🫁

Chapter 32 — Oxygenator Failure

❓

The question this chapter answers

A membrane lung does not stop working. It gets worse at working, over hours to days, while the numbers on the console stay inside the range somebody chose. So the bedside question is not has the oxygenator failed? — by the time the answer is yes the decision has been made for you. The question is how much reserve is left in this device, which of its two jobs is being lost first, and is what I am seeing the membrane at all.

This chapter is about the membrane lung as a gas exchanger: what it is doing, how to measure it, the four things that degrade it, and when the loss of gas exchange is itself the reason to change the circuit.

🗓️

Evidence search date: 23 September 2026. Two sources were read as complete full text and are the only Verified sources in this chapter: the clinical anchor (van Minnen and colleagues) and the physiology anchor (Park and colleagues). Primary sources are the ELSO Red Book 6th edition (the circuit chapter, the physiology chapter and the circuit malfunction and crisis management chapter), ECMO in the Adult Patient (circuit monitoring, transmembrane pressure and oxygenator change), the ISCCM manual (circuit hardware and the monitoring protocol), ECMO: A Practical Guide to Management (rated flow, the oxygenator comparison table and the recirculation worked example), and ECPR and Resuscitative ECMO (circuit nomenclature).

What this chapter covers — and what it does not

This chapter owns	This chapter does not own
The membrane lung as a gas exchanger — what determines oxygen transfer and carbon dioxide transfer, how each is measured, and how each degrades	Thrombus as an object. Chapter 31 owns clot — where it forms, the five channels that detect it, and the system-exchange procedure
The gas-exchange trigger for a circuit change, which Chapter 31 explicitly assigned here — the numbers, their denominators, and their limits	The haemolysis trigger, which Chapter 37 owns, and the coagulopathy trigger, which Chapter 31 owns. Each chapter owns its own trigger; Chapter 31 owns how the exchange decision is taken and how the exchange is done
The transmembrane pressure drop as a diagnostic — what it measures, what it must be divided by, and what it cannot predict	Pressure as a thrombus signal in its own right (Chapter 31 §31.4), and pump inlet and outlet pressures, which Chapter 33 owns
Non-thrombotic membrane failure — water vapour and condensation, plasma leak, interrupted sweep gas, and heat exchanger failure	Air in the circuit. Chapter 35 owns air and the circuit emergencies
The interpretation of pre-membrane and post-membrane blood gases as a statement about the DEVICE	The interpretation of the patient's arterial gas. Chapter 27 owns blood gas and oxygenation monitoring and this chapter uses it without re-deriving it
Why a measured oxygen transfer can fall in a healthy membrane	Recirculation itself. Chapter 6 owns it and Chapter 9 owns persistent hypoxaemia on venovenous support. This chapter owns only the consequence for judging the device
Nothing about how much support this patient needs	Chapter 8 owns the daily round and Chapter 10 owns liberation. This chapter states what the device can still do, not what should be asked of it

32.1 What the membrane lung actually does, and the two numbers that describe it

A membrane lung is a box with two blood ports and two gas ports. Blood flows outside thousands of hollow fibres; sweep gas flows inside them; oxygen and carbon dioxide diffuse across the fibre wall down their partial-pressure gradients. Everything that can go wrong with it is a change in one of three things: the surface available for diffusion, the resistance of the blood path, or the gas on the other side.

🧠

The structural claim of this chapter · Both numbers that describe a membrane lung are ratios in disguise

A membrane lung is judged at the bedside by two quantities, and neither is interpretable as the number the monitor displays.

The pressure drop across the device is the difference between the pressure entering it and the pressure leaving it. It rises when resistance rises — which is what clot does — but it also rises when flow rises through a perfectly clean membrane. The quantity that carries information is the pressure drop divided by the blood flow that produced it.
The oxygen transfer is the outlet minus inlet oxygen content difference, multiplied by the flow. It falls when the membrane fails — but it also falls when the inlet blood arrives already oxygenated, which is exactly what recirculation does. The quantity that carries information is the transfer set against the content difference it was offered.

So the same error appears twice: a number about the device is read without the denominator that makes it about the device. The consequence is symmetrical and it runs in both directions. A unit that reads the raw pressure drop will exchange oxygenators that are working. A unit that reads the raw oxygen transfer will keep oxygenators that are not.

This is Chapter 30 §30.2's rule applied to a single component: a set value and its measured partner belong on the same line, and the informative quantity is the relationship between them. Here the pairs are pressure drop with blood flow, and oxygen transfer with inlet saturation.

🧮

The arithmetic, once, so that later sections can use it

Oxygen transfer. The oxygen supplied by a membrane lung is the outlet minus inlet oxygen content difference, times the blood flow. The ELSO Red Book puts the normal outlet-minus-inlet difference at 5 cc per dL, and works the example: a 70 kg adult with no native lung function, circuit flow 4 L/min, haemoglobin 10.5 g/dL, drainage saturation 64 percent. The inlet content is 9 cc/dL and the outlet content at 100 percent saturation is 14 cc/dL, so the difference is 5 cc/dL; at 40 dL/min that is 200 cc of oxygen per minute — which happens to be this patient's whole oxygen consumption.

Two things follow from that single example and both matter clinically.

The outlet content is set almost entirely by the haemoglobin, because the outlet is already fully saturated. A membrane lung cannot make more oxygen-carrying capacity; it can only fill what arrives.
The inlet content is set by the patient. The same device, on the same day, at the same flow, delivers a different oxygen transfer if the venous saturation changes — and the direction is counter-intuitive: the sicker the venous blood arriving, the higher the measured transfer.

Carbon dioxide transfer. The carbon dioxide cleared is the inlet minus outlet carbon dioxide content difference. At a gas-to-blood flow ratio of 1:1 it is about the same as the oxygen transfer. Raise the ratio and much more carbon dioxide can be removed — the Red Book takes the example as high as 8:1 — which is why a device used mainly for carbon dioxide removal can work at a much lower blood flow than one used for oxygenation.

Rated flow is the summary of a device's oxygenating capacity: the flow of venous blood that still exits at 95 percent saturation. Below rated flow the outlet is fully saturated and the transfer is set by the content difference; above it the blood passes too quickly for all the red cells to be oxygenated and the outlet saturation falls.

A numeric disagreement between two of this book's own primary sources, recorded rather than resolved. The ELSO Red Book defines the standard venous blood used to specify rated flow as haemoglobin 12 g/dL and saturation 70 percent. ECMO: A Practical Guide to Management gives the same definition as haemoglobin 12 g/dL and saturation 75 percent. The two cannot both be the standard, the difference changes the rated flow a manufacturer may quote, and neither source is followed here in preference to the other. Chapter 29's convention for numeric disagreements between primary sources applies: reproduce both, assert neither, and owe the resolution to whichever chapter can retrieve the source. [VERIFICATION REQUIRED — the primary standards document defining rated flow was not retrieved.]

💡

Clinical pearl · What a healthy membrane looks like, in numbers, so that a sick one is recognisable

Post-membrane oxygen tension. ECPR and Resuscitative ECMO states that modern membrane lungs are efficient enough that the oxygen tension of returning blood is usually over 500 mmHg at a delivered oxygen fraction of 1.0. That is a useful expectation and it is not a universal one: in the physiology anchor, a clean polymethylpentene device in five animals produced post-membrane oxygen tensions of 422 (341 to 463) mmHg at baseline and 354 (188 to 441) mmHg after multiple organ failure was induced, varying with blood flow. The number to know is not a textbook value but this circuit's own value on the day it was primed.
Post-membrane saturation. ECMO in the Adult Patient gives the simplest rule in the field: blood that is less than fully saturated on 100 percent oxygen may indicate a failing oxygenator. Below rated flow there is no other reason for it.
The pressure drop. The ISCCM manual's unit protocol targets a pre-membrane to post-membrane gradient of under 30 mmHg, monitored continuously, with pre-membrane and post-membrane gases every 8 hours. That figure comes from a paediatric protocol and is reproduced as that protocol's own target, not as a general threshold.
And the methodological rule that makes any of these comparable across days. ECMO in the Adult Patient is explicit: blood gases obtained from the circuit are measured with the sweep gas set at 100 percent oxygen. A post-membrane gas drawn on a blended sweep is not comparable with yesterday's, and the fall it shows is the blender, not the membrane.

32.2 The order in which the two gases fail, and why the order is the diagnosis

🧠

Physiology · The two gases do not have the same safety margin, and that is why they fail in an order

The Red Book states the gradients in one line. With 100 percent oxygen as the sweep, oxygen diffuses down a gradient of about 600 to 40 mmHg. Carbon dioxide diffuses down a gradient of about 45 to 0 mmHg. Carbon dioxide nevertheless keeps up, because its solubility and diffusivity are so much greater that the amounts of the two gases exchanged are roughly equal when the gas-to-blood flow ratio is 1:1.

Read that as a statement about reserve and the clinical consequence appears.

Oxygen crosses with an enormous gradient to spare. Below rated flow the outlet blood is fully saturated, and it stays fully saturated while a substantial fraction of the membrane is lost. A membrane lung has to be badly damaged before its post-membrane oxygen tension moves.
Carbon dioxide crosses on a gradient of a few tens of mmHg, and it depends on that gradient being maintained by continuously flushing the gas side. Anything that occupies the gas side, wets the fibre, or removes surface takes carbon dioxide clearance away first.

The Red Book makes exactly this point for the commonest non-thrombotic failure. Water vapour condensing inside the fibres at low gas flow rates "can limit gas exchange in a similar fashion to plasma leakage in the PP devices, particularly because the CO2 diffusional gradient is much lower than O2." The ISCCM manual lists water vapour in the membrane under the determinants of carbon dioxide removal and not under the determinants of oxygenation.

So the order carries information about the cause, and this is the chapter's second structural claim.

Carbon dioxide clearance falling with a normal pressure drop, over hours, at a low sweep points at the gas side — water, condensation, or a sweep that is simply too low for the blood flow. Much of this is reversible without changing anything.
Carbon dioxide clearance falling together with a rising pressure drop, and eventually with a falling post-membrane oxygen, points at the blood side — clot occupying the fibre bundle. This is not reversible and Chapter 31 owns what to do about it.
A fall in post-membrane oxygen is not an early warning. It is a late one, and a unit that waits for it has waited too long.

The honest limit on the order. In the Chapter 31 anchor cohort, acute oxygenator thrombosis presented with both channels moving — a 53 percent fall in outlet oxygen tension in 8 of 11 patients in whom gas transfer had deteriorated, alongside an outlet carbon dioxide of 40 (32 to 48) mmHg despite sweep flows of 10 L/min or more. The order is a tendency with a mechanism behind it, not a law, and a large enough clot takes both at once. Certainty: high for the gradient physiology; moderate that carbon dioxide gives way first; low as a rule for discriminating cause in an individual patient.

⚠️

Danger · The most important sentence in this chapter is about when NOT to suspect the membrane

The ELSO Red Book states the discriminator that saves circuits: "Since a decline in ML gas exchange efficiency usually occurs over a period of hours to days, abrupt changes suggest an interruption of fresh gas flow to the ML."

An abrupt loss of gas exchange is a gas-supply problem until proved otherwise, and the proof takes thirty seconds. The Red Book gives the sequence: assess the gas supply line starting at the membrane lung and proceed to the wall or the tank. If the gas is from the wall, move it to an alternative source to exclude a hospital supply problem. More often the source is a compressed oxygen cylinder that has been exhausted.

And it names the scenario, which every unit will recognise: a patient returning from an intrahospital transport on cylinder gas, where the operator does not put the gas line back on the wall. The bottle runs for a variable time and then the patient deteriorates abruptly. This is avoided by a checklist on return from any venture outside the room, and by nothing else.

Three further causes of a sudden loss of sweep that are not the membrane: a disconnected or kinked sweep line, a blender or flowmeter knocked to zero, and a gas outlet occluded by condensate. Every one of them is corrected in under a minute, and every one of them looks exactly like a failing oxygenator on the blood gas.

The rule this book adopts: before anything else is done about a loss of gas exchange, put a hand on the gas line and follow it to its source. Chapter 90 owns the emergency sequence; this belongs here because it is the commonest reason an oxygenator is wrongly blamed.

32.3 Carbon dioxide: the sweep, the dead space, and the ceiling

Carbon dioxide clearance is the function that is titrated at the bedside every day, by one knob, in one direction. It is also the function with a ceiling that almost nobody is taught, and the ceiling is where the failing membrane hides.

🔬

Evidence · What actually determines carbon dioxide transfer, measured rather than assumed

Park M, Costa ELV, Maciel AT, Silva DP, Friedrich N, Barbosa EVS, Hirota AS, Schettino G, Azevedo LCP. "Determinants of Oxygen and Carbon Dioxide Transfer during Extracorporeal Membrane Oxygenation in an Experimental Model of Multiple Organ Dysfunction Syndrome." PLoS One 2013;8(1):e54954. DOI 10.1371/journal.pone.0054954, PMC3558498. Read as complete full text. Five pigs on venovenous ECMO with a polymethylpentene membrane, a predefined sequence of blood flows (0.5, 1.5, 3.0 L/min) and sweep flows giving gas-to-blood ratios of 1:2, 1:1 and 2:1, repeated before and after lung lavage and faecal peritonitis produced multiple organ failure. 45 measurements in each condition. Carbon dioxide transfer was measured the reference way, from the gas exhaled by the membrane.

Carbon dioxide transfer had four independent determinants (mixed generalised linear model, backward elimination, pseudo-R squared 0.89):

Variable	Coefficient, per unit	p	What it means at the bedside
Sweep gas flow	plus 32.9 mL/min per L/min	under 0.001	The expected knob, quantified
Blood flow	plus 0.017 mL/min per mL/min	under 0.001	Blood flow is not neutral for carbon dioxide
Pre-membrane carbon dioxide tension	plus 1.19 mL/min per mmHg	under 0.001	A hypercapnic patient unloads more per pass
Haemoglobin	minus 3.48 mL/min per g/dL	0.042	Less haemoglobin, more carbon dioxide carried as bicarbonate and more easily released

Two of these deserve to be read twice.

The effect of blood flow on carbon dioxide transfer was roughly 1.9 times that of gas flow. Carbon dioxide removal is not the sweep's private business. Raising the blood flow raises it; lowering the blood flow lowers it, which is precisely what happens during a weaning trial and during every intervention that reduces flow for another reason.
A 10 mmHg rise in pre-membrane carbon dioxide raised transfer by about as much as an extra 0.5 L/min of sweep. The authors note the implication directly: this supports permissive hypercapnia as a way of optimising extracorporeal carbon dioxide removal. It also means a patient who is already well controlled gives the membrane less to work with, so a device near its limit will declare itself first in the patient whose carbon dioxide is normal.

Temperature and pre-membrane oxygen saturation dropped out of the carbon dioxide model entirely.

Certainty: moderate. Five animals, a controlled protocol, an explicit statistical model with variance inflation factors reported, and effects large enough to matter — but an animal model with a single device, and the coefficients are not transferable as numbers to a human circuit. What transfers is the rank order of the determinants and the fact that blood flow outweighs gas flow.

🧮

The ceiling · A membrane lung has dead space, and its dead space fraction is set by the sweep-to-blood ratio

Every alveolus that is ventilated but not perfused is dead space. A hollow fibre is no different, and a fibre bundle flushed faster than the blood can deliver carbon dioxide to it is a bundle of dead space.

This was measured directly. Zakhary B, Sheldrake J, Pellegrino V. "Extracorporeal membrane oxygenation and V/Q ratios: an ex vivo analysis of CO2 clearance within the Maquet Quadrox-iD oxygenator." Perfusion 2020;35(1 suppl):29–33. DOI 10.1177/0267659120906767. Abstract read; full text not retrieved. A closed blood-primed ex vivo circuit with two Quadrox-iD oxygenators in series on a Rotaflow pump, the first used to set the carbon dioxide tension delivered to the second, sweep flows of 100 percent oxygen varied to give different ratios of gas flow to blood flow. Exhaust carbon dioxide was measured directly and carbon dioxide transfer and dead space fraction calculated.

The findings, as reported in the abstract.

Carbon dioxide transfer rose with gas flow and then plateaued above a gas-to-blood ratio of about 4.0.
Exhaust carbon dioxide rose linearly with the carbon dioxide tension delivered to the device, with the slope of that line falling at high ratios.
Oxygenator dead space fraction was 0.3 to 0.4 at lower ratios and rose to 0.8 to 0.9 at ratios above 4.0.

What that means with a hand on the flowmeter. Below a gas-to-blood ratio of roughly four, turning the sweep up does what it is supposed to do. Above it, most of the gas leaving the device never touched blood that had carbon dioxide left to give, and the carbon dioxide clearance has stopped responding. The knob still turns. The number on the flowmeter still rises. Nothing further is bought.

The three consequences this chapter draws.

A sweep that has risen past about four times the blood flow is a finding, not a setting. At 4 L/min of blood flow that is a sweep of 16 L/min, which few units reach; at 2 L/min of blood flow it is a sweep of 8 L/min, which many do. The ratio is what matters, and it is not displayed anywhere.
The device most likely to be at its ceiling is the one running at low blood flow — a weaning trial, a carbon dioxide removal configuration, a patient whose flow was reduced for drainage problems. That is the setting in which the ceiling will be met first, and the setting in which a rising carbon dioxide is most likely to be misread as a failing membrane.
Raising the blood flow is a carbon dioxide intervention and, in the physiology anchor, a more powerful one per unit than raising the sweep. Where the patient allows it, it is the move that is left when the sweep has stopped working.

Certainty: moderate for the dead space physiology, which is mechanistically secure and directly measured; low for the specific ratio of 4.0, which comes from one device in one ex vivo circuit read from an abstract, and which should not be transferred to another oxygenator as a number. [VERIFICATION REQUIRED — full text.]

💡

Clinical pearl · The earliest sign a membrane lung is losing capacity is free, and it is not on the blood gas

The carbon dioxide tension is a controlled variable. It is held where the team wants it by the sweep, so it stays normal while the device deteriorates — and it moves only when the sweep can no longer hold it. That is why van Minnen and colleagues, whose cohort is the clinical anchor of this chapter, describe a deteriorating gas exchange as "a late marker for functional decline of the oxygenator."

The uncontrolled variable is the effort it takes. This book therefore asks for the pair, which is Chapter 30 §30.2's rule in its most useful single instance:

Record the sweep flow and the arterial carbon dioxide tension on the same line, with the blood flow beside them, every time.

A sweep that has crept from 3 to 6 L/min over four days to hold the same carbon dioxide, at the same blood flow, in a patient whose metabolic rate has not risen, is a membrane finding — and the carbon dioxide never moved. A chart that records only the carbon dioxide cannot show it. A chart that records only the sweep cannot show it either.

Three things must be excluded before the trend is called a membrane finding, and all three are commoner than membrane failure: the patient is producing more carbon dioxide (fever, agitation, sepsis, overfeeding, a resolving sedation); the blood flow has fallen, and over the ranges the physiology anchor studied the effect of blood flow on carbon dioxide transfer was roughly 1.9 times that of gas flow; and the native lung has got worse, which on venovenous support removes the share of clearance it was contributing.

This is reasoning, and it is labelled as such. Certainty: low as a validated monitor. The physiology is secure and the direction is supported by the anchor's own statement that gas exchange deterioration is late, but no study retrieved for this chapter has evaluated the sweep requirement as a monitored variable, with a threshold, against oxygenator failure. It is offered as a free observation that costs one extra column, not as a test with operating characteristics.

32.4 Oxygen: why a falling oxygen transfer is not a failing membrane until the inlet is checked

🔬

Evidence · What determines oxygen transfer, and the one variable that does not

From the physiology anchor (Park and colleagues, full text read), the oxygen transfer model had three independent determinants (pseudo-R squared 0.81):

Variable	Coefficient, per unit	p	Direction
Blood flow	plus 0.066 mL/min per mL/min	under 0.001	More flow, more transfer
Pre-membrane oxygen saturation	minus 1.67 mL/min per percent	under 0.001	Better blood arriving, LESS transfer
Pre-membrane carbon dioxide tension	minus 0.96 mL/min per mmHg	0.001	Unexplained by the authors, see below

Sweep gas flow, temperature and haemoglobin all dropped out of the oxygen model. Stratified by the ratio of sweep to blood flow, the oxygen transmembrane gradient and the oxygen transfer were the same at every ratio tested — the authors state there was no effect of gas flow on oxygenation within the range studied. Across the blood flow range, oxygen transfer rose about six-fold from 0.5 to 3.0 L/min, while carbon dioxide transfer rose about two and a half fold.

This is Chapter 27's rule arrived at from the opposite direction. Chapter 27 established that the blender is not the oxygen knob because a working membrane saturates outlet blood on very little oxygen. Here the same conclusion falls out of a measurement: turning the sweep up changes carbon dioxide and does not change oxygen transfer. Two independent lines, one rule. Blood flow is the oxygen knob.

One finding the authors could not explain, reproduced as printed. Higher oxygen transfer at lower pre-membrane carbon dioxide runs against the expected direction, because a low carbon dioxide raises haemoglobin's affinity for oxygen. The authors say plainly that they were unable to bring forth a physiological explanation, offer a speculative one, and concede they cannot exclude residual confounding with blood flow, since lower pre-membrane carbon dioxide values occurred at higher blood flows. A source that prints an unexplained coefficient and says so is the kind this book prefers; the coefficient is reproduced here and is not built on.

⚠️

Danger · The exchange threshold everyone quotes has a denominator, and the denominator is the patient

The ELSO Red Book gives the field's one numeric gas-exchange trigger, in its circuit malfunction chapter: "Membrane lung efficiency is determined by calculating oxygen transfer. When oxygen transfer is below 150 ml/min, ML exchange should be considered."

That is a useful number and it is not a property of the device. Oxygen transfer is the outlet minus inlet content difference times the flow, and two of those three terms belong to the patient.

Worked from the Red Book's own method, with its own example patient at 4 L/min.

Situation	Inlet content	Outlet content	Difference	Transfer at 4 L/min
Red Book example: haemoglobin 10.5 g/dL, inlet saturation 64 percent	9.0 cc/dL	14.0 cc/dL	5.0 cc/dL	200 mL/min
Same membrane, recirculation raises inlet saturation to 90 percent (the Practical Guide example, at haemoglobin 10)	12.4 cc/dL	15.2 cc/dL	2.8 cc/dL	112 mL/min
Same membrane, anaemia to haemoglobin 7 g/dL, inlet saturation 64 percent	6.0 cc/dL	9.4 cc/dL	3.4 cc/dL	136 mL/min
A membrane that has genuinely failed — outlet saturation only 85 percent — in a patient with haemoglobin 12 and an inlet saturation of 40 percent	6.4 cc/dL	13.7 cc/dL	7.3 cc/dL	292 mL/min

Both errors are in that table and both are common. Recirculation and anaemia each take a perfectly good membrane below the exchange threshold. A badly desaturated patient keeps a demonstrably failing membrane comfortably above it — the outlet saturation of 85 percent is itself diagnostic, and the transfer figure conceals it.

The first row is the Red Book's own worked example and the second is the Practical Guide's own recirculation example. The third and fourth rows are this book's arithmetic using the Red Book's method and the standard content relation, and they are calculations rather than findings.

What this book concludes, and it is the correction Chapter 31 assigned here.

Calculate the oxygen transfer, and never quote it without the inlet saturation and the haemoglobin on the same line.
The statement that is genuinely about the device is the post-membrane saturation and oxygen tension at a known blood flow, with the sweep at 100 percent oxygen and the flow below rated flow. Under those conditions, blood leaving at less than full saturation has only one explanation.
Treat 150 mL/min as a prompt to work out why, not as an indication. It is one of the very few numbers in this subject and it deserves to survive being used correctly.

32.5 The pressure drop, and what it can and cannot tell you

Chapter 31 hands this chapter the pressure drop as a measurement and keeps it as a thrombus signal. The division is worth stating once more because it decides what each number is for: Chapter 31 asks whether there is clot; this chapter asks whether the device can still do its job.

🧮

What the pressure drop is, and the three ways it moves without any clot

The pressure drop is the pressure entering the membrane lung minus the pressure leaving it. ECPR and Resuscitative ECMO defines it in the circuit nomenclature as the difference between the pressure after the pump and before the membrane lung, and the pressure after the membrane lung, reflective of resistance across the membrane lung.

Resistance is not the only thing that sets it. The ISCCM manual lists the determinants explicitly: the device's own physical characteristics, and variation of its internal resistance during use, including blood temperature, viscosity and thrombus formation on the membrane. Add the one everybody knows and nobody writes down, and there are four:

Blood flow. The same membrane at 5 L/min shows a higher pressure drop than at 3 L/min. This is why the ratio, not the number, is followed — the rule Chapter 31 §31.4 established from the clot side.
Temperature. A patient cooled for neuroprotection has more viscous blood and a higher pressure drop, with nothing wrong with the device. Chapter 22 and Chapter 68 own the temperature decision; what belongs here is that it moves this measurement.
Haematocrit. Transfusing a patient from a haemoglobin of 7 to 10 raises viscosity and raises the pressure drop. A transfusion and a pressure-drop rise on the same day are not independent observations.
Clot. Which is the one being looked for, and which is therefore easiest to find when the other three have been accounted for.

And the observation that makes this measurable in a unit without pressure transducers. The Red Book: "If pressures are not measured, increased resistance can be inferred when increased pump speed is required to maintain the same flow." Revolutions per minute against achieved flow is the same information, obtained free, and it is already on every console.

One differential that belongs to the pressure drop and to nothing else. The Red Book notes that monitoring resistance detects obstruction due to clot formation in the membrane lung "or aspiration of a clot from the venous system into the circuit." A pressure drop that steps up abruptly rather than climbing is more likely to be something that arrived than something that grew — and that distinction changes what should be looked for in the patient.

💡

Clinical pearl · How to make the pressure drop into a usable measurement in one week

Record the pressure drop and the blood flow on the same line, every time, and calculate the ratio. Nothing else in this chapter costs so little.
Establish this circuit's own baseline ratio in the first 24 hours, when the membrane is known to be clean, and write it at the top of the chart. Every later reading is compared with that, not with a textbook figure. The ISCCM unit protocol's target of a gradient under 30 mmHg, monitored continuously, is reproduced here as that protocol's own target, from a paediatric unit, and is not offered as a general threshold.
Record the temperature and the haemoglobin beside it, because those are the two confounders that move slowly enough to be mistaken for a trend.
Draw the post-membrane gas at the same time, with the sweep at 100 percent oxygen. ECMO in the Adult Patient: the transmembrane pressure trend should be observed regularly in conjunction with post-oxygenator blood gas measurements, and a deteriorating post-membrane oxygen tension "is sometimes coupled to an increase in transmembrane pressure." The word is sometimes, and the two channels are recorded together precisely because neither is reliable alone.
A step change is investigated the same hour. A slow climb is followed for a day and interpreted as a slope. Chapter 30 §30.11 makes the general version of this point: a gradual monotonic rise is a different finding from an abrupt one, and only a column shows which it is.

32.6 Non-thrombotic membrane failure: water, plasma leak, gas supply and the heat exchanger

Clot is not the only thing that stops a membrane lung working, and the non-thrombotic causes matter out of proportion to their frequency because three of the four are reversible and one of them is reversible in about a minute.

🧠

Physiology · Water is the one that is always happening

A membrane lung is a dehumidifier that nobody asked for. The Red Book: during normal operation both polymethylpentene and polypropylene devices remove a significant amount of water, and water vapour loss from the gas outlet can amount to over 1 L per day, depending on blood temperature, the fluid pressure inside the blood phase, and the gas flow rate.

That water has to go somewhere, and at low gas flows it does not leave. The Red Book again: at lower gas flow rates the water vapour condenses on the inside of the fibres, because the gas is cooler there — and that condensation "can limit gas exchange in a similar fashion to plasma leakage in the PP devices, particularly because the CO2 diffusional gradient is much lower than O2."

Two clinical consequences, and they point in opposite directions.

Over a litre a day of insensible loss through a gas outlet is a fluid balance item that appears on no chart. It rises with temperature and with sweep. Chapter 54 is owed this as a named, quantified insensible loss and the figure belongs in the balance rather than in the unexplained-deficit column.
The condition that produces condensation is a low sweep, and the condition that produces the greatest water loss is a high one. A device run for days at a low sweep — a weaning patient, a carbon dioxide removal configuration — is the one that wets, and wetting takes carbon dioxide clearance first, which is exactly the signal the weaning team is watching.

The treatment is a manoeuvre, not a decision. ECMO in the Adult Patient: "Condensation of water within the oxygenator will decrease efficiency and the oxygenator needs to be regularly flushed with a high gas flow to remove the accumulated moisture." The ISCCM manual lists water vapour in the membrane as a determinant of carbon dioxide removal.

So a rising carbon dioxide with a normal pressure drop at a low sweep has a free diagnostic test: flush the device at a high sweep for a short period and see whether the carbon dioxide clearance returns. If it does, the membrane is wet and nothing needs changing. If it does not, the finding stands.

The manoeuvre has a name and one source does specify its duration. ECPR and Resuscitative ECMO describes condensation developing inside the microtubules of the membrane lung from the sudden cooling of gas once it is no longer in contact with warm blood, and states that it "can be flushed out by increasing the gas flow rate for a few seconds, which is called 'sighing' the membrane lung." [Correction, 2026-09-23: this chapter as first published stated that no retrieved source specified the flush. That was wrong — the passage above is in a source this chapter already lists, and it was missed. The flush flow rate and the frequency remain unspecified by any source retrieved. VERIFICATION REQUIRED for those two.]

🔬

Evidence · Plasma leak is largely a historical failure mode, and the reason matters

Plasma leak is the loss of the blood-gas barrier itself: plasma crossing into the gas phase through a microporous fibre, flooding the gas side and abolishing exchange. It is the classic reason a polypropylene device had to be changed.

The material solved it. The Red Book: polymethylpentene hollow fibres — outer diameter about 380 µm, wall thickness about 90 µm — are a true membrane rather than a microporous one, and the change in durability is not incremental: polymethylpentene devices allow use for up to 30 days, compared with about 6 hours for most polypropylene devices, and one patient is recorded as surviving 605 days of extracorporeal support with only a few device changes. The ISCCM manual and ECMO in the Adult Patient both state that polymethylpentene devices show minimal or reduced plasma leakage.

And the price, stated by the same source: the gas-exchange efficiency of polymethylpentene is slightly less than that of microporous polypropylene. The field traded a little efficiency for a device that lasts weeks instead of hours, and that trade is why the failure modes in this chapter are the ones that matter now.

Two design facts worth carrying, because they explain where devices differ. The Red Book describes two blood-path geometries: transverse flow, with relatively low internal resistance but more prone to low-flow zones with stasis and needing a larger surface area; and axial flow, with relatively higher internal resistance but good blood washout and a lower surface area requirement. Both meet manufacturing standards for mass transfer. The difference is resistance against stasis, which is the same trade-off Chapter 31 §31.2 identified inside every part of a circuit.

Certainty: high that plasma leak is uncommon with current devices; moderate for the durability figures, which are the Red Book's own and are device-class statements rather than guarantees.

🚩

Pitfall · The heat exchanger fails in about one patient in fifty and gets blamed on the patient

ECMO in the Adult Patient records it plainly: failure of the heat exchanger is a recognised complication occurring in around 2 percent of patients, and "this is not always recognized immediately, as it is often wrongly attributed to the overall condition of the patient."

That is the whole pitfall in one sentence. A patient who will not stay warm on a circuit is assumed to be vasodilated, septic, or dying, and the device is not examined. The discriminating observation is Chapter 30 §30.2's rule applied to temperature: record the set water temperature and the delivered blood temperature on the same line. A widening gap between them is the heat exchanger. A narrowing patient temperature with the gap unchanged is the patient.

And a second heat-exchanger finding that is not about gas exchange at all. The same source warns that the water circulating in these systems can be contaminated by various organisms; that although the water is usually not in contact with blood, these organisms can be released into the atmosphere and may be a danger to the patient, staff and relatives; and that local decontamination procedures should not be ignored. The source does not name an organism and none is named here. Chapter 58 owns infection on ECMO and is owed this as a named environmental source.

One more configuration note, because it changes what the pressure drop means. A second oxygenator can be placed in the circuit to increase total membrane surface area. ECMO in the Adult Patient shows it and attaches the warning: the blood flow must be maintained high enough to avoid thrombus building up in the fibres. A circuit with two membranes in series has two pressure drops, two sets of gases, and twice the surface at which the low-flow rule in Chapter 31 §31.2 applies.

32.7 Controversy 1 · Does the pressure drop predict oxygenator failure, or only confirm it?

⚖️

The clinical question. The transmembrane pressure drop is the most-watched single number in Part VI and the commonest stated reason a circuit is changed. Does it give warning, or does it only describe a decision that has already been made?

Position A: it warns, and the trajectory is unusually specific. Chapter 31's anchor cohort measured the pressure drop normalised to blood flow in 16 patients with acute oxygenator thrombosis and found that the ratio rose 2.6-fold (2.1 to 4.0) before the exchange and fell 3.2-fold (2.0 to 5.2) after it, p under 0.001 in both directions, and did not rise in any other acutely exchanged patient. Specificity like that is rare in this subject. The bedside sources agree on the mechanism: ECMO in the Adult Patient states that an increase in transmembrane pressure indicates occlusion in the oxygenator, usually thrombus, and the Red Book that a rise in delta P at constant pump flow indicates increasing resistance.

Position B: it confirms, and the best-designed study of prediction failed to find it. van Minnen and colleagues set out specifically to identify predictors, in the cohort that anchors this chapter.

Higher pressure drop was an independent risk factor for exchange in the multivariable model, alongside a higher arterial carbon dioxide and a higher haemoglobin, with a Harrell's concordance statistic of 0.77 for the combined score.
But the pressure drop was also the most prevalent stated indication for exchange, for both emergency and elective exchanges. A variable that is the reason for the decision cannot also be an independent predictor of it, and the authors say so themselves: there may be a risk of a self-fulfilling prophecy, and they note that there were no emergency exchanges due to an acute shutdown of the system.
And the prospective form of the test failed. They calculated the percentage change in pressure drop, adjusted for blood flow, over the 12 and 24 hours before the exchange, and found no correlation. A sensitivity analysis using the pressure-drop-to-blood-flow ratio in place of the raw pressure drop gave similar results throughout.
They also record the prior literature's verdict without softening it: earlier studies showed the pressure-drop rise to be a late indicator of oxygenator failure.

Where the evidence actually sits, and the two positions are answering different questions.

Over the run, the normalised pressure drop discriminates. Patients whose membranes thrombosed had a different trajectory from those whose did not, in both cohorts. That is real and it is what Chapter 31 uses it for.
Over the preceding day, it does not. The one study that looked specifically at the short window before the event found nothing, which is precisely the window in which a warning would be useful.
And the deepest problem is not statistical. In both cohorts the pressure drop was the criterion clinicians used to decide. No study retrieved for this chapter has evaluated the pressure drop against an outcome that was determined independently of it — a membrane exchanged on other grounds and then examined, or a protocolised comparison with a prespecified trigger. Until one exists, the operating characteristics of this measurement are unknown, and every figure quoted for it is partly a description of local practice.

What this book concludes.

Follow the ratio of pressure drop to blood flow, from this circuit's own day-one baseline, and treat a sustained multi-fold rise as a strong finding. That is supported.
Do not expect it to give a day's notice. It did not in the study designed to test that.
Never let a normal pressure drop reassure. Chapter 31 established that device-related coagulopathy occurs in membranes whose function is nearly normal; this chapter adds that water and a low sweep degrade carbon dioxide clearance with no pressure change at all. The pressure channel is one of five and it is blind to two of the others.

What would resolve it. A prospective study recording the normalised pressure drop continuously, with the exchange decision made on prespecified criteria that do not include it, reporting sensitivity, specificity and lead time for gas-exchange failure — and reporting the exchanges that turned out to be unnecessary, which no current study does.

Certainty: moderate that the normalised pressure drop discriminates over a whole run; low that it predicts within a day; and none for any threshold value.

32.8 Controversy 2 · When the carbon dioxide rises, should the sweep go up?

⚖️

The clinical question. The arterial carbon dioxide has risen from 45 to 58 mmHg overnight. The sweep is at 4 L/min and the blood flow at 3.5 L/min. Somebody is about to turn the sweep up. Should they?

Position A: yes, immediately, because that is exactly what the sweep is for. The relationship is the best-characterised in extracorporeal gas exchange. In the physiology anchor, sweep flow was an independent determinant of carbon dioxide transfer, at about 33 mL/min of additional clearance per L/min of gas, and the effect was immediate and reproducible across every blood flow tested. The Red Book states the operating principle without qualification: the sweep gas flow rate is set by the operator based on the desired arterial carbon dioxide. Hypercapnia is not benign — it raises intracranial pressure, raises pulmonary vascular resistance and worsens right ventricular afterload in exactly the patients least able to tolerate it. And the alternative lever, raising the blood flow, is not free: it means more shear, more haemolysis, more recirculation on venovenous support and more strain on a drainage cannula. Turning a knob that works, on a patient who is acidotic, is good medicine.

Position B: yes, and then stop, because the sweep responds to everything and diagnoses nothing. Four arguments, in ascending order of force.

The differential was skipped. A rising carbon dioxide on extracorporeal support has at least four causes and the membrane is only one: the patient is producing more (fever, agitation, sepsis, overfeeding, lightening sedation); the blood flow has fallen, and over the ranges the physiology anchor studied the effect of blood flow on carbon dioxide transfer was roughly 1.9 times that of gas flow; the native lung has deteriorated, removing its share on venovenous support; or the membrane has lost capacity. Raising the sweep corrects all four identically and distinguishes none of them.
There is a ceiling, and it is closer at low blood flow. Above a gas-to-blood flow ratio of about four, the ex vivo data show carbon dioxide transfer plateauing and the dead space fraction rising to 0.8 to 0.9. At a blood flow of 2 L/min that ceiling arrives at a sweep of 8 L/min. Past it the knob still turns and nothing happens, and the team discovers this at the worst possible moment.
The correction itself is a hazard if it is fast. ECMO: A Practical Guide to Management states that rapid decreases in arterial carbon dioxide are associated with neurological complications through alterations in cerebral blood flow, and that hypercapnia should be corrected over 4 to 8 hours after ECMO initiation. A large sweep increase in a chronically hypercapnic patient is a cerebral intervention. Chapter 22 owns the post-arrest version of this and is owed the general statement; Chapter 7 and Chapter 14 own it at initiation.
And the argument that matters most: the carbon dioxide is a controlled variable, so holding it normal consumes the reserve invisibly. Every sweep increase that succeeds also removes the evidence that anything happened. A membrane can lose most of its carbon dioxide capacity over a week with a perfectly flat carbon dioxide trend and nothing recorded but a flowmeter that nobody charted.

Where the evidence actually sits. There is no disagreement about the physiology. The disagreement is about what to record, and on that the anchor cohort's own words settle the direction: van Minnen and colleagues describe deteriorating gas exchange as "a late marker for functional decline of the oxygenator", and note that an in-vitro study found alteration of the carbon dioxide concentration measured at the outlet of the oxygenator to be an early indicator of deterioration — a measurement, they add, that is not generally performed. [The in-vitro study is cited within the anchor and was not separately retrieved. VERIFICATION REQUIRED.]

What this book concludes.

Turn the sweep up. Then write down what it cost. The sweep, the blood flow and the carbon dioxide on the same line, with the gas-to-blood ratio calculable from them.
Before the second increase, do the differential — production, flow, native lung, membrane — and flush the device at a high sweep to exclude condensation.
When the gas-to-blood ratio approaches four, escalate the enquiry rather than the knob, and consider raising the blood flow instead where the patient allows it.
Correct a chronically raised carbon dioxide slowly, and treat a large fall as an event to be recorded.

What would resolve it. Continuous exhaust carbon dioxide measurement, which several devices can now provide, evaluated prospectively as an early indicator of membrane deterioration against a hard endpoint. The measurement exists. The study does not.

Certainty: high that sweep raises carbon dioxide transfer; moderate for the plateau above a ratio of about four; moderate for the harm of rapid correction; low for the concealment argument, which is this book's reasoning from the controlled-variable problem rather than a finding.

32.9 The exchange trigger this chapter owns

Chapter 31 owns the decision to change a circuit and the procedure for doing it. What it assigned here is the gas-exchange half of the indication, and this section states it in the form a team can act on.

🔬

Evidence · The clinical anchor, and what it found when it looked for predictors

van Minnen O, Oude Lansink-Hartgring A, Hoffmann RF, van den Bergh WM. "Risk factors for elective and emergency oxygenator exchanges during veno-venous extracorporeal membrane oxygenation." Perfusion 2024;39(5):911–920. DOI 10.1177/02676591231168642, PMID 36995209, PMC11191661. Read as complete full text. Retrospective observational cohort, single Dutch tertiary centre, all 45 consecutive adults supported with venovenous ECMO for at least 48 hours between October 2018 and March 2022. Cardiohelp with the HLS advanced 7.0 set throughout; unfractionated heparin to an activated partial thromboplastin time of 50 to 70 seconds; the oxygenator inspected twice daily by a perfusionist and an intensivist; weaning done by lowering the sweep rather than the blood flow, to minimise clot risk. There was no strict protocol for when to exchange.

The cohort. Median ECMO duration 16 days; 28 of 45 (62 percent) discharged alive; 23 of 45 (51 percent) supported for COVID-19 acute respiratory distress syndrome.

The exchanges. 19 of 45 patients (42 percent) had at least one oxygenator exchange, on median day 12; 10 (22 percent) had a second, on median day 9 of that circuit — 29 exchanges in all. Ten of the 29 (34 percent) were emergency exchanges, defined by timing rather than by indication: an exchange performed outside the 10 a.m. to 5 p.m. window. The most prevalent indication for both emergency and elective exchanges was a rising pressure drop.

The three independent risk factors for exchange (backward stepwise Cox regression; combined score Harrell's c equal to 0.77):

Variable	What it plausibly represents
Higher arterial carbon dioxide tension (with the corresponding lower pH)	Failing carbon dioxide clearance — the gas-exchange trigger this chapter owns
Higher pressure drop	Rising resistance, and also the criterion the clinicians were using
Higher haemoglobin	Uncertain — a known thrombotic risk factor, but the authors note haemoglobin was low in both groups and could not separate the level from the transfusion that produced it

And the finding that is easiest to misread: lower lactate dehydrogenase was the only independent risk factor for an EMERGENCY exchange — odds ratio 0.90 (0.82 to 1.00), borderline. Both lactate dehydrogenase and C-reactive protein were lower before emergency than before elective exchanges. The authors offer two speculations and label them as such: that a lower lactate dehydrogenase may mark red-cell-rich, fibrin-poor thrombus formed under low shear — and the membrane lung is the low-shear component of the circuit — which might degrade function faster; or simply that the more inflamed patients were the ones clinicians chose to exchange during working hours. Neither is established, and the whole finding rests on ten events.

What was NOT associated. D-dimer was higher in the exchange group but was not associated with exchange in the regression. Plasma free haemoglobin was associated univariately but did not survive, and the authors note it was measured only every 48 hours and attribute haemolysis more plausibly to the pump than to the membrane — which is Chapter 31 §31.3's finding arriving from a second cohort.

Certainty: low to moderate. Forty-five patients and 29 events, retrospective, a single oxygenator brand, over half the cohort COVID-19, and the authors' own acknowledgement of a self-fulfilling prophecy. What it establishes firmly is the frequency — an exchange in more than four patients in ten, a third of them out of hours — and that the carbon dioxide channel carries independent information. What it does not establish is any threshold.

💡

Clinical pearl · The four gas-exchange findings that are a trigger, and the four that are not

A trigger, when the alternatives below have been excluded.

Post-membrane blood that is less than fully saturated, on a sweep of 100 percent oxygen, at a blood flow below the device's rated flow. This is the most device-specific observation available, because below rated flow there is no other explanation.
Oxygen transfer under 150 mL/min (the Red Book's threshold), with the inlet saturation and the haemoglobin checked first — because §32.4 shows recirculation and anaemia each cross it in a healthy device.
Carbon dioxide clearance that no longer responds to sweep, with the gas-to-blood ratio at or approaching the plateau, after a high-sweep flush has failed to recover it and after production, blood flow and the native lung have been accounted for.
A sustained multi-fold rise in the pressure-drop-to-blood-flow ratio from this circuit's own baseline, with the post-membrane gases moving in step. The two channels together; neither alone.

Not a trigger.

A raw pressure drop without the flow that produced it, or a rise explained by cooling, transfusion or a flow increase.
A declining oxygen transfer without its denominators. In a retrospective comparison of four contemporary adult devices in 282 patients (Sousse LE, Hagiwara J, Liu Q, Maul TM, Michalek JE, Anderson BJ, Dellavolpe JD. "Comparative Performance of Oxygenators for Extracorporeal Membrane Oxygenation in Adults: A Retrospective Analysis." ASAIO J, published online 10 August 2026. DOI 10.1097/MAT.0000000000002805. Abstract read), oxygen transfer across the membrane declined over the course of support as a matter of course (p under 0.001), and was higher on venovenous than on venoarterial support. Decline is the expected trajectory, not a finding. And this book's own reading of that result, which the authors do not state: as a patient recovers, the venous blood arriving at the membrane is better oxygenated, which by §32.4's arithmetic lowers the measured transfer in a device that has not changed at all. Some of the reported decline over time may be the denominator problem rather than device deterioration.
A single abnormal gas. Repeat it, with the sweep at 100 percent oxygen, before anything else happens.
Visible clot alone. Chapter 31 settled that: it is a reason to look harder at the other channels, not an indication.

And the operational fact that should inform the timing. In a 2026 mixed-methods appraisal of ECMO monitoring (Yin W, Permessur A, Pittas G, Sukhlal S, Chow N, Pinheiro S, Cobos A, Sabetta G, Cahill J, Zaidi S, Longtin J, Schwartz J. "Challenges and opportunities in ECMO monitoring: A mixed-methods appraisal." J Extra Corpor Technol 2026;58(3):280–294. DOI 10.1051/ject/2026022, PMC13577623), clinicians described preparation for a membrane swap taking about 15 to 30 minutes or longer, while the swap itself typically takes under 2 minutes — during which the patient has no extracorporeal support. The expensive part is the preparation, which is exactly why a planned exchange and an emergency one are different procedures, and why the whole point of this chapter's surveillance is to convert one into the other. ECMO in the Adult Patient puts it in a sentence: "Predicting oxygenator failure is vital to ensure that the change of oxygenator is a planned procedure rather than having to be carried out as an emergency."

🚩

Pitfall · Monitoring the membrane lung without a protocol, because there isn't one

A 2026 appraisal of ECMO monitoring summarises the ELSO recommendations for gas-exchange monitoring as: continual arterial and venous blood gas monitoring; pre-membrane and post-membrane blood gas analysis daily; regular checking of the membrane lung's pressure gradient; monitoring of blood flow; maintaining adequate haemoglobin; and monitoring lactate as an indicator of tissue perfusion. That list is feasible in almost any resource setting, and it is a reasonable minimum. [These recommendations are read here at second hand through that appraisal and the ELSO guideline text was not retrieved. VERIFICATION REQUIRED.]

But the same source states the gap without softening it: "no unified protocol exists specifying how these parameters should be integrated or recorded to characterize overall circuit gas-exchange efficiency."

Three things follow, and they are the reason this chapter is written the way it is.

The list specifies what to measure and not what to do with it. Daily pre-membrane and post-membrane gases produce a pair of numbers; whether they are read as a trend, against what baseline, and beside which flow, is left entirely to the unit.
Continuous measurement of oxygen and carbon dioxide in the sweep gas exists on some devices and is not standard, and many membrane lungs still carry no integrated sensor for post-membrane oxygen tension or saturation at all. Two published approaches to continuous gas-exchange monitoring — an algorithmic one and a sensor-based one using oximetry probes, an ultrasonic flow meter and infrared exhaust carbon dioxide — are described in the same appraisal as not widely adopted, because of modelling complexity, extra sensors and interruption of the care workflow.
So the measurement that this chapter argues is the earliest available — what the sweep has to do to hold the carbon dioxide — is also the one that costs nothing to record and is recorded almost nowhere.

A caution about that appraisal, recorded because it bears on how much weight its interview findings carry. Its customer-discovery component interviewed 45 stakeholders, of whom 15 were from sales and business and a further 10 were scientists or biomedical engineers; 20 were clinicians of any kind, recruited largely through web searches and referrals, with 58 percent from a single US state. That is a market-discovery instrument, not a clinical practice survey, and its "pain points" are read here as hypotheses about what clinicians find difficult rather than as evidence of what they do.

32.10 A worked case: the carbon dioxide that never moved

🧮

Worked example · Day 10 of venovenous ECMO, and a knob that nobody charted

The patient. Day 10 of venovenous ECMO for severe pneumonia. On day 5 the blood flow was reduced from 4.0 to 2.0 L/min because of persistent drainage chatter, and it has stayed there; the patient has tolerated it and is improving slowly. Arterial saturation 93 percent. Ventilator settings unchanged and lung-protective. Temperature 37.2 degrees. Haemoglobin stable. Nobody is worried.

The trigger for the discussion. Overnight the arterial carbon dioxide rose to 52 mmHg. The night team raised the sweep from 5 to 6 L/min and it came back to 46. The morning round is asked whether the oxygenator is failing and whether it should be changed.

The chart, with the columns §32.3 and §32.5 ask for.

Value	Day 10	Day 8	Day 6	Day 5	Day 1
Blood flow (L/min)	2.0	2.0	2.0	2.0	4.0
Sweep (L/min)	6.0	4.0	2.5	1.5	4.0
Gas-to-blood ratio	3.0	2.0	1.25	0.75	1.0
Arterial carbon dioxide (mmHg)	46	45	44	45	44
Pressure drop (mmHg)	19	19	18	18	34
Pressure drop / blood flow	9.5	9.5	9.0	9.0	8.5
Post-membrane oxygen tension (mmHg)	398	402	405	404	410
Post-membrane saturation (percent)	100	100	100	100	100
Plasma free haemoglobin (mg/dL)	14	13	14	12	11
D-dimer, fibrinogen, platelets	Unchanged	Unchanged	Unchanged	Unchanged	Baseline

What the round said, and what the chart says.

"The carbon dioxide is fine." It is fine because it is held. It has not moved in ten days and it never will until the sweep runs out. The carbon dioxide row carries no information at all in this chart; the sweep row carries all of it.
"The pressure drop has halved since day one, so the membrane is clean." The raw pressure drop halved because the blood flow halved. Divided by the flow that produced it, the ratio has gone 8.5, 9.0, 9.0, 9.5, 9.5 — essentially flat. That is the correct reading and it is the right conclusion: there is no evidence of clot.
"Post-membrane oxygen is 398, so the membrane is working." It is, for oxygen. §32.2 explains why that is not reassurance: oxygen has a 600 to 40 mmHg gradient and holds up until a device is badly damaged. A normal post-membrane oxygen excludes late failure, not early failure.

The finding, which is in one row. The sweep needed to hold the same carbon dioxide has gone from 1.5 to 6 L/min in five days, at an unchanged blood flow, in a patient who is not more septic, not febrile, not agitated, and whose native lung is improving rather than deteriorating. That is a four-fold rise in the work the gas side is doing for the same result. The gas-to-blood ratio has gone from 0.75 to 3.0 and is approaching the plateau described in §32.3.

The compartment is named by what is NOT moving. Carbon dioxide clearance is degrading; oxygen transfer, the pressure ratio, the haemolysis panel and the coagulation panel are all unchanged. By §32.2 that is the gas side, not the blood side — and the device has spent five days at a gas-to-blood ratio below 1.0, which is precisely the condition the Red Book describes as producing condensation inside the fibres.

What should happen, in order.

Exclude the three non-membrane causes properly — carbon dioxide production, blood flow, and the native lung's share. All three are stable here and that has to be checked rather than assumed.
Repeat the gases with the sweep at 100 percent oxygen, so that today's numbers are comparable with day one's.
Flush the membrane at a high sweep and re-measure — "sighing" the membrane lung, in the phrase ECPR and Resuscitative ECMO uses, by raising the gas flow rate for a few seconds. If carbon dioxide clearance returns and the sweep can come back down, the membrane was wet and nothing needed changing. The flow rate to use, and how often to do it, are not specified by any source retrieved for this chapter.
If it does not return, the finding stands, and it is one of the four gas-exchange triggers in §32.9. Chapter 31 then owns the decision and the procedure.
Either way, consider whether the blood flow still needs to be 2.0 L/min — raising it is a carbon dioxide intervention in its own right, and it moves the device away from both the dead-space ceiling and the low-flow stasis Chapter 31 §31.2 warns about.

The teaching point. Nothing abnormal was ever recorded. Every number in the chart stayed in range, on every day, and a device lost most of its carbon dioxide reserve while the team watched. The failure would have declared itself at the ceiling, abruptly, at whatever hour the ceiling happened to arrive — and that is how a third of oxygenator exchanges come to be done out of hours.

This case is constructed to illustrate the mechanisms described in this chapter. The values are illustrative and are not drawn from a published report.

32.11 What this chapter could not close

🚩

Open questions, recorded rather than concealed

There is no unified protocol for integrating gas-exchange measurements, and the field says so. A 2026 appraisal states plainly that no protocol exists specifying how the monitored parameters should be combined or recorded to characterise a circuit's gas-exchange efficiency. Everything in §32.9 is therefore assembled by this book from separate sources and is not a standard.
The field's one numeric gas-exchange trigger — oxygen transfer under 150 mL/min — is quoted from the ELSO Red Book, and the primary source behind it was not retrieved. Given that this is the only number of its kind in the subject and that §32.4 shows it is crossed by anaemia and recirculation in a healthy device, retrieving and reading its origin is the single highest-value outstanding item for this chapter. [VERIFICATION REQUIRED.]
No measurement in this chapter has been evaluated against an outcome determined independently of it. In both cohorts the pressure drop and the gas exchange were the criteria clinicians used to decide on an exchange, which the anchor's authors name in their own limitations as a risk of a self-fulfilling prophecy. The operating characteristics of every parameter in Part VI are, to this extent, unknown.
The measurement this chapter argues is earliest — the sweep required to hold the carbon dioxide — has never been evaluated as a monitored variable. It is offered as reasoning with a free implementation cost, not as a validated test, and it is labelled that way in §32.3.
Continuous exhaust carbon dioxide monitoring exists on some devices and has not been prospectively evaluated. An in-vitro study finding outlet carbon dioxide to be an early indicator of deterioration is cited within the anchor and was not separately retrieved; two published continuous-monitoring approaches are described as not widely adopted. Carried forward from Chapter 29 and still open: the 13-patient study the ISCCM manual cites as having used a rise in D-dimer as a very early precursor of oxygenator failure remains unidentified, and no cut-off from it should ever be quoted.
No venoarterial-specific data on membrane lung failure were retrieved, which is the same gap Chapter 31 recorded for circuit thrombosis. A 282-patient device comparison found oxygen transfer higher on venovenous than on venoarterial support — a result that §32.4's arithmetic would predict from inlet saturation alone, and which therefore cannot be read as a device or configuration effect.
The flow rate and the frequency of the high-sweep flush for condensation are not specified by any source retrieved, although four sources state that it is necessary. Only its duration is given anywhere: ECPR and Resuscitative ECMO calls the manoeuvre "sighing" the membrane lung and says to raise the gas flow rate for a few seconds. [Correction, 2026-09-23: this item as first published said no source specified the flush at all. That was an error — the passage is in a source this chapter lists, and it was missed on the first pass.] A manoeuvre that is universally recommended and almost nowhere specified is a good target for a simple study.
The definition of rated flow is not consistent between two of this book's own primary sources — standard venous blood at 70 percent saturation in the Red Book, 75 percent in ECMO: A Practical Guide to Management. The primary standards document was not retrieved. Chapter 3 is owed this as a correction to its equipment description.
No study quantifies how often an interruption of sweep gas is mistaken for membrane failure, despite every source naming it as the commonest cause of an abrupt change. The scenario is described and the frequency is unknown.
The one contemporary head-to-head device comparison had no direct measurement of gas transfer and its authors say so, calling for prospective multicentre work incorporating direct measurements. Its finding of no difference between four current devices is therefore an absence of evidence at the level of the patient, not a demonstration of equivalence at the level of the membrane.

Corrections and additions owed to other chapters

Chapter 27 (Blood Gas and Oxygenation Monitoring) is owed §32.4's confirmation from the transfer side: in a controlled physiological experiment, sweep flow dropped out of the oxygen transfer model entirely and the oxygen transfer was identical at every sweep-to-blood ratio tested. Chapter 27 reached "the blender is not the oxygen knob" from the saturation side; this is the same rule measured.
Chapter 22 (Post-ECPR Management), Chapter 7 and Chapter 14 are owed the general statement that rapid falls in arterial carbon dioxide are a cerebral intervention, with the Practical Guide's instruction that hypercapnia be corrected over 4 to 8 hours. Chapter 22 §22.3 is already owed a change-in-carbon-dioxide ceiling alongside its absolute target; this is the mechanism half of that correction.
Chapter 54 (Fluid and Haemodynamic Management) is owed the over 1 L per day of insensible water loss through the gas outlet, which rises with temperature and sweep and appears on no fluid chart.
Chapter 58 (Infection Prevention and Treatment) is owed the heat-exchanger water circuit as a named environmental source of organisms that can reach the patient, staff and relatives — alongside Chapter 31's finding that the circuit itself can be the septic focus.
Chapter 33 (Pump Failure) is owed the Red Book's inference rule — increased pump speed required to maintain the same flow is a resistance measurement in a unit without pressure transducers — and the boundary that pump inlet and outlet pressures belong to it, not here.
Chapter 37 (Haemolysis) is owed the anchor cohort's second-cohort agreement with Chapter 31: plasma free haemoglobin did not survive multivariable analysis as a predictor of oxygenator exchange, and the authors attribute haemolysis more plausibly to the pump than to the membrane.
Chapter 10 and Chapter 26 (weaning) are owed two things. The anchor centre weaned by lowering the sweep rather than the blood flow, explicitly to minimise clot formation — a practice with a clear rationale that should be stated where weaning is decided. And the physiology that complicates it: a device held at a low sweep and a low blood flow is the one that wets, and wetting degrades carbon dioxide clearance, which is the endpoint the weaning trial is reading.
Chapter 53 (Mechanical Ventilation During ECMO) is owed the native lung's share of carbon dioxide clearance as a named confounder of any membrane assessment on venovenous support.
Chapter 84a (Extracorporeal Carbon Dioxide Removal) is owed §32.3 entire: the dead-space ceiling is most relevant exactly where blood flows are lowest, which is the configuration that chapter is about.
Chapter 87, Chapter 88 and Chapter 92 are owed three paired columns as design requirements: sweep with arterial carbon dioxide and blood flow; pressure drop with blood flow, temperature and haemoglobin; set water temperature with delivered blood temperature.
Chapter 89 (Troubleshooting Algorithms) is owed the gas-supply-first rule for any abrupt loss of gas exchange and the high-sweep flush as a free diagnostic step before any exchange is considered.

Carried forward and still open: the alpha-stat versus pH-stat question, for which Chapter 28 established that no ECMO-specific evidence exists; and Chapter 31's unanswered question of what circuit thrombus does to the patient rather than to the device.

32.12 Evidence and certainty summary

Claim	Best evidence	Certainty
Oxygen crosses the membrane on a gradient of about 600 to 40 mmHg and carbon dioxide on about 45 to 0 mmHg, with roughly equal amounts exchanged at a gas-to-blood ratio of 1:1	ELSO Red Book, stated in the circuit and physiology chapters and reproduced in two other primary sources	High — established physiology
Carbon dioxide clearance is the more fragile function and gives way before oxygen transfer; a fall in post-membrane oxygen is a late sign	The gradient argument above; the Red Book's statement that condensation limits exchange "particularly because the CO2 diffusional gradient is much lower than O2"; the ISCCM manual listing water vapour only under carbon dioxide; the clinical anchor calling deteriorating gas exchange a late marker	Moderate as a tendency with a mechanism. Not a law — in the Chapter 31 anchor, acute oxygenator thrombosis moved both channels at once
An abrupt loss of gas exchange is a gas-supply problem until proved otherwise, because membrane decline takes hours to days	ELSO Red Book, circuit malfunction chapter, with the exhausted transport cylinder named as the commonest source	High as a discriminator; unknown as a frequency, since nobody has measured how often it happens
Sweep gas flow and blood flow are both independent determinants of carbon dioxide transfer — sweep at about 33 mL/min per L/min, and blood flow with an effect the authors put at roughly 1.9 times that of gas flow over the ranges studied	Park and colleagues, 5 pigs, 45 measurements per condition, mixed model, pseudo-R squared 0.89. Full text read	Moderate for the rank order and the direction; low for the coefficients as transferable numbers
Sweep gas flow does not change oxygen transfer; blood flow and the inlet saturation do	Same study: sweep dropped out of the oxygen model, and oxygen transfer was identical at every sweep-to-blood ratio tested. Concordant with Chapter 27's conclusion from the saturation side	Moderate to high — two independent lines of argument reaching the same rule
Carbon dioxide transfer plateaus above a gas-to-blood flow ratio of about 4, where the oxygenator dead space fraction reaches 0.8 to 0.9	Zakhary and colleagues, ex vivo, one device, two oxygenators in series. Abstract read, full text not retrieved	Moderate for the existence of a ceiling with a rising dead space fraction; low for the value of 4.0, which is device-specific
Measured oxygen transfer falls when the inlet blood arrives better oxygenated, so recirculation and anaemia each take a healthy membrane below the 150 mL/min exchange threshold	The Practical Guide's own recirculation example (6.9 versus 2.8 cc/dL of oxygen added) plus this book's arithmetic using the Red Book's method and threshold	High as arithmetic; the conclusion follows necessarily from the definition of oxygen transfer
Oxygen transfer under 150 mL/min should prompt consideration of a membrane exchange	ELSO Red Book, circuit malfunction chapter	Low as a threshold. The primary source was not retrieved, and the quantity has two patient-determined terms. Treat as a prompt to work out why
The transmembrane pressure drop rises with clot but also with blood flow, blood temperature and viscosity, so only the ratio to blood flow is interpretable	ISCCM manual for the determinants; ELSO Red Book for the constant-flow rule; Chapter 31's anchor for the normalised trajectory	High for the confounders; moderate to high that the normalised ratio discriminates over a run
The pressure drop does not give a day's warning	van Minnen and colleagues found no correlation between the blood-flow-adjusted change in pressure drop over 12 or 24 hours and the exchange; earlier work cited there calls it a late indicator	Low to moderate — one small cohort, but it is the only study that asked the question directly
No parameter in this chapter has been evaluated against an independently determined outcome, because the parameters ARE the exchange criteria	Stated by the anchor's authors as a risk of a self-fulfilling prophecy in their own limitations	High as a statement about the state of the evidence
Oxygenator exchange is frequent: 42 percent of patients, 34 percent of exchanges out of hours, and a rising pressure drop the commonest stated indication	van Minnen and colleagues, 45 patients, 29 exchanges. Full text read	Low to moderate for the proportions — small, single centre, single device, half the cohort COVID-19, and centre-specific thresholds
Higher arterial carbon dioxide, higher pressure drop and higher haemoglobin were independent risk factors for exchange (concordance 0.77); lower lactate dehydrogenase was the only risk factor for an emergency exchange	Same cohort. The emergency finding rests on 10 events, with an odds ratio of 0.90 (0.82 to 1.00)	Low. The carbon dioxide association is the part with an independent mechanism behind it; the lactate dehydrogenase finding is hypothesis-generating and the authors treat it as such
A membrane lung loses over 1 L of water per day through the gas outlet, and at low gas flows that water condenses inside the fibres and limits carbon dioxide exchange; a high-sweep flush is the treatment	ELSO Red Book for the loss and the mechanism; ECMO in the Adult Patient and the ISCCM manual for the consequence and the flush	Moderate — consistent expert description across three sources, mechanistically clear, no quantification of the flush and no outcome evidence
Plasma leak is largely a historical failure mode; polymethylpentene devices last up to 30 days against about 6 hours for most polypropylene devices, at the cost of slightly lower efficiency	ELSO Red Book, with the ISCCM manual and ECMO in the Adult Patient concordant on minimal plasma leakage	High that plasma leak is now uncommon; moderate for the durability figures as device-class statements
Heat exchanger failure occurs in about 2 percent of patients and is commonly misattributed to the patient	ECMO in the Adult Patient, stated with the misattribution named	Moderate — a single expert source, but a specific and actionable figure
Four contemporary adult oxygenators showed no difference in oxygen transfer after adjustment, and transfer declined over the course of support	Sousse and colleagues, 282 adults, 2020 to 2023, single centre, propensity weighted. Abstract read; the authors state they had no direct measurement of gas transfer	Low. Absence of a difference at patient level, not equivalence at membrane level. The decline over time is partly predicted by §32.4's arithmetic as patients recover
No unified protocol exists for integrating gas-exchange parameters, and many membrane lungs carry no integrated post-membrane oxygen sensor	A 2026 mixed-methods appraisal, stated explicitly	Moderate as a statement about current practice; the appraisal's interview component is a market-discovery instrument and is weighted accordingly

32.13 Key points

A membrane lung does not stop working, it gets worse at working — over hours to days, while the controlled numbers stay in range. An abrupt change is the gas supply until proved otherwise, and the proof takes thirty seconds: put a hand on the gas line and follow it to the wall.
Both numbers that describe a membrane lung are ratios in disguise. The pressure drop means nothing without the blood flow that produced it; the oxygen transfer means nothing without the inlet saturation and the haemoglobin that were offered to it.
The two errors are symmetrical. Read the raw pressure drop and you will exchange working oxygenators. Read the raw oxygen transfer and you will keep failing ones.
Carbon dioxide crosses on a gradient of about 45 mmHg and oxygen on about 560. Carbon dioxide therefore gives way first, and a falling post-membrane oxygen tension is a late sign, not an early one.
The order of failure names the compartment. Carbon dioxide alone, with a flat pressure ratio, points at the gas side — water, condensation, a low sweep — and much of that is reversible. Carbon dioxide with a rising pressure ratio and, later, falling oxygen points at the blood side, and that is clot.
Sweep flow is the carbon dioxide knob and blood flow is the oxygen knob — and blood flow is also a carbon dioxide knob, with an effect the physiology anchor puts at roughly 1.9 times that of gas flow over the ranges it studied. Sweep flow does nothing at all for oxygen transfer.
The sweep has a ceiling. Above a gas-to-blood ratio of about four, carbon dioxide transfer plateaus and the device's dead space fraction reaches 0.8 to 0.9. At a blood flow of 2 L/min that ceiling is a sweep of 8.
The carbon dioxide tension is a controlled variable and therefore carries almost no information. The information is in what the sweep has to do to hold it. Record sweep, blood flow and carbon dioxide on the same line, every time; it costs one column.
Before calling a creeping sweep requirement a membrane finding, exclude three commoner things: rising carbon dioxide production, a fallen blood flow, and a deteriorating native lung.
A membrane lung loses over a litre of water a day, and at low sweep that water condenses inside the fibres and takes carbon dioxide clearance first. A high-sweep flush is both the treatment and a free diagnostic test.
Correct a chronically raised carbon dioxide slowly. Rapid falls alter cerebral blood flow; the Practical Guide asks for 4 to 8 hours at initiation, and the principle does not stop after the first day.
The device-specific statement is post-membrane blood that is less than fully saturated, on 100 percent oxygen, below rated flow. Below rated flow there is no other explanation.
The 150 mL/min oxygen transfer threshold is crossed by recirculation and by anaemia in a perfectly good membrane, and is not crossed by a failing membrane in a badly desaturated patient. Use it as a prompt to work out why.
Circuit gases are drawn with the sweep at 100 percent oxygen, or today's numbers are not comparable with last week's.
The pressure drop discriminates over a whole run and does not give a day's notice, and in every study so far it was also the criterion used to decide on the exchange — so its operating characteristics are unknown.
A normal pressure drop is not reassurance. It is blind to a wet membrane, blind to a low sweep, and Chapter 31 showed it is blind to device-related coagulopathy.
Three things move the pressure drop with no clot at all: a change in blood flow, cooling, and transfusion. Record temperature and haemoglobin beside it.
Without pressure transducers the same information is free: rising pump speed needed to hold the same flow is a resistance measurement.
Oxygenator exchange is common — more than four patients in ten in the anchor cohort, and a third of the exchanges out of hours. Preparation takes 15 to 30 minutes or more; the swap itself takes under two minutes, during which there is no support. That asymmetry is the entire argument for surveillance.
Chapter 32 owns the gas-exchange trigger and Chapter 31 owns the decision and the procedure. When the four findings in §32.9 are met and the alternatives are excluded, this chapter's job is done and the next one begins.

📌

Chapter answer: A membrane lung is a gas exchanger with two very different safety margins, and almost every error in reading one comes from forgetting that. Oxygen crosses on an enormous gradient and holds up until the device is nearly finished; carbon dioxide crosses on a small one and gives way early, which is why the carbon dioxide channel is where a failing membrane first speaks and the oxygen channel is where it finally shouts. But the carbon dioxide is held where the team wants it, so the number that moves is not the gas — it is the sweep required to produce it, and that number is on a flowmeter nobody charts. The same forgetting produces the other two errors: a pressure drop read without the flow that made it, and an oxygen transfer read without the blood that was offered to it. Pair each of them with its denominator, write the pair on one line, and most of this chapter becomes a matter of looking at a column. What is left over is a small number of real decisions — is this abrupt, in which case it is the gas supply; is it carbon dioxide alone, in which case flush it; is it both channels with a rising resistance, in which case the membrane is finished and Chapter 31 takes over.