⚙️

Chapter 33 — Pump Failure

❓

The question this chapter answers

A pump alarm is the loudest event in Part VI and the least likely to be about the pump. Pump failure occurs in about one patient in a hundred; a falling flow occurs most days. So the bedside question is not has the pump failed? but what is the flow reading actually measuring, which of six things has changed, and which of them is made worse by the move everyone reaches for first.

This chapter is about the pump as a machine: what it does, what the number on the console means, the six reasons that number falls, the two ways a pump head comes apart, and what happens to a patient on venoarterial support in the seconds after a pump stops.

🗓️

Evidence search date: 23 September 2026. One source is read as complete full text and is the only Verified source in this chapter. Primary sources are the ELSO Red Book 6th edition (the circuit chapter and the circuit malfunction and crisis management chapter), ECPR and Resuscitative ECMO (pump failure, decoupling and the backup equipment list), ECMO: A Practical Guide to Management (the complications and emergencies chapter, including its two-kinds-of-decoupling algorithm), Comprehensive Healthcare Simulation: ECMO Simulation (Johnston and Su) — mined here for the first time in this book (roller and centrifugal engineering, and the transport failure list), and ECMO in the Adult Patient.

What this chapter covers — and what it does not

This chapter owns	This chapter does not own
The pump as a machine — what the console sets, what it measures, and how a centrifugal pump differs mechanically from a roller pump	Clot as an object. Chapter 31 owns thrombus, including pump head thrombosis as a thrombotic diagnosis (§31.3). This chapter owns what a clot does to the machine
The differential for a falling blood flow at unchanged pump speed, and the order in which it should be worked	Chatter and access insufficiency as a drainage problem. Chapter 34 owns cannula problems and Chapter 9 owns persistent hypoxaemia and the chatter it describes
Pump stoppage and its consequences, including retrograde flow on venoarterial support	The emergency sequence itself. Chapter 35 owns air and the circuit emergencies and Chapter 90 owns the emergency checklist. This chapter owns why the clamp comes first
Decoupling, physical and magnetic, and why the two are managed differently	Haemolysis as a clinical problem and its trigger for a circuit change. Chapter 37 owns it. This chapter owns the mechanical reason the pump produces it
Console and power failure, and the two indications for the hand crank	The membrane lung. Chapter 32 owns gas exchange, the pressure drop as a diagnostic, and the gas-exchange trigger
Pump type as a choice: centrifugal against roller, and what the registry data do and do not show	Afterload physiology on venoarterial support. Chapter 13 owns the three circulations and the afterload-sensitive pump; this chapter uses that without re-deriving it
Nothing about anticoagulation	Chapter 42 owns anticoagulation monitoring and Chapter 39 owns the coagulation physiology

33.1 The pump does one thing, and the number on the screen is not it

A console has one output and one input. The output is a rotational speed, which is the only quantity the operator actually controls. The input is a measured blood flow, which is a consequence of that speed and of two things the operator does not control: what the circuit can drain, and what it has to push against.

🧠

The structural claim of this chapter · On a non-occlusive pump, the flow reading is a measurement of the patient

The Red Book describes the console's main function exactly: to set pump speed, and to measure and display the resulting flow. The word resulting is doing all the work.

The engineering distinction that everything else follows from, stated by Ündar and colleagues in this chapter's anchor study: "The roller pump is an occlusive pump that overcomes downstream resistance to maintain the same flow rate regardless of changes in downstream pressure. In contrast, the centrifugal pump is a non-occlusive blood pump where the flow rate depends on rotation speed and downstream resistance."

So on the pump almost every adult programme now uses, the displayed flow moves whenever the patient moves. A fall in circulating volume, a cannula that shifts against a vessel wall, a rise in intrathoracic pressure, a tamponade, a pneumothorax, a rising mean arterial pressure on venoarterial support, a clot growing in the return limb — each changes the flow at an entirely unchanged speed, and each is a different problem with a different treatment.

Three consequences, and the whole chapter is built on them.

A flow alarm is a symptom, not a diagnosis. It is the circuit's equivalent of a falling blood pressure.
The informative quantity is the pair — speed and achieved flow, on the same line. A falling flow at unchanged speed is a circuit or patient finding. The same flow at reduced speed is a decision somebody made. A chart carrying only one of the pair cannot tell them apart, which is Chapter 30 §30.2's rule and Chapter 32's pressure-drop rule arriving at the pump.
Their relationship over days is the pump's own performance curve, obtained free. The Red Book gives the use directly: "If pressures are not measured, increased resistance can be inferred when increased pump speed is required to maintain the same flow." Rising revolutions per minute for the same litres is a resistance measurement in a unit with no transducers at all.

And the reason a roller pump behaves differently is not a detail. An occlusive pump holds its flow against a rising downstream pressure — so on a roller pump a falling flow really does mean the pump, and on a centrifugal pump it usually does not. Half the troubleshooting instincts in this field were formed on a machine most units no longer use.

💡

Clinical pearl · What the console can tell you before anyone touches anything

The Red Book lists what a modern console displays and alarms on: low flow, pressure limit, air bubble, and back flow. Most contain backup batteries for power loss and transport, and a motor backup, manual or electric, in case of motor failure. Some contain an integrated pressure regulator allowing the operator to set a negative pressure limit, which produces an automatic reduction in pump speed as soon as that limit is reached — a feature §33.9 argues about.

Four readings, in this order, before a hand goes near a knob.

Is the impeller turning? Look at it. This single observation separates the whole of §33.6 from the whole of §33.2.
What is the speed, and has anyone changed it? A flow that fell when the speed fell is not a finding.
What is the inlet pressure? A more negative inlet pressure with a falling flow is a drainage problem (Chapter 34). An unchanged inlet pressure with a falling flow is not.
What is the pressure drop across the membrane, divided by the flow? Rising is Chapter 32's territory; flat excludes it.

And the observation that outranks all four: the sound. A change in the pump's note, or new vibration, is a pump-head finding and it precedes most of the numbers.

33.2 "Pump failure" is almost never the pump

🔬

Evidence · How often the pump is actually the problem

The figure, from ECMO: A Practical Guide to Management, in its complications and emergencies chapter: "With advances in technology, pump failure has drastically reduced. While extracorporeal pumps continue to pose risks, pump failure only occurs in 1.1% of patients." For comparison, the same chapter puts accidental decannulation at about 1.3 percent — which belongs to Chapter 34.
The list that reframes the subject, from the ECMO simulation text's transport chapter: "Common causes of pump failure are transducer errors, severe hypovolemia, pump cone failure, or obstructed cannulae. It is uncommon for the actual electrical panel and pump to fail." Two of those four are not the pump at all, and one of them is not even the circuit.
The Red Book supplies the two commonest mimics by name. For a sudden fall in flow at constant speed: "often caused by decreased blood return to the pump, referred to as access insufficiency." And for the more frightening presentation: "When a console failure is suspected, a flow probe malfunction must first be excluded as this can mimic the appearance of console failure."
And one cause that presents as pump stoppage and is not the pump at all: "A massive circuit breach will result in pump stoppage."

Certainty: moderate for the 1.1 percent figure — a single expert source quoting a registry-derived reference, with no denominator or definition given here; high that the mimics listed are real and common, since three independent primary sources name them.

⚠️

Danger · Six things make the flow fall, and raising the pump speed is the wrong first move in four of them

Work the list in this order. It is ordered by how quickly each can be excluded, not by frequency.

The measurement. A flow probe that has slipped, been re-sited, or failed reads low or zero with a perfectly functioning pump. The Red Book requires this to be excluded before console failure is diagnosed. Look at the impeller.
The patient's preload. Hypovolaemia, bleeding, diuresis, a rising intrathoracic pressure, tamponade, a pneumothorax, a Valsalva, a turn in the bed. Drainage falls, the inlet pressure becomes more negative, and the flow follows. Raising the speed here increases the negative inlet pressure, worsens haemolysis and makes chatter worse.
The drainage. Cannula malposition, migration, kinking, a clot at the tip, or a cannula sucking onto the vessel wall. Chapter 34 owns this and Chapter 9 owns the chatter it produces. Raising the speed makes it worse, for the same reason.
The afterload. On venoarterial support, a rising mean arterial pressure directly reduces flow at unchanged speed, because a non-occlusive pump cannot overcome it — Chapter 13 owns that physiology and calls it the afterload-sensitive pump. On any configuration, a clot in the return limb or membrane does the same (Chapter 31, Chapter 32). Raising the speed here does restore flow, and does nothing about the cause.
Circuit integrity. A breach, a disconnection, a damaged stopcock or pigtail. Chapter 35 owns this. Raising the speed accelerates whatever is being lost or entrained.
The pump. Decoupling, console failure, power failure, or a clot in the head. This is the last item on the list and the first word in everybody's mouth.

The two moves that are safe before the diagnosis is made: look at the impeller, and look at the inlet pressure. Everything else waits for those two.

And the one move that is never safe: raising the speed to make an alarm stop. In four of the six it makes the underlying problem worse, in one it conceals it, and in the sixth it does nothing at all.

33.3 The centrifugal pump's safety feature is its failure mode

🧠

Physiology · Two pumps, two ways of being dangerous, and each hazard is the other's solution

The roller pump is occlusive. It compresses a length of tubing in a raceway and advances whatever is in front of it. The ECMO simulation text describes what follows: it uses gravity for inflow, so it requires longer circuits and the patient elevated above the pump head; it generates negative pressure on the inflow side, which can cause haemolysis, and positive pressure on the outflow side, which can rupture the tubing if forward flow is restricted for any reason. "Tubing disruption can occur instantaneously both from roller head friction causing weakening of the tubing in the raceway or from acute occlusion of the circuit distal to the roller heads." The consequences are organisational: a person at the bedside watching the device, periodic replacement of the raceway tubing, and a long cumbersome circuit. The Red Book agrees from the other side: rupture of the main tubing "is an ever-present risk with using roller pumps that can generate extremely high pressures and expose a section of tubing to continuous deformation by the pump head."

The centrifugal pump is non-occlusive. It spins blood outward and the flow depends on speed and on what is downstream. The same source states the gain and the price in one sentence: "While the pump head will spin (and hemolyze blood in the head), tubing rupture from pressure buildup will not occur."

That sentence is the chapter's second structural claim. The field traded a pump that could not stop pushing for a pump that can. Every distinctive hazard of the machine now in use is the direct consequence of the thing that made it safer.

It cannot burst the circuit, so when flow is obstructed it grinds instead — spinning against a closed path, heating and shearing the blood in the head, with no rupture to announce it.
It can stop moving blood while still being switched on, which a roller pump essentially cannot.
It does not hold flow against a rising afterload, which is why the displayed flow tracks the patient (§33.1).
And it draws rather than being fed, which permits a short, low-prime, portable circuit — and produces high negative inlet pressures that cause haemolysis and can entrain air from any open stopcock.

The history is worth one line because it explains the residual risk. Early centrifugal heads sat on a ball bearing that heated as it turned and haemolysed blood; at low flow rates blood spends longer in the head, so haemolysis was severe, and that limited their use in small children and infants. Modern heads are magnetically levitated rather than bearing-supported, heat build-up is minimised and the heads are miniaturised with very small priming volumes, so haemolysis is greatly reduced. Reduced, not abolished — and the low-flow vulnerability is structural, not historical.

Certainty: high for the engineering contrast and its consequences, which three independent primary sources describe concordantly.

33.4 Retrograde flow: what happens on venoarterial support in the seconds after a pump stops

⚠️

Danger · A stopped pump on venoarterial support is not an absence of support, it is an open arteriovenous shunt

A centrifugal circuit is non-occlusive from end to end. There are no one-way valves. ECMO in the Adult Patient states the consequence plainly: if the pump stops, the blood can stagnate — as in a venovenous configuration — or flow in the reverse direction, as in a venoarterial configuration, because the arterial blood pressure will be higher than the venous pressure, driving the blood from artery to vein.

And ECPR and Resuscitative ECMO names what that is: "In a V-A ECMO pump failure, blood flow will reverse through the centrifugal pump causing a massive left to right shunt."

Read the plumbing and the size of the problem appears. The conduit is a short, wide-bore, low-resistance tube connecting a large artery to a large vein, and the driving pressure is the patient's own mean arterial pressure. A patient whose circulation has just lost its pump now also has a wide-open arteriovenous fistula, in a heart that was too weak to manage without help a minute earlier.

So the first action is the clamp, and the clamp is a treatment, not a preparation. The Red Book's instruction is exactly this and it gives the reason: "As soon as it is determined that the pump is not functioning, the circuit should be clamped to prevent retrograde blood flow." ECPR and Resuscitative ECMO puts the clamp first in its pump-failure sequence for the same reason, and ECMO: A Practical Guide to Management's emergency algorithm begins every pump-failure pathway with CLAMP and STOP the pump.

Three things follow.

On venoarterial support, clamping is time-critical in a way it is not on venovenous support, where the failure mode of a stopped pump is stasis and clot rather than shunt. The same event has two different diseases attached to it depending on the configuration.
A console's back flow alarm exists for this and should never be silenced without looking.
This is also why a bridge exists, and why Chapter 31 §31.2 insists a bridge must be free of blood and filled with crystalloid. A bridge is the controlled version of the same short circuit.

What nobody has published: how long a patient tolerates it. No source retrieved for this chapter quantifies the shunt fraction, the rate of haemodynamic collapse, or a safe interval. The instruction is universal and the number behind it does not exist. [VERIFICATION REQUIRED.]

33.5 Decoupling: two kinds, and reseating the head only fixes one

🧮

The distinction, and why the sources draw it

A centrifugal pump head is a disposable plastic chamber holding an impeller. The impeller is turned without any shaft passing through the chamber wall — the drive motor and the impeller are coupled magnetically across that wall. Decoupling means that coupling has been lost, and it happens in two entirely different ways.

The emergency algorithm in ECMO: A Practical Guide to Management lists them as two separate entries with two separate pathways: "PHYSICAL Pump Head DECOUPLING" and "MAGNETIC Pump Head DECOUPLING (hemological)." The word in its parentheses is the whole point.

	Physical decoupling	Magnetic decoupling
What has happened	The pump head has come out of its seat in the drive unit	Thrombosis or air inside the head has unbalanced the magnetic field between drive and impeller
What is heard and seen	ECPR and Resuscitative ECMO: "the pump can become noisy with excessive vibrations"; the head is visibly out of position	The same noise and the same vibration, with the head correctly seated
What fixes it	Stop flow, reseat the pump head, reinitiate flow. ECPR and Resuscitative ECMO gives exactly this sequence	Reseating does nothing. The Practical Guide describes a full reset — turning the console off or the flow to zero, then resetting until the magnet and flow resume — and its algorithm gives clamp and turn the speed to zero, call for help and a backup pump, then raise the speed to 1,500 and unclamp, and either return to the previous speed or change the pump
What it really is	A mechanical event	A clot or air problem wearing a mechanical costume. Chapter 31 §31.3 owns the clot; Chapter 35 owns the air

The discriminator is not the noise, because the noise is the same. It is whether the head is out of its seat — which is a thing you look at, not a thing you measure.

And the second discriminator is what preceded it. A pump that became noisy over hours, in a patient whose plasma free haemoglobin has been rising, is Chapter 31 §31.3's pump head thrombosis and not a mechanical failure at all — and in that cohort clot was documented in every pump head removed. A pump that became noisy in the second after somebody moved the cart is physical.

Certainty: moderate. The two-category distinction and both management pathways come from expert sources describing established practice; no study retrieved for this chapter reports how often either occurs, or how often reseating succeeds.

💡

Clinical pearl · The three redundancies that turn a pump failure into an inconvenience

A secondary pump head at the bedside, which the Practical Guide names as the standard mitigation for magnetic field imbalance in systems built for it.
A backup console or motor, or a hand crank — the Red Book notes that some systems employ a hand crank while others require a second console or motor, and the ECMO simulation text makes knowing where the hand crank is, and having practised the technique specific to your own circuit, an explicit objective of battery-shutdown simulation.
A pre-primed backup circuit. ECPR and Resuscitative ECMO recommends considering one in any programme; the Practical Guide notes that many centres keep one on units caring for these patients.

And the equipment list itself, reproduced from ECPR and Resuscitative ECMO because Chapter 90 will want it as a checklist: sterile tubing clamps · shears · intravenous fluid · a tie-gun and tie-straps · appropriately sized connectors and tubing · sterile gloves and gowns · facemasks, shields and bouffants · skin antisepsis sticks · a secondary console or hand crank · a secondary motor · a secondary flow probe.

The last item on that list is the one that matters most often, because §33.2 establishes that a failed flow probe is a commoner event than a failed pump and looks exactly like one.

33.6 Console and power failure, and the two indications for the hand crank

⚠️

Danger · The hand crank has exactly two indications, and using it outside them costs time

The Red Book's sequence for a suspected console or power failure is short and every step earns its place.

Exclude a flow probe malfunction first, because it mimics console failure exactly.
Clamp the circuit, to prevent retrograde blood flow (§33.4).
Confirm the impeller is not spinning. Not the display — the impeller.
Remove the pump head and transfer it to the hand crank or backup console, which restores flow and supports the patient until the problem can be resolved.

And then the rule that most units have never been told: "Loss of power and console failure are the only indications to employ the hand crank. If the pump motor is still functioning, then transitioning to the hand crank will only cause delays and distract from other troubleshooting."

That sentence contradicts the instinct a simulation curriculum builds. The hand crank is the most memorable piece of equipment on the cart and the most satisfying thing to reach for. It is the right answer to two problems and the wrong answer to the other four in §33.2's list, where it substitutes a slow manual pump for a working motor and occupies the pair of hands that should be finding the actual cause.

How likely is the scenario it is for? ECMO: A Practical Guide to Management: pump failure from power loss rarely occurs, because of built-in backup batteries — most systems revert to battery power in the event of loss of electricity. The Red Book's scenario is specifically a catastrophic console malfunction, or a loss of mains power in combination with a depleted battery. Both conditions, not either.

So the practical form of the rule is a question asked before the crank comes out of its bracket: is the impeller spinning, and is the battery indicator showing charge? If the answer to either is yes, the crank is the wrong tool.

🚩

Pitfall · The person fixing the circuit cannot also be the person looking after the patient

The Red Book states it as a requirement rather than a suggestion: "The personnel responsible for troubleshooting the ECLS circuit should not simultaneously be responsible for patient management."

This is the single most transferable sentence in the chapter and it costs nothing to implement. During a pump stoppage the patient needs ventilator escalation, vasoactive support and possibly chest compressions, while the circuit needs clamping, diagnosis and a pump head moved from one motor to another. One person attempting both does neither, and the failure mode is that the circuit work wins because it is in front of them. ECMO: A Practical Guide to Management's algorithm builds this in: every pump-failure pathway contains "INCREASE ventilator and inotropic support" as a parallel action, not a subsequent one. The ECMO simulation text makes the same point from the constraint side, describing a transport vehicle in which the physician managing the code is also the only person free to ventilate — and concluding that roles must be delineated in advance, because in the moment there is nobody to delegate to.

And the blood, because the Red Book corrects a widespread practice here. "The historical practice of having blood products available in the room is excessive and unnecessary." What is required instead: crossmatched blood available in the blood bank for every patient, and a hospital massive transfusion protocol that makes at least four units of uncrossmatched packed red cells available for immediate use. Chapter 46 owns transfusion; what belongs here is that the preparation is a protocol, not a fridge in the room.

33.7 The pump as the place clot arrives

🔬

Evidence · What 53 explanted circuits contained, and where

Figueroa Villalba CA, McMullan DM, Reed RC, Chandler WL. "Thrombosis in Extracorporeal Membrane Oxygenation (ECMO) Circuits." ASAIO J 2022;68(8):1083–1092. DOI 10.1097/MAT.0000000000001605. Abstract read; full text not retrieved. 53 ECMO circuits from 46 paediatric patients, with the tubing, pump and oxygenator examined for visible thrombi and representative samples taken for histological, immunofluorescence and immunohistochemical analysis. This is precisely the examination Chapter 31 §31.8 asks programmes to perform on every circuit they remove, and its senior author is the same Chandler who anchored Chapter 29.

Thrombi were found in 81 percent of circuits. By location and clinical significance, as reported:

Site	Proportion of circuits	What it did
Pump axle thrombi, with fragments trapped in the oxygenator	45 percent	Listed among the less clinically significant findings
Arterial tubing	30 percent	In one case, aortic and mesenteric artery thrombosis followed by bowel infarction
Venous tubing	26 percent	In 11 percent of cases these grew large enough to break off and embolise to the pump, resulting in increased haemolysis
Connectors	26 percent	Grouped with the clinically significant thrombi
Deep oxygenator membrane	15 percent	Less clinically significant
Inflow oxygenator membrane surface	11 percent	A rapidly increasing pressure drop across the oxygenator over 1 to 2 days, oxygenator failure and circuit replacement — associated with intravascular venous thrombosis and bacterial infection before ECMO began

The safety signal: antifibrinolytic therapy during ECMO was associated with an increased risk of pump thromboembolism.

The reading this chapter takes, which is not the paper's framing. The pump is not only a place where clot forms — it is the place clot arrives. A thrombus that grew somewhere else entirely, in the venous limb or at a connector, breaks off and lands in the head, where it is ground rather than passed on to the patient. The pump is the circuit's filter of last resort, and it pays for that in haemolysis. That reframes a rising plasma free haemoglobin: it may be reporting a clot that is nowhere near the pump.

Certainty: low to moderate. Forty-six patients, all paediatric, single centre, abstract read only; the proportions are of circuits examined rather than of patients supported, and no exposure denominators are given for the antifibrinolytic association. What it establishes firmly is that thrombus is near-universal, that the pump axle is one of the commonest sites, and that embolisation INTO the pump is a real and quantified route to haemolysis.

Two things it is owed to, and one it adds to. Chapter 31 is owed the site distribution as the first systematic explanted-circuit dataset in the book. Chapter 32 is owed the inflow-surface finding, which is the histological correlate of its rapidly rising pressure drop. And the antifibrinolytic signal sits directly beside Chapter 31's finding that 2 of 4 patients given activated factor VII developed acute oxygenator thrombosis: two cohorts, two different prohaemostatic drugs, two circuit-thrombosis signals, neither quantified. Chapter 42 is owed both together.

33.8 Controversy 1 · Centrifugal or roller?

⚖️

The clinical question. Almost every adult programme now uses a centrifugal pump, for the reasons §33.3 sets out. Registry data in neonates keep suggesting the older technology does better. Is that a real finding, and does it mean anything for adults?

The anchor study, read as complete full text. Ündar A, Kunselman AR, Barbaro RP, Alexander P, Patel K, Thomas NJ. "Centrifugal or Roller Blood Pumps for Neonatal Venovenous Extracorporeal Membrane Oxygenation: Extracorporeal Life Support Organization Database Comparison of Mortality and Morbidity." Pediatr Crit Care Med 2023;24(8):662–669. DOI 10.1097/PCC.0000000000003251, PMC11069414.

The design is the point. Earlier registry analyses had found centrifugal pumps associated with higher mortality and complications — one in 8,366 patients under 10 kg (2011–2016), propensity-matched, and one in 14,026 patients under 18 (2010–2015) reporting more kidney injury, hyperbilirubinaemia, circuit thrombosis and haemolysis — but both pooled pulmonary, cardiac and ECPR patients across venovenous, venoarterial and venovenoarterial support. This study was built to remove that heterogeneity: ELSO Registry 2016–2020, neonates 28 days or under, primary pulmonary failure, cannulated via the right internal jugular vein with a dual-lumen venovenous cannula, polymethylpentene oxygenators, single ECMO run only. From 1,742 raw paediatric venovenous runs, 612 neonates remained: 340 centrifugal, 272 roller. The authors state their hypothesis in advance: that centrifugal pump use would be associated with GREATER odds of survival.

Position A: the association is real, and it survived being tested properly.

In the multivariable model, centrifugal pump use was associated with LESSER odds of survival: OR 0.53 (0.33 to 0.84), p under 0.008 — the opposite of the prespecified hypothesis, in the direction two earlier and larger registry analyses had already found.
Thrombosis and clots in the circuit components were associated with lesser odds of survival: OR 0.28 (0.16 to 0.60), p under 0.001.
The authors note that several high-volume neonatal and paediatric centres returned to roller pumps only a few months after transitioning to centrifugal.

Position B: read the univariate result and the complication data, and the picture changes shape.

On univariate analysis there was no survival difference at all: 272 of 340 versus 228 of 272, OR 0.77 (0.51 to 1.17), p equal to 0.22. The significant result appears only after adjustment.
And the complications ran the other way. Centrifugal pumps had FEWER mechanical complications (85/340 versus 95/272; OR 0.62, 0.44–0.88, p equal to 0.008) and FEWER circuit thromboses (30/340 versus 48/272; OR 0.45, 0.28–0.73, p equal to 0.001), and MORE haemolysis (52/340 versus 12/272; OR 3.79, 2.07–7.49, p under 0.001).
Haemolysis itself was not independently associated with survival (OR 0.60, 0.31–1.19, p equal to 0.14), and the authors explicitly note they cannot exclude an effect in either direction.
The groups were not otherwise comparable. Roller-pump patients received more neuromuscular blockade (OR 1.73) and more systemic steroids (OR 1.98), more dopamine, dobutamine, alprostadil and surfactant, and markedly less high-frequency ventilation (OR 0.25, 0.14–0.44) — while centrifugal patients received more adrenaline, noradrenaline, sildenafil and epoprostenol. These are different patients being treated by different units.
The authors' own leading explanation is confounding by centre: roller pumps "have been used in more established high-volume ECMO centers for several years," while centrifugal pumps "were just recently introduced in new, low-volume centers with limited ECMO experience." They could not identify centre volume or experience from the registry and say so.

Where the evidence actually sits, and one methodological point this book adds.

The complication data are internally coherent and are the most useful part of the study: the centrifugal pump clots less and grinds more. That is exactly what §33.3's engineering predicts and exactly what Chapter 31 §31.3 found at the level of a single pump head. Certainty: moderate, and it is the finding to carry forward.
The survival estimate should be held much more loosely than its confidence interval suggests, for a reason the paper does not raise. The multivariable model adjusts for haemolysis, mechanical complications and circuit thrombosis. Those are not confounders — they are the mechanisms through which a pump type could plausibly affect survival at all. Adjusting for a variable on the causal path between exposure and outcome removes part of the effect being measured, and can move an estimate in either direction unpredictably. The estimate moved from a non-significant 0.77 to a significant 0.53 on adjustment, which is the direction that should prompt most caution. This is this book's reading and is not stated by the authors.
And the authors' own confounding-by-centre explanation is not a minor caveat. If roller pumps live in experienced high-volume units and centrifugal pumps arrived in new ones, then the study may be measuring programme maturity and calling it pump type.

What this book concludes.

Do not change adult practice on this. The entire evidence base is neonatal and paediatric, the mechanisms differ with body size — low flow rates and long residence times in the head are the specific centrifugal vulnerability, and they are a neonatal problem before they are an adult one — and no adult comparison was retrieved.
Do carry the complication trade-off, which is mechanistically secure: expect less circuit thrombosis and more haemolysis from a centrifugal pump. That should change what is monitored, not what is bought.
In neonatal and small-infant practice, treat this as a genuinely unresolved question and not a settled one in either direction, and note that the field's own high-volume centres have voted with their equipment in both directions.

What would resolve it. The authors' own recommendation: independent investigators directly comparing centrifugal and roller pumps with their respective complete circuits, with centre volume and experience recorded, and with brands identified — the registry did not release pump, cannula or oxygenator brand names for this analysis, and the authors note that different dual-lumen cannulas alone can differ by 100 percent in pressure drop.

Certainty: moderate that the complication profiles differ in the direction described; low for the survival association, which is unadjusted-null, adjusted-significant, possibly over-adjusted, and confounded by centre on the authors' own account; none for any inference to adult practice.

33.9 Controversy 2 · Should the console be allowed to reduce the speed by itself?

⚖️

The clinical question. Several modern consoles will automatically reduce pump speed when the inlet pressure becomes too negative, and restore it when the pressure recovers. The Red Book describes the feature as an integrated pressure regulator with an operator-set negative pressure limit; a 2026 appraisal of ECMO monitoring describes one commercial implementation that continuously monitors venous inlet pressure and automatically reduces pump revolutions if excessive negative pressure is detected, restoring them once the pressure stabilises. Should this be switched on?

Position A: yes, because the harm it prevents is the harm this chapter keeps describing. Excessive negative inlet pressure is the direct cause of haemolysis from suction, of chatter, of air entrainment from any open stopcock, and of cannula trauma against a vessel wall. It is also produced by the single commonest wrong reflex in Part VI — raising the speed to make a low-flow alarm stop (§33.2). An automatic limiter enforces the correct response faster than any human, at three in the morning, with no one in the room, and it does so before the free haemoglobin rises rather than after. Chapter 31 §31.3 lists excessive suction from cannula malposition, with high negative inflow pressures and a pump speed disproportionate to the flow achieved, among the five causes of a rising free haemoglobin that must be excluded. This feature removes one of them by construction.

Position B: it silently converts a drainage problem into a flow problem, and the conversion is the signal. Every argument in this chapter has rested on reading the pair — speed against achieved flow — because a falling flow at unchanged speed is a circuit or patient finding while a lower flow at a lower speed is a decision. An autoregulating console makes that decision, continuously, without being asked. What the chart then shows is a patient whose flow is a little lower than yesterday, at a speed a little lower than yesterday, with the inlet pressure back within limits — which is exactly the appearance of a stable patient, and exactly the appearance of a patient who is quietly becoming hypovolaemic or whose cannula is migrating. The same 2026 appraisal lists the general hazards of automated systems in this setting and they apply precisely here: skill atrophy, mode confusion, and automation bias, with the observation that it is not conclusive whether current clinical decision support systems improve patient outcomes.

Where the evidence actually sits: there is none. No study retrieved for this chapter compares outcomes, haemolysis, or missed diagnoses with the limiter enabled against disabled. The feature is described in equipment documentation and in review articles and has never been evaluated. [VERIFICATION REQUIRED.]

What this book concludes, and it is a reconciliation rather than a choice.

Leave the limiter on. The harm it prevents is real, immediate and mechanistic; the harm it may cause is a delay in recognition.
And make its interventions visible. An automatic speed reduction is an event, and by Chapter 30's rule an event that is not written down did not happen. A console that has reduced the speed by itself has just made a diagnosis on the team's behalf and should be treated as having raised an alarm, not as having solved a problem.
The practical form: record the speed, the achieved flow and the inlet pressure on the same line, and treat any speed the team did not set as a finding to be explained the same day.

Certainty: high that excessive negative inlet pressure causes the harms listed; none for any effect of autoregulation on outcome, in either direction; and the recording rule is this book's reasoning rather than a finding.

33.10 A worked case: the flow that read zero

🧮

Worked example · 03:10, venoarterial ECMO, and a hand on the crank

The patient. Day 4 of venoarterial ECMO for post-infarction cardiogenic shock. Femoral-femoral. Speed 3,400 revolutions per minute, flow 3.8 L/min, mean arterial pressure 72 mmHg, noradrenaline 0.12 µg/kg/min, distal perfusion cannula in place.

The event. At 03:10 the low-flow alarm sounds and the flow display reads 0.0 L/min. The arterial trace is unchanged. The nurse calls for help. The registrar, correctly trained, moves to the cart for the hand crank.

The four readings §33.1 asks for, which take about twenty seconds.

Reading	Value	What it excludes
Is the impeller turning?	Yes, visibly, at an unchanged note	Console failure, power failure, physical decoupling — the whole of §33.6
Speed	3,400, unchanged, battery indicator showing mains	Nobody changed it; the console has power
Inlet pressure	Minus 42 mmHg, unchanged from the evening	Drainage and preload — §33.2 items 2 and 3
Pressure drop across the membrane, and post-membrane gas	Unchanged; post-membrane oxygen tension 430 mmHg	Membrane obstruction, and any afterload rise on the circuit side

The patient's own readings, taken in parallel by the second person. Mean arterial pressure 71 mmHg. Pulse oximetry unchanged. The right radial gas drawn shortly afterwards is unchanged from six hours earlier. Nothing about the patient has changed.

The reading. A circuit in which the impeller is spinning, the inlet pressure is unchanged, the membrane pressure drop is unchanged, the post-membrane gas is unchanged, and the patient is unchanged, is a circuit that is still flowing. The only element in the chain that is not corroborated by anything else is the flow measurement itself.

The diagnosis is a flow probe malfunction, which the Red Book names as the specific mimic that "must first be excluded" before console failure is diagnosed, and which the ECMO simulation text lists first among the common causes of apparent pump failure — transducer errors. The probe had been repositioned during a dressing change two hours earlier.

What the hand crank would have done. Reaching for it would have required clamping the circuit and removing the pump head from a working motor — that is, deliberately stopping a pump that was working, in a venoarterial patient, creating the arteriovenous shunt §33.4 describes, in order to replace a functioning machine with a slower manual one. The Red Book's rule exists for precisely this moment: loss of power and console failure are the only indications, and if the motor is still functioning the crank causes delay and distraction.

What should happen, in order.

Look at the impeller. This one observation decides the next ten minutes.
Call for help and put a second person on the patient, per the Red Book's separation-of-roles requirement — the patient still needs watching whether or not the circuit is broken.
Corroborate the flow from something that is not the flow probe: the inlet and pressure-drop readings, the post-membrane gas, the patient's own arterial gas and saturation, and the colour difference between the two limbs. On a circuit that has genuinely stopped, the return limb goes dark within seconds.
Reseat or replace the flow probe — which is why §33.5's backup equipment list has a spare one on it.
Write down what happened, with a time. An alarm that resolved without an intervention is exactly the transient event Chapter 30 §30.1 says the chart loses.

The counterfactual that makes this case worth reading. Change one reading — the impeller is NOT turning — and every subsequent step reverses. Clamp immediately, because on venoarterial support a stopped pump is an open arteriovenous shunt and the clamp is the treatment. Confirm the impeller is still. Then, and only then, the crank or the backup console.

The teaching point. The first question is never "what do we do?" It is "is the impeller turning?" — and the honest reason that question gets skipped is that the alarm is loud, the display says zero, and the hand crank is the thing everybody has practised.

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

33.11 What this chapter could not close

🚩

Open questions, recorded rather than concealed

No study quantifies how long a patient on venoarterial support tolerates retrograde flow through a stopped pump. Every source instructs an immediate clamp; none gives a shunt fraction, a rate of collapse, or a safe interval. The instruction is universal and the number behind it does not exist. This is the single most consequential unmeasured quantity in the chapter.
The 1.1 percent figure for pump failure comes from one expert source citing a reference, with no denominator, definition or date range given. Whether it counts decoupling, console failure, flow-probe failure, or only the motor is not stated. Every incidence claim about pump failure in this chapter should be read with that in mind. [VERIFICATION REQUIRED.]
How often does a flow probe fail, and how often is it mistaken for the pump? Three sources name it as the mimic to exclude first. No source retrieved reports how often it happens. Given that the wrong answer leads to deliberately stopping a working pump, this absence is out of proportion to the difficulty of measuring it.
Nobody has reported how often physical decoupling occurs, how often reseating succeeds, or how often a magnetic decoupling turns out to be clot rather than air. The two-pathway management is universal expert practice with no denominator behind it.
Autoregulating consoles that reduce pump speed automatically have never been evaluated against any outcome, in either direction. §33.9's conclusion is a reconciliation of mechanisms, not a reading of evidence.
The centrifugal-versus-roller question has no adult data at all. The entire literature is neonatal and paediatric; the mechanism most implicated — prolonged residence time in the head at low flow — is a small-patient problem by construction. An adult comparison was not retrieved and may not exist.
The antifibrinolytic and pump-thromboembolism association rests on one paediatric explanted-circuit series with no exposure denominators, and sits beside Chapter 31's four-patient activated factor VII signal. Two prohaemostatic drugs, two circuit-thrombosis signals, neither quantified.
No adult explanted-circuit series was retrieved. The site distribution in §33.7 is from 46 paediatric patients; Chapter 31 asked programmes to examine every explanted circuit precisely because this data barely exists.

Corrections and additions owed to other chapters

Chapter 31 (Circuit Thrombosis) is owed §33.7's site distribution as the first systematic explanted-circuit dataset in this book — including that pump axle thrombi were present in 45 percent of circuits and that clot embolising INTO the pump from the venous limb occurred in 11 percent and produced haemolysis. This is the direct answer to what §31.8's "look at what came out" produces when a programme actually does it.
Chapter 32 (Oxygenator Failure) is owed the inflow oxygenator membrane surface thrombus finding — present in 11 percent of circuits and producing a rapidly increasing pressure drop over 1 to 2 days followed by oxygenator failure — which is the histological correlate of its own pressure-drop diagnostic, and an argument that a rapid rise means surface thrombus specifically.
Chapter 34 (Cannula Problems) is owed §33.2 items 2 and 3 entire — that the commonest reason a flow falls is upstream of the pump — and the 1.3 percent accidental decannulation figure.
Chapter 37 (Haemolysis) is owed the mechanical account: a centrifugal pump grinds because it cannot burst, its vulnerability is greatest at low flow where residence time in the head is longest, and clot arriving from elsewhere in the circuit is a named cause of a rising free haemoglobin with nothing wrong at the pump itself.
Chapter 42 (Anticoagulation Monitoring) is owed the antifibrinolytic and pump-thromboembolism association, recorded alongside Chapter 31's activated factor VII signal and the protamine hazard. Three prohaemostatic interventions, three circuit-thrombosis signals, none quantified.
Chapter 46 (Transfusion) is owed the Red Book's correction: blood products in the room are "excessive and unnecessary"; crossmatched blood in the bank plus a massive transfusion protocol delivering four units of uncrossmatched red cells immediately is the requirement.
Chapter 35 and Chapter 90 are owed the backup equipment list in §33.5, and the finding that a massive circuit breach presents as pump stoppage.
Chapter 60 (ECMO Transport) is owed the simulation text's transport-specific findings: the most common time for haemodynamic instability on transport is the circuit change, especially when converting from a roller to a centrifugal pump, because centrifugal pumps are markedly afterload-sensitive and flow may be poor in a patient managed with high-dose vasoconstrictors.
Chapter 87, Chapter 88 and Chapter 92 are owed the paired column this chapter rests on: pump speed and achieved flow on the same line, with the inlet pressure beside them — and any speed the team did not set recorded as an event.
Chapter 89 (Troubleshooting Algorithms) is owed §33.2's six-item ordered differential and §33.6's two-indication rule for the hand crank, which is the single most correctable piece of received practice in Part VI.
Chapters 88 to 90 are also owed the Red Book's staffing requirement: the person troubleshooting the circuit must not be the person managing the patient.

Carried forward and still open: Chapter 31's unanswered question of what circuit thrombus does to the patient rather than to the device; Chapter 32's unretrieved primary source for the 150 mL/min oxygen transfer threshold; and the absence of any venoarterial-specific circuit-complication dataset, now recorded by three consecutive chapters.

33.12 Evidence and certainty summary

Claim	Best evidence	Certainty
A centrifugal pump is non-occlusive, so its flow depends on rotation speed AND downstream resistance; a roller pump is occlusive and holds its flow against a rising downstream pressure	Stated in those terms by Ündar and colleagues; concordant in the Red Book, ECMO in the Adult Patient and the ECMO simulation text	High — established engineering
On a centrifugal circuit the displayed flow is therefore a measurement of the patient as much as of the pump, and only speed paired with achieved flow carries information	Follows from the above; the Red Book gives the practical form — rising speed needed for the same flow is a resistance measurement	High as a deduction; the paired-column rule is this book's, from Chapter 30 §30.2
Pump failure occurs in about 1.1 percent of patients	ECMO: A Practical Guide to Management, citing a reference	Low. No denominator, definition or date range given. Directionally useful, not quotable as a rate
Most apparent pump failure is not the pump: transducer errors, severe hypovolaemia, pump cone failure and obstructed cannulae are the common causes, and failure of the electrical panel and pump is uncommon	ECMO simulation text, transport chapter, stated directly	Moderate — a single expert source, but concordant with the Red Book's two named mimics
A flow probe malfunction mimics console failure and must be excluded first; a sudden flow fall at constant speed is usually access insufficiency; a massive circuit breach presents as pump stoppage	ELSO Red Book, circuit malfunction chapter, all three stated explicitly	High as expert practice with clear mechanisms; unknown as frequencies
Raising the pump speed to abolish a low-flow alarm worsens the problem in most of the differential	Follows from the mechanisms: more negative inlet pressure worsens haemolysis, chatter, air entrainment and cannula suction. Chapter 31 §31.3 lists suction with a disproportionate speed among the five causes of a rising free haemoglobin	High for the mechanism; this book's ordering of the differential is reasoning, not a finding
A centrifugal pump cannot rupture the circuit from pressure build-up, and therefore grinds instead — spinning and haemolysing against an obstruction	ECMO simulation text, stated in one sentence; the Red Book gives the converse for roller pumps	High
Early centrifugal heads sat on a ball bearing and haemolysed severely at low flow because of long residence time in the head, which limited their use in infants; modern heads are magnetically levitated and haemolyse far less	ECMO simulation text	High for the history; moderate that the low-flow vulnerability remains structural rather than historical
When a centrifugal pump stops on venoarterial support, blood reverses through it, producing a massive arteriovenous shunt — so the clamp is a treatment, not a preparation	ECPR and Resuscitative ECMO names the shunt; the Red Book instructs clamping specifically to prevent retrograde flow; ECMO in the Adult Patient gives the physics	High that it happens and that clamping is correct; none for how fast it matters — no source quantifies the shunt or a tolerable interval
Decoupling comes in two kinds — physical, fixed by reseating the head, and magnetic, caused by thrombus or air and not fixed by reseating — and they sound the same	ECMO: A Practical Guide to Management's algorithm lists them separately; ECPR and Resuscitative ECMO gives the reseating sequence and the noise-and-vibration presentation	Moderate — consistent expert practice; no frequencies and no success rates reported for either pathway
Loss of power and console failure are the ONLY indications for the hand crank; if the motor is still turning it causes delay and distraction	ELSO Red Book, stated in those words	High as a recommendation, and it is the correction this chapter most wants read
Power-loss pump failure is rare because consoles revert to battery; the Red Book's scenario requires console malfunction OR mains loss together with a depleted battery	Practical Guide and Red Book, concordant	Moderate to high
The person troubleshooting the circuit must not be the person managing the patient	ELSO Red Book, stated as a requirement; the Practical Guide's algorithms build ventilator and inotropic escalation in as a parallel action	High as expert consensus; no study has tested it
Blood products kept in the room are unnecessary; crossmatched blood in the bank plus a massive transfusion protocol giving four units of uncrossmatched red cells immediately is what is required	ELSO Red Book, correcting a named historical practice	Moderate — a clear expert recommendation reversing common practice
Thrombus is present in 81 percent of explanted circuits; pump axle thrombi in 45 percent; in 11 percent a venous-limb thrombus embolised INTO the pump and produced haemolysis	Figueroa Villalba and colleagues, 53 circuits from 46 paediatric patients, histologically examined. Abstract read	Low to moderate. Paediatric, single centre, 46 patients; proportions are of circuits examined, not of patients supported
Antifibrinolytic therapy during ECMO was associated with increased risk of pump thromboembolism	Same series	Very low — no exposure denominators. Recorded as a signal beside Chapter 31's activated factor VII finding
In neonatal venovenous ECMO, centrifugal pumps had FEWER mechanical complications (OR 0.62) and FEWER circuit thromboses (OR 0.45) but MORE haemolysis (OR 3.79)	Ündar and colleagues, 612 neonates, ELSO Registry 2016–2020. Full text read	Moderate, and mechanistically exactly what §33.3 predicts. The finding to carry forward from this study
Centrifugal pump use was associated with lesser odds of survival in neonatal venovenous ECMO (adjusted OR 0.53, 0.33–0.84)	Same study — but univariate OR 0.77 (0.51–1.17), p equal to 0.22; the association appears only after adjustment for haemolysis, mechanical complications and thrombosis	Low. Adjustment was made for variables on the causal path (this book's reading, not the authors'); the groups differed markedly in co-treatment; and the authors' own leading explanation is confounding by centre volume and experience, which the registry did not let them measure
Nothing in the centrifugal-versus-roller literature applies to adults	All three registry analyses are neonatal or paediatric; no adult comparison was retrieved	High as a statement about the state of the evidence
Consoles that automatically reduce speed on excessive negative inlet pressure have never been evaluated against any outcome	The feature is described by the Red Book and by a 2026 monitoring appraisal; no evaluation was retrieved	High as a statement about the evidence; §33.9's conclusion is a reconciliation of mechanisms

33.13 Key points

A pump alarm is a symptom, not a diagnosis. Pump failure occurs in about one patient in a hundred; a falling flow occurs most days.
On a centrifugal pump the console sets one thing — speed — and measures another that it does not control. The flow reading moves whenever the patient's preload, drainage or afterload moves.
Record pump speed and achieved flow on the same line, with the inlet pressure beside them. A falling flow at unchanged speed is a finding; the same flow at a lower speed is a decision somebody made, and a chart with only one of the pair cannot tell them apart.
Rising speed needed to hold the same flow is a resistance measurement, free, on any console — the same information a transmembrane pressure transducer gives (Chapter 32).
Six things make the flow fall, in this order of exclusion: the measurement, the preload, the drainage, the afterload, circuit integrity, and last of all the pump. Two of them are not the circuit at all.
Raising the speed to silence a low-flow alarm is the wrong first move in four of the six. It makes negative inlet pressure worse, which means more haemolysis, more chatter, more air entrainment and more cannula suction.
The two moves that are always safe before a diagnosis: look at the impeller, and look at the inlet pressure.
A flow probe failure mimics console failure exactly and is commoner than it. The Red Book requires it to be excluded first, and it is first on the simulation literature's list of causes of apparent pump failure.
The centrifugal pump's safety feature is its failure mode. It cannot burst the circuit, so when obstructed it grinds; it can stop moving blood while still switched on; it does not hold flow against afterload; and it draws rather than being fed, so it produces high negative inlet pressures.
Its haemolytic vulnerability is greatest at LOW flow, because residence time in the head is longest there — which is why it was a neonatal problem before it was anything else, and why it matters during weaning.
On venoarterial support a stopped pump is not an absence of support. It is an open arteriovenous shunt, driven by the patient's own arterial pressure through a short wide-bore tube. Clamp first, and clamp because of the shunt.
The same event has two different diseases attached depending on configuration: on venovenous support a stopped pump causes stasis and clot; on venoarterial it causes shunt.
Decoupling comes in two kinds and they sound identical. Physical decoupling is fixed by stopping flow and reseating the head. Magnetic decoupling is thrombus or air wearing a mechanical costume, and reseating will not fix it.
A pump that became noisy over hours in a patient whose free haemoglobin has been rising is pump head thrombosis (Chapter 31 §31.3), not a mechanical failure.
Loss of power and console failure are the only two indications for the hand crank. If the motor is still turning, using it means deliberately stopping a working pump — and on venoarterial support, opening a shunt to do it.
Before the crank leaves its bracket, ask two questions: is the impeller spinning, and does the battery show charge? A yes to either means the crank is the wrong tool.
The person troubleshooting the circuit must not be the person managing the patient. Ventilator and vasoactive escalation is a parallel action, not a subsequent one.
The pump is not only where clot forms, it is where clot arrives. In 11 percent of explanted circuits a venous-limb thrombus broke off and embolised into the pump, producing haemolysis — so a rising free haemoglobin may be reporting a clot nowhere near the pump.
Centrifugal pumps clot less and grind more. That trade-off is the secure finding from the registry comparison; the survival signal in neonates is not secure, and none of it applies to adults.
A console that reduced the pump speed by itself has made a diagnosis on your behalf. Treat it as an alarm that has been answered, not a problem that has been solved.

📌

Chapter answer: The pump is the component named in the alarm and the least likely thing to have gone wrong. What the console actually controls is a speed; what it displays is a flow, and on a non-occlusive pump that flow is a measurement of the patient — of what can be drained and what must be pushed against. So the falling number has six possible owners and the pump is the sixth, while the move everyone reaches for first, turning the speed up, makes four of the other five worse. Two observations settle almost all of it and neither requires touching anything: is the impeller turning, and what is the inlet pressure. When the impeller genuinely is not turning, the priorities invert completely — because a centrifugal circuit has no valves, and on venoarterial support the patient's own arterial pressure will drive blood backwards through the dead pump into a vein. The clamp then is not preparation for a repair. It is the treatment for a shunt that was opened the moment the impeller stopped.