Mortality and morbidity in acutely ill adults treated with liberal versus conservative oxygen therapy (IOTA): a systematic review and meta-analysis

Chu, The Lancet April 28, 2018 Vol 391 1693-705

Clinical Question

  • In patients with sepsis, critical illness, stroke, trauma, myocardial infarction, cardiac arrest and those who had emergency surgery does a liberal oxygen strategy compared with a conservative oxygen strategy increase hospital mortality?


  • Oxygen therapy has traditionally been thought of as a harmless adjunct to patient care and is often given without much thought. Hypoxemia is dangerous but by the same token hyperoxia can cause absorption atelectasis, acute lung injury, central nervous system toxicity, reduced cardiac output and cerebral and coronary vasoconstriction
  • In the pursuit to, above all, do no harm, it is important to question what is the optimal amount of oxygen our intensive care patients should receive
  • AVOID demonstrated no benefit in giving oxygen to normoxic patients with STEMI undergoing PCI, DETO2X showed no difference in one year mortality between room air and oxygen therapy in normoxic patients with suspected AMI, OXYGEN-ICU found a striking mortality benefit in those critically ill patients given a conservative oxygen regimen and several stroke trials have suggested a conservative oxygen strategy is non-inferior (the largest being the SO2S trial) to the liberal strategy
  • The ICU-ROX trial has just completed recruiting and results will be reported early next year. This 1000 patient trial of mechanically ventilated ICU patients will compare the effect of giving the smallest amount of inspired oxygen versus standard care on Ventilator Free Days


  • Systematic review and meta-analysis
  • 2 authors conducted a comprehensive search of databases (Cochrane, MEDLINE, Embase, HealthSTAR, LILACS, PapersFirst, WHO trials registry) + reference lists and author contact for missing data
  • Standardised data collection form
  • Intention-to-treat analysis
  • Morbidity was assessed using a random-effects meta-analysis allowing for differences in the treatment effect from study to study (differences in study population, intervention received, follow-up length etc.)
  • Results were re-analysed using various statistical techniques, for example, assuming a worse-case scenario for missing participants, reweighing trials using fixed-effect meta-analysis, excluding unpublished trials
  • The quality of the included trials was assessed using the Cochrane Risk of Bias assessment tool in an attempt to avoid Type I (false-positive) and Type II (false-negative) errors.

Trials Included

  • 25 randomised controlled trials, including 16,037 patients
    • Stroke patients: n=7,795
    • Myocardial infarct/cardiac arrest: n=6,306
    • Critical care: n=582
    • Emergency surgery: n=338
    • Sepsis: n=50
  • 12 trials (n=13,389) excluded patients with hypoxaemia at baseline, whereas all other trials excluded patients if baseline hypoxaemia was severe i.e P/F ratio <100
  • Median follow-up: 3 months (IQR 2-6 months)
  • Risk of bias
    • 18 trials deemed low risk
    • 7 trials deemed high risk


  • Inclusion: acutely ill adults requiring hospital admission who were randomised to liberal or conservative oxygen strategies and reported outcome of interest (mortality, disability, hospital acquired infections/pneumonia, hospital length of stay)
  • Exclusion: <18 years, pregnant, chronic respiratory disease trials, psychiatric disease, patients on ECMO, patients with hyperbaric oxygen therapy or elective surgery. Studies solely comparing different oxygen delivery devices
  • Baseline characteristics
    • Median age: 64
    • Male: 64%
    • Baseline SpO2 (10 studies): 96.4% in liberal group vs. 96.7% in conservative


  • Liberal oxygen
    • The treatment arm with the higher oxygen target, measured by FiO2, PaO2, SaO2 or SpO2
    • Median FiO2 administered 0.52 (IQR 0.39-0.85) for median duration of 8 h (IQR 4-24)


  • Conservative oxygen
    • The group with the lower oxygen target (including room air)
    • Median FiO2 administered 0.21 (IQR 0.21-0.25)

Management common to both groups

  • Room air or oxygen was delivered by:
    • Nasal prongs in 4 trials
    • Facemask in 13 trials
    • Invasive ventilation in 8 trials
  • These pragmatic trials meant other management was not prescribed


  • Mortality – significantly increased in liberal oxygen group
    • Comparing liberal vs. conservative groups
      • In hospital mortality (available from 19 trials, n=15,071)
        •  3.7% vs. 3.0%, RR = 1.21 (95% C.I. 1.03-1.43), p=0.020, high quality evidence
      • 30 day mortality (available from 14 trials, n=15,053)
        • 6.4% vs. 5.6%, RR = 1.14 (95% C.I. 1.01-1.28), p=0.033, high quality evidence
      • Mortality at longest available follow-up (23 trials, n=15,754)
        • 10.5% vs. 9.5%, RR = 1.1 (95% C.I. 1.00-1.20), p=0.044, high quality evidence
    • Meta-regression demonstrated that as SpO2 increased, liberal oxygen was associated with higher RR in-hospital mortality (14 trials) and higher RR of mortality at longest follow up (15 trials)
      • For every 1% increase in SpO2, the relative risk of
        • In-hospital mortality increased by 25%
        • Mortality at longest follow-up increased by 17%.
    • Mean number needed to harm using a liberal oxygen strategy was 71
    • There was low hetereogeneity for the assessment of death with I2=0
  • Disability after stroke or traumatic brain injury (5 trials) – no significant difference
    • Proportion of patients with mRS >2: RR 1 (95% C.I. 0.92-1.09)
  • Hospital acquired infections
    • In medical patients (7 trials) – no significant difference
      • RR 1.04 (95% C.I. 0.94-1.16)
    • Following emergency surgery (2 trials) – significantly reduced in liberal group
      • RR 0.50 (95% C.I. 0.36-0.69)
      • Trial sequential analysis showed that the required information size was not reached to conclusively determine the effect of the intervention
  • Length of hospital stay (12 studies) – no significant difference
    • Mean difference -0.25 days (95% C.I. -0.68 to 0.18)
  • The hetereogeneity for outcomes other than mortality, were more variable with I2 ranging from 58-67% making the conclusions difficult to assess
  • Subgroup analysis
    • No significant difference based on:
      • ICU vs. non-ICU setting
      • Invasive ventilation vs. non-invasive ventilation

Authors’ Conclusions

  • This systematic review and meta-analysis provides evidence that hyperoxia caused by a liberal oxygen strategy increases mortality


  • The analysis asked a focused clinical question
  • The criteria of the included articles was of a high standard
  • It is unlikely that important studies have been missed because of the comprehensive database trawl
  • The validity of the trials was praised with 2 out of 25 excluded: 1 as it duplicated data from another trial (but at a different timepoint) and 1 as patients could be randomised at different times to different treatment groups
  • The results were similar from study to study, although, alone none of the trials were statistically significant. It is only with the meta-analysis that significance is reached
  • The result is biologically plausible and has a dose response such that higher SpO2 is associated with higher in-hospital and longest follow-up mortality
  • The outcome assessed is clinically important


  • Wide variety of patients included and not all were ICU patients
  • Wide variety in intervention and control, with FiO2 in the liberal arm at 100% in some cases
  • The relative weightings placed on each trial is determined by the number of events (in this case deaths): in particular the OXYGEN-ICU trial (Giardis et al) is given 32.1% weight (478 patients) vs say the DETO2X trial at 16.8% (6629 patients). The OXYGEN-ICU trial was imperfect: single-centre, early stopping.
  • The heterogeneity of the secondary outcomes makes it difficult to establish any conclusion regarding these measures

The Bottom Line

  • This systematic review and meta-analysis adds to the growing body of literature that hyperoxia is dangerous. I will continue to aim for the lowest safest inspired oxygen in my ICU patients and encourage others to titrate FiO2 down if SpO2 >95%.
  • I eagerly await the results of the ICU-ROX trial which will be the largest trial of liberal versus restricted O2 in ICU mechanically ventilated patients.

External Links


Summary author: Celia Bradford @celiabradford
Summary date: June 6 2018
Peer-review editor: @davidslessor


  • Timo

    A) Mortality – significantly increased in conservative oxygen group
    B) Meta-regression demonstrated that as SpO2 increased, liberal oxygen was associated with higher RR in-hospital mortality (14 trials) and higher RR of mortality at longest follow up (15 trials)

    Liberal oxygen therapy increases mortality.

    Am I missing something?

    • David Slessor David Slessor

      thanks for spotting my mistake! I have now changed it. The correct conclusion is liberal oxygen therapy increases mortality!

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.