Hypoxaemia and respiratory failure

• Hypoxaemia means an abnormally low partial pressure of oxygen in arterial blood. It is the final common result of five mechanisms: low inspired oxygen, hypoventilation, impaired diffusion, ventilation-perfusion (V/Q) mismatch, and shunt.
• Type 1 respiratory failure is hypoxaemia with a normal or low carbon dioxide (PaCO2). Type 2 respiratory failure is hypoxaemia with raised PaCO2, reflecting inadequate alveolar ventilation.
• The alveolar gas equation converts the inspired oxygen tension into an alveolar oxygen tension (PAO2), and the A-a gradient (PAO2 minus PaO2) tells you whether hypoxaemia comes from a lung problem or from pure hypoventilation.
• Oxygen must be prescribed like a drug: most stable patients need an SpO2 of about 92% to 96%, while people at risk of retaining carbon dioxide (for example with COPD) need 88% to 92%.
• Shunt does not improve much with supplemental oxygen; in contrast, V/Q mismatch, diffusion limitation, hypoventilation and low inspired oxygen tension usually do. A shunt fraction above 50% makes PaO2 essentially independent of the inspired oxygen concentration.

By the end of this page you should be able to:

1. Define hypoxaemia and distinguish type 1 from type 2 respiratory failure, using PaO2, SaO2, PaCO2 and pH.
2. Explain the five mechanisms of hypoxaemia from cause to blood gas pattern, including the alveolar gas equation and the A-a gradient.
3. Interpret a peripheral oxygen saturation and an arterial blood gas, and state the limits of pulse oximetry in carbon monoxide exposure and darker skin tones.
4. Choose and adjust oxygen delivery devices, high-flow nasal oxygen, noninvasive ventilation and intubation in hypoxaemic and hypercapnic respiratory failure.
5. Recognise chronic hypoxaemic respiratory failure and the thresholds used to prescribe long-term oxygen therapy.
6. Outline pragmatic steps for oxygen and pulse oximetry where resources are limited.

Respiratory failure is the failure of the lungs to carry out one or both of their two jobs: taking oxygen up and eliminating carbon dioxide. It is usually split by the gas that is deranged.

• Type 1 (hypoxaemic) respiratory failure. PaO2 under 60 mmHg or SaO2 under 88%, with a normal or low PaCO2. ^[1] The problem is oxygen transfer, not carbon dioxide clearance.
• Type 2 (hypercapnic or ventilatory) respiratory failure. PaCO2 at or above 45 mmHg with an acidemic pH below 7.35 in the acute form. ^[1] The problem is inadequate alveolar ventilation. If the process is chronic, the PaCO2 may be raised while the pH is normal or near normal because the kidneys retain bicarbonate. ^[1]
• PaO2/FiO2 (P/F ratio). Dividing the arterial oxygen tension by the fraction of inspired oxygen (FiO2, as a decimal) removes the effect of how much oxygen is being given and allows comparison between patients and across time. It is the index used to grade hypoxaemia in conditions such as acute respiratory distress syndrome; a lower value means worse oxygenation. ^[1]
• Mixed respiratory failure. Hypoxaemia and hypercapnia can occur together, for example in severe chronic obstructive pulmonary disease (COPD), because the mechanisms overlap.

Data on how often respiratory failure occurs are sparse and depend heavily on the setting; the proportion of intensive care patients with hypoxaemic failure varies between studies and countries. What is consistent everywhere is that hypoxaemia is an independent marker of serious underlying disease and a key reason for admission, escalation of care and death from lung, cardiac and septic illness. ^[2][4]

The cause of hypoxaemia is best understood through its mechanism rather than through a list of diagnoses. The five mechanisms below are the underlying biology; the diseases are simply where they occur. ^[1]

Oxygen in the alveoli is not the same as oxygen in the air outside. Some alveolar oxygen is displaced by water vapour and by carbon dioxide, which shares the alveolar space. The alveolar gas equation converts inspired conditions into alveolar oxygen tension (PAO2):

PAO2 = FiO2 x (PB - PH2O) - (PaCO2 / R)

PB is barometric pressure, PH2O is water vapour pressure in the airways at body temperature, and R is the respiratory quotient. The exact constants vary with altitude and temperature, but the principle is that alveolar oxygen falls as alveolar carbon dioxide rises. ^[1] If inspired oxygen falls, as at altitude or on an aeroplane, PAO2 falls in direct proportion.

The A-a gradient is the difference between the PAO2 (calculated) and the PaO2 (measured from arterial blood): PAO2 - PaO2. It tells you whether oxygen is crossing the alveolar membrane normally.

• A normal A-a gradient with hypoxaemia points to pure hypoventilation, or to low inspired oxygen, because the lung membrane itself is working; simply increasing minute ventilation or raising the inspired oxygen fixes it. ^[1]
• A widened A-a gradient with hypoxaemia points to a lung problem: V/Q mismatch, shunt or diffusion limitation. ^[1]

A small A-a difference is normal because the lung is not a perfect uniform exchanger. The acceptable value rises with age as lung units become less uniformly matched; Lagina and Valley (2024) provide the age-adjustment equation in their figure 3. ^[1]

Giving 100% oxygen drives PAO2 to several hundred mmHg. Whether the blood oxygen responds tells you which mechanism is operating:

• Low inspired oxygen, hypoventilation and diffusion limitation: PaO2 rises substantially, often to normal. ^[1]
• V/Q mismatch: PaO2 rises, usually markedly, because the well-ventilated units receive enough oxygen to correct the average. ^[1]
• Shunt: PaO2 barely rises, because the shunted blood never meets alveolar oxygen at all. The rise is disappointing even with 100% oxygen. ^[1]

This response is the physiological basis of both the clinical "oxygen challenge test" and the distinction drawn in the table below.

A sixth pattern, oxygen delivery-consumption mismatch, is not a problem of oxygen uptake at all but of supply falling behind demand (anaemia, low cardiac output, hypermetabolic states). It widens the A-a gradient and is treated by correcting the underlying balance. ^[1]

Key points of reasoning:

• V/Q mismatch is the most common cause of hypoxaemia in critically ill patients. ^[1]
• Shunt is the reason severe hypoxaemia can be refractory. As the shunt fraction rises past 50%, PaO2 stops responding to higher FiO2 and supplemental oxygen alone is no longer the answer. ^[1]
• Hypoventilation always raises PaCO2 along with lowering PAO2, and oxygen improves the hypoxaemia but does not fix the hypercapnia.

Breathlessness is the symptom patients report, but it is non-specific and can be disproportionate to the degree of hypoxaemia. Important questions include how suddenly it began, whether there is cough, fever, chest pain, wheeze, ankle swelling, weight loss, pregnancy, recent travel to altitude, drug or alcohol use, sleep disruption, and any neuromuscular symptoms such as weakness. A rapid rise over hours suggests an acute intrapulmonary or cardiac process; a slowly progressive course suggests chronic lung disease decompensating.

• Cyanosis is a late and unreliable sign. It depends on the amount of deoxygenated haemoglobin in the blood rather than on saturation, and it is hard to see in anaemia or in pigmented skin.
• Tachypnoea, use of accessory muscles, and paradoxical abdominal movement point to increased work of breathing.
• Confusion, drowsiness or a fall in Glasgow Coma Scale can be the first sign of severe hypoxaemia or hypercapnia, particularly in older adults.
• Neck vein distension, a raised jugular venous pressure, and peripheral oedema suggest cor pulmonale or fluid overload.
• Chest signs (crackles, wheeze, bronchial breath sounds, dullness to percussion) give clues to the underlying mechanism.

Pulse oximetry is the standard bedside screen because it is continuous and noninvasive, but it has substantial limits:

• It cannot detect hypercapnia or hypoventilation. A normal SpO2 can coexist with a rising PaCO2, so patients on supplemental oxygen need capnography or a blood gas if ventilation is a concern. ^[3]
• It becomes less reliable at low saturation (below about 80%) and is not designed for the severe range. ^[3]
• Carbon monoxide binds haemoglobin with a spectrum similar to oxyhaemoglobin, so standard two-wavelength oximetry reads falsely high or normal in carbon monoxide poisoning. Co-oximetry is required. ^[3]
• Methaemoglobin, intravenous dyes and severe motion or low perfusion can all produce misleading values. ^[3]
• Darker skin pigmentation can cause overestimation of oxygenation, particularly during hypoxaemia; occult hypoxaemia (a normal SpO2 with an arterial saturation below 88%) was more common in Black than in White patients in a pooled analysis. ^[1][3]

In remote or low-resource settings, pulse oximetry may be the only available oxygen monitor and is still extremely valuable; where it is absent or the signal is unreliable, clinical judgement and, when possible, arterial blood gas sampling are essential.

The differential is driven by the mechanism suggested by the A-a gradient and the clinical picture:

• Normal A-a gradient with hypoxaemia: low inspired oxygen (altitude), hypoventilation (drug overdose, stroke, neuromuscular or chest wall disease). ^[1]
• Wide A-a gradient, responsive to oxygen: V/Q mismatch (pneumonia, oedema, atelectasis, asthma, COPD), diffusion limitation. ^[1]
• Wide A-a gradient, refractory to oxygen: shunt (ARDS, large shunt defects, massive atelectasis). ^[1]
• Mixed hypoxaemic and hypercapnic failure: severe COPD, obesity hypoventilation, neuromuscular weakness. ^[1]

Cardiogenic pulmonary oedema and diffuse alveolar haemorrhage can look identical at the bedside and are separated by cardiac imaging and bronchoscopy.

• Pulse oximetry (SpO2). Continuous, noninvasive and essential, but see the limits above. ^[1][3]
• Arterial blood gas (ABG). The gold standard for grading hypoxaemia and hypercapnia. ^[1][2] It gives PaO2, SaO2, PaCO2, pH and bicarbonate in one sample. Samples should be taken at rest and interpreted after a change in position, FiO2 or flow rate has settled.
• Venous blood gas (VBG). Useful for pH, lactate and electrolytes and has evidence for estimating PaCO2 in stable patients, but it does not give an accurate PaO2, so it should not replace an ABG when hypoxaemic failure is suspected. ^[2]
• Capillary (earlobe or fingertip) blood gas. A reasonable alternative when arterial puncture is difficult; it reflects pH and PaCO2 well but tends to underestimate PaO2. ^[2]
• Alveolar-arterial gradient calculation. Calculated from the ABG as PAO2 - PaO2 using the alveolar gas equation above. ^[1]

• Chest radiograph to look for pneumothorax, alveolar filling, interstitial change or effusion.
• CT of the chest when the diagnosis remains unclear, particularly in immunocompromised patients, suspected fungal or interstitial disease, or to evaluate for shunt (CT angiography) or pulmonary embolism.
• Echocardiography to exclude cardiogenic pulmonary oedema and to look for intracardiac shunt.
• Laboratory tests guided by the suspected mechanism (microbiology, cardiac biomarkers, haemoglobin, renal and hepatic function).

There is no universal staging system for hypoxaemia alone; the relevant cut-offs are physiological: ^[1]

• Hypoxaemic respiratory failure: PaO2 under 60 mmHg or SaO2 under 88%. ^[1]
• Hypercapnic respiratory failure: PaCO2 at least 45 mmHg with pH below 7.35 (acute), or raised PaCO2 with near-normal pH (chronic). ^[1]
• ARDS and similar research definitions additionally use the PaO2/FiO2 ratio with positive end-expiratory pressure (PEEP) thresholds, but that is a syndrome-specific scale rather than a general measure of hypoxaemia.
• Pulse oximetry can approximate arterial saturation, but a normal reading does not exclude significant hypoxaemia, especially at lower saturations and in darker skin. ^[3]

• In acute hypercapnia, pH falls quickly; in chronic hypercapnia, renal compensation has had days to act.
• Once PaCO2 rises above about 75 mmHg in a normally eucapnic patient, mentation becomes depressed and respiratory drive falls, which worsens the problem. ^[1]

• Give oxygen to all hypoxaemic patients, titrated to the appropriate target, while treating the underlying cause in parallel. ^[1]
• Do not treat breathlessness with oxygen if the patient is not hypoxaemic. There is no evidence of benefit and oxygen is a drug with hazards.
• Assess airway, breathing and circulation together, including the ability to protect the airway, work of breathing and haemodynamic status. ^[1]

Sources differ, and the correct target depends on the risk of hypercapnia:

• Standard (low risk of hypercapnia): BTS and AARC guidance recommend SpO2 94% to 98%; TSANZ and the German National S3 guideline recommend 92% to 96%. ^[2]
• High risk of hypercapnia (COPD, obesity hypoventilation, neuromuscular and chest wall disease): consensus target is SpO2 88% to 92%. ^[1][2]
• Conservative (recent ICU literature): PaO2 55 to 80 mmHg, or SpO2 90% to 98%, is likely safe, and there are signals of harm from prolonged hyperoxaemia. ^[1]
• Darker skin tone: where SaO2 or PaO2 cannot be measured, consider the more liberal end of the range (about 92% to 98%). ^[1]
• Resuscitation: WHO guidance in COVID-19 was greater than 94% during resuscitation, then greater than 90% once stable. ^[2]
• In ARDS-style care, avoid prolonged hyperoxia even when a higher target is used.

These guidelines genuinely differ, and the right choice is the one your local protocol and the patient's clinical situation support.

• Each additional litre per minute by nasal cannula raises the delivered FiO2 by about 4% above room air (21%). ^[1]
• When minute ventilation exceeds device flow, the delivered oxygen is diluted by room air; this is why conventional devices fail in severe hypoxaemia. ^[1]
• Start with the least invasive device that achieves the target, and escalate if the patient needs more than about 6 L/min by mask or about 40% FiO2 to stay in range. ^[1]

HFNO delivers warmed, humidified oxygen at flow rates up to 60 L/min. It improves washout of dead space, reduces work of breathing and tolerates higher flows than a mask. ^[1]

• Use HFNO for hypoxaemia above the capability of conventional oxygen, and for any patient with increased work of breathing or dyspnoea. ^[1]
• If work of breathing remains high despite correction of hypoxaemia, consider a trial of noninvasive ventilation first in appropriate patients. ^[1]

• First-line for hypercapnic failure in acute exacerbation of COPD and for cardiogenic pulmonary oedema, where NIV improves distress and reduces mortality. ^[1]
• Candidates must be breathing spontaneously, able to protect the airway and to remove the mask if they vomit. ^[1]
• In hypercapnic failure, reassess within four hours of starting NIV; proceed to intubation if pH does not correct, PaCO2 does not fall or the patient deteriorates. ^[1]
• In pure hypoxaemic failure without acidosis, the evidence for NIV is weaker and there is a concern about self-inflicted lung injury from large tidal volumes; HFNO is usually preferred. ^[1]
• Higher FiO2 with NIV or HFNO can worsen hypercapnia in COPD by removing hypoxic vasoconstriction, so keep the SpO2 at 88% to 92% in anyone at risk of retaining CO2. ^[1]

There is no standardised PaO2 or SpO2 threshold for intubation. The decision rests on whether noninvasive support is failing or the patient shows tissue hypoperfusion, haemodynamic instability or an inability to protect the airway. ^[1]

• HFNO for more than 48 hours before intubation has been associated with higher mortality in some observational data, so reassess frequently rather than leaving support unchanged. ^[1]
• Preoxygenate for at least three minutes using HFNO or NIV, which builds an alveolar oxygen reserve. ^[1]
• Apnoeic oxygenation with bag-valve mask ventilation during rapid sequence induction reduces desaturation and aspiration risk is not increased by it. ^[1]
• Consider video laryngoscopy as the primary method, and awake fibre-optic intubation when hypoxaemia would make apnoea dangerous. ^[1]

Chronic hypercapnic respiratory failure is defined by PaCO2 at or above 45 mmHg with a normal or near-normal pH, reflecting renal compensation over time. ^[1]

Long-term oxygen therapy (LTOT) has strong evidence in selected patients with chronic hypoxaemic COPD. The landmark trials that shaped modern practice (Nocturnal Oxygen Therapy Trial, 1980; MRC working party, 1981) enrolled patients with severe resting hypoxaemia and showed a survival benefit. ^[5]

Thresholds used for prescribing LTOT (from those trials, still reflected in most guidelines) include: ^[5]

• PaO2 at or below 55 mmHg on room air, or
• PaO2 56 to 59 mmHg with evidence of end-organ effect: haematocrit above 55%, peripheral oedema, or P pulmonale on the electrocardiogram.

The physiological idea is that continuous oxygen reduces hypoxic pulmonary vasoconstriction and the strain on the right ventricle, but the evidence that mortality tracks hypoxaemia (rather than comorbidity) has always been incomplete. Recent trials have not shown benefit from long-term oxygen for moderate, exertional or nocturnal hypoxaemia in COPD, nor a dose-response between 15 and 24 hours per day in severe hypoxaemia. ^[5] LTOT should therefore be prescribed to the patient who meets objective criteria, not to someone who is simply breathless.

Practical points:

• Confirm hypoxaemia by ABG on room air at rest on at least two occasions before committing.
• Lower flow rates given over the appropriate hours each day are effective; more is not better.
• Reassess at a clinically appropriate interval and consider withdrawal if oxygenation has recovered.
• Counsel about fire risk with open flames and smoking, and about tripping over tubing.

• Hyperoxia can produce free radical lung injury, absorption atelectasis and coronary or cerebral vasoconstriction; prolonged high oxygen is particularly risky in arterial disease and in neonates.
• Oxygen-induced hypercapnia in COPD through removal of hypoxic vasoconstriction and through the Haldane effect.
• Mask and nasal trauma with prolonged high flow: dry nose, sinus pain and skin breakdown.
• Ventilator-related complications when intubation is needed: barotrauma, infection, critical illness weakness.
• Occult hypoxaemia in darker skin, carbon monoxide poisoning or low-perfusion states, leading to a false sense of security.

There is no population screening for hypoxaemia. Prevention means identifying and treating the conditions that cause it early: pneumonia and sepsis, early recognition of drug overdose and stroke, safe oxygen prescribing, smoking cessation, and vaccines against influenza and pneumococcus.

• Pregnancy. Maternal oxygen targets are chosen to keep both mother and fetus well oxygenated; resuscitation guidance in pregnancy targets higher saturations early.
• Children. Lower thresholds of acceptable SpO2 apply in paediatrics and the causes differ (bronchiolitis, congenital heart disease); specialist paediatric oxygen protocols should be followed.
• Older adults. Hypoxaemia often presents as delirium or a fall rather than breathlessness, and respiratory drive may be blunted.
• Immunocompromised patients. Lung disease can be atypical and hypoxaemia may progress without warning; lower thresholds for imaging and arterial blood gas sampling are appropriate.

Oxygen is an essential medicine and is often the single most useful therapy available. ^[4]

• Oxygen concentrators extract oxygen from room air and are the preferred source in remote and low-resource hospitals and home care; they avoid the cost and transport problem of cylinders. ^[4] WHO and partners distributed more than 30,000 concentrators across 121 countries as of early 2021; 37 were classified as fragile settings. ^[4]
• Reliability. Concentrators need clean power, correct filter maintenance and technical support. In one Kenyan survey, 18 of 57 patients (32%) had their oxygen interrupted by power outages with a median duration of 11 minutes, contributing to a case fatality of 19% among those affected. Across 230 low-resource facilities, 43.8% had uninterrupted oxygen access and 24.8% had none. ^[4]
• Storage options. Solar-powered systems and low-pressure oxygen storage have shown promise in rural Africa to bridge power gaps, although upfront cost and maintenance support remain barriers. ^[4]
• Pulse oximetry is the key triage tool and should be available wherever oxygen is given, but remember its limits in dark skin, carbon monoxide and low perfusion; arterial blood gas testing or careful clinical assessment must substitute if the signal is unreliable. ^[3][4]
• Clinical practice. Treat the sickest patients first: use the lowest effective flow to reach the target, reserve high-flow for severe hypoxaemia and, where NIV is used, ensure staff training and close monitoring.

• Hypoxaemia is low oxygen in arterial blood; it is a sign of serious underlying disease, not a diagnosis in itself.
• Type 1 respiratory failure is hypoxaemia with normal or low PaCO2; type 2 is hypoxaemia with raised PaCO2 and acute acidaemia.
• The alveolar gas equation converts inspired oxygen into alveolar oxygen; the A-a gradient separates a lung problem from pure hypoventilation.
• Low inspired oxygen, hypoventilation, diffusion limitation, V/Q mismatch and shunt are the five mechanisms; V/Q mismatch is the commonest; shunt is the least responsive to oxygen.
• Cyanosis is late and unreliable; pulse oximetry is useful but is falsely reassuring in carbon monoxide poisoning, unreliable below about 80% and prone to overestimation in darker skin.
• Arterial blood gas remains the gold standard for oxygenation and ventilation.
• Oxygen targets differ: 92-96% or 94-98% for most patients not at risk of hypercapnia; 88-92% where hypercapnia is a risk; around 90-98% is likely safe in the critically ill.
• Escalate from conventional devices to high-flow nasal oxygen, then NIV (especially for COPD and cardiogenic oedema), then intubation when support is failing.
• Chronic respiratory failure with PaCO2 at least 45 mmHg and near-normal pH is compensated; long-term oxygen improves survival only in selected severely hypoxaemic patients.
• In low-resource settings, concentrators and pulse oximetry save lives, but only with reliable power, maintenance and trained staff.

Acute respiratory distress syndrome, asthma, bronchiectasis, carbon monoxide poisoning, chronic obstructive pulmonary disease, cystic fibrosis, interstitial lung disease, obesity hypoventilation syndrome, pneumonia, pulmonary embolism, sleep apnoea.