tools to reliably determine the efficacy of different treatment strategies. Meanwhile, clinicians should avoid targeting excessively low or high Pao₂ levels in critically ill patients until the results of further trials become available.