Hypoxemia refers to low oxygen tension arterial blood gases or partial pressure of oxygen (PaO2) in the blood, occurring when oxygen supplies fail to meet oxygen demands.1 Hypoxemia is common and generally viewed as deleterious, especially in critically ill patients. Supplementary oxygen is the main strategy for the prevention and treatment of hypoxemia, either by invasive ventilation or non-invasive ventilation, and is wildly used in a hospital setting.

In clinical practice, providing supplemental oxygen for almost all acutely or critically ill patients, regardless of blood oxygen levels, is a longstanding cultural norm. However, this practice is not based on clinical evidence.2,3 A significant proportion of patients are exposed to an excessive oxygen administration. Undoubtedly, hypoxia can lead to cell injury and even death, and adequate oxygen supplementation is necessary,4 while hyperoxia may also cause cell, tissue or organ injury due to enhanced oxidative stress and inflammation.5,6 In recent years, more studies have investigated the relevant between hyperoxia and clinical outcomes in acutely or critically ill patients, while the results are contradictory. Some studies indicated that hyperoxia can be associated with poor clinical outcomes in different patients, such as patients with mechanically ventilation,7 traumatic brain injury (TBI),8 after resuscitation from cardiac arrest,9 and myocardial infarction,10 whilst other studies have not.11–14

In the intensive care unit (ICU), oxygen therapy is administrated for most patients. A number of studies have focused on the fraction of inspired oxygen or targets of arterial oxygenation in these patients, however, the management of oxygenation targets remains challenging in critically ill patients. Four previous systematic reviews all reached the conclusions that a liberal oxygen therapy strategy in adult patients admitted to the ICU could increase mortality and the number of adverse events compared with a conservative oxygen therapy strategy.15–18 However, two multicenter randomized control trials (RCTs) recently published in the New England Journal of Medicine involving ICU patients with hypoxemic respiratory failure, or mechanical ventilation found no significant differences in clinical outcomes between the conservative oxygen therapy groups and liberal oxygen therapy groups.19,20 Additionally, another similar RCT involving ICU patients with acute respiratory distress syndrome (ARDS) found conservative oxygen therapy may increase 90-day mortality and mesenteric ischemic events.21

As new high-quality RCTs have been published and the results are inconsistent with previous studies, we performed this systematic review of RCTs with meta-analysis and trial sequential analysis (TSA) to evaluate the benefits and harms of conservative versus liberal oxygen therapy in critically ill patients in the ICU.

This study was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA statement) guidelines.22 This meta-analysis was registered on PROSPERO (Registration number: CRD42021234555).

According to our search strategy, 5597 potentially studies were identified. After removing duplicates, 4291 studies were screened by titles and abstracts and 42 studies were further screened by reading full-text. Finally, ten studies involving 5429 adult patients admitted to the ICU with critical illness,19,29–31 septic shock,32 ARDS,21 out-of-hospital cardiac arrest (OHCA),33 traumatic brain injury (TBI),34 severe acute stroke,35 or acute hypoxemic respiratory failure20 were included in this meta-analysis (Fig. 1). The studies were published from 2015 to 2021. The number of participants ranged from 65 to 2888. Most studies were assessed as low risk of bias (see Additional file 2). Six studies only included patients who received invasive mechanical ventilation at randomization.19,21,30,32–34 All other studies randomized patients to liberal versus conservative oxygen therapy using oxygenation target, except for two studies using FiO2.34,35 Details of the included individual studies characteristics were shown in Table 1 and Additional file 3.

Figure 1.

Flow diagram for the identification of eligible studies.

(0,28MB).

Mortality

Data on mortality were available for all ten studies. The mortality in the conservative oxygen therapy group and the liberal oxygen therapy group were 38.4% (1034 of 2692 patients) and 38.7% (1055 of 2723 patients) at the longest follow-up, respectively. No significant difference was detected between two groups (RR 0.97; 95% CI 0.88–1.08; p=0.63; I2=28%; Fig. 2a). Moderate heterogeneity was detected. There was no obvious asymmetry in funnel plots by visually inspecting, while Egger's test indicated that publication bias may exist (p=0.014, see Additional file 4a). TSA result showed that the required information size was 4148. The cumulative Z curve reached the required information size and crossed the futility boundary, suggesting that a RRR of 20% or greater could be rejected (Fig. 3a).

Figure 2.

Forest plots of comparison: conservative oxygen therapy versus liberal oxygen therapy. (a) Mortality at the longest follow-up; (b) mortality at 28 days; (c) mortality at 90 days; (d) ICU mortality.

(0,83MB).

Figure 3.

Trial sequential analysis. (a–c) The cumulative Z curve (complete blue line) was constructed using a random effect model. Etched red line shows conventional test boundary. Complete red line represents the trial sequential monitoring boundary. (a) TSA for mortality at the longest follow-up. A diversity-adjusted information size of 4148 patients were calculated on the basis of using alfa=0.05 (two sided), beta=0.10 (power 90%), an anticipated relative risk reduction (RRR) of 20.0%, and a control event rate of 38.7%. The cumulative Z curve crossed the futility boundary and reached the required information size. (b) TSA mortality at 28 days. A diversity-adjusted information size of 9367 patients was calculated on the basis of using alfa=0.05 (two sided), beta=0.10 (power 90%), an anticipated relative risk reduction (RRR) of 20.0%, and a control event rate of 34.4%. The cumulative Z curve did not cross any boundaries, and did not reach the required information size. c. TSA for mortality at 90 days. A diversity-adjusted information size of 3325 patients was calculated on the basis of using alfa=0.05 (two sided), beta=0.10 (power 90%), an anticipated relative risk reduction (RRR) of 20.0%, and a control event rate of 40.0%. The cumulative Z curve crossed the futility boundary and reached the required information size.

(0,54MB).

The pooled results showed that a conservative oxygen therapy strategy could not decrease mortality compared with a liberal oxygen therapy strategy at 28 days (RR 0.90; 95% CI 0.75–1.09; p=0.28; I2=54%; 4245 participants, 6 studies, Fig. 2b), 90 days (RR 1.02; 95% CI 0.92–1.13; p=0.71; I2=26%; 4705 participants, 6 studies, Fig. 2c), and in ICU (RR 0.85; 95% CI 0.53–1.37; p=0.51; I2=68%; 906 participants, 4 studies, Fig. 2d). For 28-day all-cause mortality, TSA showed that the cumulative Z-curve did not cross any boundaries for benefit and harm, nor the futility boundary (Fig. 3b). For 90-day all-cause mortality, the cumulative Z curve reached the required information size and crossed the futility boundary (Fig. 3c).

In this meta-analysis for adult ICU patients, we found no beneficial or harmful effects of conservative oxygen therapy compared with liberal oxygen therapy. Both primary outcomes and secondary outcomes were comparable between two groups. TSA results indicated a RRR of 20% or greater could be rejected with respect to mortality at 90 days and at the longest follow-up, but in terms of mortality at 28 days, the required information size to detect or reject a RRR of 20% was not achieved.

Our results were at variance with the results of previous meta-analyses on this topic. Damiani et al. and Helmerhorst et al. have conducted meta-analyses including observational studies and drawn a similar conclusion that hyperoxia may increase mortality in critically ill patients.15,16 The pooled results of study conducted by Barbateskovic et al. including 10 RCTs indicated that higher oxygen supplementation was associated with increased mortality and the incidence of serious adverse events. However, the authors were very uncertain about the results due to very low-certainty evidence.17 Similarly, Hirase et al. have found conservative oxygen therapy administrated in the ICU could reduce mortality and new-onset non-respiratory organ failure compared to liberal oxygen therapy.18 We updated the meta-analysis on this topic, including results from three recently published high-quality RCTs, none of which found that ICU patients would benefit from conservative oxygen therapy as compared to liberal oxygen therapy, which may support the fact that our conclusions differ from those previous meta-analyses.19–21

More studies have gained increasing interest in investigating the effects of exposure to hyperoxia, and have found that excessive oxygenation had deleterious properties in various pathophysiological processes.36 In a recent meta-analysis including 25 RCTs involving 16,037 acutely ill adults, Chu and colleagues found that liberal oxygen therapy was associated with higher mortality than conservative oxygen therapy with no improvement on other important clinical outcomes.37 Our results showed no significant difference between the two groups. The different results between the two studies may be due to the following reasons-first, we only included studies involving critically ill patients admitted to the ICU. To our knowledge, both hypoxia and hyperoxia were independent risk factors of mortality in ICU patients.38,39 Patients assigned to the liberal oxygen therapy may be at higher risk of exposure to hyperoxia, while patients assigned to the conservative oxygen therapy may be at higher risk of exposure to hypoxia. Secondly, in our study, the mortality at the longest follow-up were 38.4% (1034 of 2692 patients) in the conservative oxygen therapy group and 38.7% (1055 of 2723 patients) in the liberal oxygen therapy group, while in the study conducted by Chu and colleagues, the mortality at the longest follow-up in the two groups were 9.5% (749 of 7857 patients) and 10.5% (828 of 7897 patients), respectively. It is reasonable to assume that the severity of disease in our study is higher. In addition, a significant proportion of patients in our study received invasive mechanical ventilation or had acute hypoxemic respiratory failure.19–21,30,32–34 For these patients, a liberal oxygen therapy strategy to ensure adequate oxygen supplementation may be necessary.

The strengths of our study are as follows: First, we used a comprehensive, up-to-date search strategy, which have identified three recently published well designed RCTs. Moreover, all studies were published in recent years and most of them were assessed as low risk of bias. Second, the methodology used in this study was rigorous. TSA was performed to control the risk of random errors. A best-worst scenario analysis and a worst–best scenario analysis were performed to assess the potential impact of the missing participants for the primary outcomes. Limitations existed in this study must also be considered. As with previous meta-analyses on this topic, the primary limitation was that the definitions of liberal and conservative oxygen therapy varied widely from study to study. For example, some studies used a fixed FiO2, while others used a particular oxygenation target by measuring PaO2, SaO2, or SpO2. Considering the differences in patients’ conditions and lung function, higher FiO2 oxygen supplementation does not necessarily lead to higher tissue oxygen saturation, so it may be more reasonable to define liberal and conservative oxygen therapy by a particular oxygenation target. We conducted sensitive analyses by excluding two studies using a fixed FiO2 and we found the results were consistent with main analyses. Second, we only included studies involving ICU patients, while some studies included mixed populations with specific conditions, such as TBI and severe acute stroke. And duration of intervention also existed differences among included studies. Nonetheless, the results of subgroup analyses and sensitive analyses were consistent.

Oxygen administration is part of a routine treatment in the ICU. Although no beneficial or harmful effects of conservative versus liberal oxygen therapy were detected in this study, it is possible that different oxygen therapy strategies have effect on clinical outcomes. More studies are required to find an appropriate oxygen therapy strategy for ICU patients. Considering the complexity of the ICU patient's condition, oxygen therapy strategy in future studies should be designed according to patients’ conditions.

In conclusion, no beneficial or harmful effects of conservative oxygen therapy were found compared with liberal oxygen therapy in adult ICU patients. Conservative oxygen therapy did not reduce all-cause mortality at 28 days, 90 days, and the longest follow-up. Other important clinical outcomes were also comparable between two groups.

XML and DSL conceived of the study, participated in the design, collected the data, performed statistical analyses and drafted the manuscript. CL participated in the design and performed statistical analyses. ZM helped to revise the manuscript critically for important intellectual content. YL collected the data and performed statistical analyses. HYY performed statistical analyses. FHZ conceived of the study, participated in the design and revised the manuscript critically for important intellectual content. All authors read and approved the final manuscript.

Innovation Research of Chinese PLA general hospital (CX19010). Special Research of Military Medical Innovation (18CXZ026).

The authors declare that they have no conflict of interest.