Chapter question: A patient on ECMO can have four different blood gases drawn within the same minute, from four different places, all of them correct and all of them saying different things. Which sample answers which question — and what has changed about the physiology of oxygen and carbon dioxide once a machine, rather than a lung, is doing the exchange?
Evidence search date: 20 September 2026.
Primary sources: ELSO Red Book 6th edition — Chapter 25 (venovenous management: sweep gas weaning, the sweep-gas oxygen fraction argument, extracorporeal blood flow ranges), Chapter 28 (adult cardiac management: the gas-exchange and oxygen-delivery troubleshooting table), Chapter 45 (the apnoea test on ECMO, owned by Chapter 24 and cited only). ISCCM Manual of RRT and ECMO in ICU — Chapter 26 (extracorporeal physiology, rated flow, the inlet-outlet content difference) and Chapter 33 (metabolic monitoring). ECMO: A Practical Guide to Management (Taha, Caridi-Scheible, Leiendecker) — Chapter 5, the gas-transfer figure, the rated-flow table and the worked recirculation table; Chapter 9, the monitoring-modality table and the pulse-oximetry error. ECMO in the Adult Patient (Core Critical Care, 2017) — circuit blood gas sampling technique and the transmembrane-pressure pairing.
External evidence: a physiological review of extracorporeal gas exchange by the field's founding investigator; a monitoring review of the adult venoarterial patient; an international standardisation document for in vitro evaluation of extracorporeal devices; an experimental study of carbon dioxide removal in 12 pigs; a review of extracorporeal carbon dioxide removal; a pharmacotherapy review of ventilatory strategy on extracorporeal support; a 33-patient study of carboxyhaemoglobin as a marker of haemolysis on ECMO; a registry analysis of pulse-oximeter accuracy by patient race in an ECMO population, read second-hand; and two general reviews of blood gas and oximetry measurement.
Every external reference was read as an indexed full-text passage or, where stated, second-hand through a journal-watch summary. None was retrieved as a complete article. Digital object identifiers are reproduced from the search record and have not been independently resolved. Every database row seeded from this chapter is marked Verified = No.
What this chapter covers — and what it does not
Chapter 25 established that every haemodynamic instrument on a circuit was calibrated in a circulation that no longer exists. Chapter 26 established that every echocardiographic number is a measurement at a stated flow. This chapter completes the argument for the two gases, and the conclusion is sharper than for either of the others: on ECMO a blood gas result is not interpretable at all unless you know where the sample came from.
This chapter owns | Owned elsewhere — cite, do not re-argue |
Where blood gases are drawn from on a circuit, and what each site measures | The physiology of differential oxygenation and the mixing point — Chapter 13, Chapter 16 |
Gas transfer across the membrane lung: rated flow, the oxygen and carbon dioxide levers and their independence | The circuit and oxygenator as hardware, and device selection — Chapter 3; oxygen delivery in the intact circulation — Chapter 2; effective flow and the mixing equation on venovenous support — Chapter 6 |
Oxygenation and carbon dioxide targets, and the disagreement between this book's own primary sources | Managing persistent hypoxaemia on venovenous support — Chapter 9; differential hypoxaemia — Chapter 16 |
Recirculation as a measured oxygen-transfer loss, and the saturation rule that detects it | Recirculation physiology and effective flow — Chapter 6; its place in the daily round — Chapter 8; its calculation — Appendix 5; the cannula geometry that causes it — Chapter 26, Chapter 34 |
The pre-membrane and post-membrane pair as physiological information | The same pair as an oxygenator failure diagnosis, with transmembrane pressure — Chapter 32 |
Pulse oximetry and co-oximetry on a circuit, including occult hypoxaemia and dyshaemoglobins | Haemolysis as a complication, its causes and its management — Chapter 37 |
Haemoglobin as an oxygen-delivery lever, and the argument about transfusing to raise it | The transfusion strategy itself, thresholds and products — Chapter 46 |
The rate at which carbon dioxide should be corrected | The neurological consequences of rapid correction after cardiac arrest — Chapter 22; cerebral monitoring — Chapter 28 |
Assembling a blood gas round, and the errors of interpretation | Ventilator settings on ECMO — Chapter 53; lactate and the laboratory panel — Chapter 29; oxygen content and delivery calculators — Appendix 2, Appendix 3 |
The boundary with Chapter 32 is the one to hold. A post-membrane gas that has deteriorated is a machine diagnosis and belongs to Chapter 32. This chapter owns the same sample read as physiology: what the membrane is delivering to this patient, now. The two readings use the same number and answer different questions — which is this Part's recurring theme.
27.1 On ECMO the two gases have been unbundled
This is the organising idea, and like Chapter 25's it is mechanical rather than clinical.
Physiology · One organ becomes two independent knobs
In a native lung, oxygenation and carbon dioxide clearance are performed by the same structure under the same settings. They are not independent: recruit the lung and both improve; lose it and both deteriorate. Clinicians reason about "gas exchange" as one thing because, in a patient, it mostly is one thing.
A membrane lung separates them completely, onto two knobs that do not interact.
- Oxygen transfer is limited by how much haemoglobin passes the membrane per minute. Blood entering the device is already 60 to 80 percent saturated, so only the remaining fraction can be filled. The levers are therefore extracorporeal blood flow and haemoglobin concentration, together with the inlet saturation. The oxygen fraction of the sweep gas is not a meaningful lever in a working device (§27.3).
- Carbon dioxide clearance is limited by how much gas passes the other side of the membrane per minute. Carbon dioxide is far more soluble and more diffusible than oxygen; the gradient runs from roughly 45 mmHg in blood to essentially zero in the sweep gas. The lever is sweep gas flow, with blood flow playing a minor role except at the very low end.
The clinical consequences follow immediately and are the whole chapter in miniature.
- The knobs cannot substitute for one another. Raising the sweep gas for hypoxaemia does nothing. Raising blood flow for hypercapnia does very little.
- The two gases can be driven to opposite ends of the physiological range at the same time. Complete removal of a resting adult's carbon dioxide production is achievable at blood flows as low as 500 mL/min, against the roughly 6 L/min required for full oxygenation. That gap is the entire rationale for extracorporeal carbon dioxide removal — and it is also the trap, because a patient whose carbon dioxide and pH look perfect may be receiving almost no oxygenation support.
- A normal pH on ECMO is not evidence that the circuit is oxygenating. It is evidence that the sweep gas is flowing.
Source for the physiology and the 500 mL/min figure: Bartlett RH. Comprehensive Physiology 2020;10(3):879–891. DOI 10.1002/j.2040-4603.2020.tb00130.x.
Variable | Affects oxygenation | Affects carbon dioxide elimination |
Diffusion gradient | Yes | Yes |
Membrane surface area | Yes | Yes |
Oxygen fraction of the sweep gas | Yes in principle — and see §27.3, where a working membrane saturates the outlet blood at a sweep oxygen fraction of 0.21 | No |
Extracorporeal blood flow | Yes — the dominant lever | No, except at the low end of the blood flow range, where it becomes a major determinant |
Sweep gas flow rate | No | Yes — the dominant lever |
Haemoglobin concentration | Yes — the second lever, and the one most often forgotten (§27.4) | Indirectly, through carbon dioxide carriage |
Inlet (pre-membrane) saturation | Yes, inversely — a higher inlet saturation means less oxygen can be added (§27.5) | Through the inlet carbon dioxide tension, which sets the gradient |
Table adapted from the gas-exchange table in Chung M, Shiloh AL, Carlese A, Butt WW, Spapen H. The Scientific World Journal 2014. DOI 10.1155/2014/393258, with the haemoglobin and inlet-saturation rows added from Bartlett 2020 and from this book's primary textbooks.
Clinical pearl · The sweep gas is the on-off switch; the blood flow is the volume control
ELSO Red Book Chapter 25 states it in one line, about weaning: "Irrespective of the extracorporeal blood flow, once the sweep gas flow ceases, ECMO support effectively ceases."
This is worth holding onto because it inverts the intuition. Most clinicians watch the blood flow number on the console as the measure of "how much ECMO" a patient is receiving. A circuit running at 5 L/min with the sweep gas off is providing no gas exchange at all — it is an expensive, thrombogenic bypass loop. Conversely, a circuit at 2 L/min with an open sweep is still clearing carbon dioxide efficiently.
When someone asks how much support a patient is on, the answer requires two numbers, and the sweep is the one more often omitted from the handover.
Pitfall · Turning up the blender for hypoxaemia
The commonest reflex error in this domain, and it is a category error rather than a dosing error. On a functioning membrane lung the outlet blood is already fully saturated (§27.3). Raising the sweep gas oxygen fraction from 0.6 to 1.0 in a hypoxaemic patient on venovenous support adds essentially nothing, because there is no unsaturated haemoglobin left in the outlet blood to fill.
The hypoxaemia is coming from somewhere else — insufficient blood flow relative to cardiac output, recirculation, anaemia, a failing membrane, or native shunt (Chapter 9 owns the differential). Turning the blender up wastes the minutes in which the real cause could have been found.
27.2 Four samples, four questions
A conventional ICU patient has one arterial sampling site and the word "the blood gas" is unambiguous. On ECMO there are at least four, and they are not interchangeable.
Sample | What it physically is | The question it answers | What it is NOT |
Pre-membrane (pre-oxygenator) port | Blood drained from the patient, after the pump, before the gas exchanger | On venoarterial support: the true mixed venous sample. The ISCCM Manual states it directly — venous saturation on venoarterial ECMO is a true reflection of venous oxygen content and can be taken from a central line or pre-membrane gases. This is the oxygen-delivery-versus-consumption measurement (Chapter 25) | On venovenous support it is NOT a mixed venous sample, because it contains recirculated, already-oxygenated blood. It is a mixture whose composition is the recirculation diagnostic (§27.5) |
Post-membrane (post-oxygenator) port | Blood leaving the gas exchanger, about to re-enter the patient | What the machine is delivering. Should be fully saturated with a high partial pressure of oxygen. Paired with the pre-membrane sample it gives the oxygen and carbon dioxide the device actually transferred | Not the patient's arterial blood. On venovenous support it never is; on venoarterial support it is what reaches the lower body but not necessarily the brain |
Right radial or right brachial arterial line | Blood supplied to the innominate, right carotid and right subclavian territory | On peripheral venoarterial support: what the brain and the coronary arteries are receiving. Both the ISCCM Manual and the practical-guide textbook state that the arterial line should be on the right arm and that gases should be taken from that side | Not representative of the lower body on venoarterial support, and its divergence from a femoral sample is the definition of the problem Chapter 16 owns |
Left arm, femoral or other distal arterial site | Blood in the territory supplied predominantly by the circuit on femoral venoarterial support | What the machine's territory is receiving, and by comparison with the right radial sample, where the mixing point is sitting | Not the cerebral supply. A reassuring femoral gas on peripheral venoarterial ECMO is compatible with a hypoxaemic brain |
Danger · The sampling site must be written on the result, every time
This chapter's version of the rule Chapters 25 and 26 established for circuit flow. "PaO₂ 58" is not a result on ECMO. It is a result on a patient. On a circuit it becomes a result only when accompanied by: the site, the extracorporeal blood flow, the sweep gas flow, the sweep gas oxygen fraction, and the ventilator settings at the time.
The specific harm is not theoretical. A post-membrane sample charted as an arterial gas reads as spectacular oxygenation in a profoundly hypoxaemic patient. A femoral arterial sample charted as "arterial" on peripheral venoarterial support conceals differential hypoxaemia — the exact scenario Chapter 16 exists for, and the one in which the brain is injured while the chart looks normal.
Clinical pearl · Draw circuit gases with the sweep gas at 100 percent oxygen
A practical convention from ECMO in the Adult Patient, already adopted as part of the daily circuit round in Chapter 8: to ensure reliable and comparable results, blood gases obtained from the circuit are measured with the sweep gas set at 100 percent oxygen. The reason is that a post-membrane partial pressure of oxygen means nothing as a device measurement unless the inspired fraction on the gas side is known and constant — exactly as an arterial partial pressure of oxygen means nothing without the inspired fraction on a ventilator.
The same source adds the practical warning that sampling is done carefully to avoid excessive loss of blood, as the pressure at the sampling point can be very high. Post-pump ports are under positive pressure; an unguarded stopcock sprays.
Set the sweep oxygen fraction to 1.0, wait for equilibration, draw the pair, then return it. And record that you did, because the next person comparing to your value needs to know the conditions were the same.
Worked example · The same minute, four samples, four different conclusions
Peripheral femoro-femoral venoarterial ECMO for post-cardiotomy shock. Circuit flow 3.8 L/min, sweep 3.5 L/min at an oxygen fraction of 1.0. Native lungs poorly compliant, ventilator on rest settings. The heart has begun to eject.
Site | Result | Conclusion |
Pre-membrane | Saturation 61%, PO₂ 34 mmHg, PCO₂ 47 mmHg | True mixed venous. Extraction is high but not extreme; oxygen delivery is roughly adequate to consumption. Chapter 25 owns what to do about it |
Post-membrane | Saturation 100%, PO₂ 421 mmHg, PCO₂ 36 mmHg | The membrane is working. Oxygen added and carbon dioxide removed as expected at this sweep-to-blood ratio |
Femoral arterial | Saturation 99%, PO₂ 187 mmHg, PCO₂ 37 mmHg | The lower body is being perfused by the circuit. Looks superb |
Right radial arterial | Saturation 84%, PO₂ 51 mmHg, PCO₂ 44 mmHg | The brain is being perfused by the patient's own failing lungs. This is differential hypoxaemia and it is an emergency (Chapter 16) |
Three of the four samples are reassuring and one is not, and the one that is not is the only one that describes the brain. The recovering heart is the cause: as native output rises, it pushes the mixing point distally, and the blood reaching the arch vessels is increasingly the patient's own, deoxygenated by a lung that cannot oxygenate it.
The lesson is not that the right radial sample is the important one. It is that a patient on peripheral venoarterial support requires the right radial sample to exist at all — and in this case the left-sided arterial line that was already in situ would have read 99%.
This case is constructed to illustrate the sampling map in this section and the physiology Chapter 16 owns. It is not drawn from a published report.
27.3 The oxygen side: rated flow, and why the blender is not the oxygen knob
Rated flow is the central engineering concept and it is worth stating precisely, because it is what a post-membrane gas is implicitly being compared against.
Evidence · Rated flow, and a discrepancy in its definition
The definition. Rated flow is the flow of standardised venous blood, in litres per minute, that exits a membrane lung at 95 percent saturation. The concept combines surface area, mixing geometry, residence time, inlet saturation, haemoglobin and blood flow into a single number, because as flow rises the transit time falls and eventually the exiting blood is no longer fully saturated.
The test. When inlet blood is at standardised conditions, outlet saturation should be 95 percent at rated flow. If outlet saturation is less than 95 percent at a flow below rated flow, oxygenator function is reduced — usually because of clot in the blood path.
A NUMERICAL DISCREPANCY, NOW THREE-WAY, PRINTED AND NOT RESOLVED. Chapter 6 already recorded a two-way version of this — the standardised inlet saturation is 70 percent in the ELSO Red Book and the ISCCM Manual and 75 percent in the practical-guide textbook, and Chapter 3 was corrected in place to record both. Chapter 27 adds two further observations that widen it.
- The anchor physiological source contains both values internally. Bartlett 2020 gives "hemoglobin 12 g/dL and 70% saturated" in the body of the paper and "hemoglobin 12 gr/dL, saturation 75%" in the same paper's summary section. The disagreement is not only between textbooks; it is inside the field's canonical physiological review.
- A third value comes from the standard itself. The 2026 international standardisation document cites ISO 7199 for an inlet condition of 65 percent plus or minus 5 percent, at haemoglobin 120 g/L.
So the field has 65, 70 and 75 percent in circulation for the same defined quantity. This matters rather than being pedantry. Inlet saturation determines how much oxygen the device must add per unit of blood, so it determines the rated flow figure a manufacturer may print. Two devices tested at 65 percent and 75 percent inlet saturation are not comparable, and a bedside clinician comparing an observed outlet saturation against a printed rated flow is making exactly that comparison. All three values are recorded here; none is asserted.
Sources: Bartlett RH. Comprehensive Physiology 2020;10(3):879–891. DOI 10.1002/j.2040-4603.2020.tb00130.x. Arens J, Belliato M, Broman LM, et al. Artificial Organs 2026;50(5):674–688. DOI 10.1111/aor.70124.
Physiology · Why the sweep gas oxygen fraction is not the lever
The ELSO Red Book states the consequence in its venovenous management chapter, and it is counter-intuitive enough to be worth quoting: "in a properly functioning membrane lung even very low flows (sweep gas flow 1 to 2 litres per minute or less) of unenriched air as sweep gas (sweep oxygen fraction 0.21) should result in a saturation close to 100% in the blood leaving the membrane lung."
The reason is the shape of the problem. Air at one atmosphere delivers an alveolar-equivalent oxygen tension of roughly 150 mmHg against venous blood at 40 mmHg — a gradient of over 100 mmHg across a membrane with an enormous surface area and no diffusion barrier comparable to a diseased alveolus. The membrane is so efficient that the gradient available from room air is already more than sufficient to saturate the haemoglobin present. Raising the fraction to 1.0 raises the dissolved oxygen and the outlet partial pressure dramatically, but adds almost nothing to the oxygen content, because content is haemoglobin-bound and the haemoglobin is already full.
Therefore: the Red Book concludes it is not considered necessary to titrate the sweep gas oxygen fraction at any time, and some centres always use 1.0 for safety and simplicity. The oxygen levers are extracorporeal blood flow and haemoglobin; these, with the patient's total cardiac output, determine arterial saturation.
Certainty: high for the physiology, expert consensus for the practice recommendation.
Pitfall · The post-membrane partial pressure of oxygen is a device number, not a dose
A post-membrane PO₂ of 450 mmHg looks alarming to anyone trained on ventilated patients, and it triggers a reflex to turn the blender down. It is not hyperoxia in any meaningful sense. The oxygen content of that blood is set by its haemoglobin and its 100 percent saturation; the dissolved fraction contributed by a PO₂ of 450 is roughly 1.35 mL/dL on top of about 20 mL/dL bound — under 7 percent of the total, and it is immediately given up to the first venous blood it mixes with.
What a falling post-membrane PO₂ at a constant sweep oxygen fraction does mean is a membrane that is failing, and that is Chapter 32's territory. Keeping the sweep oxygen fraction at 1.0 preserves that diagnostic signal. Turning it down for fear of hyperoxia destroys a surveillance measurement and achieves nothing physiologically — which is a reasonable summary of why the Red Book stopped recommending titration.
There is a separate and legitimate concern about systemic hyperoxia on venoarterial support, where the circuit can drive arterial PO₂ to supraphysiological levels in the territory it perfuses. That is a flow and arterial-target question, not a blender question, and Chapter 16 and Chapter 22 own it.
27.3.1 How much blood flow does this patient actually need?
The arithmetic is worth doing once, because it makes the haemoglobin lever visible.
Worked example · Blood flow required for full oxygenation, from first principles
The source example: an 80 kg adult with a resting oxygen requirement of 240 mL/min, venous saturation 60 percent, haemoglobin 11 g/dL, membrane outlet saturation 100 percent.
Step | Calculation | Result |
Oxygen capacity per gram of haemoglobin | Standard constant | 1.34 mL O₂ per g Hb |
Oxygen content of fully saturated blood at Hb 11 g/dL | 1.34 × 11 × 1.00 | 14.7 mL/dL |
Oxygen content of inlet blood at 60% saturation | 1.34 × 11 × 0.60 | 8.8 mL/dL |
Oxygen added per decilitre across the membrane (the outlet-inlet difference) | 14.7 − 8.8 | 5.9 mL/dL |
Blood flow needed to deliver 240 mL/min | 240 ÷ 5.9 = 40.7 dL/min | About 4.1 L/min |
The published source gives 4.1 L/min for these conditions and the arithmetic above reproduces it. Dissolved oxygen is ignored throughout, as it is in the source and in clinical practice.
Now change one variable and nothing else. At haemoglobin 14 g/dL the content difference becomes 1.34 × 14 × 0.40 = 7.5 mL/dL, and the flow required falls to 240 ÷ 7.5 = 32 dL/min, or 3.2 L/min. A 3 g/dL rise in haemoglobin has removed roughly 0.9 L/min of required blood flow without touching the pump.
This is the calculation behind §27.4, and it is why the haemoglobin on an ECMO chart is a gas-exchange variable and not only a transfusion variable.
Source of the worked conditions and the 4.1 L/min figure: Bartlett RH 2020, DOI 10.1002/j.2040-4603.2020.tb00130.x. The 14 g/dL variant is this book's arithmetic on the same method.
Clinical pearl · Choose the circuit for the patient's metabolic rate, not for the patient's weight
The source rule: for a typical 80 kg adult the circuit components must be able to provide around 4 L/min and the membrane oxygenator must have a rated flow greater than 5 L/min, with the usual adult target around 6 L/min rated flow. The reason for the margin is that the requirement is not static: once the patient is stabilised and sedation is stopped, activity raises oxygen consumption and carbon dioxide production, and flow and sweep must rise to match.
An awake, mobilising ECMO patient needs more extracorporeal support than a sedated one, not less. That is the opposite of the intuition built on ventilated patients, and it is a recurring source of unexplained desaturation on the day sedation is lightened.
27.4 Haemoglobin is the forgotten oxygen lever
Controversy 1 — Should an anaemic, hypoxaemic patient on venovenous ECMO be transfused rather than run at higher flow?
The problem, stated exactly. §27.3.1 shows that haemoglobin and blood flow are interchangeable levers on oxygen delivery: a 3 g/dL rise in haemoglobin substitutes for roughly 0.9 L/min of flow in an 80 kg adult. The field's founding physiological review takes an explicit position on which to move, and it is a position most modern intensivists would not adopt without argument.
The case for transfusing. Stated in the source's own words, as the legend to its figure on blood flow requirement: "The risks of transfusion (mismatch, hepatitis) are rare and minimal compared to the risk of high flow (blowout, hemolysis, high pressure). Therefore, if the patient is anemic, transfusion is favored over high flow when optimizing oxygen delivery." The mechanical argument behind it is real and is supported elsewhere in this book's sources: higher pump speeds and higher extracorporeal blood flow are associated with haemolysis (Red Book Chapter 25), free haemoglobin rises with more negative access pressure (ECMO in the Adult Patient), and access insufficiency, cavitation and cannula problems all scale with the flow demanded of a given cannula. Raising haemoglobin achieves the same oxygen delivery with a gentler circuit.
The case against. Three objections, and they are not answered in the source.
- The stated harms of transfusion are dated and incomplete. "Mismatch and hepatitis" is a 1980s risk framing. The modern concerns in this population are transfusion-associated circulatory overload, transfusion-related acute lung injury in patients whose lungs are the problem, alloimmunisation in patients who may go on to transplantation, immunomodulation, and the association between transfusion volume and mortality reported across critical care.
- Restrictive transfusion has been tested in general critical care and liberal transfusion has not been shown to be better. The ECMO population was not in those trials, but the burden of proof for departing from restrictive practice sits with the departure.
- The comparison is not transfusion versus high flow; it is transfusion versus accepting a lower saturation. §27.8 shows that the targets themselves are permissive. A patient at 82 percent saturation with adequate delivery, clearing lactate, is not obviously a patient who needs either intervention.
What is not in dispute. Two things.
- Haemoglobin genuinely is an oxygen-transfer variable on ECMO, in a way it is not on a ventilator, because the membrane's transfer is capped by the unsaturated haemoglobin presented to it. A falling haemoglobin causes desaturation on venovenous support with nothing else changed, and the mechanism should be recognised before the pump is turned up.
- High flow has specific circuit costs that do not exist in a native circulation, and the mechanical argument for preferring the haemoglobin lever is sound as far as it goes.
Where this book lands.
- Recognise the mechanism. When arterial saturation falls on venovenous ECMO, put haemoglobin on the differential alongside recirculation, rising cardiac output, and membrane failure (§27.12). This is the part of the source's argument that is unambiguously correct and under-used.
- Do not adopt the transfusion recommendation as written. The risk statement supporting it is a generation out of date. Chapter 46 owns the threshold and should state one explicitly for this population.
- Consider the haemoglobin lever specifically when the alternative is a flow the cannula cannot deliver without chatter, or a flow already associated with rising plasma free haemoglobin — that is, when the mechanical objection to high flow is concrete rather than theoretical in this patient.
- Record the reasoning either way, because a transfusion given for oxygen delivery on ECMO will otherwise be audited against a general ICU threshold and look like an error.
What would settle it. A randomised comparison, in venovenous ECMO patients with a haemoglobin below a defined threshold and inadequate oxygen delivery, of transfusion against increased extracorporeal blood flow, with haemolysis markers, circuit lifespan and survival as endpoints. No such trial was identified. The recommendation as it stands rests on a mechanical argument and a risk comparison, not on comparative data.
Pitfall · A falling haemoglobin looks exactly like a failing oxygenator
Both present as a gradual fall in arterial saturation over hours on venovenous support with the console unchanged. The pre-membrane and post-membrane pair separates them in one draw:
- Failing membrane: post-membrane saturation falls below 100 percent, post-membrane PO₂ falls, and transmembrane pressure usually rises (Chapter 32).
- Falling haemoglobin: post-membrane saturation is still 100 percent and the post-membrane PO₂ is unchanged. The device is still doing everything it can; there is simply less to carry.
The second is a laboratory diagnosis that a blood gas makes before the full blood count returns, because most blood gas analysers report a haemoglobin.
27.5 Recirculation is a measurable loss of oxygen transfer
Chapter 6 owns recirculation physiology and effective flow, and Appendix 5 owns its calculation. This section owns what recirculation does to the numbers in front of you, because it is the clearest demonstration in the book that the console flow is not the support delivered.
Evidence · The same circuit flow, less than half the oxygen transferred
A worked table in this book's practical-guide textbook holds every variable constant except the pre-oxygenator saturation. Haemoglobin 10 g/dL; post-oxygenator saturation 100 percent with PO₂ 500 mmHg in both conditions.
Situation | Pre-oxygenator oxygen content | Post-oxygenator oxygen content | Oxygen added |
Baseline, pre-oxygenator saturation 60%, PO₂ 40 mmHg | 8.3 mL/dL | 15.2 mL/dL | 6.9 mL/dL |
Recirculation, pre-oxygenator saturation 90%, PO₂ 60 mmHg | 12.4 mL/dL | 15.2 mL/dL | 2.8 mL/dL |
Oxygen transfer falls from 6.9 to 2.8 mL/dL — a 59 percent loss — with the pump speed, the blood flow, the sweep and the membrane all unchanged. The console reports the same support throughout.
The source's framing: if blood flow and membrane lung function remain constant, the main determinant of oxygenation is the oxygen-carrying capacity of the blood passing through the device, which depends on haemoglobin concentration and inlet saturation. An increase in inlet saturation decreases overall oxygen transfer efficiency and can lead to desaturation.
Certainty: the arithmetic is internally consistent; the scenario is illustrative. Real recirculation fractions vary with configuration, cannula size and position, jet direction, flow and intracavitary pressures.
A CONSTANT DISCREPANCY, ALREADY FLAGGED BY CHAPTER 6 AND REPEATED HERE RATHER THAN SILENTLY CORRECTED. Reproducing these contents requires an oxygen-carrying constant of approximately 1.36 mL/g plus the dissolved fraction, not the 1.34 mL/g used elsewhere in this book and in §27.3.1. Chapter 6 recorded the same finding when it reproduced this table. The published values are printed here unaltered. The difference is immaterial to the conclusion — a 59 percent transfer loss either way — but it is the reason two worked examples in the same chapter do not use the same constant, and a reader recomputing them should know why.
Source: ECMO: A Practical Guide to Management, Chapter 5, Table 5.2.
Clinical pearl · The one recirculation rule with no exceptions
From the same source, and it is a logical rule rather than an empirical one: "An increase in pre-oxygenator saturation with a fall in arterial saturation may be an indicator for recirculation; pre-oxygenator saturation can only be higher than arterial saturation in the setting of recirculation."
Think about why. On venovenous support, blood leaving the membrane at 100 percent saturation mixes in the right heart with native venous return before being ejected through the lungs to the arterial circulation. The drainage cannula should be sampling blood that has been round the body — that is, blood that is by definition more desaturated than arterial blood. If the drainage sample is more saturated than the arterial sample, some of what the cannula is draining never went round the body at all. There is no other explanation.
So: trend the pre-oxygenator saturation alongside the arterial saturation. Divergence in that specific direction is diagnostic, and it costs nothing beyond gases you are already drawing.
Pitfall · Recirculation is normal and only sometimes a problem
The same source is careful about this and it is worth repeating. Because blood is drained and reinfused into the same venous compartment, some recirculation is always present and expected. It should be addressed only when it is clinically relevant — that is, when oxygen delivery is compromised.
Chasing a recirculation percentage toward zero is not a treatment goal. Repositioning a cannula in a stable, adequately oxygenated patient exposes them to the risks of §26.5 for a number on a chart.
27.6 The carbon dioxide side: the most powerful knob at the bedside
Physiology · Why carbon dioxide clearance is almost too easy
Three facts combine, and the result is a lever with far more authority than anything on a ventilator.
- The gradient is nearly maximal. Pre-membrane blood carries a carbon dioxide tension of roughly 45 to 50 mmHg; the sweep gas contains essentially none. The gradient is therefore the full blood tension, always, and it cannot be improved upon.
- Carbon dioxide is far more diffusible. One source puts the transfer efficiency at roughly ten times that of oxygen. This book's practical-guide textbook makes the same point from the other direction: the oxygen gradient (about 600 to 0 mmHg) is much greater than the carbon dioxide gradient (about 45 to 0 mmHg), yet the amounts exchanged are roughly equal at a gas-to-blood flow ratio of 1, because solubility and diffusivity favour carbon dioxide so heavily.
- Most of the carbon dioxide is in a rapidly available reservoir. Venous blood carries roughly 45 mL/dL of total carbon dioxide, the great majority as bicarbonate, which dissociates rapidly as dissolved carbon dioxide is stripped. The membrane is not limited to the dissolved fraction; it drains the bicarbonate pool.
Consequences. At a 1:1 sweep-to-blood ratio, outlet blood carbon dioxide tension is typically 35 to 40 mmHg and carbon dioxide removal roughly equals oxygen delivery. Raise the sweep-to-blood ratio to 2, 3 or 10 and the outlet tension can fall to 5 or 10 mmHg while oxygen delivery does not change at all. Complete removal of a resting adult's carbon dioxide production is achievable at blood flows around 500 mL/min.
A NOTE ON ONE PASSAGE. The retrieved text of the physiological source states that carbon dioxide in blood "is in the form of carboxyhemoglobin" and lists carboxyhaemoglobin as a component of total carbon dioxide content. The correct term for haemoglobin-bound carbon dioxide is carbamino-haemoglobin; carboxyhaemoglobin is haemoglobin bound to carbon monoxide — a distinct species that matters independently on ECMO and is the subject of §27.10. The retrieved passage may contain an indexing or transcription artefact rather than an authorial error; the discrepancy is recorded and the substantive claim — that most carbon dioxide removed by a membrane lung comes from the bicarbonate pool — is unaffected and is used.
Sources: Bartlett RH 2020, DOI 10.1002/j.2040-4603.2020.tb00130.x; Chung M et al. 2014, DOI 10.1155/2014/393258; ECMO: A Practical Guide to Management, Chapter 5, Figure 5.5.
Sweep-to-blood flow ratio | What it is for | Effect on carbon dioxide | Effect on oxygen |
Sweep off | A weaning trial, or a disaster | None. Support has ceased | Essentially none |
Below 1:1 | Deliberate permissive hypercapnia; the apnoea test (Chapter 24) | Partial clearance | Unchanged |
1:1 | The default. Equivalent to a normal lung's ventilation-perfusion ratio | Outlet tension typically 35 to 40 mmHg; clearance roughly equals oxygen delivery | Unchanged |
2:1 to 4:1 | Hypercapnia not corrected at 1:1; high carbon dioxide production | Substantially greater clearance | Unchanged |
Up to 8:1 or 10:1 | Selective carbon dioxide removal, as in status asthmaticus | Outlet tension may fall to 5 to 10 mmHg | Unchanged |
Sweep with added carbon dioxide | Preventing iatrogenic hypocapnia when the membrane is too efficient for the patient's production | Reduces clearance by reducing the gradient | Unchanged |
Evidence · At very low blood flow, the sweep stops being the dominant lever
The independence of the two knobs is not absolute, and the exception is specific. In 12 healthy pigs on a low-flow extracorporeal carbon dioxide removal circuit (membrane surface area 1.8 m², two degrees of induced hypercapnia), carbon dioxide removal rose with extracorporeal blood flow at every gas flow tested. But:
- In severe hypercapnia, raising the sweep from 4 to 10 L/min increased removal from 86.4 ± 7.1 to 96.5 ± 8.7 mL/min at a blood flow of 350 mL/min — and was ineffective at a blood flow of 200 mL/min.
- Carbon dioxide removal rose with the degree of hypercapnia at 350 mL/min but not at 200 mL/min, which the authors interpret as the device having already reached maximal removal at the lower flow.
- Post-membrane carbon dioxide tension was similar regardless of pre-membrane tension (p = 0.08), which is the same observation as Bartlett's: the membrane strips what reaches it almost completely, so the limit is delivery, not exchange.
Certainty: low. Twelve animals, one device, one surface area, an experimental model of hypercapnia rather than disease. The authors also note that published thresholds beyond which more sweep adds nothing differ between devices — one study found a plateau above 2 L/min, another found no effect above 4 L/min, and in exacerbations of chronic obstructive pulmonary disease the reported threshold is 4 to 5 L/min. There is no transferable number here, only the principle.
The transferable principle: at full ECMO blood flows the sweep is sovereign over carbon dioxide. At the low blood flows used for carbon dioxide removal alone, blood flow re-enters as a co-determinant, and a sweep increase that achieves nothing means the blood flow is the limit.
Source: Qian Z, He H, Wang Y, et al. Artificial Organs 2024;48(6):586–594. DOI 10.1111/aor.14718.
Clinical pearl · Titrate the sweep to pH, not to carbon dioxide tension
The ELSO Red Book records the practice and the reason: many centres judge adequate carbon dioxide clearance on a normal arterial pH rather than a normal arterial carbon dioxide tension, because metabolic compensation of hypercapnia is seen frequently in patients requiring venovenous ECMO.
A patient who has been hypercapnic for days on the ward arrives with a raised bicarbonate. Driving their carbon dioxide tension to 40 mmHg with the sweep makes them profoundly alkalotic, shifts the oxyhaemoglobin dissociation curve leftward at the moment oxygen delivery is most precarious, and is achieved by a knob that will do it in minutes. The pH is the variable with the physiological meaning; the carbon dioxide tension is the variable with the round number.
Pitfall · Nobody knows what fraction of the clearance the membrane is doing
A limitation the Red Book states plainly and that deserves more attention than it gets: the proportion of total carbon dioxide clearance performed by the membrane is not routinely measured, and may not be represented solely by the sweep gas flow setting, so the effects of any reduction in sweep are somewhat uncertain.
The patient's own lungs are clearing some carbon dioxide, and how much changes as they recover. A sweep reduction that a recovering patient tolerates easily on Tuesday may not be tolerated on Wednesday if a lobe collapses overnight. This is the reason the Red Book treats regular sweep weaning as routine care for stable or improving patients rather than as an occasional test — the only way to find out how much the native lung is contributing is to take some sweep away and watch.
27.6.1 Water in the gas phase
Physiology · The membrane's second failure mode, and it is a carbon dioxide failure
Membrane lungs fail in two ways and they present differently, which is why the pre-post pair is drawn for both gases and not just for oxygen.
- Clot in the blood phase. Occurs gradually despite anticoagulation. Detected by increased blood flow resistance and decreasing oxygenation. Treated by replacing the membrane lung.
- Water accumulation in the gas phase. Condensate from the humidified blood side crosses into the fibres and blocks the gas path. It blocks carbon dioxide transfer specifically. Treated by increasing the gas flow — the classic "sigh" or flush manoeuvre, which clears the water without replacing anything.
So a rising carbon dioxide tension with preserved oxygenation is not the same problem as a falling oxygenation with a rising transmembrane pressure. The first may be reversible in seconds at the bedside; the second means a circuit change. Chapter 32 owns both diagnoses; this section exists so that the gas result is read as pointing at one of them rather than at "the oxygenator".
Source: Bartlett RH 2020, DOI 10.1002/j.2040-4603.2020.tb00130.x.
27.7 The rate of correction is a separate decision from the target
Chapter 22 established, for the post-arrest patient, that the first two minutes on a circuit move carbon dioxide, oxygen and arterial pressure simultaneously in the direction of maximum cerebral insult, by default, and that all three are set by knobs. That argument generalises to every ECMO initiation, and this book's primary textbooks give it a number.
Evidence · Correct hypercapnia over 4 to 8 hours
From the gas-transfer figure legend in this book's practical-guide textbook: the gas-to-blood flow ratio is set according to the objective carbon dioxide tension, and "rapid decreases of PaCO₂ are associated with neurological complications due to alterations in cerebral blood flow, so hypercapnia should be corrected over 4 to 8 hours after ECMO initiation."
Certainty: expert consensus in a textbook figure legend, with no supporting study cited at that point. The underlying physiology is not in doubt — between 20 and 60 mmHg, each mmHg of carbon dioxide changes cerebral blood flow by 2 to 3 percent, beginning within 2 minutes and equilibrating within about 12 (Chapter 22 §22.3) — and the direction of the recommendation is supported by the registry and porcine evidence assembled in Chapter 22. The specific 4-to-8-hour window is a convention, not a validated interval, and no source reaching this book states how it was derived.
Source: ECMO: A Practical Guide to Management, Chapter 5, Figure 5.5 legend.
Danger · The sweep gas will correct a chronic hypercapnia in under ten minutes if you let it
This is the single most avoidable iatrogenic injury in this chapter. A patient arriving on ECMO with a carbon dioxide tension of 95 mmHg, accumulated over days, has a cerebral circulation equilibrated to that tension. A sweep set at a conventional 1:1 ratio at 4 L/min of blood flow will strip that to near-normal within minutes, producing an abrupt cerebral vasoconstriction of the order of 100 percent or more in relative terms, in a brain that may already be at risk.
The protective act is a setting on the initiation checklist, not a decision made later. Chapter 22 §22.3's rule for the post-arrest patient generalises: set the initial sweep low, leave it, and titrate upward against serial gases toward a target reached over hours. The starting sweep is one of the very few numbers on an ECMO circuit where the safe value is deliberately the wrong value.
And: this applies to any large sweep increase, not only to initiation — for example, a sweep raised to clear a carbon dioxide tension that had been allowed to rise during a weaning trial.
Clinical pearl · Two numbers, written on the chart, at cannulation
- The carbon dioxide tension the patient arrived with. It is the anchor for the whole correction, and it is routinely lost because the pre-cannulation gas is on a different chart, in a different unit, taken by a different team.
- The target for the first four hours — conventionally not a normal value, but a value roughly a third to a half of the way down from the starting point, with the full correction deferred.
If neither number is written, the default is fast correction, because the default sweep setting is the one that corrects fast.
27.8 What arterial saturation should this patient have?
Controversy 2 — What is the arterial oxygen saturation target on venovenous ECMO?
The problem, stated exactly. Four of this book's primary sources give four overlapping but non-identical answers, and none cites evidence for its figure.
- "A minimum target goal SaO₂ of 80% and PaO₂ over 50 mmHg should be achieved for patients on ECMO" — the practical-guide textbook, Chapter 9.
- "During VV ECMO, oxygen saturation above 85% is acceptable as long as oxygen delivery is adequate", with oxygen delivery around three times oxygen consumption — the ISCCM Manual, Chapter 33.
- "The target arterial saturation is 80% to 90%. ECMO flow and hemoglobin are titrated to maintain this level of support" — the physiological review.
- And, from the same review's description of practice: "the flow is decreased until the arterial saturation is 85% to 90%" — a narrower band than the target it states two paragraphs later.
A patient at 82 percent is simultaneously above one threshold, below another, inside a third and outside a fourth.
The case for accepting a low saturation (80 percent and above). Oxygen content and delivery, not saturation, determine whether tissue is oxygenated. A patient at 82 percent saturation with a haemoglobin of 13 g/dL has a higher arterial oxygen content than a patient at 96 percent with a haemoglobin of 9 g/dL. Pushing saturation higher requires either more flow — with its haemolysis, access and cannula costs — or transfusion, with its own costs (Controversy 1). And the alternative to permissive hypoxaemia in the pre-ECMO era was injurious ventilation, which is what the therapy exists to avoid. The whole logic of lung rest depends on tolerating a saturation that would be unacceptable in any other ICU patient.
The case for a higher target. The tolerance is asserted, not demonstrated. No trial has randomised ECMO patients to different saturation targets, and the neurological consequences of days at 80 percent in patients who are frequently also at risk from emboli, haemorrhage and altered cerebral autoregulation are unknown. Chapter 28 inherits a porcine carotid flow deficit that persisted after arterial pressure was restored; permissive hypoxaemia is being applied to a cerebral circulation this book has repeatedly found to behave abnormally on a circuit. And the measurement is least reliable exactly where the target sits: pulse oximetry is accepted as accurate to within 3 to 5 percent above roughly 70 percent saturation and less reliable below it, with a systematic tendency to underestimate at low saturation — so a displayed 80 percent carries an uncertainty band that straddles every threshold above.
What is not in dispute. Three things.
- Delivery, not saturation, is the physiological quantity. Every source that gives a saturation figure qualifies it with a delivery condition, and the ISCCM Manual is explicit that saturation above 85 percent is acceptable as long as oxygen delivery is adequate.
- The ratio matters and this book now holds three versions of it. The physiological review describes the normal, homeostatically maintained ratio of delivery to consumption as 5:1, with about 20 percent of available oxygen extracted at rest. The ISCCM Manual gives a minimum acceptable target on ECMO of around 3:1. And Chapter 6's verified source (Douflé and Katira 2024) argues for titrating to mixed venous saturation and a delivery-to-consumption ratio of at least 2:1 rather than to raw flow. These are not straightforwardly in conflict — one is a normal value and two are floors — but they are routinely quoted as though they were interchangeable, and the gap between them is large: a patient at 2:1 is extracting half the oxygen delivered, two and a half times the normal fraction. The floor a unit adopts should be written down, because 2:1 and 3:1 imply materially different flows and transfusion thresholds in the same patient.
- A saturation is uninterpretable without a haemoglobin. Two sources make delivery the criterion; delivery cannot be computed without it.
Where this book lands.
- Do not set a saturation target. Set a delivery target and read the saturation as one of its inputs. This is Chapter 6's position arriving by a different route — titrate to mixed venous saturation and the delivery-to-consumption ratio, not to raw flow or raw saturation. The bedside form: saturation, haemoglobin, circuit flow, native cardiac output, lactate trend, and pre-membrane (mixed venous on venoarterial) saturation, read together.
- 80 percent is a floor, not a goal. Every source that quotes it quotes it as a minimum. A patient sitting at 80 with a downward trend is not at target; they are at the edge of the range the field is willing to defend.
- Treat a saturation reported below about 85 percent as a number requiring arterial confirmation by co-oximetry, for the measurement reasons in §27.9 — not because the target is wrong but because the instrument is worst there.
- Rising lactate, falling mixed venous saturation, or a widening extraction ratio overrides any saturation number, in either direction. Chapter 25 owns the interpretation; Chapter 54 owns the response.
- State the target, with its delivery condition, in the daily plan. "Sats 80s, acceptable" written without a haemoglobin and a lactate is not a plan; it is a description.
What would settle it. A randomised comparison of arterial saturation targets in venovenous ECMO — for example 80 to 85 percent against 90 to 95 percent — with neurological outcome, transfusion burden, circuit complications and survival as endpoints. No such trial was identified. All four targets above appear to be conventions; two of the four sources state theirs without any supporting citation at the point of statement.
[VERIFICATION REQUIRED] — none of the four targets was traced to a primary study.
Physiology · Why a venovenous ECMO patient is hypoxaemic on purpose, and what sets the number
The ISCCM Manual gives the cleanest account: arterial oxygenation on venovenous support is a combination of mixed venous saturation and intrapulmonary shunt. In the initial phase, with non-functional lungs, the shunt fraction approaches 100 percent and arterial oxygenation approaches the mixed venous value.
Follow that through. Fully saturated blood leaves the membrane, is returned to the right atrium, and mixes with the native venous return that bypassed the circuit. The mixture crosses a lung that is not exchanging. Arterial blood is therefore the post-mixing venous blood, unchanged. Arterial saturation on venovenous ECMO with no native lung function is a mixing ratio, and the ratio is set by extracorporeal blood flow against native cardiac output.
Two consequences follow, both established in Chapter 25 and both worth restating here as gas-exchange facts.
- Arterial saturation is a cardiac output monitor. A febrile, agitated or septic patient with a rising cardiac output will desaturate at a fixed circuit flow, with a perfect membrane. Cooling the patient can raise the arterial saturation more effectively than anything done to the machine.
- There is a ceiling. Even at full circuit flow, if native cardiac output exceeds it, some venous blood must bypass the circuit and some desaturation is unavoidable. A demand for 95 percent saturation in a hyperdynamic patient with no lung function is a demand for a flow the cannulae cannot deliver.
Clinical pearl · The bedside test that separates lung recovery from everything else
The ISCCM Manual records the manoeuvre: lung recovery is assessed daily by raising the ventilator oxygen fraction to 1.0 and checking whether the arterial partial pressure of oxygen improves; a rise in end-tidal carbon dioxide is also taken as lung recovery.
The logic is exact. On venovenous ECMO with a dead lung, the arterial gas is the mixing ratio and is completely insensitive to what the ventilator does. The moment the ventilator's oxygen fraction starts to change the arterial gas, some alveolar units have re-opened and are exchanging. A response to the ventilator is, on this therapy, the definition of lung recovery — and it is a ten-second test.
End-tidal carbon dioxide behaves the same way for a different reason: it reflects transpulmonary blood flow and alveolar ventilation, and Red Book Chapter 28 notes it as a useful surrogate estimate of transpulmonary blood flow on venoarterial support.
27.9 The instruments: pulse oximetry on a circuit
Every argument above assumes that the saturation on the monitor is the saturation in the patient. On ECMO that assumption fails in four independent ways at once, and the failures do not cancel.
Mechanism | Effect | Why it is worse on ECMO |
Reduced pulsatility | Pulse oximetry requires a pulsatile arterial signal to separate arterial from venous and tissue absorbance. Low perfusion states degrade it | Venoarterial support may abolish pulsatility entirely. The ISCCM Manual notes that the tracing may be difficult in patients with severely depressed cardiac function and non-pulsatile flow |
Dyshaemoglobins | Conventional oximeters cannot distinguish the four major haemoglobin species and read carboxyhaemoglobin largely as oxyhaemoglobin | The circuit generates carbon monoxide through haemolysis (§27.10). This is an ECMO-specific source of a dyshaemoglobin that rises in proportion to the problem you are trying to detect |
Low saturation range | Oximetry is accepted as accurate to within 3 to 5 percent above roughly 70 percent saturation and is less reliable below it, with a tendency to underestimate | The targets in §27.8 sit in the 80s, which is where the instrument's accepted error band is widest and where a few percentage points change management |
Skin pigmentation | Systematically higher rates of occult hypoxaemia in patients with darker skin | Demonstrated in an ECMO population specifically — see below |
Evidence · The measured error on ECMO, and an equity finding in this population
The magnitude. This book's practical-guide textbook states that pulse oximetry is typically accurate to within 2 to 3 percent of arterial saturation, but that in patients on ECMO this range is 3 to 7.2 percent, higher and more inaccurate, attributing it to endogenous carbon monoxide from circuit-induced haemolysis combined with less pulsatile flow. Its recommendation follows directly: frequent arterial saturation monitoring is recommended even with normal pulse oximeter readings.
The equity finding. A registry analysis of adults with respiratory failure on ECMO between January 2019 and July 2020 examined 372 paired pulse-oximeter and arterial saturation readings in the range 92 to 96 percent. Among patients whose pre-ECMO pulse oximeter readings were between 92 and 96 percent, 10.2 percent of White patients had an arterial oxygen saturation below 88 percent, compared with 21.5 percent of Black patients — roughly double the risk of occult hypoxaemia. Around three-quarters of the cohort had confirmed COVID-19.
Certainty: low, and the provenance is weak. The registry analysis was read second-hand, through a journal-watch summary in an emergency medicine journal, not from the primary paper. The summarising authors note its own limitations: the mechanism is unknown, it relies on documented race rather than objective skin-tone measurement, and it is not known whether all oximeter devices are affected equally. The 3 to 7.2 percent ECMO error figure is quoted from a textbook citing a single reference that was not retrieved.
What survives: the direction and the clinical instruction, which both sources agree on. The pulse oximeter is less trustworthy on ECMO than off it, and it is least trustworthy in the patients in whom hypoxaemia is most likely to be missed.
Sources: ECMO: A Practical Guide to Management, Chapter 9. Valbuena V et al. Chest 2021, DOI 10.1016/j.chest.2021.09.025, as summarised in Beck S, Honan B, Mallows JL, Ting J. Emergency Medicine Australasia 2022;34(2):303–305. DOI 10.1111/1742-6723.13957.
Danger · Two rules that follow from the oximetry data
- On peripheral venoarterial ECMO, the pulse oximeter probe goes on the right hand or right ear, and the ISCCM Manual says so explicitly — especially when an arterial line on the right cannot be placed. A probe on the left hand or a toe is monitoring the machine's territory and will not fall when the brain desaturates (§27.2).
- Do not use a pulse oximeter to exclude hypoxaemia on ECMO. The textbook recommendation is that frequent arterial saturation measurement is required even when the oximeter reads normally, and the registry data show that a reading in the 92 to 96 percent band is compatible with an arterial saturation below 88 percent in one patient in five in some groups.
The oximeter's legitimate role on a circuit is as a continuous trend and an alarm, not as a measurement. The measurement is co-oximetry on a sample whose site is recorded.
Pitfall · A calculated saturation is not a measured saturation
Blood gas analysers report saturation in two ways and the difference matters here more than anywhere else in the ICU. A calculated saturation is derived from the partial pressure of oxygen using a standard dissociation curve — it assumes normal pH, temperature, 2,3-diphosphoglycerate and, critically, no dyshaemoglobins. A co-oximeter measures the species spectrophotometrically and reports oxyhaemoglobin, deoxyhaemoglobin, carboxyhaemoglobin and methaemoglobin separately.
On ECMO the assumptions behind the calculated value are the specific ones that fail. A patient with a carboxyhaemoglobin of 7 percent has a calculated saturation that is close to normal and a measured oxyhaemoglobin fraction that is not.
Use co-oximetry, and read the dyshaemoglobin fractions rather than only the oxygen saturation line.
27.10 Carboxyhaemoglobin: the dyshaemoglobin the circuit manufactures
This is the most ECMO-specific measurement in the chapter, it is available on most modern point-of-care analysers, and it is largely unused.
Physiology · Why a machine that has no carbon monoxide produces carboxyhaemoglobin
The circuit haemolyses red cells through shear, through negative access pressure and through contact with the pump. Free haemoglobin released into plasma is degraded by haem oxygenase into free iron, biliverdin and carbon monoxide — carbon monoxide is a stoichiometric product of haem catabolism, not a contaminant. The carbon monoxide binds haemoglobin and appears as carboxyhaemoglobin.
Two amplifying features matter. Haem oxygenase-1 is inducible by haem itself, by oxidative stress and by inhaled nitric oxide, so the process is self-perpetuating once it starts. And carboxyhaemoglobin worsens oxygen delivery twice over: it removes binding sites and it shifts the dissociation curve leftward, impairing offloading.
So on ECMO, carboxyhaemoglobin is simultaneously a marker of circuit-induced haemolysis, a cause of impaired oxygen delivery, and a source of pulse-oximeter error — three different problems with one measurement.
Evidence · Carboxyhaemoglobin above 3 percent as a haemolysis signal
In 33 consecutive ECMO patients at one centre over roughly five years, with 110 paired measurements of plasma free haemoglobin and carboxyhaemoglobin:
- Haemolysis was defined as plasma free haemoglobin above 50 mg/dL, following the ELSO reference.
- All carboxyhaemoglobin values were above 3 percent when haemolysis was present. All but two were below 3 percent when it was absent, and both exceptions (4.4 and 7.1 percent) came from a single patient.
- A 3 percent cut-off discriminated haemolysis, Fisher's exact test p = 0.006.
- In the two patients with clinical haemolysis, carboxyhaemoglobin rose in proportion to plasma free haemoglobin, reaching a peak of 9.1 percent in the patient whose free haemoglobin became unmeasurably high.
- Background context from the same source: carboxyhaemoglobin is typically mildly elevated in critically ill patients, around 0.9 to 1.9 percent, and ELSO-reported haemolysis rates are around 5.8 percent on venovenous and 5.6 percent on venoarterial support.
Certainty: very low. Thirty-three patients, single centre, retrospective, only two with haemolysis, no multivariable adjustment — the authors themselves call it hypothesis-generating and call for larger prospective evaluation.
NUMERICAL PROBLEMS FLAGGED AND NOT CORRECTED. The retrieved text mixes units for plasma free haemoglobin. The figure legend defines the threshold as 500 mg/L, equal to 50 mg/dL, consistent with the ELSO reference; but the individual patient values are given as "1600 mg/dL" and "greater than 2000 mg/dL", which would be 16 and 20 g/dL — more free haemoglobin in plasma than most patients have in their red cells. These are almost certainly mg/L values mislabelled, but both forms appear in the source and neither is asserted here. The conclusion used — that carboxyhaemoglobin rises with plasma free haemoglobin and that values above 3 percent were associated with haemolysis — does not depend on the disputed magnitudes.
Source: Kai Man C, Koon Ngai L. Clinical Case Reports 2018;6(2):346–349. DOI 10.1002/ccr3.1351.
Clinical pearl · Read the carboxyhaemoglobin line on every ECMO co-oximetry result
It is already on the printout. It costs nothing. It rises before the plasma free haemoglobin result returns from a laboratory that may not run the assay daily — which is the whole argument the source makes.
A practical reading, with the certainty attached:
- Under about 2 percent: consistent with the background elevation of critical illness.
- Rising, or above about 3 percent, in a patient not exposed to environmental carbon monoxide and not a smoker: look for haemolysis — plasma free haemoglobin, lactate dehydrogenase, haptoglobin, bilirubin, urine colour, and the circuit itself (access pressure, pump speed, transmembrane pressure, visible clot). Chapter 37 owns what to do about it.
- Remember it is also an oximetry-error term (§27.9) and a delivery problem in its own right.
Certainty for the 3 percent threshold: very low — one small retrospective series. It is offered as a prompt to investigate, not as a diagnostic criterion.
Chapter 23 established a parallel obligation for a different assay: neuron-specific enolase requires a haemolysis index with every sample. The two rules have the same root — on a circuit, haemolysis contaminates measurements that are not about haemolysis.
27.11 Reading the pre-membrane and post-membrane pair
The pair is drawn routinely in most units and interpreted narrowly, as a pass-or-fail check on the oxygenator. It carries more than that, and separating the readings is what this section is for.
Observation | Read as physiology (this chapter) | Read as a device diagnosis (Chapter 32) |
Post-membrane saturation 100 percent, high PO₂ | The membrane is delivering everything it can at this blood flow. Any remaining hypoxaemia is a patient or mixing problem, not a device problem | Device normal |
Post-membrane saturation below 100 percent at a flow below rated flow | Oxygen delivery per unit blood is reduced; the flow required for a given delivery has risen | Reduced oxygenator function, usually clot in the blood path. Pair with transmembrane pressure |
Pre-membrane saturation low (40 to 60 percent) | On venoarterial support: high extraction. The delivery-versus-consumption question (Chapter 25). On venovenous support: little recirculation | Not a device finding |
Pre-membrane saturation high (over 80 percent), arterial saturation falling | Recirculation (§27.5), and the transfer loss can be calculated | Not a device finding, though it may be caused by cannula position (Chapter 26, Chapter 34) |
Pre-membrane saturation higher than arterial saturation | Recirculation, with no other possible explanation | Not a device finding |
Carbon dioxide falls appropriately across the membrane; oxygen rises appropriately | Both exchange functions intact | Device normal |
Oxygen rises normally; carbon dioxide barely falls | Carbon dioxide clearance is limited | Suggests water in the gas phase — treat by increasing gas flow before assuming a circuit change is needed (§27.6.1) |
Neither gas changes appreciably across the membrane | No extracorporeal gas exchange is occurring | Membrane malfunction. Both the 2014 monitoring review and the ELSO Red Book give this as the definition: if the expected rise in oxygen and fall in carbon dioxide is not seen, membrane malfunction should be suspected. Also check that the sweep gas line is actually connected and flowing — the commonest cause of no gas exchange is no gas |
Pitfall · The pre-membrane sample is only a mixed venous sample on venoarterial support
Stated twice in this chapter because it is the error that most often turns a correct number into a wrong conclusion.
- Venoarterial: blood returns to the aorta and must traverse the systemic circulation before it can be drained again. The drainage cannula therefore samples genuine mixed venous blood, and the ISCCM Manual endorses it as such.
- Venovenous: blood returns to the venous compartment a few centimetres from where it is drained. Some of what the cannula draws never left the circuit. A pre-membrane saturation of 78 percent on venovenous support may mean a well-perfused patient with low extraction, or a poorly perfused patient with 30 percent recirculation. It does not distinguish them, which is why the arterial comparison in §27.5 exists.
Clinical pearl · Draw the pair and the arterial sample together, or the pair means less
Three sites, one moment. Pre-membrane, post-membrane and right radial arterial, drawn within the same minute, at documented circuit settings, answer between them almost every gas question this chapter raises: is the device working, is there recirculation, what is extraction doing, and what is the brain receiving.
Drawn hours apart they answer none of them, because on this therapy every one of those numbers moves with circuit flow, sweep, cardiac output, temperature and sedation — all of which change between draws.
27.12 A structured blood gas round
Assembled from the sources in this chapter and stated as this book's own synthesis.
Pitfall · This is not the hypoxaemia algorithm — Chapter 9 owns that
Chapter 9 owns the differential for hypoxaemia on established venovenous ECMO: six causes, distinguished by drainage, post-membrane and mixed venous saturations, with its own algorithm, an escalation ladder costed step by step, and findings this chapter does not repeat — including that reducing pump speed may raise arterial saturation when recirculation is the cause.
The tree below is the measurement layer beneath it: which sample to draw next, and what each one excludes. It ends where Chapter 9 begins. Drawn here because the sampling sequence is this chapter's subject and because a differential cannot be worked without the samples that populate it.
Step | What to record | Why |
1. The settings, first, before any result | Extracorporeal blood flow; sweep gas flow; sweep oxygen fraction; pump speed; ventilator settings; temperature; vasoactive infusions | Every number below is conditional on these. Chapters 25 and 26 established the rule for pressure and imaging; it applies identically here |
2. The three gases, same minute, sites named | Pre-membrane, post-membrane, right radial arterial (venoarterial) or arterial (venovenous) | §27.2, §27.11 |
3. Oxygen: is the device delivering? | Post-membrane saturation and PO₂ against the expected values at this sweep oxygen fraction | §27.3. A falling trend at constant settings is the surveillance signal Chapter 32 uses |
4. Oxygen: is the delivery reaching the patient? | Pre-membrane saturation against arterial saturation; haemoglobin; circuit flow against estimated native cardiac output | §27.4, §27.5, §27.8 |
5. Carbon dioxide: is it where it should be, and did it get there at the right speed? | pH before carbon dioxide tension; the trend since the last gas; the sweep setting that produced it | §27.6, §27.7 |
6. The dyshaemoglobins | Carboxyhaemoglobin and methaemoglobin from the co-oximeter | §27.9, §27.10 |
7. The perfusion cross-check | Lactate trend; mixed venous or pre-membrane saturation; urine output; the extraction ratio if computable | A saturation target is defensible only with a delivery condition attached (§27.8). Chapter 25 owns the haemodynamic reading; Chapter 29 owns lactate as a laboratory value |
8. The lung recovery test, once daily | Raise the ventilator oxygen fraction to 1.0 briefly and see whether the arterial gas responds; note end-tidal carbon dioxide | §27.8. On this therapy a response to the ventilator is lung recovery |
9. The sweep weaning question, once daily in stable or improving patients | Can the sweep come down, and what happened when it did | Red Book Chapter 25 treats this as routine care, not an occasional test — because the native lung's contribution is unmeasured and can only be discovered by removing support (§27.6) |
Clinical pearl · The one-line handover
A complete gas-exchange handover on ECMO fits in a sentence and contains six numbers: "Flow 4.2, sweep 4 at 100 percent, arterial sats 88 on the right radial, pre-membrane 62, haemoglobin 9.8, lactate 1.4 and falling."
Compare with what is usually said: "Sats are 88." The second version is not a shorter form of the first. It is a different and much smaller quantity of information, and it cannot be acted on.
27.13 Three situations where the ordinary rules are suspended
Situation | What changes | Owned by |
The first gas after extracorporeal cardiopulmonary resuscitation | Carbon dioxide, oxygen and arterial pressure all move toward maximum cerebral insult within two minutes, by default, because the sweep is at a conventional setting and the blender is still at 1.0 from cannulation. The first right-radial gas is both the most important and the most likely to be omitted. §27.7's correction rule applies at its most stringent | Chapter 22 owns the post-arrest argument; Chapter 19 §19.5 owns reperfusion as an intervention |
The apnoea test for death by neurological criteria | The membrane clears carbon dioxide so efficiently that apnoea alone will not raise it. The sweep must be reduced, not the blood flow — to 0.5 to 1 L/min, with added carbon dioxide if needed, the sweep oxygen fraction set to 1.0, blood flow adjusted to keep oxygenation adequate, and baseline gases collected from both the peripheral cannula and post-membrane, because on venoarterial support not all vascular sites reflect cerebral blood composition. Thresholds: carbon dioxide tension above 60 mmHg, a rise above 20 mmHg, pH below 7.28 | Chapter 24, which owns the determination pathway in full. Reproduced here only because it is the extreme case of §27.6: the sweep is the knob, and a clinician who reaches for blood flow will not raise the carbon dioxide and may injure the patient |
Extracorporeal carbon dioxide removal | The two knobs are deliberately decoupled to the extreme: full carbon dioxide clearance at blood flows of 0.5 L/min or less, with no meaningful oxygenation support. At these flows the sweep is no longer sovereign and blood flow re-enters as a co-determinant (§27.6). A device can remove up to around 25 percent of carbon dioxide production at the low end of the flow range | NOBODY. A tracker query run for this chapter found that no chapter in this book's 92-chapter map owns extracorporeal carbon dioxide removal — Chapter 79 is ECMO with a left ventricular assist device, and Part XII contains no carbon dioxide removal chapter. This is recorded as an unhoused gap; the nearest homes are Chapter 3 (Configurations) or Chapter 78 (Hybrid Configurations). The physiology is retained here because it is the same physiology, and because the trap — normal gases with no oxygenation reserve — is a gas-interpretation trap |
Danger · Reduce the sweep, never the blood flow
Chapter 24 established this rule for the apnoea test and it generalises to every situation in which carbon dioxide must be allowed to rise. Reducing blood flow to raise carbon dioxide reduces oxygen delivery at the same time, and does so disproportionately — because carbon dioxide clearance is relatively preserved at low blood flow (that is the entire basis of carbon dioxide removal devices) while oxygen delivery falls in direct proportion to flow.
A clinician who turns the pump down to raise the carbon dioxide will desaturate the patient long before the carbon dioxide has risen enough to be useful. The sweep is the knob. It is the knob in the apnoea test, in permissive hypercapnia, in weaning, and in any deliberate hypercapnia.
27.14 What we do not know
- There is no randomised evidence for any arterial oxygen saturation target on ECMO. Four primary sources give four overlapping figures and two state theirs without a citation at the point of statement (§27.8).
- The 4-to-8-hour window for correcting hypercapnia is a convention. No source reaching this book states how it was derived, and no trial has compared correction rates with a neurological endpoint. Chapter 22 assembled the supporting physiology and registry signals; the interval itself is unvalidated.
- The proportion of carbon dioxide clearance performed by the membrane is not routinely measured, so the effect of any sweep reduction is uncertain by the Red Book's own admission (§27.6). Most devices are not equipped to measure carbon dioxide removal at all.
- "Standardised venous blood" has at least three published definitions, differing in inlet saturation between 65 and 75 percent, which makes printed rated-flow figures imperfectly comparable between devices (§27.3).
- The transfusion-versus-flow question has never been tested. The recommendation to prefer transfusion rests on a mechanical argument and a risk comparison whose stated harms are a generation out of date (Controversy 1).
- The carboxyhaemoglobin threshold for haemolysis is from 33 patients, two of whom had haemolysis (§27.10). It has not been prospectively validated and the source's own plasma free haemoglobin units are internally inconsistent.
- The magnitude of pulse-oximeter error on ECMO rests on a single cited reference that was not retrieved, and the equity finding in the ECMO population was read second-hand through a journal-watch summary (§27.9). Neither has been independently confirmed for this book.
- The threshold sweep gas flow beyond which additional sweep achieves nothing differs between devices — reported plateaus of 2 L/min, 4 L/min and 4 to 5 L/min in three different settings. There is no transferable number (§27.6).
- Nothing in this chapter addresses temperature correction of blood gases on a circuit whose heat exchanger sets the patient's temperature. The alpha-stat and pH-stat question, which is consequential for cerebral blood flow and which on ECMO is entangled with a machine setting, was not covered by any source retrieved for this chapter and is recorded here as an unclosed gap.
- No source reaching this chapter gives a validated interval for how often gases should be drawn, on either configuration, at any stage.
- A scope gap, found by this chapter and not closed by it: no chapter in this book owns extracorporeal carbon dioxide removal as a therapy. The physiology appears in §27.6 and §27.13 because it is the physiology of every membrane lung, but the indications, devices, evidence and complications of low-flow carbon dioxide removal are currently unassigned. A decision is needed before Part XII is written.
Key points
- On ECMO the two gases have been unbundled onto two knobs that do not interact. Oxygen transfer is limited by haemoglobin presented to the membrane per minute — the levers are blood flow and haemoglobin. Carbon dioxide clearance is limited by gas passing the other side — the lever is sweep gas flow. Every common error in this domain is one gas treated with the other's knob.
- The sweep gas is the on-off switch. The ELSO Red Book: irrespective of extracorporeal blood flow, once the sweep gas ceases, ECMO support effectively ceases. A handover that gives the blood flow and not the sweep has given half the dose.
- A blood gas on ECMO is uninterpretable without its site. There are four distinct samples — pre-membrane, post-membrane, right radial, and distal arterial — and they answer four different questions. A femoral arterial gas on peripheral venoarterial support can be perfect while the brain is hypoxaemic.
- The blender is not the oxygen knob. A working membrane saturates the outlet blood to nearly 100 percent on room air at 1 to 2 L/min of sweep. Raising the sweep oxygen fraction for hypoxaemia adds almost nothing, because there is no unsaturated haemoglobin left to fill. Keep it at 1.0 and preserve the post-membrane partial pressure of oxygen as a surveillance signal.
- Haemoglobin is an oxygen-transfer variable on ECMO in a way it is not on a ventilator. In the worked example, raising haemoglobin from 11 to 14 g/dL removes roughly 0.9 L/min of required blood flow in an 80 kg adult. A falling haemoglobin causes desaturation with a perfect membrane.
- Recirculation is a calculable loss. In the published worked table, a pre-oxygenator saturation rising from 60 to 90 percent cut oxygen transfer from 6.9 to 2.8 mL/dL — a 59 percent loss with the console unchanged.
- Pre-membrane saturation can only exceed arterial saturation in the setting of recirculation. There is no other explanation, and the comparison costs nothing.
- The pre-membrane sample is a true mixed venous sample on venoarterial support and is not one on venovenous support. On venovenous it is a mixture whose composition is the recirculation diagnostic.
- Correct hypercapnia over 4 to 8 hours, not in ten minutes. The sweep will do it in ten minutes if allowed to. Set the starting sweep low on the initiation checklist — one of the few numbers on a circuit where the safe value is deliberately the wrong value.
- Titrate the sweep to pH, not to carbon dioxide tension, because metabolic compensation of chronic hypercapnia is common in this population and a normal carbon dioxide tension in a compensated patient is an alkalosis.
- Reduce the sweep, never the blood flow, when carbon dioxide must be allowed to rise. Reducing flow desaturates the patient long before it raises the carbon dioxide, because carbon dioxide clearance is relatively preserved at low flow and oxygen delivery is not.
- Set a delivery target, not a saturation target. The four published saturation figures — 80, above 85, 80 to 90, and 85 to 90 percent — do not agree and none is traced to a primary study. 80 percent is a floor that every source quotes as a minimum, not a goal.
- Arterial saturation on venovenous ECMO is a mixing ratio, and therefore a cardiac output monitor. Cooling a febrile patient can raise it more effectively than anything done to the machine.
- Lung recovery on venovenous ECMO is defined by a response to the ventilator. Raise the ventilator oxygen fraction to 1.0 for a moment: if the arterial gas moves, alveoli have re-opened. It is a ten-second daily test.
- The pulse oximeter is less accurate on ECMO than off it — a reported error range of 3 to 7.2 percent rather than 2 to 3 — from endogenous carbon monoxide plus reduced pulsatility, and it is least accurate in the 80s where the targets sit. Confirm with co-oximetry; the probe goes on the right side.
- Occult hypoxaemia is unequally distributed. In an ECMO registry cohort with pulse-oximeter readings of 92 to 96 percent, 10.2 percent of White and 21.5 percent of Black patients had an arterial saturation below 88 percent — read second-hand, and enough to change where the burden of proof sits when an oximeter reads reassuringly.
- Read the carboxyhaemoglobin line. The circuit manufactures carbon monoxide by haem catabolism. Above roughly 3 percent, in a non-smoker, it points at haemolysis — and it is simultaneously a cause of impaired delivery and a source of oximeter error.
- A carbon dioxide tension that will not fall despite more sweep, with normal oxygen transfer, suggests water in the gas phase — treated by increasing gas flow, not by changing the circuit.
What this chapter hands to the next
- Chapter 6 (VV ECMO Physiology) is owed two corrections of the record: the rated-flow inlet-saturation discrepancy it opened is now three-way, not two-way, and the anchor physiological review disagrees with itself inside one paper (§27.3); and its Hüfner-constant flag recurs in the same reproduced table (§27.5). Neither changes any conclusion in Chapter 6; both should be recorded there.
- Chapter 28 (Cerebral Monitoring) inherits the carbon dioxide correction-rate argument in its general form, and §27.14's unclosed gap on temperature correction of blood gases, which is a cerebral blood flow question before it is a chemistry question.
- Chapter 29 (Laboratory Monitoring) inherits the dyshaemoglobin panel, the plasma free haemoglobin threshold of 50 mg/dL and its unit ambiguity, and the general rule this chapter shares with Chapter 23: on a circuit, haemolysis contaminates measurements that are not about haemolysis.
- Chapter 30 (Data Interpretation) inherits a new class of defect to sit alongside calibration-versus-discrimination: the same measurement read as physiology and as a device diagnosis, which is the pre-post pair in §27.11 and the reason two chapters can own one number without conflict.
- Chapter 32 (Oxygenator Failure) inherits §27.11 entire, plus the distinction between clot in the blood phase (falling oxygenation, rising resistance, change the circuit) and water in the gas phase (failing carbon dioxide clearance, treated by flushing gas).
- Chapter 37 (Haemolysis) inherits the carboxyhaemoglobin signal, the 3 percent prompt with its certainty label, and the ELSO plasma free haemoglobin reference.
- Chapter 46 (Transfusion Strategy) inherits Controversy 1 and is owed an explicit threshold for this population, with the observation that a transfusion given for oxygen delivery on ECMO will otherwise be audited against a general ICU threshold and look like an error.
- Chapter 53 (Mechanical Ventilation During ECMO) inherits the lung-recovery test, the fact that arterial gases are insensitive to the ventilator when the shunt fraction approaches 100 percent, and the sweep-weaning routine.
- Chapter 9 (Persistent Hypoxaemia on VV ECMO) already owns the six-cause differential and its algorithm. It is owed from here only the arithmetic that quantifies each branch — the recirculation transfer-loss calculation in §27.5, the haemoglobin substitution in §27.3.1 — and the measurement caveats in §27.9 that determine whether the saturation triggering the algorithm was real.
- UNHOUSED — extracorporeal carbon dioxide removal. A tracker query for this chapter established that no chapter in the 92-chapter map owns it. The low-flow physiology in §27.6 and the finding that the sweep stops being the dominant lever at low blood flow are held here until a home is chosen; Chapter 3 or Chapter 78 are the candidates, and the decision is owed before Part XII.
- Appendix 2 and Appendix 3 (Oxygen Content and Delivery Calculators) are owed the two worked calculations in §27.3.1 and §27.5 as their test cases.
References
Primary sources
- Brodie D, Peek G, MacLaren G, et al., eds. Extracorporeal Life Support: The ELSO Red Book, 6th edition. Chapter 25, adult respiratory management — sweep gas weaning as routine care; the statement that once sweep gas ceases support effectively ceases; the sweep-oxygen-fraction argument (a properly functioning membrane saturates outlet blood at a sweep oxygen fraction of 0.21 and 1 to 2 L/min); extracorporeal blood flow of 3 to 6 L/min as a general adult safe range; the pH-rather-than-carbon-dioxide-tension convention; the admission that the membrane's share of total carbon dioxide clearance is not routinely measured. Chapter 28, adult cardiac management — the gas-exchange and oxygen-delivery troubleshooting table, including the membrane-malfunction rule, the transfuse-and-increase-flow responses to low venous saturation with rising lactate, and end-tidal carbon dioxide as a surrogate for transpulmonary blood flow. Chapter 45 — the apnoea test protocol, cited only; Chapter 24 owns it. Verified = Yes (retrieved through the project document store).
- ISCCM Manual of RRT and ECMO in ICU, Chapter 26, Physiology during ECMO Support — rated flow, the inlet-outlet oxygen content difference, the sweep-to-blood ratio figures up to 8:1. Chapter 33, Monitoring during ECMO — metabolic monitoring; the shunt-fraction account of arterial oxygenation on venovenous support; saturation above 85 percent acceptable with adequate delivery; oxygen delivery around three times consumption; the right-arm arterial line and pulse oximeter rule; venous saturation on venoarterial support as a true reflection of venous oxygen content; the daily lung-recovery test. Verified = Yes (retrieved through the project document store).
- Taha AR, Caridi-Scheible M, Leiendecker E, et al. ECMO: A Practical Guide to Management. Chapter 5 — Figure 5.5 (the oxygen and carbon dioxide gradients and the 4-to-8-hour correction rule), Table 5.1 (rated flows and surface areas of five commercial devices), Table 5.2 (the worked recirculation example), and the pre-oxygenator saturation rule for detecting recirculation. Chapter 9 — Table 9.1, the monitoring modalities; the pulse oximetry error of 3 to 7.2 percent on ECMO; the minimum saturation target of 80 percent and partial pressure of oxygen over 50 mmHg; the right radial or brachial arterial line and the mixing cloud. Verified = Yes (retrieved through the project document store).
- ECMO in the Adult Patient (Core Critical Care, 2017) — circuit blood gas sampling, including the instruction to measure circuit gases with the sweep gas at 100 percent oxygen and the warning about sampling pressure; transmembrane pressure paired with post-oxygenator gases for oxygenator surveillance; free haemoglobin as an indicator of circuit stress rising with more negative access pressure; the flow floor of 1.5 L/min through any tubing. Verified = Yes (retrieved through the project document store).
External references — all read as indexed full-text passages unless otherwise stated; none retrieved as a complete article; no identifier independently resolved
- Bartlett RH. Physiology of extracorporeal gas exchange. Comprehensive Physiology 2020;10(3):879–891. DOI 10.1002/j.2040-4603.2020.tb00130.x. The anchor external source: rated flow and its two internally inconsistent definitions; the outlet-inlet content difference; the oxygen kinetics including delivery at five times consumption; the sweep-to-blood ratio account and the 500 mL/min versus 6 L/min contrast; the two membrane failure modes (clot in the blood phase, water in the gas phase); the 80 kg worked example giving 4.1 L/min; the target saturation of 80 to 90 percent; the transfusion-over-flow recommendation quoted in Controversy 1; and the carboxyhaemoglobin terminology discrepancy flagged in §27.6. Verified = No.
- Chung M, Shiloh AL, Carlese A, Butt WW, Spapen H. Monitoring of the adult patient on venoarterial extracorporeal membrane oxygenation. The Scientific World Journal 2014. DOI 10.1155/2014/393258. The variable-by-variable gas exchange table adapted in §27.1; carbon dioxide transferring roughly ten times more efficiently than oxygen; the need for carbon dioxide-enriched sweep gas to prevent hypocapnia; the pre-versus-post membrane malfunction rule; circuit flow of 50 to 75 mL/kg/min for adults. Verified = No. NOTE: this article was returned twice in the search record under two different identifiers — 10.1155/2014/393258 and a compound form 10.1155/2014/393258@10.1155/8086.si.571070 — with identical text. The discrepancy is recorded and the plain identifier is used.
- Arens J, Belliato M, Broman LM, Clauser JC, Donker DW, Lorusso R, Malfertheiner M, Rajsic S, Swol J, Vercaemst L, Wittberg LP, et al. Standardization of in vitro evaluation of extracorporeal life support devices for research and development. Artificial Organs 2026;50(5):674–688. DOI 10.1111/aor.70124. The ISO 7199 inlet condition of 65 ± 5 percent saturation; the minimum exchange requirement of 50 mL oxygen per litre of blood at haemoglobin 120 g/L from 65 to 95 percent saturation; the respiratory-quotient argument giving a minimum of 40 mL carbon dioxide per litre for an adult mixed diet; the recommendation to report gas exchange as millilitres of gas per litre of blood across a range of flows. Verified = No.
- Qian Z, He H, Wang Y, Geng S, Li Y, Yuan X, Zhang R, Yang Y, Qiu H, Liu S, Liu L. Evaluation of CO₂ removal rate of ECCO₂R for a renal replacement therapy platform in an experimental setting. Artificial Organs 2024;48(6):586–594. DOI 10.1111/aor.14718. Twelve pigs; the interaction between sweep and blood flow at low flows; the removal figures at 4, 6 and 10 L/min of gas flow at blood flows of 200 and 350 mL/min; the finding that post-membrane carbon dioxide tension was similar regardless of pre-membrane tension (p = 0.08); and the device-dependent sweep thresholds quoted from three other studies. Verified = No.
- Camporota L, Barrett N, Guervilly C. Current applications for the use of extracorporeal carbon dioxide removal in critically ill patients. BioMed Research International 2016. DOI 10.1155/2016/9781695. The physiological separation of oxygenation and carbon dioxide removal; the arithmetic that 1 L of blood carries roughly 500 mL of carbon dioxide against a production of 200 to 250 mL/min; the biphasic removal kinetics (dissolved first, then bicarbonate); the statement that carbon dioxide removal devices typically clear up to about 25 percent of production. Verified = No.
- Dzierba AL, Abrams D, Muir J, Brodie D. Ventilatory and pharmacotherapeutic strategies for management of adult patients on extracorporeal life support. Pharmacotherapy 2019;39(3):355–368. DOI 10.1002/phar.2230. The determinants of oxygen delivery through the circuit; recirculation as a source of inefficiency; the statement that blood flow is a lesser determinant of carbon dioxide removal except at the low end of the range, where it becomes a major one. Verified = No.
- Kai Man C, Koon Ngai L. Endogenous carbon monoxide production in extracorporeal membrane oxygenation-related hemolysis: potential use of point-of-care CO-oximetry carboxyhemoglobin to detect hemolysis. Clinical Case Reports 2018;6(2):346–349. DOI 10.1002/ccr3.1351. Thirty-three patients, 110 paired measurements; the 3 percent carboxyhaemoglobin cut-off (p = 0.006); the peak of 9.1 percent; the haem oxygenase mechanism; the background critical-illness range of 0.9 to 1.9 percent; ELSO haemolysis rates of 5.8 percent venovenous and 5.6 percent venoarterial. Verified = No. The plasma free haemoglobin unit inconsistency is flagged in §27.10.
- Valbuena V, et al. Chest 2021. DOI 10.1016/j.chest.2021.09.025. The ECMO registry analysis of occult hypoxaemia by race: 372 paired readings in the 92 to 96 percent band; 10.2 percent of White and 21.5 percent of Black patients with an arterial saturation below 88 percent. Verified = No, and read SECOND-HAND ONLY, through the journal-watch summary at reference 13. The author list, methods and exact cohort definition were not seen. This is the weakest provenance of any substantive figure in this chapter and is labelled as such in the text.
- Beck S, Honan B, Mallows JL, Ting J. From other journals, April 2022. Emergency Medicine Australasia 2022;34(2):303–305. DOI 10.1111/1742-6723.13957. The journal-watch column through which reference 12 was read, including its own caveats about mechanism, documented race versus skin tone, and device heterogeneity. Verified = No.
- Syabbalo N. Measurement and interpretation of arterial blood gases. International Journal of Clinical Practice 1997;51(3):173–176. DOI 10.1111/j.1742-1241.1997.tb11419.x. The general oximetry limitation table; the statement that oximetry is accurate within 3 to 5 percent above roughly 70 percent saturation and less reliable below; the point that current oximeters cannot distinguish the four major haemoglobin species. Verified = No. A 1997 general review, used only for measurement principles that have not changed.
- Kiekkas P, Alimoutsi A, Tseko F, Bakalis N, Stefanopoulos N, Fotis T, Konstantinou E. Knowledge of pulse oximetry. Journal of Clinical Nursing 2013;22(5-6):828–837. DOI 10.1111/j.1365-2702.2012.04151.x. Used only for its literature review of oximetry limitations: low-perfusion states, dyshaemoglobins, the tendency to underestimate below 70 percent saturation, and the 30-to-60-second response delay at a fingernail site against under 10 seconds at a central site. Verified = No.
- Beasley R, Chien J, Douglas J, et al. Thoracic Society of Australia and New Zealand oxygen guidelines for acute oxygen use in adults. Respirology 2015;20(8):1182–1191. DOI 10.1111/resp.12620. Used only for its statement of variable pulse oximetry accuracy in acutely ill patients with wide limits of agreement in both directions, and its list of interfering factors. Not an ECMO document and no ECMO recommendation is taken from it. Verified = No.
Inherited from an earlier chapter and cited rather than re-retrieved
- Douflé G, Katira BH, et al. American Journal of Respiratory and Critical Care Medicine 2024. Retrieved and marked Verified = Yes by Chapter 6, which took from it the position that extracorporeal support should be titrated to mixed venous saturation and a delivery-to-consumption ratio of at least 2:1, not to raw flow. Cited in Controversy 2 as the third of this book's three delivery-to-consumption figures. Not re-retrieved for this chapter; its verification status is Chapter 6's.
Named in the text and not retrieved
- The primary reference behind the 3 to 7.2 percent pulse oximetry error on ECMO, cited by the practical-guide textbook as its reference 32. Not identified. This is the highest-value outstanding retrieval for this chapter, because the figure is quoted throughout the field.
- ISO 7199 itself, the standard defining oxygenator test conditions, known only through the 2026 standardisation document.
- The Valbuena 2021 primary paper (reference 12), read only through a journal-watch column.
- The three device-specific sweep-gas plateau studies cited by Qian 2024 (plateaus at 2 L/min, 4 L/min and 4 to 5 L/min). Not retrieved; no transferable threshold is asserted.
- Any source on temperature correction of blood gases during extracorporeal support — the alpha-stat and pH-stat question. Nothing was retrieved, and §27.14 records it as an unclosed gap rather than an omission.
Chapter status
Published 20 September 2026. Third chapter of Part V. Chapter 27 of 92.
What was verified. The ELSO Red Book Chapters 25, 28 and 45, the ISCCM Manual Chapters 26 and 33, ECMO: A Practical Guide to Management Chapters 5 and 9, and ECMO in the Adult Patient were retrieved through this project's document store and are marked Verified = Yes. Every external reference was read as an indexed full-text passage, and one — the occult hypoxaemia registry analysis — was read second-hand through a journal-watch summary and is labelled as such wherever it appears. No digital object identifier was independently resolved. Every database row seeded from this chapter is marked Verified = No.
Numerical and bibliographic problems flagged and not corrected.
- The rated-flow inlet saturation discrepancy is now three-way. Chapter 6 recorded 70 percent (ELSO Red Book, ISCCM Manual) against 75 percent (practical-guide textbook). This chapter adds that the anchor physiological review contains both 70 and 75 percent internally, in the same paper, and that the 2026 standardisation document cites ISO 7199 for 65 ± 5 percent (§27.3). All three printed; none asserted.
- The Hüfner constant discrepancy flagged by Chapter 6 recurs, in the recirculation table reproduced at §27.5, which requires approximately 1.36 mL/g plus dissolved oxygen where §27.3.1 uses 1.34 mL/g. Printed unaltered rather than silently recalculated, and the reason stated in the text.
- Three delivery-to-consumption ratios now stand in this book — a normal of 5:1, an ECMO floor of 3:1, and Chapter 6's verified source arguing for at least 2:1. Separated explicitly in Controversy 2 because they are routinely quoted as interchangeable.
- Four non-identical arterial saturation targets across four primary sources — 80 percent minimum, above 85 percent, 80 to 90 percent, and 85 to 90 percent — none traced to a primary study, labelled [VERIFICATION REQUIRED] in Controversy 2.
- A terminology discrepancy in the anchor physiological source, which names haemoglobin-bound carbon dioxide "carboxyhemoglobin" where the correct term is carbamino-haemoglobin (§27.6). Possibly an indexing artefact; recorded, and the substantive claim it sits in is unaffected and used.
- Internally inconsistent plasma free haemoglobin units in the carboxyhaemoglobin study — a threshold given as 500 mg/L equal to 50 mg/dL alongside patient values of 1600 mg/dL and over 2000 mg/dL, which would be 16 and 20 g/dL (§27.10). Both forms printed; neither asserted.
- A duplicated identifier for the 2014 monitoring review, returned under both a plain and a compound digital object identifier with identical text (reference 6).
- A SCOPE GAP, not a numerical one: no chapter in this book owns extracorporeal carbon dioxide removal, established by tracker query while writing §27.13. Recorded in §27.14 and in the handover.
Running total of flagged published numerical or bibliographic problems: Ch 15 (1), Ch 17 (3), Ch 18 (2), Ch 19 (3), Ch 24 (6), Ch 26 (4), Ch 27 (7, plus one scope gap).
Audits passed: clinical, physiology, evidence, citation, numerical, safety, contradiction, redundancy, bedside utility, literature update.