Oxygen therapy for acutely ill medical patients: a clinical practice guidelineBMJ 2018; 363 doi: https://doi.org/10.1136/bmj.k4169 (Published 24 October 2018) Cite this as: BMJ 2018;363:k4169
Overview of recommendations
Recommendation 1 - upper limit
Recommendation 2 - lower limit (90-92%)
Recommendation 3 - lower limit (>92%)
©BMJ Publishing Group Limited.
Disclaimer: This infographic is not a validated clinical decision aid. This information is provided without any representations, conditions or warranties that it is accurate or up to date. BMJ and its licensors assume no responsibility for any aspect of treatment administered with the aid of this information. Any reliance placed on this information is strictly at the user's own risk. For the full disclaimer wording see BMJ's terms and conditions: https://www.bmj.com/company/legal-information/
- Reed A C Siemieniuk, methods co-chair, general internist1,
- Derek K Chu, general internist2,
- Lisa Ha-Yeon Kim, clinical fellow2,
- Maria-Rosa Güell-Rous, senior pulmonology consultant3,
- Waleed Alhazzani, critical care clinician12,
- Paola M Soccal, pulmonologist45,
- Paul J Karanicolas, associate professor of surgery6,
- Pauline D Farhoumand, general internist7,
- Jillian L K Siemieniuk, registered nurse8,
- Imran Satia, respiratory physician2,
- Elvis M Irusen, professor of pulmonology and intensive care9,
- Marwan M Refaat, cardiologist10,
- J Stephen Mikita, patient partner11,
- Maureen Smith, patient partner12,
- Dian N Cohen, patient partner13,
- Per O Vandvik, general internist14,
- Thomas Agoritsas, general internist1715,
- Lyubov Lytvyn, patient partnership liaison1,
- Gordon H Guyatt, chair, distinguished professor12
- 1Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton L8S 4K1, Canada
- 2Department of Medicine, McMaster University, Hamilton L8S 4K1, Canada
- 3Departament de Pneumologia, Hospital de la Santa Creu I Sant Pau. Barcelona, Catalonia 08041, Spain
- 4Division of Pulmonary Medicine, Geneva University Hospitals, 1211 Geneva, Switzerland
- 5Faculty of Medicine, Geneva University, 1206 Geneva, Switzerland
- 6Department of Surgery, Sunnybrook Health Sciences Centre, University of Toronto, Toronto, Ontario M4N 3M5, Canada
- 7Division General Internal Medicine, University Hospitals of Geneva, 1205 Geneva, Switzerland
- 8Alberta Health Services, Calgary, Alberta T1Y 6J4, Canada
- 9Divisions of Pulmonology and Medical Intensive Care, Stellenbosch University, Cape Town 7505, South Africa
- 10Departments of Internal Medicine and Biochemistry & Molecular Genetics, American University of Beirut Faculty of Medicine and Medical Center, Beirut 1107 2020, Lebanon
- 11Salt Lake City, Utah 84106, USA
- 12Ottawa, Ontario K2P 1C8, Canada
- 13Hatley, Quebec J0B 4B0, Canada
- 14Institute of Health and Society, Faculty of Medicine, University of Oslo, Oslo, Norway
- 15Division Clinical Epidemiology, University Hospitals of Geneva, 1205 Geneva, 1205, Switzerland
- Correspondence: R A C Siemieniuk
What you need to know
It is a longstanding cultural norm to provide supplemental oxygen to sick patients regardless of their blood oxygen saturation
A recent systematic review and meta-analysis has shown that too much supplemental oxygen increases mortality for medical patients in hospital
For patients receiving oxygen therapy, aim for peripheral capillary oxygen saturation (SpO2) of ≤96% (strong recommendation)
For patients with acute myocardial infarction or stroke, do not initiate oxygen therapy in patients with SpO2 ≥90% (for ≥93% strong recommendation, for 90-92% weak recommendation)
A target SpO2 range of 90-94% seems reasonable for most patients and 88-92% for patients at risk of hypercapnic respiratory failure; use the minimum amount of oxygen necessary
What is the best way to use oxygen therapy for patients with an acute medical illness? A systematic review published in the Lancet in April 2018 found that supplemental oxygen in inpatients with normal oxygen saturation increases mortality.1 Its authors concluded that oxygen should be administered conservatively, but they did not make specific recommendations on how to do it. An international expert panel used that review to inform this guideline. It aims to promptly and transparently translate potentially practice-changing evidence to usable recommendations for clinicians and patients.2 The panel used the GRADE framework and following standards for trustworthy guidelines.3
The panel asked;
In acutely ill patients, when should oxygen therapy be started? (What is the lower limit of peripheral capillary oxygen saturation (SpO2)?)
In acutely ill patients receiving oxygen therapy, how much oxygen should be given? (What is the upper limit of SpO2?)
The panel makes a strong recommendation for maintaining an oxygen saturation of no more than 96% in acutely ill medical patients (upper limit). The panel did not make a recommendation on when to start (the lower limit) for all medical patients because there was not enough evidence. Instead, the panel suggests that patients with acute stroke or myocardial infarction and a SpO2 ≥90% not receive supplemental oxygen (a weak recommendation if SpO2 is 90-92% and a strong recommendation if 93-100%). Box 1 shows the article and evidence linked to this Rapid Recommendation. The infographic provides an overview of the key absolute benefits and harms, as well as the quality of evidence that informed each of the recommendations.
Linked resources in this BMJ Rapid Recommendations cluster
Siemieniuk RAC, Chu DK, Kim LH-Y, et al. Oxygen therapy for acutely ill medical patients: a clinical practice guideline. BMJ 2018;363:k4169
Summary of the results from the Rapid Recommendation process
Chu DK, Kim LH, Young PJ, et al. Mortality and morbidity in acutely ill adults treated with liberal versus conservative oxygen therapy (IOTA): a systematic review and meta-analysis. Lancet 2018;391:1693-705.
Review and meta-analysis of all available randomised trials that assessed oxygen therapy for acutely ill adults
Expanded version of the results with multilayered recommendations, evidence summaries, and decision aids for use on all devices
The panel was confident that the recommendation against letting oxygen saturation rise above 96% applies to almost all patients in hospital with a medical problem. The recommendation also applies to pre-hospital care. The evidence may apply to surgical and obstetric patients, but the panel did not review the evidence on postoperative healing and infections and therefore decided not to comment on these patients. Similarly, the panel did not review the evidence on oxygen therapy in neonates and infants.
Supplemental oxygen therapy is widely used in hospitals: 25% or more of patients who visit the emergency department receive oxygen.4 Clinicians often give oxygen to many patients presenting with stroke without hypoxaemia, and to almost all patients presenting with myocardial infarction.5 Until recently, many healthcare professionals believed that oxygen had little or no harm for acutely ill adults. In addition to mortality, other difficulties caused by oxygen can include nasal or throat irritation and hampered mobility. Doctors first used oxygen for medical purposes in the 19th century,6 and its use became routine in the early 20th century.7 Modern guidelines vary in their advice on when to give oxygen for acute medical conditions and how much to give (see table 1).
When to start oxygen—Peripheral capillary oxygen saturation (SpO2) thresholds typically trigger the use of oxygen treatment. Thresholds range from SpO2 <90% to <95% in guidelines. Recommendations for starting oxygen in specific groups vary: patients with stroke with SpO2 <95%,9 and, regardless of SpO2, those experiencing an acute myocardial infarction who feel breathless, are offered oxygen.11
When to stop oxygen—Many guidelines do not say how much is too much. Healthcare workers may respond to this advice by keeping a buffer between a patient’s SpO2 and the lower limit (for example, by keeping the SpO2 close to 100%). Some guidelines advocate targeting a SpO2 range. Proposed limits range from 98% for most patients, to an upper limit of 92% for patients with risk of hypercapnic respiratory failure, such as patients with chronic obstructive pulmonary disease.15
How this recommendation was created
Our international panel included methodologists, a respiratory therapist/technician, a nurse, patient partners who have been hospitalised for an acute medical condition, pulmonologists, intensivists, internists, an anaesthesiologist, a cardiologist, emergency physicians, and a surgeon (see appendix 1 on bmj.com for details of panel members). They decided on the scope of the recommendation and the outcomes most important to patients. The panel identified three key patient-important outcomes: mortality, hospital acquired infections, and length of hospitalisation. For two specific populations for which there was substantial randomised evidence available, the panel noted additional key outcomes: for patients with stroke, disability; and for patients with acute myocardial infarction, recurrent myocardial infarction, revascularisation, and chest pain.
The panel met to discuss the evidence and formulate a recommendation. No member had financial conflicts of interest; intellectual and professional conflicts were minimised and are transparently described (appendix 2 on bmj.com). The panel followed the BMJ Rapid Recommendations procedures for creating a trustworthy recommendation,2 including using the GRADE approach to critically appraise the evidence and create recommendations (appendix 3 on bmj.com).3 The panel considered the benefits, as well as any harms and burdens, of oxygen therapy, the certainty (quality) of the evidence for each outcome, typical and expected variations in patient values and preferences, acceptability, and feasibility.22 Within the GRADE framework, recommendations can be either strong or weak (also known as conditional), and for or against a specific course of action.23
The panel considered several key practical issues: psychological comfort from oxygen, discomfort (such as nasal irritation), and feasibility (such as impact on nursing resources). The panel was interested in knowing whether the impacts of oxygen were different in different medical conditions or study populations.
A recent systematic review and meta-analysis of randomised controlled trials of acutely ill adults quantified whether inpatients were at greater risk of death with liberal or conservative oxygen therapy.1 Patients randomised to liberal oxygen therapy were more likely to die (risk ratio 1.21 (95% confidence interval 1.03 to 1.43)). The increase in mortality was highest in the trials with the greatest increase in SpO2; this suggests a dose-response relation and strengthens the inference that excessive oxygen is a cause of death. The review included 25 randomised controlled trials. Figure 2 outlines key study and participant characteristics. This shows that the results apply to a wide variety of patient groups.
Upper limit of oxygen therapy
The panel had moderate certainty that oxygen increases mortality when the SpO2 is above 96%. Providing supplemental oxygen above a SpO2 of 96% probably increases mortality by around 1%. There is probably no difference in length of hospitalisation or risk of hospital acquired infections. Average (median) SpO2 was 96% in participants randomised to none or limited oxygen therapy. The evidence was rated down from high to moderate certainty for indirectness (uncertain applicability) because the trials used varying SpO2 thresholds, leaving some uncertainty regarding the value above which mortality increases.
Lower limit of oxygen therapy
The evidence regarding the lower limit comes from the patients who were included in the clinical trials with baseline SpO2 over 90%. The evidence in patients with initially higher SpO2 (>92%) is more certain because most patients in the trials had a baseline SpO2 above 92%. For example, in the largest of eight trials of patients with stroke only 240 patients (3.1% of 7677 participants) had an initial SpO2 of 90-93.9%.16 For myocardial infarction, six trials enrolled 7898 patients: in the largest trial, 1062 patients (16.0%) had an initial SpO2 ≤94%.17 For all outcomes, the panel rated down the quality of the evidence for indirectness (uncertain applicability) in patients with a SpO2 of 90-92%. Because trials informing the lower limit of when to start oxygen were restricted to patients with stroke and myocardial infarction, whether the evidence applies to patients without these conditions is uncertain.
The confidence intervals around the absolute effects in both stroke and myocardial infarction demonstrate that administering supplemental oxygen in patients with these conditions is unlikely to result in an important reduction in mortality. For stroke, supplemental oxygen probably does not reduce disability. In patients with acute myocardial infarction, supplemental oxygen probably does not reduce chest pain, recurrent myocardial infarction, or the need for a coronary revascularisation intervention.
Understanding the recommendations
The infographic summarises the benefits and harms of oxygen therapy.
Scope of recommendations
Our recommendations apply to critically ill or surgical patients with sepsis. They also apply to patients who are en route to hospital in an ambulance and to those who are hospitalised.
We did not consider patients with uncomplicated surgery. There is a separate body of evidence, mostly in the elective surgical setting.18 There is an unresolved debate about whether supplemental oxygen reduces the risk of surgical site infections. Our recommendations may not apply to young children (particularly neonates). There is a separate body of evidence and considerations such as necrotising enterocolitis and retinopathy of prematurity.19
Upper limit of oxygen therapy
The panel makes a strong recommendation that, if supplemental oxygen is administered, clinicians ensure a maximum SpO2 of 96%
This is because saturation above this level likely causes a small but important increased risk of death without plausible benefit. It is probable that the optimal upper SpO2 limit is lower than 96%, but exactly how much lower is unknown. Patients randomised to more liberal oxygen therapy typically achieved a SpO2 >96%. The data from the trials provide only limited support for any particular upper threshold, including the 96% chosen by the panel.
Lower limit of oxygen therapy
For patients with myocardial infarction or stroke, the panel makes a strong recommendation against initiating supplemental oxygen when the initial SpO2 is >92%
In patients with myocardial infarction or stroke, there are probably no benefits to initiating oxygen therapy when SpO2 is >92%, and it may cause harm.
The panel makes a weak recommendation against initiating oxygen in these patients with a SpO2 of 90-92%
There may not be any benefits for patients with this lower SpO2 (90-92%). Fewer patients with this SpO2 range at baseline were included in the trials, so the panel had less certainty in the results. There is no evidence of benefit from supplemental oxygen initiated in patients with myocardial infarction and stroke whose SpO2 is ≥90%, but there exists at least a modest risk of harm.
The panel did not issue recommendations for all patients or for other conditions because there were too few participants in the clinical trials who had a baseline SpO2 <95%.
Values and preferences
The panel believes that almost all patients would value avoiding even a small increased risk of death with supplemental oxygen. Although the panel viewed nasal and throat irritation and a decrease in mobility from oxygen therapy as unimportant, they felt that most patients would not choose to endure even a minor inconvenience if there is probably no benefit.
Figure 3 outlines the key practical issues about the use of oxygen therapy for patients.
A target SpO2 range of 90-94% seems wide enough to allow for normal fluctuation, and is likely low enough to avoid harm.
Upper thresholds for SpO2 in patients at risk of hypercapnic respiratory failure should be lower than for other patients (see box 2 for some common examples). Excessive oxygen could increase the risk of needing mechanical ventilation in these patients. Other existing evidence supports a target SpO2 of about 88-92% in such patients.20 Box 2 also shows a small number of acute illnesses with specific evidence to support more oxygen.
Examples of conditions that might benefit from higher or lower oxygen saturation thresholds
Lower target (such as SpO2 88-92%)
Patients at risk of hypercapnic respiratory failure, for example:
Chronic obstructive pulmonary disease
Neuromuscular respiratory diseases
Obstructive sleep apnoea
Decreased central respiratory drive (such as sedative overdose, stroke, encephalitis)
Higher target (such as SpO2 approaching 100%)
Carbon monoxide poisoning
Sickle cell crisis
Shared decision making
The patient panellists said that oxygen therapy is often given to patients with insufficient discussion and explanation. Clearer information may reduce anxiety and improve patient satisfaction in patients where oxygen is needed.
Costs and resources
Patients are unlikely to view the modest cost of oxygen as excessive, particularly in settings where they do not directly pay for their care.
A target SpO2 range (rather than a lower limit without an upper limit) will need closer monitoring by the healthcare team. Our recommendations do not consider healthcare payer considerations. We suggest a target SpO2 range that is sufficiently wide that it does not require excessive attention (such as 90-94%). Some patients will have wider SpO2 fluctuations and may therefore require a wider target range; these patients may also benefit from closer monitoring.
There were no robust data comparing supplemental oxygen to no oxygen in patients with a SpO2 <90%, so the impact of oxygen therapy in such patients is uncertain.
Addressing the following gaps in our knowledge may inform decision makers and future guideline recommendations:
Does supplemental oxygen provide benefit to patients experiencing a stroke or myocardial infarction with a SpO2 <92% (such as 85-92%)?
Is supplemental oxygen harmful in patients with medical conditions other than stroke or myocardial infarction with a SpO2 85-94%?
The reasons why excessive supplemental oxygen increases mortality are uncertain. Excessive oxygen can lead to reduced cardiac output, vasoconstriction, inflammation, and oxidative stress.21 In addition, excessive oxygen might lead to falsely reassuring SpO2 values and make it difficult to recognise when a patient’s condition worsens.
Education in practice
How do you use supplemental oxygen in medical patients?
Based on this article, how do you think your practice might change? Is there anything that you would say to your patient or do differently?
How might you share this information with your organisation or review local policies on oxygen targets?
How patients were involved in the creation of this article
Three people with lived experience of acute medical conditions requiring hospitalisation were members of the panel. They identified and rated outcomes, and helped lead the discussion on values and preferences in a videoconference and email discussions before the full panel meetings. They noted that patients are often underinformed about the reason for and implications of supplemental oxygen therapy.
Updates to this article
Table 2 shows evidence that has emerged since the publication of this article. As new evidence is published, a group will assess the new evidence and make a judgment on to what extent it is expected to alter the recommendation.
This BMJ Rapid Recommendation article is one of a series that provides clinicians with trustworthy recommendations for potentially practice changing evidence. BMJ Rapid Recommendations represent a collaborative effort between the MAGIC group (http://magicproject.org/) and The BMJ. A summary is offered here and the full version including decision aids is on the MAGICapp (https://app.magicapp.org), for all devices in multilayered formats. Those reading and using these recommendations should consider individual patient circumstances, and their values and preferences and may want to use consultation decision aids in MAGICapp to facilitate shared decision making with patients. We encourage adaptation and contextualisation of our recommendations to local or other contexts. Those considering use or adaptation of content may go to MAGICapp to link or extract its content or contact The BMJ for permission to reuse content in this article. Series adviser Rafael Perera-Salazar.
Contributors: All panel members participated in the teleconferences or email discussions and met all authorship criteria.
Competing interests: All authors have completed the BMJ Rapid Recommendations interests disclosure form, and a detailed description of all disclosures is reported in appendix 2 on bmj.com. As with all BMJ Rapid Recommendations, the executive team and The BMJ judged that no panel member had any financial conflict of interest. Professional and academic interests are minimised as much as possible, while maintaining necessary expertise on the panel to make fully informed decisions. DK Chu, LH-Y Kim, and W Alhazzani co-authored the systematic review that formed the evidence base for this guideline. RAC Siemieniuk, T Agoritsas, PO Vandvik, L Lytvyn, and GH Guyatt are members of the GRADE Working Group: BMJ Rapid Recommendations adheres to GRADE methods.
Funding: This guideline was not funded.
Transparency: RAC Siemieniuk affirms that the manuscript is an honest, accurate, and transparent account of the recommendation being reported; that no important aspects of the recommendation have been omitted; and that any discrepancies from the recommendation as planned (and, if relevant, registered) have been explained.
Provenance and peer review: Commissioned; externally peer reviewed