Perioperative hyperoxia versus normoxia during cardiopulmonary bypass in adult cardiac surgery: a systematic review and meta-analysis of randomized controlled trials Authors: Najat Y. AlSejari, Batoul H. Aljaber, Saud A. Alyahya, Abdullah M. Alharran, Hamza A. Abdul-Hafez, Amal A. Alsubaiei Published: 2026 ABSTRACT BackgroundWhile hyperoxia has conventionally been the standard during cardiopulmonary bypass (CPB) to prevent hypoxemia, recent concerns regarding oxygen-induced oxidative stress have sparked debate over whether a normoxic strategy provides a safer clinical alternative.MethodsA comprehensive search of PubMed, Web of Science, CENTRAL, and Scopus was conducted to identify randomized controlled trials (RCTs) through December 2025. Primary outcomes included short-term and long-term mortality, and acute kidney injury (AKI). Secondary outcomes included assessment of clinical complications and healthcare resource utilization. Risk ratios (RR) and mean differences (MD) were pooled using a random-effects model.ResultsTwelve RCTs involving 2,732 patients were included. The analysis found no significant difference between hyperoxia and normoxia in short-term mortality (RR: 1.23, 95% CI [0.60, 2.54]; p = 0.57), long-term mortality (RR: 1.10, 95% CI [0.84, 1.43]; p = 0.48), or the incidence of AKI (RR: 0.94, 95% CI [0.82, 1.07]; p = 0.32). Similarly, no significant differences were observed in secondary clinical outcomes, including myocardial infarction, stroke, arrhythmia, or surgical site infections. Hospital resource utilization was comparable, with no significant differences in mechanical ventilation duration (p = 0.06), ICU stay (p = 0.32), or length of hospital stay (p = 0.08).ConclusionThis meta-analysis detected no statistically significant differences between hyperoxic and normoxic strategies during CPB regarding mortality, clinical outcomes, or hospital resource utilization. Still, the current evidence remains uncertain, warranting further research.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13019-026-04344-x. Introduction Cardiopulmonary bypass (CPB) is essential for most cardiac surgical procedures, but it inherently provokes a systemic inflammatory response and ischemia-reperfusion injury [1]. To mitigate these risks, managing arterial oxygen tension (PaO2) is crucial; however, it remains an inconsistent aspect of care during cardiac surgery [2]. Historically, the standard of care during CPB has included the administration of hyperoxia, defined as supraphysiological oxygen levels (PaO2 > 200–300 mmHg) [3, 4]. This liberal approach was primarily justified by the need to prevent accidental hypoxemia, maintain adequate tissue oxygenation during periods of diminished flow, and protect against gaseous micro-emboli [5]. Additionally, some evidence suggests that hyperoxia may reduce the incidence of surgical site infections (SSIs), although this potential benefit remains debated [6]. Although oxygen is critical, the optimal arterial oxygen target during the high-risk time of extracorporeal circulation remains unclear [7]. New physiological findings challenge the notion that more oxygen is always beneficial, suggesting that excessive oxygen may exacerbate organ damage [8]. A high amount of oxygen can trigger the production of reactive oxygen species (ROS), leading to oxidative stress [9]. This mechanism is particularly relevant to CPB, where hemoglobin-mediated vasoconstriction and ROS-induced cytotoxicity can compromise microvascular perfusion [10]. Specifically, hyperoxia causes systemic vasoconstriction, which can potentially reduce cardiac output and regional blood flow to critical organs, such as the brain and myocardium [11]. Kidney vulnerability to hyperoxic damage is a primary concern. Acute kidney injury (AKI) is a common and severe consequence of cardiac surgery, with oxidative stress during reperfusion being a major pathophysiological factor [12, 13]. Accordingly, avoiding hyperoxia could theoretically attenuate reperfusion injury and preserve renal function [14]. Randomized controlled trials (RCTs) comparing hyperoxic and normoxic strategies have yielded conflicting results [3–5, 15–23]. Also, while previous meta-analyses have investigated this topic, they were often limited by combining on-pump and off-pump cardiac procedures in their analyses and by relying on smaller trials with significant methodological heterogeneity [24, 25]. However, the recent publication of large-scale, high-quality RCTs, such as the GLORIOUS and ROCS trials, has significantly expanded the evidence base [16, 22]. Given this novel, high-quality data and the fact that current international guidelines offer no strong recommendations regarding intraoperative oxygen targets [26, 27], an updated synthesis focused exclusively on the highly specific physiological environment of on-pump cardiac surgery is required. Therefore, this systematic review and meta-analysis aims to provide a more precise clinical evaluation by comparing the efficacy and safety of hyperoxic versus normoxic strategies strictly in adult patients undergoing cardiac surgery on CPB. Methodology Protocol registration This systematic review and meta-analysis was prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO) under the identification number [ID: CRD420261280031]. The conduct and reporting of this study were strictly guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [28] and the methodological recommendations established in the Cochrane Handbook for Systematic Reviews of Interventions [29]. Data sources & search strategy A systematic literature search was conducted on December 30, 2025, in major electronic databases, including PubMed, Web of Science, CENTRAL, and Scopus. To identify relevant studies, we employed a search strategy combining terms related to the intervention and population, as follows: “(“cardiopulmonary Bypass” OR “Thoracic Surgery” OR “cardiac surgery” OR “heart surgery” OR “CABG” OR “valve surgery” OR “CPB” OR “on-pump” OR “on pump” OR “aorta surgery” OR “artery bypass” OR “coronary bypass”) AND (“hyperoxia” OR “Oxygen inhalation therapy” OR “hyperox*” OR “high oxygen” OR “liberal oxygen” OR “oxygen treatment” OR “oxygen therapy” OR “oxidative”) AND (“randomized controlled trial” OR “randomized” OR “randomised” OR “RCT”).” To ensure comprehensive coverage, the reference lists of all identified randomized controlled trials (RCTs) were manually cross-checked for potential additional studies. Full details of the search strings and output for each database are provided in Table S1. Eligibility criteria We selected RCTs that compared hyperoxic versus normoxic oxygenation strategies during CPB based on the following PICO criteria: Population (P): adult patients undergoing elective or urgent cardiac surgery on CPB, including coronary artery bypass grafting (CABG), valve surgery, or combined procedures.Intervention (I): hyperoxic or liberal oxygenation strategy, defined as a high fraction of inspired oxygen (FiO2, typically 1.0) or a high target for arterial partial pressure of oxygen (PaO2 > 180 mmHg).Control (C): normoxic, conservative, or restrictive oxygenation strategy, targeting physiological oxygen levels (FiO2 of < 1 or PaO2 of < 180 mmHg).Outcomes (O): The primary outcomes were short-term mortality (up to 30 days), long-term mortality, and AKI. Secondary outcomes include clinical outcomes (myocardial infarction (MI), stroke, need for renal replacement therapy (RRT), arrhythmia, pneumonia, delirium, and SSIs) and healthcare resources utilization (duration of mechanical ventilation, length of intensive care unit (ICU) stay, and length of hospital stay (LoS)). Population (P): adult patients undergoing elective or urgent cardiac surgery on CPB, including coronary artery bypass grafting (CABG), valve surgery, or combined procedures. Population (P): adult patients undergoing elective or urgent cardiac surgery on CPB, including coronary artery bypass grafting (CABG), valve surgery, or combined procedures. Intervention (I): hyperoxic or liberal oxygenation strategy, defined as a high fraction of inspired oxygen (FiO2, typically 1.0) or a high target for arterial partial pressure of oxygen (PaO2 > 180 mmHg). Intervention (I): hyperoxic or liberal oxygenation strategy, defined as a high fraction of inspired oxygen (FiO2, typically 1.0) or a high target for arterial partial pressure of oxygen (PaO2 > 180 mmHg). Control (C): normoxic, conservative, or restrictive oxygenation strategy, targeting physiological oxygen levels (FiO2 of < 1 or PaO2 of < 180 mmHg). Control (C): normoxic, conservative, or restrictive oxygenation strategy, targeting physiological oxygen levels (FiO2 of < 1 or PaO2 of < 180 mmHg). Outcomes (O): The primary outcomes were short-term mortality (up to 30 days), long-term mortality, and AKI. Secondary outcomes include clinical outcomes (myocardial infarction (MI), stroke, need for renal replacement therapy (RRT), arrhythmia, pneumonia, delirium, and SSIs) and healthcare resources utilization (duration of mechanical ventilation, length of intensive care unit (ICU) stay, and length of hospital stay (LoS)). Outcomes (O): The primary outcomes were short-term mortality (up to 30 days), long-term mortality, and AKI. Secondary outcomes include clinical outcomes (myocardial infarction (MI), stroke, need for renal replacement therapy (RRT), arrhythmia, pneumonia, delirium, and SSIs) and healthcare resources utilization (duration of mechanical ventilation, length of intensive care unit (ICU) stay, and length of hospital stay (LoS)). We excluded observational studies, non-randomized trials, and studies involving pediatric populations or off-pump surgery. Study selection The retrieved records were imported into Covidence for automatic duplicate removal. Subsequently, two independent reviewers (N.Y.A. and B.H.A.) screened titles and abstracts for relevance. Then, potentially eligible studies underwent a full-text assessment against the inclusion criteria. Any discrepancies arising during the selection process were resolved through discussion or by consulting a third senior investigator. Data extraction Data was extracted independently by two investigators (S.A.A. and A.M.A.) using a pre-piloted Excel spreadsheet. Any discrepancies arising during the extraction process were resolved through discussion or by consulting a third senior investigator. The following data points were collected: Study Characteristics: Author, publication year, country, study design, sample size, specific oxygenation targets for both intervention and control groups, type of surgery, timing of intervention, temperature management, surgery type, main inclusion criteria, primary outcome, and follow-up duration.Participant Demographics and Clinical Data: Age, gender, body mass index (BMI), baseline left ventricular ejection fraction (LVEF), and comorbidities.Outcome Data: Event counts and total participants for dichotomous outcomes (AKI, mortality, MI., stroke, RRT, arrhythmia, pneumonia, delirium, SSI) and mean/standard deviation (SD) for continuous outcomes (duration of mechanical ventilation, length of ICU stay, and LoS). Data reported as medians and interquartile ranges (IQR) were converted to means and SDs using the methods described by Wan et al. [30]. Study Characteristics: Author, publication year, country, study design, sample size, specific oxygenation targets for both intervention and control groups, type of surgery, timing of intervention, temperature management, surgery type, main inclusion criteria, primary outcome, and follow-up duration. Study Characteristics: Author, publication year, country, study design, sample size, specific oxygenation targets for both intervention and control groups, type of surgery, timing of intervention, temperature management, surgery type, main inclusion criteria, primary outcome, and follow-up duration. Participant Demographics and Clinical Data: Age, gender, body mass index (BMI), baseline left ventricular ejection fraction (LVEF), and comorbidities. Participant Demographics and Clinical Data: Age, gender, body mass index (BMI), baseline left ventricular ejection fraction (LVEF), and comorbidities. Outcome Data: Event counts and total participants for dichotomous outcomes (AKI, mortality, MI., stroke, RRT, arrhythmia, pneumonia, delirium, SSI) and mean/standard deviation (SD) for continuous outcomes (duration of mechanical ventilation, length of ICU stay, and LoS). Data reported as medians and interquartile ranges (IQR) were converted to means and SDs using the methods described by Wan et al. [30]. Outcome Data: Event counts and total participants for dichotomous outcomes (AKI, mortality, MI., stroke, RRT, arrhythmia, pneumonia, delirium, SSI) and mean/standard deviation (SD) for continuous outcomes (duration of mechanical ventilation, length of ICU stay, and LoS). Data reported as medians and interquartile ranges (IQR) were converted to means and SDs using the methods described by Wan et al. [30]. Risk of bias and certainty of evidence The risk of bias for individual RCTs was assessed independently by at least two reviewers (N.Y.A. and A.A.A.) using the Cochrane Risk of Bias 2 (RoB 2) tool [31]. This assessment covered five domains: the randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Visualizations of the risk-of-bias assessments were generated using the RobVis tool. Additionally, the certainty of the pooled evidence was evaluated using the GRADE (Grading of Recommendations Assessment, Development, and Evaluation) approach [32, 33], which accounts for the risk of bias, inconsistency, indirectness, imprecision, and publication bias. Disagreements were resolved through consensus. Any discrepancies arising during either process were resolved through discussion or by consulting a senior investigator. Statistical analysis Statistical analyses were conducted using Stata/SE software, version 19 (StataCorp). For dichotomous outcomes, we calculated risk ratios (RR) with 95% confidence intervals (CIs). Continuous outcomes were analyzed using mean differences (MD) with 95% CIs. A random-effects model (REML) was used for all outcomes to account for the observed variance in the interventions’ oxygenation thresholds. To include studies with both-arm zero events, a continuity correction of 0.5 was applied [34]. Heterogeneity was assessed using the Chi-squared test and I2 statistic. Sources of heterogeneity were investigated using sensitivity analyses (the leave-one-out method) and, where appropriate, Galbraith plots. Publication bias was assessed via funnel plots and Egger’s test. Results Search results and study selection A total of 1,937 records were identified through database searching. After removing 1,045 duplicate or ineligible records, 892 records remained for screening. During the screening process, 872 records were excluded. Then, 20 full texts were assessed for eligibility. Of these, eight studies were excluded for different reasons (Table S2). Finally, 12 studies [3–5, 15–23] met the inclusion criteria and were included in the systematic review (Fig. 1). Fig. 1PRISMA flow chart of the screening process PRISMA flow chart of the screening process Characteristics of included studies This review synthesized evidence from 12 RCTs involving 2,732 patients [3–5, 15–23]. The studies were conducted across multiple countries, primarily in Europe and the United States. The duration of follow-up varied significantly, ranging from hospital discharge to approximately six years, and the specific oxygenation targets differed between studies. Further details on the study design of the included trials are outlined in (Table 1). Also, details on the included patients’ baseline data are outlined in (Table 2). Table 1Summary characteristics of the included RCTsStudy IDStudy designCountryTotal participantsHyperoxia strategyNormoxia strategyTiming of interventionTemperature managementSurgery typeMain inclusion criteriaAKI definitionPrimary outcomeFollow-up duration AbouArab et al. 2019 RCTFrance324FiO2​ 1.0 (PaO2​​ 447 ± 98 mmHg)PaO2​ < 150 mmHgDuring CPBNormothermia (>36 °C)Elective cardiac surgery (CABG, Valve, Combined)Adults >18y, elective cardiac surgery with CPBKDIGO criteriaPOAF and/or VT/VF within 15 days15 days Belboul et al. 1991 RCTSweden48PaO2​​ 190–300 mmHgPaO2​​ 75–112 mmHgDuring CPBHypothermia (28–30 °C)Elective CABGPatients undergoing elective CABG with CPBSevere postoperative renal dysfunction / arbitrary creatinine thresholdsBlood cell rheology (Red cell filtration rate)Post-op hospital stay Ihnken et al. 1998 RCTGermany40PaO2​ 400 mmHgPaO2​ 140 mmHgDuring CPBModerate Hypothermia (28–32 °C)Elective CABG40 consecutive patients for elective CABGNRPMN elastase, oxidative stress markers5 days post-op Lopez et al. 2024 RCTUSA200FiO2​ 1.0 intra-op (FdO2​ 0.8 on CPB)FiO2 0.21; FdO2 0.5 (CPB PaO2​ 80–110 mmHg)Pre-CPB, During CPB, Post-CPB (Intra-op)Not specified (Standard care)Elective open cardiac surgery (CABG, Valve, Combined)Adults >18y, elective open cardiac surgery, no pre-op O2KDIGO criteriaOxidative stress (F2​-isoprostanes) and AKI (creatinine change)12 months McGuinness et al. 2016 RCTNew Zealand, Australia298PaO2​​ > 150 mmHgPaO2​​ 75–90 mmHgDuring CPBMild Hypothermia (~32–34 °C)Elective cardiac surgery with CPBAdults >16y, scheduled cardiac surgery with CPBKDIGO criteriaAcute Kidney Injury (KDIGO criteria)90 days Onur et al. 2022 RCTTurkey100FiO2​ 1.0FiO2​ 0.4 (PaO2​ 100–180 mmHg)During CPBHypothermia (31–32 °C)Isolated CABGElective isolated CABG, ASA III, Age 18-80Postoperative renal dysfunction is defined by serum creatinine elevationMMSE scores (Cognitive function)6 months Shaefi et al. 2021 RCTUSA100FiO2​ 1.0FiO20.35 (PaO2​ 100–150 mmHg on CPB)Pre-CPB, During CPB, Post-CPB (Intra-op)Not specifiedCABG with CPBAge ≥ 65y, elective/urgent CABG with CPBKDIGO criteriaNeurocognition (Telephonic MoCA) on Day 26 months Smit et al. 2016 RCTNetherlands50FiO2 0.4-0.6 (PaO2​ 200-220 mmHg)FiO2 0.4 (PaO2​ 130-150 mmHg)During CPB, Post-CPB (ICU)Moderate Hypothermia (34–36 °C)Elective isolated CABGAdults >18y, elective isolated CABG, Hb ≥ 12.1 g/dLNRMyocardial injury (CK-MB, Troponin-T)12 h post-ICU Topcu et al. 2020 RCTTurkey48FiO2​ 0.7 (CPB), 0.5 (Rewarming)FiO2​ 0.35 (CPB), 0.45 (Rewarming)Pre-CPB, During CPBHypothermia (32 °C)Elective isolated CABGAdults, elective isolated on-pump CABGPostoperative renal dysfunction is defined by an increase in serum creatinine > 0.5 mg/dL from baseline or > 2.0 mg/dLMyocardial Total Oxidant Status (TOS)30 days Toraman et al. 2007 RCTTurkey90FiO2​ 0.4 (CPB), 0.6 (Rewarming)FiO2​ 0.35 (CPB), 0.45 (Rewarming)During CPBHypothermia (28 °C)Isolated CABGElective isolated CABGPostoperative renal dysfunction / 50% increase in baseline creatinineOptimal FiO2​ to avoid hyperoxemia (PaO2​ levels)Hospital discharge Von Zeppelin et al. 2024 RCTGermany45PaO2​​ 250–300 mmHgPaO2​ 100-150 mmHg on cardioplegia; gradual SaO2 from 90% to 100% on reperfusionDuring CPBHypothermiaElective cardiac surgeryAdults >50y, elective surgery, aortic clamp >75 minNRMyocardial injury markers (CK, CK-MB, Troponin)72 h Wiberg et al. 2025 RCTDenmark1389FiO21.0FiO2​ 0.5During CPB, Post-CPBNormothermia (≥ 36.5 °C)CABG and/or AVR (CPB-assisted)Adults >18y undergoing CABG and/or AVRSevere renal failure requiring dialysisComposite: Death, stroke, renal failure, or heart failureMedian 5.9 yearsAKI: Acute Kidney Injury, ASA: American Society of Anesthesiologists, AVR: Aortic Valve Replacement, CABG: Coronary Artery Bypass Grafting, CK: Creatine Kinase, CK-MB: Creatine Kinase-Myocardial Band, CPB: Cardiopulmonary Bypass, FiO₂: Fraction of Inspired Oxygen, Hb: Hemoglobin, ICU: Intensive Care Unit, KDIGO: Kidney Disease: Improving Global Outcomes, MMSE: Mini-Mental State Examination, MoCA: Montreal Cognitive Assessment, mmHg: Millimeters of Mercury, PaO₂: Partial Pressure of Oxygen, PMN: Polymorphonuclear, POAF: Postoperative Atrial Fibrillation, RCT: Randomized Controlled Trial, SaO₂: Arterial Oxygen Saturation, TOS: Total Oxidant Status, VT/VF: Ventricular Tachycardia / Ventricular Fibrillation Table 1Summary characteristics of the included RCTsStudy IDStudy designCountryTotal participantsHyperoxia strategyNormoxia strategyTiming of interventionTemperature managementSurgery typeMain inclusion criteriaAKI definitionPrimary outcomeFollow-up duration AbouArab et al. 2019 RCTFrance324FiO2​ 1.0 (PaO2​​ 447 ± 98 mmHg)PaO2​ < 150 mmHgDuring CPBNormothermia (>36 °C)Elective cardiac surgery (CABG, Valve, Combined)Adults >18y, elective cardiac surgery with CPBKDIGO criteriaPOAF and/or VT/VF within 15 days15 days Belboul et al. 1991 RCTSweden48PaO2​​ 190–300 mmHgPaO2​​ 75–112 mmHgDuring CPBHypothermia (28–30 °C)Elective CABGPatients undergoing elective CABG with CPBSevere postoperative renal dysfunction / arbitrary creatinine thresholdsBlood cell rheology (Red cell filtration rate)Post-op hospital stay Ihnken et al. 1998 RCTGermany40PaO2​ 400 mmHgPaO2​ 140 mmHgDuring CPBModerate Hypothermia (28–32 °C)Elective CABG40 consecutive patients for elective CABGNRPMN elastase, oxidative stress markers5 days post-op Lopez et al. 2024 RCTUSA200FiO2​ 1.0 intra-op (FdO2​ 0.8 on CPB)FiO2 0.21; FdO2 0.5 (CPB PaO2​ 80–110 mmHg)Pre-CPB, During CPB, Post-CPB (Intra-op)Not specified (Standard care)Elective open cardiac surgery (CABG, Valve, Combined)Adults >18y, elective open cardiac surgery, no pre-op O2KDIGO criteriaOxidative stress (F2​-isoprostanes) and AKI (creatinine change)12 months McGuinness et al. 2016 RCTNew Zealand, Australia298PaO2​​ > 150 mmHgPaO2​​ 75–90 mmHgDuring CPBMild Hypothermia (~32–34 °C)Elective cardiac surgery with CPBAdults >16y, scheduled cardiac surgery with CPBKDIGO criteriaAcute Kidney Injury (KDIGO criteria)90 days Onur et al. 2022 RCTTurkey100FiO2​ 1.0FiO2​ 0.4 (PaO2​ 100–180 mmHg)During CPBHypothermia (31–32 °C)Isolated CABGElective isolated CABG, ASA III, Age 18-80Postoperative renal dysfunction is defined by serum creatinine elevationMMSE scores (Cognitive function)6 months Shaefi et al. 2021 RCTUSA100FiO2​ 1.0FiO20.35 (PaO2​ 100–150 mmHg on CPB)Pre-CPB, During CPB, Post-CPB (Intra-op)Not specifiedCABG with CPBAge ≥ 65y, elective/urgent CABG with CPBKDIGO criteriaNeurocognition (Telephonic MoCA) on Day 26 months Smit et al. 2016 RCTNetherlands50FiO2 0.4-0.6 (PaO2​ 200-220 mmHg)FiO2 0.4 (PaO2​ 130-150 mmHg)During CPB, Post-CPB (ICU)Moderate Hypothermia (34–36 °C)Elective isolated CABGAdults >18y, elective isolated CABG, Hb ≥ 12.1 g/dLNRMyocardial injury (CK-MB, Troponin-T)12 h post-ICU Topcu et al. 2020 RCTTurkey48FiO2​ 0.7 (CPB), 0.5 (Rewarming)FiO2​ 0.35 (CPB), 0.45 (Rewarming)Pre-CPB, During CPBHypothermia (32 °C)Elective isolated CABGAdults, elective isolated on-pump CABGPostoperative renal dysfunction is defined by an increase in serum creatinine > 0.5 mg/dL from baseline or > 2.0 mg/dLMyocardial Total Oxidant Status (TOS)30 days Toraman et al. 2007 RCTTurkey90FiO2​ 0.4 (CPB), 0.6 (Rewarming)FiO2​ 0.35 (CPB), 0.45 (Rewarming)During CPBHypothermia (28 °C)Isolated CABGElective isolated CABGPostoperative renal dysfunction / 50% increase in baseline creatinineOptimal FiO2​ to avoid hyperoxemia (PaO2​ levels)Hospital discharge Von Zeppelin et al. 2024 RCTGermany45PaO2​​ 250–300 mmHgPaO2​ 100-150 mmHg on cardioplegia; gradual SaO2 from 90% to 100% on reperfusionDuring CPBHypothermiaElective cardiac surgeryAdults >50y, elective surgery, aortic clamp >75 minNRMyocardial injury markers (CK, CK-MB, Troponin)72 h Wiberg et al. 2025 RCTDenmark1389FiO21.0FiO2​ 0.5During CPB, Post-CPBNormothermia (≥ 36.5 °C)CABG and/or AVR (CPB-assisted)Adults >18y undergoing CABG and/or AVRSevere renal failure requiring dialysisComposite: Death, stroke, renal failure, or heart failureMedian 5.9 yearsAKI: Acute Kidney Injury, ASA: American Society of Anesthesiologists, AVR: Aortic Valve Replacement, CABG: Coronary Artery Bypass Grafting, CK: Creatine Kinase, CK-MB: Creatine Kinase-Myocardial Band, CPB: Cardiopulmonary Bypass, FiO₂: Fraction of Inspired Oxygen, Hb: Hemoglobin, ICU: Intensive Care Unit, KDIGO: Kidney Disease: Improving Global Outcomes, MMSE: Mini-Mental State Examination, MoCA: Montreal Cognitive Assessment, mmHg: Millimeters of Mercury, PaO₂: Partial Pressure of Oxygen, PMN: Polymorphonuclear, POAF: Postoperative Atrial Fibrillation, RCT: Randomized Controlled Trial, SaO₂: Arterial Oxygen Saturation, TOS: Total Oxidant Status, VT/VF: Ventricular Tachycardia / Ventricular Fibrillation Summary characteristics of the included RCTs AKI: Acute Kidney Injury, ASA: American Society of Anesthesiologists, AVR: Aortic Valve Replacement, CABG: Coronary Artery Bypass Grafting, CK: Creatine Kinase, CK-MB: Creatine Kinase-Myocardial Band, CPB: Cardiopulmonary Bypass, FiO₂: Fraction of Inspired Oxygen, Hb: Hemoglobin, ICU: Intensive Care Unit, KDIGO: Kidney Disease: Improving Global Outcomes, MMSE: Mini-Mental State Examination, MoCA: Montreal Cognitive Assessment, mmHg: Millimeters of Mercury, PaO₂: Partial Pressure of Oxygen, PMN: Polymorphonuclear, POAF: Postoperative Atrial Fibrillation, RCT: Randomized Controlled Trial, SaO₂: Arterial Oxygen Saturation, TOS: Total Oxidant Status, VT/VF: Ventricular Tachycardia / Ventricular Fibrillation Table 2Baseline characteristics of the participantsStudy IDNumber of patients in each groupAge (years), mean (SD)Gender (male), N. (%)LVEF, mean (SD)BMI, mean (SD)Comorbidities N. (%)HyperoxiaNormoxiaHyperoxiaNormoxiaHyperoxiaNormoxiaHyperoxiaNormoxiaHyperoxiaNormoxiaHyperoxiaNormoxiaAbouArab et al. 201916116368 ± 1068 ± 9129/161 (80.1%)123/163 (75.5%)27 ± 427 ± 456 ± 1057 ± 9HTN: 111 (69%), DM: 48 (30%), Dyslipidemia: 93 (58%), CKD: 18 (11%), PVD: 14 (9%), COPD: 13 (8%)HTN: 112 (69%), DM: 39 (24%), Dyslipidemia: 91 (56%), CKD: 19 (12%), PVD: 25 (15%), COPD: 11 (7%)Belboul et al. 1991242462 ± 461 ± 719/24 (79.2%)21/24 (87.5%)NRNR60 ± 1060 ± 20NRNRIhnken et al. 1998202061 ± 263 ± 216/20 (80.0%)17/20 (85.0%)NRNR61 ± 561 ± 6Smoking: 14 (70%), DM: 5 (25%)Smoking: 12 (60%), DM: 7 (35%)Lopez et al. 202410010065 [58–71]*67 [60–73]*73/100 (73.0%)67/100 (67.0%)30 [26–35]*30 [26–34]*NRNRHTN: 88 (88%), DM: 44 (44%), CKD: 28 (28%), COPD: 15 (15%), Cerebrovascular Dis: 7 (7%), PVD: 8 (8%)HTN: 81 (81%), DM: 38 (38%), CKD: 30 (30%), COPD: 13 (13%), Cerebrovascular Dis: 11 (11%), PVD: 5 (5%)McGuinness et al. 201614914967.4 ± 11.267.1 ± 10.9120/149 (80.5%)114/149 (76.5%)29.2 ± 5.729.3 ± 5.5NRNRHTN: 104 (70%), Dyslipidemia: 104 (70%), DM: 42 (28%), Previous MI: 39 (26%), PVD: 15 (10%), COPD: 15 (10%), Cerebrovascular Dis: 11 (7%)HTN: 108 (73%), Dyslipidemia: 112 (75%), DM: 39 (26%), Previous MI: 44 (30%), PVD: 19 (13%), COPD: 11 (7%), Cerebrovascular Dis: 10 (7%)Onur et al. 2022505061.4 ± 8.262.1 ± 8.438/50 (76.0%)36/50 (72.0%)NRNR53.6 ± 9.254.1 ± 8.4HTN: 31 (62%), DM: 22 (44%), Smoking: 18 (36%), COPD: 11 (22%)HTN: 29 (58%), DM: 23 (46%), Smoking: 16 (32%), COPD: 13 (26%)Shaefi et al. 2021495171 [68–75]*72 [68–76]*40/49 (81.6%)44/51 (86.3%)29 [26–32]*29 [26–33]*55 [45–60]55 [50–60]HTN: 44 (90%), Dyslipidemia: 41 (84%), DM: 19 (39%), Previous MI: 11 (22%), Atrial Fibrillation: 10 (20%), COPD: 5 (10%), Stroke: 3 (6%)HTN: 47 (92%), Dyslipidemia: 45 (88%), DM: 21 (41%), Previous MI: 10 (20%), Atrial Fibrillation: 10 (20%), COPD: 5 (10%), Stroke: 4 (8%)Smit et al. 2016252567 [59–73]*67 [61–72]*22/25 (88.0%)22/25 (88.0%)27 [25–30]*28 [26–30]*NRNRHTN: 14 (56%), Previous MI: 8 (32%), DM: 3 (12%), PVD: 2 (8%)HTN: 16 (64%), Previous MI: 13 (52%), DM: 5 (20%), PVD: 3 (12%)Topcu et al. 2020262260 ± 956 ± 922/26 (84.6%)18/22 (81.8%)NRNR55 [35–60]55 [40–60]Smoking: 15 (58%), DM: 10 (38%), HTN: 10 (38%), Hyperlipidemia: 11 (42%), COPD: 5 (19%)Smoking: 11 (50%), DM: 11 (50%), HTN: 9 (41%), Hyperlipidemia: 8 (36%), COPD: 2 (9%)Toraman et al. 2007603058.5 ± 11.060.0 ± 10.030/60 (50.0%)18/30 (60.0%)27.0 ± 3.028.0 ± 4.0NRNRHTN: 24 (40%), DM: 18 (30%), Previous MI: 20 (33%), Smoking: 22 (37%)HTN: 10 (33%), DM: 10 (33%), Previous MI: 8 (27%), Smoking: 12 (40%)Von Zeppelin et al. 2024252068.4 ± 9.867.1 ± 8.6NRNR30.0 ± 5.827.9 ± 7.1NRNRHTN: 20 (80%), Hyperlipidemia: 10 (40%), Previous MI: 9 (36%), PVD: 7 (28%), DM: 6 (24%), COPD: 3 (12%)HTN: 14 (70%), Hyperlipidemia: 6 (30%), Previous MI: 3 (15%), DM: 7 (35%), PVD: 1 (5%), COPD: 1 (5%)Wiberg et al. 202569469567 (30–85)**66 (29–85)**573/694 (82.6%)583/695 (83.9%)27 [25–30]*27 [25–31]*NRNRHTN: 455 (66%), Heart Failure: 110 (16%), DM: 101 (15%), Atrial Fibrillation: 82 (12%), COPD: 50 (7%), Prev Stroke: 38 (5%), PVD: 31 (4%), Dialysis: 2 (0.3%)HTN: 443 (64%), Heart Failure: 115 (17%), DM: 105 (15%), Atrial Fibrillation: 98 (14%), COPD: 56 (8%), Prev Stroke: 40 (6%), PVD: 44 (6%), Dialysis: 4 (0.6%)*[…] denote Median [Interquartile Range]. ** denotes Mean (Range); BMI: Body Mass Index, CKD: Chronic Kidney Disease, COPD: Chronic Obstructive Pulmonary Disease, Dis: Disease, DM: Diabetes Mellitus, HTN: Hypertension, LVEF: Left Ventricular Ejection Fraction, MI: Myocardial Infarction, N: Number, NR: Not Reported, Prev: Previous, PVD: Peripheral Vascular Disease, SD: Standard Deviation Table 2Baseline characteristics of the participantsStudy IDNumber of patients in each groupAge (years), mean (SD)Gender (male), N. (%)LVEF, mean (SD)BMI, mean (SD)Comorbidities N. (%)HyperoxiaNormoxiaHyperoxiaNormoxiaHyperoxiaNormoxiaHyperoxiaNormoxiaHyperoxiaNormoxiaHyperoxiaNormoxiaAbouArab et al. 201916116368 ± 1068 ± 9129/161 (80.1%)123/163 (75.5%)27 ± 427 ± 456 ± 1057 ± 9HTN: 111 (69%), DM: 48 (30%), Dyslipidemia: 93 (58%), CKD: 18 (11%), PVD: 14 (9%), COPD: 13 (8%)HTN: 112 (69%), DM: 39 (24%), Dyslipidemia: 91 (56%), CKD: 19 (12%), PVD: 25 (15%), COPD: 11 (7%)Belboul et al. 1991242462 ± 461 ± 719/24 (79.2%)21/24 (87.5%)NRNR60 ± 1060 ± 20NRNRIhnken et al. 1998202061 ± 263 ± 216/20 (80.0%)17/20 (85.0%)NRNR61 ± 561 ± 6Smoking: 14 (70%), DM: 5 (25%)Smoking: 12 (60%), DM: 7 (35%)Lopez et al. 202410010065 [58–71]*67 [60–73]*73/100 (73.0%)67/100 (67.0%)30 [26–35]*30 [26–34]*NRNRHTN: 88 (88%), DM: 44 (44%), CKD: 28 (28%), COPD: 15 (15%), Cerebrovascular Dis: 7 (7%), PVD: 8 (8%)HTN: 81 (81%), DM: 38 (38%), CKD: 30 (30%), COPD: 13 (13%), Cerebrovascular Dis: 11 (11%), PVD: 5 (5%)McGuinness et al. 201614914967.4 ± 11.267.1 ± 10.9120/149 (80.5%)114/149 (76.5%)29.2 ± 5.729.3 ± 5.5NRNRHTN: 104 (70%), Dyslipidemia: 104 (70%), DM: 42 (28%), Previous MI: 39 (26%), PVD: 15 (10%), COPD: 15 (10%), Cerebrovascular Dis: 11 (7%)HTN: 108 (73%), Dyslipidemia: 112 (75%), DM: 39 (26%), Previous MI: 44 (30%), PVD: 19 (13%), COPD: 11 (7%), Cerebrovascular Dis: 10 (7%)Onur et al. 2022505061.4 ± 8.262.1 ± 8.438/50 (76.0%)36/50 (72.0%)NRNR53.6 ± 9.254.1 ± 8.4HTN: 31 (62%), DM: 22 (44%), Smoking: 18 (36%), COPD: 11 (22%)HTN: 29 (58%), DM: 23 (46%), Smoking: 16 (32%), COPD: 13 (26%)Shaefi et al. 2021495171 [68–75]*72 [68–76]*40/49 (81.6%)44/51 (86.3%)29 [26–32]*29 [26–33]*55 [45–60]55 [50–60]HTN: 44 (90%), Dyslipidemia: 41 (84%), DM: 19 (39%), Previous MI: 11 (22%), Atrial Fibrillation: 10 (20%), COPD: 5 (10%), Stroke: 3 (6%)HTN: 47 (92%), Dyslipidemia: 45 (88%), DM: 21 (41%), Previous MI: 10 (20%), Atrial Fibrillation: 10 (20%), COPD: 5 (10%), Stroke: 4 (8%)Smit et al. 2016252567 [59–73]*67 [61–72]*22/25 (88.0%)22/25 (88.0%)27 [25–30]*28 [26–30]*NRNRHTN: 14 (56%), Previous MI: 8 (32%), DM: 3 (12%), PVD: 2 (8%)HTN: 16 (64%), Previous MI: 13 (52%), DM: 5 (20%), PVD: 3 (12%)Topcu et al. 2020262260 ± 956 ± 922/26 (84.6%)18/22 (81.8%)NRNR55 [35–60]55 [40–60]Smoking: 15 (58%), DM: 10 (38%), HTN: 10 (38%), Hyperlipidemia: 11 (42%), COPD: 5 (19%)Smoking: 11 (50%), DM: 11 (50%), HTN: 9 (41%), Hyperlipidemia: 8 (36%), COPD: 2 (9%)Toraman et al. 2007603058.5 ± 11.060.0 ± 10.030/60 (50.0%)18/30 (60.0%)27.0 ± 3.028.0 ± 4.0NRNRHTN: 24 (40%), DM: 18 (30%), Previous MI: 20 (33%), Smoking: 22 (37%)HTN: 10 (33%), DM: 10 (33%), Previous MI: 8 (27%), Smoking: 12 (40%)Von Zeppelin et al. 2024252068.4 ± 9.867.1 ± 8.6NRNR30.0 ± 5.827.9 ± 7.1NRNRHTN: 20 (80%), Hyperlipidemia: 10 (40%), Previous MI: 9 (36%), PVD: 7 (28%), DM: 6 (24%), COPD: 3 (12%)HTN: 14 (70%), Hyperlipidemia: 6 (30%), Previous MI: 3 (15%), DM: 7 (35%), PVD: 1 (5%), COPD: 1 (5%)Wiberg et al. 202569469567 (30–85)**66 (29–85)**573/694 (82.6%)583/695 (83.9%)27 [25–30]*27 [25–31]*NRNRHTN: 455 (66%), Heart Failure: 110 (16%), DM: 101 (15%), Atrial Fibrillation: 82 (12%), COPD: 50 (7%), Prev Stroke: 38 (5%), PVD: 31 (4%), Dialysis: 2 (0.3%)HTN: 443 (64%), Heart Failure: 115 (17%), DM: 105 (15%), Atrial Fibrillation: 98 (14%), COPD: 56 (8%), Prev Stroke: 40 (6%), PVD: 44 (6%), Dialysis: 4 (0.6%)*[…] denote Median [Interquartile Range]. ** denotes Mean (Range); BMI: Body Mass Index, CKD: Chronic Kidney Disease, COPD: Chronic Obstructive Pulmonary Disease, Dis: Disease, DM: Diabetes Mellitus, HTN: Hypertension, LVEF: Left Ventricular Ejection Fraction, MI: Myocardial Infarction, N: Number, NR: Not Reported, Prev: Previous, PVD: Peripheral Vascular Disease, SD: Standard Deviation Baseline characteristics of the participants *[…] denote Median [Interquartile Range]. ** denotes Mean (Range); BMI: Body Mass Index, CKD: Chronic Kidney Disease, COPD: Chronic Obstructive Pulmonary Disease, Dis: Disease, DM: Diabetes Mellitus, HTN: Hypertension, LVEF: Left Ventricular Ejection Fraction, MI: Myocardial Infarction, N: Number, NR: Not Reported, Prev: Previous, PVD: Peripheral Vascular Disease, SD: Standard Deviation Risk of bias and certainty of evidence Six RCTs showed an overall low risk of bias [4, 5, 16, 18, 22, 23], four showed some concerns [3, 19–21], and two showed a high risk of bias [15, 17] (Fig. 2). The concerns were primarily due to the lack of blinding, which raised concerns about performance bias. Von Zeppelin et al. (2024) showed a high risk of selection bias, as they used lot randomization without providing further details about how the lots were generated or concealed [17]. Also, the full details of the certainty of evidence levels for each outcome are presented in (Table 3). Fig. 2Quality assessment of risk of bias in the included trials. The upper panel presents a schematic representation of risk levels (low = green, unclear = yellow, and high = red) for specific types of bias in the studies reviewed. The lower panel presents risk levels (low = green, unclear = yellow, and high = red) for the bias subtypes across the studies included in this review Quality assessment of risk of bias in the included trials. The upper panel presents a schematic representation of risk levels (low = green, unclear = yellow, and high = red) for specific types of bias in the studies reviewed. The lower panel presents risk levels (low = green, unclear = yellow, and high = red) for the bias subtypes across the studies included in this review Table 3GRADE certainty of evidence assessmentCertainty assessmentSummary of findingsParticipants(studies) follow-upRisk of biasInconsistencyIndirectnessImprecisionPublication biasOverall certainty of evidenceStudy event rates (%)Relative effect(95% CI)Anticipated absolute effectsWith [Normoxia]With [Hyperoxia]Risk with [Normoxia]Risk difference with [Hyperoxia] Short-term mortality 2592(10 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c16/1280 (1.3%)24/1312 (1.8%) RR 1.23 (0.60 to 2.54)16/1280 (1.3%)3 more per 1,000(from 5 fewer to 19 more) Long-term mortality 2111(4 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c95/1059 (9.0%)102/1052 (9.7%) RR 1.10 (0.84 to 1.43)95/1059 (9.0%)9 more per 1,000(from 14 fewer to 39 more) Acute kidney injury 2567(9 RCTs)not seriousnot seriousseriousbnot seriousnone⨁⨁◯◯Lowb, c204/1284 (15.9%)196/1283 (15.3%) RR 0.94 (0.82 to 1.07)204/1284 (15.9%)10 fewer per 1,000(from 29 fewer to 11 more) Myocardial infarction 1732(5 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c8/864 (0.9%)10/868 (1.2%) RR 1.13 (0.47 to 2.72)8/864 (0.9%)1 more per 1,000(from 5 fewer to 16 more) Stroke 825(7 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c3/410 (0.7%)4/415 (1.0%) RR 1.14 (0.35 to 3.70)3/410 (0.7%)1 more per 1,000(from 5 fewer to 20 more) Renal replacement therapy 706(5 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c7/354 (2.0%)5/352 (1.4%) RR 0.78 (0.27 to 2.29)7/354 (2.0%)4 fewer per 1,000(from 14 fewer to 26 more) Arrhythmia 820(6 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c125/410 (30.5%)141/410 (34.4%) RR 1.13 (0.93 to 1.38)125/410 (30.5%)40 more per 1,000(from 21 fewer to 116 more) Pneumonia 348(3 RCTs)seriousanot seriousseriousbseriouscnone⨁◯◯◯Very lowa, b,c5/173 (2.9%)8/175 (4.6%) RR 1.44 (0.41 to 5.01)5/173 (2.9%)13 more per 1,000(from 17 fewer to 116 more) Delirium 400(3 RCTs)seriousanot seriousseriousbseriouscnone⨁◯◯◯Very lowa, b,c35/201 (17.4%)41/199 (20.6%) RR 1.17 (0.79 to 1.74)35/201 (17.4%)30 more per 1,000(from 37 fewer to 129 more) Surgical site infections 1785(5 RCTs)not seriousnot seriousseriousbseriouscnone⨁◯◯◯Very lowa, b,c32/891 (3.6%)24/894 (2.7%) RR 0.83 (0.43 to 1.62)32/891 (3.6%)6 fewer per 1,000(from 20 fewer to 22 more) Duration of mechanical ventilation 2273(9 RCTs)seriousavery seriousdseriousbseriouscnone⨁◯◯◯Very lowa, b,c, d11371136-1137MD 1.02 h higher(0.06 lower to 2.09 higher) Length of ICU stay 2654(11 RCTs)seriousavery seriousdseriousbnot seriousnone⨁◯◯◯Very lowa, b,d13261328-1326MD 0.05 days higher(0.05 lower to 0.15 higher) Length of hospital stay 2404(9 RCTs)seriousavery seriousdseriousbseriouscnone⨁◯◯◯Very lowa, b,c, d12011203-1201MD 0.39 days higher(0.04 lower to 0.82 higher)CI: confidence interval; MD: mean difference; RR: risk ratio; Explanations:a. Half of the trials either showed some concern or a high risk of biasb. The oxygenation thresholds varied across the included trialsc. A wide confidence interval that does not exclude the appreciable risk of harm or benefitd. I2 > 75% Table 3GRADE certainty of evidence assessmentCertainty assessmentSummary of findingsParticipants(studies) follow-upRisk of biasInconsistencyIndirectnessImprecisionPublication biasOverall certainty of evidenceStudy event rates (%)Relative effect(95% CI)Anticipated absolute effectsWith [Normoxia]With [Hyperoxia]Risk with [Normoxia]Risk difference with [Hyperoxia] Short-term mortality 2592(10 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c16/1280 (1.3%)24/1312 (1.8%) RR 1.23 (0.60 to 2.54)16/1280 (1.3%)3 more per 1,000(from 5 fewer to 19 more) Long-term mortality 2111(4 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c95/1059 (9.0%)102/1052 (9.7%) RR 1.10 (0.84 to 1.43)95/1059 (9.0%)9 more per 1,000(from 14 fewer to 39 more) Acute kidney injury 2567(9 RCTs)not seriousnot seriousseriousbnot seriousnone⨁⨁◯◯Lowb, c204/1284 (15.9%)196/1283 (15.3%) RR 0.94 (0.82 to 1.07)204/1284 (15.9%)10 fewer per 1,000(from 29 fewer to 11 more) Myocardial infarction 1732(5 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c8/864 (0.9%)10/868 (1.2%) RR 1.13 (0.47 to 2.72)8/864 (0.9%)1 more per 1,000(from 5 fewer to 16 more) Stroke 825(7 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c3/410 (0.7%)4/415 (1.0%) RR 1.14 (0.35 to 3.70)3/410 (0.7%)1 more per 1,000(from 5 fewer to 20 more) Renal replacement therapy 706(5 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c7/354 (2.0%)5/352 (1.4%) RR 0.78 (0.27 to 2.29)7/354 (2.0%)4 fewer per 1,000(from 14 fewer to 26 more) Arrhythmia 820(6 RCTs)not seriousnot seriousseriousbseriouscnone⨁⨁◯◯Lowb, c125/410 (30.5%)141/410 (34.4%) RR 1.13 (0.93 to 1.38)125/410 (30.5%)40 more per 1,000(from 21 fewer to 116 more) Pneumonia 348(3 RCTs)seriousanot seriousseriousbseriouscnone⨁◯◯◯Very lowa, b,c5/173 (2.9%)8/175 (4.6%) RR 1.44 (0.41 to 5.01)5/173 (2.9%)13 more per 1,000(from 17 fewer to 116 more) Delirium 400(3 RCTs)seriousanot seriousseriousbseriouscnone⨁◯◯◯Very lowa, b,c35/201 (17.4%)41/199 (20.6%) RR 1.17 (0.79 to 1.74)35/201 (17.4%)30 more per 1,000(from 37 fewer to 129 more) Surgical site infections 1785(5 RCTs)not seriousnot seriousseriousbseriouscnone⨁◯◯◯Very lowa, b,c32/891 (3.6%)24/894 (2.7%) RR 0.83 (0.43 to 1.62)32/891 (3.6%)6 fewer per 1,000(from 20 fewer to 22 more) Duration of mechanical ventilation 2273(9 RCTs)seriousavery seriousdseriousbseriouscnone⨁◯◯◯Very lowa, b,c, d11371136-1137MD 1.02 h higher(0.06 lower to 2.09 higher) Length of ICU stay 2654(11 RCTs)seriousavery seriousdseriousbnot seriousnone⨁◯◯◯Very lowa, b,d13261328-1326MD 0.05 days higher(0.05 lower to 0.15 higher) Length of hospital stay 2404(9 RCTs)seriousavery seriousdseriousbseriouscnone⨁◯◯◯Very lowa, b,c, d12011203-1201MD 0.39 days higher(0.04 lower to 0.82 higher)CI: confidence interval; MD: mean difference; RR: risk ratio; Explanations:a. Half of the trials either showed some concern or a high risk of biasb. The oxygenation thresholds varied across the included trialsc. A wide confidence interval that does not exclude the appreciable risk of harm or benefitd. I2 > 75% GRADE certainty of evidence assessment 2592 (10 RCTs) ⨁⨁◯◯ Lowb, c RR 1.23 (0.60 to 2.54) 3 more per 1,000 (from 5 fewer to 19 more) 2111 (4 RCTs) ⨁⨁◯◯ Lowb, c RR 1.10 (0.84 to 1.43) 9 more per 1,000 (from 14 fewer to 39 more) 2567 (9 RCTs) ⨁⨁◯◯ Lowb, c RR 0.94 (0.82 to 1.07) 10 fewer per 1,000 (from 29 fewer to 11 more) 1732 (5 RCTs) ⨁⨁◯◯ Lowb, c RR 1.13 (0.47 to 2.72) 1 more per 1,000 (from 5 fewer to 16 more) 825 (7 RCTs) ⨁⨁◯◯ Lowb, c RR 1.14 (0.35 to 3.70) 1 more per 1,000 (from 5 fewer to 20 more) 706 (5 RCTs) ⨁⨁◯◯ Lowb, c RR 0.78 (0.27 to 2.29) 4 fewer per 1,000 (from 14 fewer to 26 more) 820 (6 RCTs) ⨁⨁◯◯ Lowb, c RR 1.13 (0.93 to 1.38) 40 more per 1,000 (from 21 fewer to 116 more) 348 (3 RCTs) ⨁◯◯◯ Very lowa, b,c RR 1.44 (0.41 to 5.01) 13 more per 1,000 (from 17 fewer to 116 more) 400 (3 RCTs) ⨁◯◯◯ Very lowa, b,c RR 1.17 (0.79 to 1.74) 30 more per 1,000 (from 37 fewer to 129 more) 1785 (5 RCTs) ⨁◯◯◯ Very lowa, b,c RR 0.83 (0.43 to 1.62) 6 fewer per 1,000 (from 20 fewer to 22 more) 2273 (9 RCTs) ⨁◯◯◯ Very lowa, b,c, d MD 1.02 h higher (0.06 lower to 2.09 higher) 2654 (11 RCTs) ⨁◯◯◯ Very lowa, b,d MD 0.05 days higher (0.05 lower to 0.15 higher) 2404 (9 RCTs) ⨁◯◯◯ Very lowa, b,c, d MD 0.39 days higher (0.04 lower to 0.82 higher) CI: confidence interval; MD: mean difference; RR: risk ratio; Explanations: a. Half of the trials either showed some concern or a high risk of bias b. The oxygenation thresholds varied across the included trials c. A wide confidence interval that does not exclude the appreciable risk of harm or benefit d. I2 > 75% Primary outcomes There was no significant difference between both groups in the incidence of short-term mortality (RR: 1.23, 95% CI [0.60, 2.54], p = 0.57, I2 = 9.19%) (Fig. 3-A), long-term mortality (RR: 1.10, 95% CI [0.84, 1.43], p = 0.48, I2 = 0%) (Fig. 3-B) or AKI (RR: 0.94, 95% CI [0.82, 1.07], p = 0.32, I2 = 0%) (Fig. 3-C). Visual inspection of the funnel plot of short-term mortality (Figure S1) suggested a relatively symmetrical distribution of studies. While Egger’s test did not indicate significant publication bias (p = 0.35), this result should be interpreted with caution, given the limited number of included studies. Additionally, the test for subgroup differences based on the definition of hyperoxia (FiO2-Based vs. PaO2-Based) was not significant in short-term mortality (p = 0.18) (Figure S2) and AKI (p = 0.54) (Figure S3). Fig. 3Forest plots of the primary outcomes, CI: confidence interval Forest plots of the primary outcomes, CI: confidence interval Secondary clinical outcomes There was no significant difference between both groups in the incidence of MI (RR: 1.13, 95% CI [0.47, 2.72], p = 0.78, I2 = 0%) (Fig. 4-A), stroke (RR: 1.14, 95% CI [0.35, 3.70], p = 0.83, I2 = 0%) (Fig. 4-B), RRT (RR: 0.78, 95% CI [0.27, 2.29], p = 0.65, I2 = 0%) (Fig. 4-C), arrhythmia (RR: 1.13, 95% CI [0.93, 1.38], p = 0.21, I2 = 0%) (Fig. 5-A), pneumonia (RR: 1.44, 95% CI [0.41, 5.01], p = 0.57, I2 = 12.66%) (Fig. 5-B), delirium (RR: 1.17, 95% CI [0.79, 1.74], p = 0.43, I2 = 0%) (Fig. 5-C), or SSIs (RR: 0.83, 95% CI [0.43, 1.62], p = 0.59, I2 = 20.45%) (Fig. 5-D). Also, the test for subgroup differences based on the definition of hyperoxia was not significant for MI (p = 0.63) (Figure S4), stroke (p = 0.68) (Figure S5), RRT (p = 0.48) (Figure S6), arrhythmia (p = 0.49) (Figure S7), or SSIs (p = 0.28) (Figure S8). Fig. 4Forest plots of the secondary clinical outcomes, CI: confidence interval Forest plots of the secondary clinical outcomes, CI: confidence interval Fig. 5Forest plots of the secondary clinical outcomes, CI: confidence interval Forest plots of the secondary clinical outcomes, CI: confidence interval Hospital resources utilization outcomes There was no significant difference between both groups regarding the duration of mechanical ventilation (MD: 1.02, 95% CI [-0.06, 2.09], p = 0.06, I2 = 91.74%) (Fig. 6-A), length of ICU stay (MD: 0.05, 95% CI [-0.05, 0.15], p = 0.32, I2 = 84.12%) (Fig. 6-B), or LoS (MD: 0.39, 95% CI [-0.04, 0.82], p = 0.08, I2 = 80.17%) (Fig. 6-C). For the duration of mechanical ventilation, a leave-one-out sensitivity analysis showed that the result became significant after omitting Shaefi et al., favoring normoxia (p = 0.035) (Figure S9). Similarly, for the LoS, the result became significant after omitting Onur et al., favoring hyperoxia (p = 0.013) (Figure S10). However, the results for ICU length of stay remained non-significant across all scenarios (Figure S11). Fig. 6Forest plots of the hospital resources’ utilization outcomes, CI: confidence interval Forest plots of the hospital resources’ utilization outcomes, CI: confidence interval Additionally, the Galbraith plot for the duration of mechanical ventilation showed that Onur et al., Ihnken et al., and Shaefi et al. were identified as potential sources of heterogeneity (Figure S12). The Galbraith plot for the length of ICU stay identified Shaefi et al. and Ihnken et al. as possible sources of the observed heterogeneity (Figure S13). For the length of hospital stay, the Galbraith plot suggested that Ihnken et al., Abou-Arab et al., and Onur et al. were outliers (Figure S14). Also, the test for subgroup differences based on the definition of hyperoxia was not significant for the duration of mechanical ventilation (p = 0.18) (Figure S15), length of ICU stay (p = 0.52) (Figure S16), or the length of hospital stay (p = 0.92) (Figure S17). Finally, a visual inspection of the funnel plot for ICU length of stay suggested a relatively asymmetrical distribution (Figure S18). Although Egger’s test was not statistically significant (p = 0.38), the statistical power of this test to exclude publication bias is limited by the small number of studies. Finally, a comprehensive summary of all meta-analysis outcomes, including event rates, pooled effect sizes, and heterogeneity statistics, is provided in (Table S3). Discussion After synthesizing 12 RCTs involving 2,732 patients, we observed no significant differences between hyperoxic and normoxic strategies in clinical and hospital utilization outcomes. Still, the current evidence certainty remains very low according to our GRADE assessment. In comparison with earlier meta-analyses (7,24,25), our results provide a novel perspective on oxygen management during CPB. Earlier reviews were often inconclusive, primarily driven by small, single-center studies with high heterogeneity and insufficient statistical power to detect meaningful differences in rare clinical outcomes, such as mortality or severe AKI [7, 24, 25]. Our findings addressed these limitations by incorporating new high-quality trials, specifically the GLORIOUS trial [16] and the ROCS trial [22]. The large GLORIOUS trial offered the statistical power to confirm that the safety signals (particularly the potential risks of hyperoxia observed in previous observational studies or general surgery trials) may not be applicable in the context of adult cardiac surgery [16, 35]. Additionally, our pooled analysis is limited to patients undergoing on-pump surgery, compared to previous reviews [7, 24]. AKI remains one of the most frequent and prognostically severe complications following cardiac surgery. Recent data demonstrate that even in standardized elective procedures, the incidence of postoperative AKI remains significantly associated with chronic kidney injury, increased overall morbidity, and prolonged ICU stay [36, 37]. The physiological disruptions inherent to the CPB circuit play a central role in this pathogenesis [36, 38]. The renal medulla is highly sensitive to hypoxia, making the kidney vulnerable to the combination of hemodilution, hypoperfusion, ischemia-reperfusion, and the profound oxidative stress generated during extracorporeal circulation [36, 38]. Because the systemic inflammatory response to the surgical insult often confounds the early biomarker prediction of AKI [37], optimizing modifiable intraoperative parameters, such as arterial oxygen tension, remains a critical area of investigation to mitigate this serious complication. A primary physiological argument against hyperoxia has been the risk of renal vasoconstriction. Hyperoxia is known to reduce renal blood flow by directly constricting afferent arterioles, thereby exacerbating the renal insult inherent to CPB [39, 40]. However, our meta-analysis yielded a null finding regarding AKI. Two mechanisms may explain why this physiological risk did not translate into clinical AKI. First, the vasoconstrictive effect of hyperoxia may be transient or clinically insignificant in the face of the profound physiological alterations associated with CPB, such as hemodilution, non-pulsatile flow, and the systemic inflammatory response, which can overwhelm subtle changes in vascular tone [41]. Second, the definition of normoxia in some included trials may still represent a hyperoxic state relative to normal physiology. Furthermore, a significant concern related to hyperoxic CPB is the generation of ROS and the resulting oxidative stress. While several included trials demonstrated that hyperoxia significantly increased oxidative stress biomarkers, this biochemical stress did not translate into increased clinical morbidity in our pooled analysis. This conflict highlights a potential dissociation between surrogate markers of oxidative stress and patient-centered outcomes. It is possible that the elevated oxidative markers do not exceed the threshold required for irreversible tissue dysfunction, or that the safety margin conferred by hyperoxia against hypoxic microembolic events counterbalances ROS-mediated damage [42]. Additionally, the lack of any differences in postoperative arrhythmias suggests that the heart can tolerate hyperoxic CPB. This tolerance is likely a result of the combined robust myocardial protection afforded by cardioplegia and hypothermia, which reduces metabolic demands and the vulnerability to ROS damage during aortic clamping [43]. Beyond primary clinical endpoints, hospital resource utilization can serve as a highly sensitive, cumulative marker for overall perioperative morbidity [44]. Our analysis demonstrated no statistically significant differences in these secondary outcomes between the normoxic and hyperoxic groups. Still, these findings should be interpreted with caution due to substantial statistical heterogeneity and sensitivity to single-study effects in leave-one-out analyses. This heterogeneity likely reflects differences in perioperative care protocols, discharge criteria, and patient populations across studies. Additionally, given the wide confidence intervals, we cannot definitively conclude that either strategy imposes an excess burden of minor or subclinical complications. Similarly, regarding overall safety, while we did not detect a statistically significant difference in complication rates, this lack of significance should not be equated with definitive proof of equivalent safety. For instance, the World Health Organization (WHO) has previously recommended high FiO2 during surgery to prevent SSIs [45]. However, our meta-analysis did not find a significant reduction in SSIs with hyperoxia in cardiac surgery, though the imprecision of the estimate leaves the possibility of a clinically relevant effect unresolved. This implies that while hyperoxia may be beneficial in general abdominal surgery, it is not the case for infection prevention in the clean, high-antibiotic-coverage setting of cardiac surgery. Conversely, while generally safe, the normoxic strategy carries a risk of inadvertent desaturation. Wiberg et al. noted that approximately 10% of patients in the normoxic group required rescue oxygen due to desaturation (SpO2 < 92%) [16]. This highlights a practical consideration regarding oxygen management: while normoxic strategies require routine monitoring to prevent inadvertent desaturation, hyperoxic strategies, while providing a higher oxygen reserve, carry the theoretical risks of ROS generation and oxidative stress [46]. In comparison to earlier meta-analyses [7, 24, 25], our results provide a novel and highly powered perspective on oxygen management during CPB. In general, our results align with prior reviews indicating a lack of significant benefit associated with either strategy; however, our analysis expands upon these results by incorporating the most recently published, large-scale RCTs (the GLORIOUS and ROCS trials [16, 22]), and by strictly limiting our pooled analysis to patients undergoing on-pump surgery. Additionally, a recently presented meta-analysis by Kasbati et al. [47] evaluated 18 studies comparing hyperoxia to normoxia in adult cardiac surgery. Consistent with our results, they reported no significant differences in the incidence of postoperative arrhythmias, in-hospital mortality, or major clinical complications. While Kasbati et al. [47] provide a valuable snapshot, their work was presented as a conference abstract, thereby limiting access to their full methodological assessment. Our manuscript addresses these gaps by providing rigorous GRADE certainty-of-evidence assessments and comprehensive analyses to evaluate the stability of our clinical and hospital utilization outcomes. Also, a recent meta-analysis by Grillo et al. compared intraoperative hyperoxia with normoxia in cardiac surgery and similarly found no significant differences in mortality or major clinical outcomes [24]. However, the present study extends prior work in several important ways. First, we restricted our analysis to RCTs, thereby minimizing confounding inherent to observational designs. Second, we incorporated recently published large-scale trials, including the GLORIOUS and ROCS studies, which substantially increase the available evidence base. Third, we provide a comprehensive assessment of the certainty of evidence using the GRADE framework, allowing a more careful interpretation of the findings. Together, these factors position our analysis as an updated and methodologically rigorous synthesis of the current evidence. Strengths & limitations The main strength of this review is the inclusion of 12 RCTs that adhere to strict PRISMA and Cochrane standards, and the incorporation of the most recent large, low-bias trials, which provide the gold-standard evidence in this regard so far. However, some limitations persist. First, there is significant heterogeneity in the oxygenation thresholds; for example, a PaO2 of 150 mmHg was considered normoxia in some trials but hyperoxia in others. Thus, as demonstrated by the achieved PaO₂ values in Table 1, comparisons across several included trials may reflect moderate rather than high hyperoxia, rather than a true normoxia versus hyperoxia comparison. This variability likely dilutes potential treatment effects and makes it difficult to assess a specific safety threshold. Although subgroup or sensitivity analyses based on achieved PaO₂ intensity or stricter exposure thresholds would be helpful, this was not feasible due to inconsistent reporting of PaO₂ values, variability in measurement timing, and the lack of standardized summary data across trials. Second, clinical heterogeneity was present regarding CPB management (temperature, cardioplegia type), intervention timing (pre-CPB, during CPB, or post-CPB), and patient comorbidities. While some studies applied the intervention primarily during CPB, others extended oxygenation strategies to the entire intraoperative period or into the early post-CPB phase. Given that the physiological effects of oxygen may differ between ischemic, reperfusion, and postoperative phases, this variability may have attenuated potential differences between treatment groups. However, this approach reflects real-world clinical practice, in which oxygen therapy is typically managed as a continuous perioperative strategy rather than restricted to a single operative phase. Third, most included patients underwent elective CABG or valve surgery, leaving a data gap regarding emergency procedures, deep hypothermic circulatory arrest, or complex aortic surgery, where ischemia-reperfusion injury is more severe. Also, hard clinical outcomes can be influenced by baseline patient comorbidities and surgical procedure. However, we could not adjust our analysis based on that, as we relied on aggregate study-level data. Similarly, the lack of IPD and the low overall incidence of mortality events precluded the ability to perform robust subgroup analyses stratified by the specific type of surgical procedure. Fourth, most trials focused on short-term outcomes; data on long-term survival and neurocognitive outcomes remain scarce. Fifth, some studies were judged to have concerns or a high risk of bias, especially regarding lack of blinding and allocation concealment, which are common challenges in intraoperative trials. These limitations contributed to the overall low certainty of evidence and may have influenced the observed effect estimates. Therefore, the findings should be interpreted with caution. Sixth, a considerable heterogeneity in the definition of AKI was noticed across the included trials. While several studies used standardized criteria, such as KDIGO, others relied on non-uniform definitions of postoperative renal dysfunction. It is important to note that these results cannot be interchanged without consideration, as the occurrence of renal failure requiring dialysis indicates a more critical stage of kidney damage than typically defined by AKI. This variability may have introduced outcome misclassification and limited the interpretability of pooled estimates. Collectively, the previous limitations rendered our GRADE assessment of most outcomes to be low or very low certainty; thus, our results should be interpreted with caution. Implications for research and clinical practice Clinically, the absence of clear superiority for either the normoxic or hyperoxic strategy, combined with the low to very low certainty of the evidence, underscores imprecision rather than definitive equivalence. Because the current data cannot confidently exclude clinically important differences, they do not strongly support establishing a single oxygenation target. Consequently, the choice of strategy may currently depend on institutional protocols and individual patient risk profiles, while ensuring appropriate standard monitoring is maintained [48]. Consequently, resources may be better allocated to optimizing other perfusion parameters, such as flow rates and hematocrit, rather than aggressively titrating oxygen. For future research, applying oxygenation thresholds to all patients may appear outdated. Future trials should prioritize precision perfusion, exploring whether specific subpopulations, such as those with pre-existing renal insufficiency or undergoing deep hypothermic circulatory arrest, might show a differential susceptibility to hyperoxic injury [22]. Additionally, the discrepancy observed in some trials between elevated oxidative biomarkers and clinical outcomes underscores a critical knowledge gap [19, 20, 22]. Subsequent trials should incorporate real-time tissue oximetry (rSO2) and novel biomarkers of endothelial glycocalyx shedding to identify the specific physiological thresholds at which oxidative stress transitions from a biochemical phenomenon to irreversible tissue damage, thereby defining the actual upper limit of safe oxygenation. Additionally, there is a critical need for well-designed RCTs that move beyond broad FiO₂ adjustments and instead utilize tightly regulated, goal-directed PaO₂ regimens. To answer the physiological questions surrounding hyperoxemic injury during CPB, future trials should incorporate stricter, highly sensitive physiological and surrogate endpoints, such as real-time microcirculatory monitoring, and specific biomarkers of oxidative stress & ischemia-reperfusion injury, rather than relying only on hard clinical outcomes that require large sample sizes. Finally, future trials should report postoperative oxygenation parameters, such as the incidence of postoperative hypoxemia and the need for rescue oxygen, as these direct metrics remain underreported. Conclusion In patients undergoing cardiac surgery on CPB, our meta-analysis detected no statistically significant differences in clinical outcomes between hyperoxic and normoxic strategies. However, these findings should not be interpreted as evidence of equivalence. Despite a large, pooled sample size, the certainty of evidence remains low to very low, driven by substantial heterogeneity in oxygenation thresholds and wide confidence intervals that cannot exclude clinically meaningful benefit or harm. This underscores a persistent state of clinical uncertainty, warranting further research to identify an optimal oxygenation threshold and to determine whether a personalized approach is more beneficial for high-risk patients. Supplementary Information Supplementary Material 1. Supplementary Material 1.