Respiratory failure is perhaps the single most common reason for the emergency admission of patients to Intensive Care. It can be defined as the failure of the respiratory system to deliver adequate oxygen to, and remove adequate carbon dioxide from, the arterial blood, to a degree that causes a threat to life.
Respiratory failure is commonly classified into two types:
Type 1, or ‘acute hypoxaemic’, failure is commonly seen acutely in diseases that involve damage to the lung tissue itself, e.g. pulmonary oedema, pneumonia, and fibrosing alveolitis. Type 2, or ‘ventilatory’, failure is seen when the alveolar ventilation is insufficient to allow the excretion of the volumes of CO2 that are being produced, e.g. chronic bronchitis or central depression.
The absolute levels of PaO2 and PaCO2 that define respiratory failure are not strictly defined, but PaO2 < 8 kPa and PaCO2 > 8 kPa are accepted widely.
The causes of both types of respiratory failure are legion. They can conveniently be grouped into four broad categories:
- • Hypoventilation
Neurological impairment, e.g. Guillian-Barré syndrome or myaesthenia gravis. High spinal injury. Overdoses, e.g. Opioids or benzodiazepines.
- • Diffusional impairment
Fibrotic lung disease. Cardiac failure.
- • Shunt
Consolidation. Collapse/atelectasis. Guillian-Barré.
- • V/Q mismatch
ARDS. Pulmonary embolus. Excessive PEEP. Abrupt fall in cardiac output.
In the first group, the problem lies outside the lungs, so there is often no impairment to gas exchange itself. This means that any ventilatory treatment will purely be supportive, while the extra-pulmonary source of the problem is addressed. The last two groups contain some of the most commonly encountered pathologies in the intensive care environment. They also contain a number of conditions which cause severe impairment to respiratory function, but which, with appropriate, directed management, can readily be corrected.
If respiratory failure persists, a chain of events will follow which will lead inexorably to the patient's death. The consequences of acute hypoxia can be illustrated by the effects on higher mental function. As SaO2 decreases to 85%, noticeable mental impairment is seen. This becomes severe <75% and once the SaO2 reaches 65%, unconsciousness occurs.1 Once the saturation decreases below this level, the oxygen being delivered is insufficient to meet cellular metabolic needs and cerebral, hepatic and renal cell death will inevitably follow. Acidosis will also affect cell function, leading to damage to key organs such as the cardiac muscle and the brain.
Therefore, the primary treatment aim for such patients is the maintenance of adequate oxygenation. This must be balanced against the known detrimental effects of excess O2 delivery (CO2 retention, atelectasis, and pulmonary oxygen toxicity) and mechanical ventilation (reduced cardiac output, fluid retention) and direct lung damage (from ventilator associated lung injury).
In order to improve oxygenation, it may be helpful to have an index to determine the nature of the impairment that is causing the respiratory failure in the first place, be it shunt related, V/Q mismatch, or diffusional impairment. Clinically, this information would aid in making the initial diagnosis (e.g. ARDS if PaO2/FiO2 ratio ≤26.6 kPa) and could help in deciding whether the patient would benefit from admission to ICU based on the likelihood of successfully reversing the condition. It would also be invaluable in the initiation of specific management strategies, the assessment of efficacies of these strategies as the condition develops, and determining the cost/benefit balance of more expensive therapies. An oxygenation index would also be helpful in stratifying research done on new treatment modalities.
Therefore, a clinically useful oxygenation index would allow: The aim therefore, or ‘Holy Grail’, has been described as: ‘a simple index that will quantify the degree of pulmonary oxygen transfer in any given patient, and which can also be used to monitor the progress of the patient as lung function changes over time’.2
assessment of severity and possible cause;
selection of the most appropriate treatment strategy;
monitoring changes in the condition as treatment is instigated or the disease process progresses;
a possible indication of prognosis.
This article will review the oxygenation indices currently available.
Oxygenation indices
A number of indices and variables have been suggested as accurate assessors of pulmonary oxygen exchange deficit. However, the extreme complexity of the respiratory system (a dynamic system with 23 generations and >108 alveoli) and the fundamentally non-linear nature of the relationship between oxygen tension and haemoglobin saturation in the oxyhaemoglobin dissociation curve (ODC) have caused many difficulties. Many indices oversimplify the system in an attempt to easily be understood, as one author put it when referring to the equally complex issue of septic shock; ‘For every complex problem, there is a solution that is simple … and wrong’.3
The key features of an oxygenation index4 are to:
be reliable and stable under constant physiological conditions, i.e. the value generated by the index should not alter when external variables such as FiO2 are the only changes made;
measure what it purports to measure;
be responsive to changes in the true value of the measured physiological parameter;
reflect the degree of changes with acceptable sensitivity;
provide clinically useful diagnostic or prognostic information.
Common indices
The simple, single variable measurements, such as SaO2 or PaO2 are unreliable as the measures of oxygen exchange deficit, showing poor correlation as clinical conditions change. Currently, there are five commonly used indices of oxygenation deficit, one based on arterial O2 content, the others based on arterial O2 tension (Table 1).
Currently available indices of oxygenation separated into that based on arterial oxygen content and those based on arterial oxygen tension
| Content based | Qs/Qt | Venous admixture |
|---|---|---|
| Tension based | PaO2/FiO2 ratio | Arterial oxygen tension to fraction of inspired oxygen ratio |
| P(A–a)O2 | Alveolar–arterial oxygen tension gradient | |
| P(A–a)O2/PaO2 | Respiratory index | |
| PaO2/PAO2 | Arterial–alveolar oxygen tension ratio |
| Content based | Qs/Qt | Venous admixture |
|---|---|---|
| Tension based | PaO2/FiO2 ratio | Arterial oxygen tension to fraction of inspired oxygen ratio |
| P(A–a)O2 | Alveolar–arterial oxygen tension gradient | |
| P(A–a)O2/PaO2 | Respiratory index | |
| PaO2/PAO2 | Arterial–alveolar oxygen tension ratio |
Currently available indices of oxygenation separated into that based on arterial oxygen content and those based on arterial oxygen tension
| Content based | Qs/Qt | Venous admixture |
|---|---|---|
| Tension based | PaO2/FiO2 ratio | Arterial oxygen tension to fraction of inspired oxygen ratio |
| P(A–a)O2 | Alveolar–arterial oxygen tension gradient | |
| P(A–a)O2/PaO2 | Respiratory index | |
| PaO2/PAO2 | Arterial–alveolar oxygen tension ratio |
| Content based | Qs/Qt | Venous admixture |
|---|---|---|
| Tension based | PaO2/FiO2 ratio | Arterial oxygen tension to fraction of inspired oxygen ratio |
| P(A–a)O2 | Alveolar–arterial oxygen tension gradient | |
| P(A–a)O2/PaO2 | Respiratory index | |
| PaO2/PAO2 | Arterial–alveolar oxygen tension ratio |
In the main, these indices are used as research tools; however, some are beginning to be used more commonly in day-to-day clinical settings. For example, PaO2/FiO2 ratio is now a fundamental part in the definition and severity grading of ARDS (where ratios of: ≤39.9 kPa define acute lung injury, and: ≤26.6 kPa define full ARDS). The problem with all of these is their reliability in the face of changing physiological parameters, most notably changes in FiO2.
Modelling systems
To examine each of these indices for reliability, and compare them with each other, a number of modelling systems have been devised. The original models, developed in the 1960s, had a simple, steady-state system, based on a fish-gill model. These models used continuous ventilation and perfusion systems, which bore little resemblance to the highly complex, dynamic respiratory system, where it is uncertain if the ‘steady-state’ ever actually exists. It is doubtful, therefore, how applicable any conclusions drawn from these older models would be to use in clinical practice. Many of them are criticized in the literature for favouring simplicity over accuracy, and then equating simplicity with truth.
More modern, computerized models aim to create ‘virtual patients’ where all aspects of ventilation and perfusion can be controlled. This allows factors such as tidal ventilation, pulsatile perfusion, dead space, and hypoxic pulmonary vasoconstriction to be incorporated. They also allow these variables to be controlled to create the conditions seen in a variety of disease states, and healthy lungs. These dynamic systems aim to give the closest representation possible of what actually occurs in the human respiratory system, and allow for an examination of the effects of different ventilation modes, I:E ratios and lung volumes in different disease states.
Venous admixture (Qs/Qt)
Venous admixture is widely regarded as the ‘gold standard’ of pulmonary oxygen transfer measurement, and is often used as the benchmark by which other tests are assessed.5 This is a ‘content-based’ index, which requires a pulmonary artery flotation catheter (PAFC) to collect the samples of mixed venous blood. Placement of such a device comes with its own intrinsic risks, both of central vein cannulation and passage of the catheter through the heart, which must be balanced against the benefits of the information that could be gained.
(This equation most accurately reflects the degree of true shunt if the patient is breathing pure oxygen, although even then it can be skewed by areas with low V/Q levels. Calculations at FiO2 < 1 will increasingly reflect hypoxaemia caused by V/Q mismatch and diffusion limitation and shunt.)
Venous admixture is believed by many to be an accurate indicator of efficiency of oxygenation, regardless of the cause of the underlying disturbance or the FiO2. Thus, it should accurately reflect the degree of impairment to oxygenation, regardless of the degree of shunt, V/Q mismatch or diffusion impairment. It is also favoured as it negates some of the problems of the non-linearity of the ODC.
Initially, this index was considered to be stable in healthy subjects and when hypoxaemia was as a result of true shunt alone.4,6 Most recent work has begun to call these assumptions into question.6 This is especially true with regard to its accuracy with changing FiO2 levels, with one study showing Qs/Qt to be highest when FiO2 = 0.21, least when it was between 0.4 and 0.6, and increasing again at levels greater than 0.6 (when all other variables were constant).7
Investigations using more sophisticated computer models suggest significant variation in Qs/Qt values as FiO2 is altered in otherwise identical patients, in all types of disease. There is some resistance to change when the underlying disturbance is as a result of high dead-space or true shunt, but the high degrees of variability are seen when the prevailing pathology is V/Q mismatch, e.g. in ARDS.6
The tension-based indices
These indices use PaO2, SaO2, PAO2, and FiO2 in a variety of combinations to attempt to quantify oxygenation independent of FiO2 and minimize the inherent non-linearity of the ODC. The advantage of this strategy is that only arterial blood sampling and inspired gas measurement are needed, reducing the risks associated with PAFC insertion.
The disadvantages in these indices are that: Each of these assumptions holds true to varying, and unpredictable, degrees in disease states.
they assume a constant C(a–v)O2, level, which has been shown to vary in many diseases;
they are inherently affected by changes in FiO2 as result of the non-linear relationship between O2 tension and O2 content;
the indices using PAO2 rely on further assumptions inherent in the alveolar gas equation;
that PACO2 = PaCO2;
that RQ = 0.8;
that PAH2O = 6.3 kPa;
that PaCO2 is in steady state.
All these factors add possible sources of inaccuracy to the data generated by these indices and question the reliability of all the tension-based indices.
PaO2/FiO2
The PaO2/FiO2 ratio was developed in 1974 in an attempt to eliminate the inaccuracies caused by the assumptions of the alveolar gas equation. For this reason, it was thought by many to be the most resistant to changes in FiO2 of all the tension-based indices.
Therefore, it was included in the American–European Consensus Conference on ARDS definition for ARDS.8 This inclusion resulted in widespread the use of the index clinically, and an impression of superiority over the other tension-based indices.9
The work done on the PaO2/FiO2 ratio by some researchers has shown a degree of stability, although often only when limits are applied, e.g. FiO2 > 0.5 or PaO2 > 0.5 and <13 kPa.6 More recent studies with complex computer modelling have shown that there is a considerable variation with changes in FiO2, especially in patients with haemodynamic or metabolic instability and to a high degree in ARDS patients.9 Again, these variations are reduced if limits are imposed.
These newer studies, using computer modelling to simulate ‘virtual’ patients with characteristics of ARDS (ventilation/perfusion mismatch, increased shunt, low haemoglobin, and base excess and high PEEP), question the validity of using this index as part of ARDS diagnosis. Variation has been shown to be of sufficient magnitude to result in misclassification of patients thought to have ARDS. For example, in one study, a PaO2/FiO2 ratio of around 18 kPa was generated with an FiO2 of 0.4, classifying the patient as ARDS, but when the FiO2 was decreased to 0.21, the ratio rose to 29 kPa, falling outside the ARDS definition.9 This conclusion also calls into question the results of the numerous other studies that have used PaO2/FiO2 ratios as part of their assessment of oxygenation, e.g. work on inhaled nitric oxide.
P(A–a)O2, P(A–a)O2/PaO2, and PaO2/PAO2
These three indices follow the limitations stated earlier of tension-based indices that are also subject to the alveolar gas equation assumptions.
P(A–a)O2
The alveolar–arterial oxygen tension gradient was the first index developed to asses oxygenation without need for pulmonary venous blood sampling, and was seen as an improvement on using PaO2 alone as it makes some effort to account for the influence of PaCO2. It was found to be useful in stable patients breathing room air, and showed some correlation with disease severity, increasing as severity increased10 (normal value is <2 kPa). However, considerably higher values were generated in ventilated patients breathing 100% oxygen compared with those on 40%, and at FiO2 <0.5, there was a poor prediction of shunt.11
P(A–a)O2/PaO2
The respiratory index was introduced in 1973. It aimed to reduce the variation shown with FiO2 changes seen with the earlier indices, and did demonstrate improved correlation compared wih P(A–a)O2 alone. There also appeared to be some benefit in using it as a clinically useful prognostic indicator. However, subsequent work has found the respiratory index does not give stable readings in ARDS and the relationship to pulmonary shunting was different in different diseases.12
Both the above indices show greatest resistance to change with changes in FiO2 when true shunting is low and V/Q mismatch is less widespread, e.g. in health or pulmonary embolus, not in ARDS or pneumonia.
PaO2/PAO2
Developed in 1965, early studies using the arterial–alveolar oxygen tension ratio suggested it was relatively unchanged as FiO2 varied, and also found that it could be used to predict the PaO2 that would be generated by any given FiO2. It also seemed better able to discriminate hypoxaemia caused by shunting alone to that caused by shunting combined with V/Q mismatch. Other studies found that stability could only be achieved within set limits, although these were often contradictory, e.g. FiO2 < 0.55 in one study,5 or FiO2 > 0.6 in another.10 More recent modelling has indicated that any degree of stability is only really seen in healthy lungs, rather than those with a high degree of V/Q mismatch.6
Despite these limitations, this third index appears to be more resistant to change with changing FiO2 than the others in this group.
Conclusion
It can be seen that, despite the potential usefulness of an index of patient oxygenation, there is currently no comprehensive, realistic, yet still clinically useful index of oxygenation available that meets all the ‘key features’. Many of the currently available indices are used as a result of their mathematical simplicity, rather than their validity.
The most commonly used index is the PaO2/FiO2 ratio. This is as a result of the relative ease of obtaining the measurements and it generating reasonably reliable/prognostic values (one reason it was chosen for the ARDS definition). Despite this, it still has important limitations that must be understood.
The search for a reliable index is now being driven by the development of newer, vastly more complex, computer models that allow dynamic patient interactions to be examined, moving away from the older idea of ‘steady state’. The results of these investigations may generate a reliable oxygenation index, however, this may require specific software to assimilate real-time data which may jeopardise its clinical usefuleness.
