1 INTRODUCTION

Oxygen therapy has been employed in clinical settings for over a century.¹ It has become a standard intervention to administer supplemental oxygen to a wide array of patients experiencing acute and critical illnesses, irrespective of their baseline oxygen saturation levels. The administration of oxygen is often considered a routine part of patient care, with the primary aim of ensuring adequate oxygen delivery to tissues and vital organs. However, despite its widespread use, oxygen therapy may not always confer benefits and can, in fact, be harmful in certain patient populations. Without vigilant monitoring and appropriate titration, the delivery of oxygen can lead to either hypoxia or hyperoxia, both of which pose significant risks to patient health and recovery.²

Experimental studies show that the sword of oxygen is double-edged and either edge cuts. Hypoxia can lead to cell injury and organ failure, as the lack of sufficient oxygen impairs cellular respiration and energy production, ultimately resulting in cellular dysfunction and, if not promptly corrected, organ system collapse.³ On the other hand, hyperoxia may also cause cell, tissue, or organ injury due to enhanced oxidative stress and inflammation.^{4, 5} The excess of oxygen can lead to the formation of reactive oxygen species, which are highly reactive molecules capable of damaging cellular components, including lipids, proteins, and DNA.⁶ This oxidative damage can trigger apoptosis, or programmed cell death, and can also activate inflammatory pathways, leading to a pro-inflammatory state that can exacerbate tissue injury and impede the healing process. The promotion of oxygen free radical generation overwhelms the body's antioxidant defenses, leading to the activation of both anti-inflammatory and pro-inflammatory factors.^{7, 8} This dual activation can create a detrimental environment for cells, tipping the balance toward cell death through apoptosis or necrosis; moreover, the change in the internal immune state induced by hyperoxia can lead to an increased risk of infection.^6-8 The implications of these findings are significant for the clinical management of patients requiring oxygen therapy.

To optimize oxygen therapy, clinical studies have been conducted. Some studies have demonstrated that excessive oxygen therapy may worsen outcomes in patients with conditions such as myocardial infarction⁹ and resuscitation from cardiac arrest.¹⁰ For general patients in the intensive care unit (ICU), retrospective studies showed that hypoxemia was associated with an increased risk of ventilator -associated pneumonia¹¹ and mortality.^12-14 Through retrospective analyses of data from two electronic medical record databases, van den Boom et al. found that the optimal saturation of pulse oximetry (SpO₂) for critically ill patients was between 94% and 98%.¹⁵ In the 2016 Effect of Conservative versus Conventional Oxygen Therapy on Mortality Among Patients in an Intensive Care Unit (Oxygen-ICU) study, Girardis et al. showed that ICU mortality was lower in the conservative oxygen therapy group (target arterial oxyhemoglobin saturation between 94% and 98% or partial pressure of arterial oxygen (PaO₂) between 70 and 100 mmHg) than in the conventional oxygen group (target arterial oxyhemoglobin saturation ≥ 97%).¹⁶ However, the sample size of this study was relatively small. In 2017, the British Thoracic Society Emergency Oxygen Guideline recommended that the SpO₂ of most acute patients should be set to 94%–98%.¹⁷ However, it has been reported that the target value of the British Thoracic Society guidelines will still lead to a large number of hyperxemia.¹⁸ The 2018 clinical practice guidelines for acute patients suggest that the SpO₂ of patients receiving oxygen therapy should not exceed 96%, which applies to almost all hospitalized patients.¹⁹ Large-scale randomized studies are needed to guide oxygen therapy in general ICU patients.

In recent years, several such studies have been reported: the Intensive Care Unit Randomized Trial Comparing Two Approaches to Oxygen Therapy (ICU-ROX) study,²⁰ the Pragmatic Investigation of Optimal Oxygen Targets Trial (PILOT) study,²¹ the Liberal Oxygenation versus Conservative Oxygenation in ARDS (LOCO₂) study,²² the Handling Oxygenation Targets in the ICU (HOT-ICU) study,²³ the Optimal Oxygenation in the Intensive Care Unit (O₂-ICU) study,²⁴ and the Conservative versus Liberal Oxygenation Targets in Intensive Care Unit Patients (ICONIC) study.²⁵ Almost concurrently with these studies, we were conducting the Peripheral Oxygen Saturation Directed Oxygen Therapy (POSDOT) study to determine whether lower SpO₂ targets could provide better outcomes for critically ill patients than higher SpO₂ targets.

2 RESULTS

2.1 Characteristics of the patients

From May 2019 to May 2022, we enrolled 1740 patients in 11 general ICUs from 10 tertiary hospitals in China. However, the enrollment was heavily influenced by the COVID-19 pandemic, we decided to stop the study, and our decision was approved by the ethics committee of our institution. Loss to follow-up occurred in 34 patients, which left an intention-to-treat (ITT) population of 1706, with 857 assigned to the low SpO₂ group and 849 to the high SpO₂ group (Figure 1). Their mean age was 59.7 ± 15.6 years, 1087 (63.7%) were male, and 1138 (66.7%) were medical patients (Table 1). At admission, 479 (28.1%) patients and 1067 (62.5%) were experiencing shock and respiratory failure, respectively. Their Acute Physiology and Chronic Health Evaluation II (APACHE II) score was 15.9 ± 7.3, and Charlson Comorbidity Index was 1 [0, 3]. The interval between ICU admission and randomization was 17.2 ± 13.6 h. A total of 1445 (84.7%) patients had an ICU stay of no less than 72 hours.

TABLE 1. Baseline characteristics of the patients included.

Demographic	All patients (n = 1706)	Low SpO2 group (n = 857)	High SpO2 group (n = 849)
Age (years)	59.7 ± 15.6	60.2 ± 15.1	59.2 ± 16.0
Sex: male (%)	1087 (63.7%)	548 (63.9%)	539 (63.9%)
Admission
Medical	1138 (66.7%)	566 (66.0%)	572 (67.4%)
Planned surgery	165 (9.7%)	82 (9.6%)	83 (9.8%)
Emergency surgery	403 (23.6%)	209 (24.4%)	194 (22.9%)
Organ failure
Shock	479 (28.1%)	237 (27.7%)	242 (28.5%)
Septic	299 (17.5%)	145 (16.9%)	154 (18.1%)
Hypovolemia	137 (8.0%)	72 (8.4%)	65 (7.7%)
Cardiac	35 (2.1%)	16 (1.9%)	19 (2.2%)
Obstructive	8 (0.5%)	4 (0.5%)	4 (0.5%)
APACHE II score	15.9 ± 7.3	15.9 ± 7.4	16.0 ± 7.2
Charlson Comorbidity Index	1 [0, 3]	1 [0, 2]	1 [0, 3]

• Note: Data were presented as mean ± standard deviation, count (percentage) or median [interquartile range].
• Abbreviations: APACHE II, Acute Physiology and Chronic Health Evaluation II; ICU, intensive care unit; SpO₂, peripheral oxygen saturation.

2.3 Outcomes

Within 28 days after randomization, 172 (20.1%) patients in the low SpO₂ group and 193 (22.7%) in the high SpO₂ group died (risk ratio [RR] 0.88, 95% confidence interval [CI] [0.74–1.06], p = 0.180) (Table 3). The log-rank test of the probability of survival between the two groups since randomization was not different either (p = 0.181, Figure 2). As for the secondary outcomes, ventilator-free time (59 [24, 105] h in the low SpO₂ group vs. 63 [24, 114] h in the high SpO₂ group, p = 0.737) and renal replacement therapy (RRT)-free time (120 [74, 195.5] h in the low SpO₂ group vs. 120 [75, 200] h in the high SpO₂ group, p = 0.670) in 14 days were not different (Table 3). Subgroup analyses based on baseline characteristics showed a borderline non-significant decrease in 28-day mortality in female patients, and a borderline significant decrease in 28-day mortality in patients with a Charlson Comorbidity Index greater than 0. Univariate and multivariate Cox regression analyses showed age, shock at mission, and respiratory failure at admission were risk factors for death (Table S1). In either analysis, high SpO₂ was not a risk factor for death.

3 DISCUSSION

Our multicenter, randomized clinical trial showed that low SpO₂-directed oxygen therapy did not decrease 28-day mortality, 14-day ventilator-free time, or 14-day RRT-free time in critically ill patients.

SpO₂ is employed routinely in clinical practice and commonly in trials exploring the correlation between oxygen therapy and clinical outcomes in critically ill patients. Based on data from more than 30,000 patients from two electronic medical record databases, van den Boom and colleagues revealed that compared patients with 40% of the time SpO₂ between 94% and 98%, patients with 80% of the time SpO₂ within the same range had a nearly halved risk of hospital death in a retrospective study.¹⁵ Although, in our study, the time-weighted average SpO₂ was between 94% and 98% in the low SpO₂ group, we did not find a difference in mortality in 28 days, ventilator-free time in 14 days, or RRT-free time in 14 days compared with the high SpO₂ group. Our findings are consistent with findings from the ICU-ROX study²⁰ and the PILOT study.²¹ The ICU-ROX investigators randomized 1000 adult mechanically ventilated patients into the conservative-oxygen group (97% as upper limit of SpO₂ and FiO₂ as low as possible to 21%) and the usual oxygen group (no upper limit of SpO₂ and FiO₂ above 30%), and they found no difference in ventilator-free days or 180-day mortality.²⁰ In the PILOT study, Semler and colleagues assigned mechanically ventilated patients into three groups based on SpO₂ target, the low SpO₂ group (88%–92%), the intermediate SpO₂ group (92%–96%), and the high SpO₂ group (96%–100%); they found no difference in ventilator-free days or 28-day mortality among these groups.²¹

Other two commonly used targets in oxygenation studies in ICUs are PaO₂ and arterial oxyhemoglobin saturation (SaO₂). Compared with patients with SaO₂ ≥ 97%, Girardis and colleagues found that patients with SaO₂ between 94% and 98% or PaO₂ of 70–100 mmHg had a significantly reduced chance of ICU death from 20.2% to 11.6%, of shock from 10.6% to 3.7%, and of bacteremia from 10.1% to 5.1%.¹⁶ However, Oxygen-ICU was an early-terminated single-center randomized controlled study with only 480 patients included. These findings were not replicated by the LOCO₂ study²² and the HOT-ICU study.²³ In the LOCO₂ study, Barrot and colleagues included 205 patients with acute respiratory distress syndrome and found no difference in 28-day mortality between patients with PaO₂ from 55 to 70 mmHg and patients with PaO₂ from 90 to 105 mmHg.²² In the HOT-ICU study on adult patients with acute respiratory failure, 1441 were randomized into the lower oxygenation group (target PaO₂ 60 mmHg) and 1447 into the higher oxygenation group (target PaO₂ 90 mmHg), and no difference in 90-day mortality was found.²³

The difference between mean/median FiO₂ in the high target group and mean/median FiO₂ in the low target group varies. It increases from 3% in the Oxygen-ICU study,¹⁶ approximately 5% in the ICU-ROX study,²⁰ approximately 6% in our study, approximately 10% in the O₂-ICU study²⁴ and the LOCO₂ study,²² approximately 12.6% in the ICONIC study,²⁵ and approximate 15% in the PILOT study²¹ and the HOT-ICU study.²³ The different differences in FiO₂ between the high target group and the low target group in these studies make it difficult to conduct a meta-analysis.

In the low SpO₂ group/conservative group, the actual SpO₂ levels were commonly higher than the predefined upper limit of the SpO₂ target. In the low SpO₂ group/conservative group, the time-weighted average of SpO₂ was 93.4% in the study of Panwar and colleagues,²⁶ 95.9% in our study, and approximately 94% in the PILOT study²¹; the predefined upper limit of the SpO₂ target was 92%, 95%, and 92%, respectively.

Even in pO₂-targeted study, the actual SpO₂ levels were commonly higher than the anticipated SpO₂ levels in the lower target group. The actual SpO₂ was approximately 93% in the LOCO₂ study²² and in the HOT-ICU study,²³ and the anticipated SpO₂ levels were 88%–92% and approximately 90%, respectively. Whether compliance with the study protocol will make a difference is unknown. The bottom line is an extremely large number of participants should be included to make a firm conclusion on whether there will be a significant difference in mortality between conservative oxygen therapy and liberal oxygen therapy in critically ill patients, and we are hoping that the Mega-ROX trial (ACTRN12620000391976) will shed some light on the question.

The strengths of our study lie in the pragmatic protocol allowing us to maintain routine practice except for oxygenation targets and differences in SpO₂ and FiO₂ between the two groups. However, our study had some limitations. First, our study stopped prematurely because of the COVID-19 pandemic. However, it was unlikely the planned sample size would make a difference in the primary outcome. Second, a large proportion of patients screened were not randomized. For one thing, one liaison at each study center was asked to screen as many patients as possible, and patients admitted into the ICU during weekends or holidays were easily categorized as being unable to be randomized. For another, the plan to include patients with an ICU stay of at least 72 hours aimed at a longer duration of oxygen treatment in the ICU, but it also led to the unnecessary exclusion of potentially eligible patients with an uncertain length of ICU stay at the time of screening. The screening process is difficult to regulate and monitor.²⁷ Third, the SpO₂ target was not achieved in the low SpO₂ group, which may attenuate the treatment effect. Fourth, data on arterial blood analysis were not mandatorily collected. The relationship between arterial SpO₂ and PaO₂ is not linear, and compared with the PaO₂ target, SpO₂ targe is a relatively less precise titration of oxygenation. But SpO₂ has the advantage of real-time continuous monitoring. Since no compensation was given to the patients, our ethics committee deemed it would be immoral to conduct arterial blood gas analyses only for the study. In the present trial, PaO₂ and FiO₂ were obtained from arterial blood gas analyses at the discretion of study physicians. Because pO₂ was measured at different times and could not be measured continuously, SpO₂ was used to titrate oxygen.^{23, 25} In the ICONIC study, only 68.3% and 52.9% of patients in the low oxygenation group and the high oxygenation group, respectively, had 50% or more of pO₂ measurements within the predetermined ranges.²⁵ This information was not available in the HOT-ICU study.²⁶ The SpO₂ values concomitant to pO₂ measurements of only three to four times per day were sufficient to differentiate between the low oxygenation group and the high oxygenation group.²⁵ Fifth, all participating study centers were tertiary hospitals from China, which would restrict the generalizability of this study. Sixth, data on ischemic events, including myocardial infarction, ischemic stroke, and intestinal ischemia, were unavailable. Seventh, although the free time in hours of respiratory and of renal support with machines wwas planned and explored as possible outcome measures for future trials, we failed to find differences in these two aspects. For convenience, a fixed time frame of 14 days for the secondary outcomes was used, which was the same as the duration of the oxygen fraction manipulation in the present study. Secondary outcomes within 14 days of enrollment, such as ventilator-free days, were also used by Gelissen and colleagues in the O₂-ICU study.²⁴

4 CONCLUSION

In critically ill patients, the low SpO₂-directed oxygen therapy did not lead to a lower 28-day mortality, a longer duration free from ventilator, or a longer duration free from RRT in 14 days.

5 MATERIALS AND METHODS

AUTHOR CONTRIBUTIONS

Drafting of the manuscript: Xiaobo Yang, Xuehui Gao, Xiang Zheng, Xu Zhao, Yanli Liu, Lu Zhang, Junli Sun, Peng Wang, Zhengqin Xu, Ronghua Hu, and Hongbin Li. Critical revision of the manuscript: Hong Qi, Yin Yuan, Wei Chen, Jie Liu, Guangqing Huang, Fengsheng Cao, and Keke Xin. Administrative and technical support: Li Yu, Min Yu, Xiaoyun Liu, Li Zhang, Siyuan Chang, Xiaojing Zou, Hong Liu, Zhaohui Fu, Huaqing Shu, Yuan Yu, Jiqian Xu, Shiying Yuan, and You Shang. Acquisition, analysis, and interpretation of data: All Authors. All authors have read and approved the final manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

ETHICS STATEMENT

This study was approved by the ethics committee of Tongji Medical College (No. 2017-S024) and ethics committees of other study sites. Informed consent was obtained from all participants. This trial was registered on Clinicaltrials.gov (NCT02999932).