Humans are aerobic animals; we need oxygen to support metabolism and generate energy. Simply put, breathing oxygen is necessary for life. Consequently, it would make sense that diseases that inhibit the transfer of oxygen from the lung to the blood should be treated with supplementary oxygen to overcome problems associated with hypoxemia, and one should strive to achieve 鈥渘ormal鈥� levels of arterial oxygen (Pao₂). But is that reasoning supported by the understanding of physiology and the weight of clinical evidence?

Stating that human tissues are dependent on oxygen refers to the need for oxygen delivery to vital organs; oxygen delivery encompasses the oxygen content of the blood as well as cardiac output. Because total oxygen content consists of oxygen bound to hemoglobin as well as oxygen dissolved in plasma, with the former far exceeding the latter, one might reason that transfusing patients with anemia to a normal level of hemoglobin would result in better outcomes. The evidence suggests otherwise, however, and the threshold for transfusion of red blood cells is well below 鈥渘ormal鈥� values of hemoglobin.^1-3

From an evolutionary perspective, it might be argued that humans are built to tolerate mild degrees of hypoxemia. The oxygen-hemoglobin saturation curve has a characteristic sigmoid shape. A decrease in Pao₂ from 90 mm Hg to 60 mm Hg still leaves one with near-normal oxygen saturation and oxygen content of the blood; long before supplemental oxygen was available, humans had pneumonia, asthma, or heart failure and tolerated mild to moderate decrements in Pao₂. Nevertheless, the sickest patients, those with acute hypoxemic respiratory failure, must benefit from restoring oxygen saturation to normal values, or at least that is what clinical intuition might tell us.

The study by Nielsen et al⁴ reported in this issue of JAMA builds on prior work of the HOT-COVID trial⁵ to assess the impact of targeting Pao₂ to 90 mm Hg vs 60 mm Hg in patients with COVID-19 pneumonia and hypoxemic respiratory failure. This study was a large multicenter trial involving medical intensive care units (ICUs) in 5 European countries. Acute hypoxemic respiratory failure was defined as necessitating supplemental oxygen of at least 10 L/min in an open system or mechanical ventilatory support (either invasive or noninvasive). The protocol required randomization of participants to supplemental oxygen to achieve high Pao₂ (90 mm Hg) or low Pao₂ (60 mm Hg) targets. Arterial lines were placed in all patients; this is one of the strengths of the study given the technical issues that often confound interpretation of pulse oximetry (eg, inadequate waveforms; erroneous measurements based on patient skin color). Given prior reports that mortality was no different between higher and lower oxygen targets,⁵ the primary outcome measure in this analysis was the number of days patients were alive without life support (including ventilatory and circulatory support, as well as kidney replacement therapy) over 90 days; secondary outcomes included complications that might be associated with hypoxemia (eg, cerebral and intestinal ischemia, myocardial infarction). With respect to the primary outcome of days alive without life support, there was a significant difference between patients assigned to the low and high oxygen targets. However, the 2 groups did not differ with respect to secondary outcomes.

How might one explain these findings? Perhaps other aspects of care differed between the groups? While the study did not prescribe other aspects of care beyond oxygen targets, physicians were instructed to make decisions about discontinuation of mechanical ventilation based on overall clinical status, and the findings were fairly consistent across all clinical sites (eFigure 11 in the article鈥檚 Supplement 2), arguing against variation in practice by group assignment as a contributing factor to the observed outcomes; this was very much a 鈥渞eal-world鈥� study. What if more patients in the high target group were intubated and started on mechanical ventilation because the physicians caring for them could not achieve the target with noninvasive ventilation? This could have contributed to the finding (not statistically significant) that more patients in the high target group were receiving mechanical ventilation at day 90. The observation that initiation of mechanical ventilation to achieve a high target Pao₂ may have occurred is less a failing of the study design than a consequence of using a high Pao₂ target; more patients may require mechanical ventilation to achieve the target and potentially experience the negative consequences associated with that intervention. The use of prone positioning, which has been touted as beneficial for averting lung injury, was slightly more common in the group with the higher oxygen target (41% of those with the higher oxygen target vs 31% of participants with the lower target); thus, any benefits attributable to prone positioning would have accrued to the patients in the high target group. Peak airway pressure, tidal volume, and positive end-expiratory pressure levels were not different between the groups receiving mechanical ventilation. Use of red blood cell transfusions was marginally higher in the group assigned to higher oxygen targets.

These results add to a number of studies in ICU as well as non-ICU settings.^6-9 A multicenter randomized trial of high vs low normal oxygen levels in patients with systemic inflammatory response syndrome showed no differences in Sequential Organ Failure Assessment scores, duration of mechanical ventilation, or in-hospital mortality.¹⁰ The ICU-ROX investigators studied 1000 patients with acute hypoxemic respiratory failure randomized to 鈥渃onservative鈥� (90% saturation) and 鈥渦sual鈥� high oxygen targets and also found no difference in ventilator-free days or mortality.¹¹ Use of supplemental oxygen for patients with suspected myocardial infarction has also been shown to be of no benefit.¹² The oxygenation targets in these studies were similar to those chosen by Nielsen et al,⁴ although their high oxygen target likely correlated with an oxygen saturation of 97% to 99%, which is higher than some other studies and provides a good separation of the 2 target groups.^5,11,13

Could higher oxygen targets help in some ways (eg, greater oxygen delivery) but hurt or harm patients in other ways, thereby negating any advantage?⁹ High levels of supplemental oxygen resulting in a fraction of inspired oxygen greater than 0.6 have been shown to contribute to lung injury.¹⁴ In addition, high concentrations of oxygen may lead to the formation of reactive oxygen species, which may cause harm throughout the body, including inflammation, bronchial epithelial injury, and disruption of mitochondrial respiration.¹⁵ Use of high concentrations of inspired oxygen may also contribute to washout of nitrogen from alveoli, which predisposes to atelectasis and worsens ventilation-perfusion mismatch. Patients with acute hypoxemic respiratory failure and acute respiratory distress syndrome who survive may develop pulmonary fibrosis. It is possible that some of this damage and associated abnormalities in pulmonary function testing and decrement in quality of life in this group may be due to lung toxicity associated with use of high oxygen targets during the acute illness.

Where does this leave clinicians in the understanding of oxygen targets in critically ill patients? While the study by Nielsen et al⁴ was limited to patients with COVID-19 pneumonia, it is one more piece in a growing collection of evidence suggesting that there is no clinical benefit and possibly harm associated with use of supplemental oxygen to achieve oxygen saturation of the blood beyond 90% to 93%.¹⁵ Furthermore, experimental data and some clinical reports suggest that there may be harm in higher targets. Mega-Rox,¹⁶ a large multinational study using nested parallel randomized clinical trials of response-adaptive randomization of patients who receive unplanned mechanical ventilation, aims to recruit 40鈥�000 patients and is now underway. The study will include patients with sepsis, individuals resuscitated after cardiac arrest, and patients with nonhypoxic ischemic encephalopathy as well as other conditions; investigators will assign patients to 鈥渓iberal鈥� vs 鈥渃onservative鈥� oxygen therapy. The primary outcome will be 90-day all-cause mortality. At this point, if there is a benefit to specific oxygen targets, it is likely small. Given the number of critically ill patients globally who receive mechanical ventilation, however, a small benefit, if demonstrated clinically, may still be important.

Corresponding Author: Richard M. Schwartzstein, MD, Division of Pulmonary, Critical Care, and Sleep Medicine, Harvard Medical School, Beth Israel Deaconess Medical Center, 330 Brookline Ave, Boston, MA 02215 (rschwart@bidmc.harvard.edu).

Published Online: March 19, 2024. doi:10.1001/jama.2024.2935

Conflict of Interest Disclosures: Dr Schwartzstein reported receipt of royalties from UpToDate and Lippincott.