ECMO

What it is, and what it is not

ECMO is plumbing. Strip away the acronym and you have five components in series.

That is the entire machine. Everything else is a pressure transducer, a bubble detector, a bridge, a set of clamps and a very expensive cart.

Which brings you to the only question that really matters before cannulation, and it should be asked out loud, in the room, by the most senior person present.

VV versus VA

The fundamental difference is one sentence. VV supports the lungs and leaves the native heart doing all of the work. VA supports the lungs and the circulation, and charges the heart afterload for the privilege.

Veno-venous

Blood is drained from a central vein, oxygenated, and returned to a central vein. The whole circuit sits upstream of the right heart, so native cardiac output is unchanged by the machine. That has a consequence people miss: arterial saturation on VV is a mixing problem. Circuit blood mixes in the right atrium with whatever venous return escaped the drainage cannula, and the mixture crosses lungs that add little. So saturation depends on the ratio of ECMO flow to native cardiac output. Let a VV patient become septic and hyperdynamic and their saturation falls, not because the circuit failed but because it is now capturing a smaller fraction of a larger output.

Practically: on VV you accept saturations in the 80s. A patient at 85 percent with a normal lactate, warm peripheries, good urine output and a clear sensorium is adequately oxygenated. Chasing 95 percent by cranking flow is how you produce hemolysis.

Veno-arterial

Blood is drained from a vein and returned to an artery, so the circuit sits in parallel with the heart and provides gas exchange and cardiac output. At full flow it can support a patient with essentially no ejection at all.

The price is afterload. Blood returned to the femoral artery travels retrograde up the descending aorta, and the left ventricle must eject against that column. Often it cannot, which produces the two problems that define VA management: differential hypoxia and left ventricular distension. Both have their own sections because both kill people.

Indications and contraindications

VV

Severe respiratory failure, usually ARDS, with refractory hypoxemia or refractory hypercapnic acidosis despite optimized conventional management. "Optimized" means something specific: lung protective tidal volumes, an adequate PEEP strategy, neuromuscular blockade if indicated, and a trial of prone positioning. If the patient has not been proned, they are usually not a referral, they are a patient who has not been proned. Also: severe air leak, status asthmaticus with unmanageable hypercapnia, and bridge to lung transplant.

Severity is often described with the Murray lung injury score, which combines the PaO2 to FiO2 ratio, PEEP, respiratory system compliance and the number of involved quadrants on the chest film. Its value is not the number. It is that it forces you to weigh four things at once rather than staring at one bad gas.

Two trials frame the modern conversation, and know them qualitatively rather than quoting figures. CESAR randomized severe respiratory failure to conventional care or to transfer to an ECMO center for consideration of ECMO, and the referral strategy did better. The fair criticism is that it tested a referral pathway including expert protocolized care, not the cannulas themselves. EOLIA randomized very severe ARDS to early ECMO or to conventional management with rescue ECMO available, and did not meet its primary endpoint. It also had substantial crossover of dying control patients to rescue ECMO, which is exactly the crossover you want ethically and exactly the crossover that dilutes a mortality signal. Later reanalyses have been read as favorable. The honest summary: the effect is smaller than enthusiasts imply, and ECMO is a reasonable rescue therapy in properly selected severe ARDS at an experienced center.

VA

• Refractory cardiogenic shock not responding to inotropes and vasopressors, from infarction, fulminant myocarditis, decompensated chronic heart failure, cardiotoxic overdose or peripartum cardiomyopathy. Myocarditis and overdose deserve special mention: both have some of the best outcomes in all of ECMO, because both recover completely if you can hold the circulation together for a few days.
• Massive pulmonary embolism with obstructive shock or arrest, as a bridge to thrombolysis, catheter-directed therapy or embolectomy.
• Extracorporeal CPR for refractory arrest. This works only under narrow conditions: witnessed arrest, immediate good bystander CPR, short low-flow time, a plausibly reversible cause, and a program that can cannulate in minutes. Outside those conditions it is futile, and every program that has done it badly learned that expensively. See running a code for what has to have gone right first.
• Post-cardiotomy shock, failure to wean from bypass, usually cannulated centrally.
• Refractory ventricular arrhythmia, maintaining perfusion while you ablate, cool or load antiarrhythmics.
• Bridge to decision, transplant or durable VAD.

Contraindications

Circuit checks on rounds

This is the part you will actually be asked to do. Walk the circuit from the patient, through the machine, and back to the patient, with your hands and your eyes, every time.

Why. Blood leaving a functioning membrane is essentially fully saturated. The hemoglobin dissociation curve flattens at a modest partial pressure, so once the sweep contains enough oxygen to saturate the blood touching the fibers, more sweep adds almost nothing to oxygen content. The only remaining variable is how much of that blood you deliver, which is blood flow. On VV that is the fraction of native cardiac output you capture and re-oxygenate; on VA it is the fraction of the systemic circulation coming from the circuit rather than the sick heart. Either way, oxygen delivery scales with flow.

Carbon dioxide behaves completely differently. It is far more diffusible across the membrane, and its transfer is limited by the partial pressure gradient, not by the membrane. The sweep contains essentially no CO2, so it continuously washes away what has just crossed, holding the gas-side partial pressure near zero and the gradient maximal. Turn the sweep up and you wash harder, the gradient stays steep, more CO2 leaves. Turn it down and the gas side equilibrates with the blood, the gradient collapses, and clearance falls. CO2 removal is so efficient that a modest sweep at modest blood flow clears an adult's entire metabolic CO2 production, which is the whole principle behind low-flow extracorporeal CO2 removal.

The corollary residents get wrong: you cannot fix a saturation of 78 percent by turning the sweep up, and you cannot fix a PaCO2 of 70 by turning the flow up. If someone asks you for an order and you are reaching for the wrong knob, stop and say the sentence again.

Pre and post oxygenator gases

• Post-oxygenator PaO2 should be high, typically several hundred, at a high sweep oxygen fraction. It measures whether the membrane still transfers oxygen. A post-membrane PaO2 that has fallen substantially over days with unchanged sweep settings is oxygenator failure, and it is a reason to plan a circuit exchange rather than fiddle with the blender.
• Pre-oxygenator (drainage) saturation is your venous sample. On VA it is a mixed venous surrogate: falling drainage saturation with a rising lactate says delivery is not meeting demand and flow should probably go up.
• A drainage saturation that is inappropriately high on VV, in the 90s, while the patient's arterial saturation is poor, is recirculation. You are draining the oxygenated blood you just returned and doing a beautiful job of oxygenating the machine. Echo the cannula positions, increase the separation between ports, and consider that you have simply set the flow above what those positions support.

Transmembrane pressure gradient

Pressure before the oxygenator minus pressure after it, at a stated blood flow. In a clean circuit it is low, and what matters is the trend at constant flow. A gradient climbing day on day means the fiber bundle is filling with clot and resistance is rising. Always chart the flow next to it, because a gradient that rose because someone increased the flow is not the same finding at all. Read it alongside the consumption labs: rising D-dimer, falling fibrinogen, falling platelets and rising plasma free hemoglobin, with a rising gradient and a falling post-membrane PaO2, is a circuit that is dying.

Looking for clot

Take a flashlight. No monitor replaces this. Shine it through the fiber bundle from the side, looking for dark streaks and wedge-shaped areas of stasis. Inspect the headers, the manifolds at each end of the bundle, where clot preferentially forms. Look at the pump head, where clot around the impeller often announces itself as new hemolysis first. Run the light along the tubing, connectors and bridge. Then document what you saw and where, because "clot in the venous header" means nothing without "which was smaller yesterday."

Chatter, also called chugging

The drainage line shakes or slaps, the flow trace becomes erratic, and the machine may alarm for negative inlet pressure. What it means is that the pump is trying to pull more blood than the patient can give it. The vein collapses onto the fenestrations, drainage stops, the vein refills, and the cycle repeats several times a second. That is the shaking you are watching.

• Hypovolemia. Bleeding, aggressive ultrafiltration, diuresis, vasodilation. The response to a fluid bolus is diagnostic; see IV fluids for what to give.
• Cannula malposition or migration. Tip against a wall, in a hepatic vein, or too far in. Echo it.
• Mechanical obstruction. A kinked line, a flexed hip, a line trapped under a limb or a bed rail.
• Anything reducing venous return: tamponade, tension pneumothorax, abdominal compartment syndrome, high PEEP, breath stacking, coughing, agitation.

A patient who chatters every time they cough is telling you something benign. A patient who has been chattering steadily for an hour is telling you something you need to find.

Daily management

Anticoagulation

The circuit is a large foreign surface at high shear and it will clot if you leave it alone. Standard practice for most patients is unfractionated heparin, bolused at cannulation and then infused, titrated to an anti-Xa level or an aPTT, with both the target and the assay varying substantially by institution. Anti-Xa is increasingly favored because it measures heparin effect directly and is less confounded by the coagulopathy, acute phase response and factor abnormalities these patients accumulate. Activated clotting time is bedside-available and crude. Know which assay your unit titrates to, and never titrate to two at once without a rule for what to do when they disagree, because they will.

Bivalirudin, a direct thrombin inhibitor, is increasingly used as a primary agent rather than only as rescue. It is the obvious choice in suspected or confirmed heparin-induced thrombocytopenia, and it is attractive in heparin resistance and antithrombin deficiency because it does not require antithrombin to work. Its effect is more predictable in critical illness, and it has no reversal agent. Argatroban is the alternative in some programs. General heparin dosing principles are on the heparin drip page, though ECMO targets are their own animal.

Component thresholds are usually written into the protocol. Common practice is platelets above roughly 50,000 in a patient who is not bleeding and higher if they are, and fibrinogen kept in the region of 150 to 200 mg/dL with cryoprecipitate or fibrinogen concentrate, on the reasoning that fibrinogen is consumed by the circuit and often falls first. Red cell transfusion has moved toward the restrictive approach used elsewhere in the ICU, discussed on the transfusion thresholds page, with the caveat that oxygen delivery on VV depends fairly directly on hemoglobin.

Why they bleed and clot at the same time

There is a clean mechanism, so learn it rather than shrugging. High shear in the pump head and across the membrane unfolds von Willebrand factor multimers, exposing the cleavage site to ADAMTS13, which chops the largest and most hemostatically active high molecular weight multimers into smaller, less useful fragments. The result is an acquired von Willebrand syndrome, present within hours of cannulation and reversing within days of decannulation. Same mechanism as the gastrointestinal bleeding of severe aortic stenosis and continuous flow LVADs.

So the patient has fragmented von Willebrand factor, thrombocytopenia, dysfunctional platelets, consumed fibrinogen, a systemic anticoagulant infusion and several large holes in large vessels. That is the bleeding half. They also have a huge thrombogenic non-endothelial surface, stasis in the headers and the bridge, and an inflammatory state. That is the clotting half. Both are true at once. A patient can ooze from every cannula site while their oxygenator fills with clot, and neither observation invalidates the other.

Hemolysis surveillance

Check plasma free hemoglobin on a schedule, with LDH, haptoglobin, bilirubin and the hemoglobin trend. Look at the urine: pink or tea-colored urine with a stable hemoglobin is hemolysis until proven otherwise. A rising plasma free hemoglobin means red cells are being destroyed in the circuit, and the usual culprits are clot in the pump head, excessively negative inlet pressure from a chattering or undersized drainage cannula, a kinked line, or a flow demand the cannula cannot support. Free hemoglobin scavenges nitric oxide and is directly nephrotoxic, so this is not a laboratory curiosity. Rising hemolysis is a circuit exchange conversation.

Ventilator lung rest on VV

Once the machine does the gas exchange, the purpose of the ventilator changes completely. You are no longer ventilating to normalize a number. You are ventilating to keep the lung open at the lowest mechanical cost while it heals. Typical rest settings, with wide local variation, mean a plateau pressure under about 25 cm H2O, a driving pressure as low as you can get it and generally under 15, a moderate PEEP of around 10 to prevent derecruitment, a slow rate, and FiO2 turned down as far as the patient tolerates.

One specific trap at cannulation. A patient sitting at a PaCO2 of 100 for two days has a compensated intracellular and cerebrospinal fluid pH. Put them on a circuit, open the sweep wide, and you will normalize the CO2 in minutes. That abrupt drop causes cerebral vasoconstriction and rapid pH shifts, and is associated with neurologic injury. Bring the CO2 down slowly and deliberately over hours by starting the sweep low and titrating. Common early error, entirely preventable.

Sedation, mobility and nutrition

The deeply sedated, paralyzed, immobile ECMO patient is largely gone as a model, and should be. Deep sedation buys you delirium, ICU-acquired weakness, longer runs and a patient who cannot participate in their own weaning. Aim for the lightest sedation that keeps the cannulas safe. On VV, particularly with a dual lumen neck cannula, extubation on ECMO and an awake, ambulating patient are achievable and are the standard in bridge-to-transplant programs, where the difference between a bedbound candidate and a walking one is the difference between transplantable and not. Screen for delirium as you would in anyone else, as on the ICU delirium page.

The circuit also changes drug handling. Tubing and membrane sequester lipophilic and highly protein-bound drugs, and the priming volume increases the volume of distribution. Sedatives, analgesics and some antimicrobials often need higher doses than you expect. That is not tolerance, it is plastic. Dose to effect and to levels where you have them, and involve your pharmacist.

Feed them. Enteral nutrition is safe on ECMO, including on VA, and ECMO by itself is not a reason to withhold it or go parenteral. Start early, use the gut, expect a hypermetabolic patient, be cautious with high-dose vasopressors plus feeding in a low-flow gut, and remember that propofol carries a lipid load that has to be counted.

North-South syndrome

Also called Harlequin syndrome or differential hypoxia. Unique to peripheral VA ECMO with femoral arterial return, and a beautiful piece of physiology that will destroy a brain if you do not look for it.

The setup: oxygenated blood enters the femoral artery and travels retrograde up the descending aorta. Meanwhile the native heart is recovering and starts ejecting again, but the lungs are still terrible, so what the left ventricle ejects into the aortic root is deoxygenated. Somewhere in the aorta the two opposing streams meet, at a mixing zone whose position depends on the balance between native output and ECMO flow. As the heart recovers, that zone marches distally. Everything proximal to it, meaning the coronaries, the innominate, the right carotid, the right subclavian and then the left carotid, is perfused entirely by deoxygenated native ejection.

So the lower body is pink and beautifully oxygenated by the circuit while the head and the heart get venous blood. The cruel part is the paradox: this problem is caused by the heart getting better. The improvement you were hoping for is the thing producing the injury.

The fixes, in the order you usually try them:

• Improve the native lungs. The problem exists because blood crossing the pulmonary circuit is deoxygenated. Recruit, raise PEEP, raise ventilator FiO2, diurese the edema, treat the pneumonia. Make the native ejection oxygenated and there is no north-south problem, only an unusual flow pattern.
• Convert to VAV. Split the return limb with a Y connector so oxygenated blood goes to both the femoral artery and a cannula in the right internal jugular. The right atrium now receives oxygenated blood, so what the left ventricle ejects into the arch is already oxygenated. VAV needs flow monitoring on both limbs and usually a partial clamp to balance them, and your perfusion team manages it actively.
• Move the return cannula to an axillary or subclavian artery, or to central cannulation in the ascending aorta. Both abolish the problem by removing the retrograde stream.
• Or decannulate. If the heart has recovered enough to cause differential hypoxia, ask seriously whether it has recovered enough to come off VA altogether, possibly onto VV if the lungs remain the problem. Sometimes the answer to a VA problem is that this is now a VV patient.

Deliberately reducing native ejection to move the mixing zone, by deepening sedation or backing off inotropes, gets mentioned. Be careful: it treats the number and worsens LV distension, which is the other problem you are already fighting.

LV distension and venting

The second consequence of retrograde flow, and the one that quietly destroys the myocardium you were trying to save.

Femoral VA ECMO fills the aorta with a retrograde column at systemic pressure, and the failing left ventricle must now generate more pressure than that column to open the aortic valve. Often it cannot, so it barely ejects or does not eject at all. But blood keeps arriving: bronchial return to the left atrium, any aortic regurgitation, and any pulmonary flow that escaped the drainage cannula. Volume goes in and nothing comes out. The ventricle dilates, end-diastolic pressure rises, that pressure transmits back through the left atrium into the pulmonary veins, and the patient develops pulmonary edema and sometimes frank pulmonary hemorrhage. Meanwhile wall stress rises and subendocardial perfusion falls, so the myocardium you were resting is being stretched and starved instead. And with no ejection, blood in the ventricle and root is stagnant, and stagnant blood clots. An LV or aortic root thrombus ends the run and often the patient.

How to recognize it

Venting options, escalating

• Reduce ECMO flow if the hemodynamics tolerate it. Less retrograde flow, less afterload, a better chance the valve opens. Cheapest and fastest partial answer, and frequently not enough.
• Inotrope. Low-dose dobutamine, milrinone or epinephrine to give the ventricle enough contractility to open the valve at least intermittently. The goal is not more cardiac output, it is valve opening and the avoidance of stasis. It costs myocardial oxygen consumption in a heart you are trying to rest, which is exactly the tension in the decision.
• Intra-aortic balloon pump. Deflation immediately before systole drops aortic end-diastolic pressure, reduces the afterload the ventricle must overcome, and restores pulsatility to a flat circuit-driven trace. A modest unloading effect, but quick, familiar and available everywhere. Counterpulsation and timing are on the balloon pump page.
• Impella. A microaxial pump across the aortic valve pulling blood directly out of the ventricle into the aorta. True active unloading and the most effective percutaneous vent. VA ECMO plus Impella is widely called ECPELLA. It costs a second large arterial access, more hemolysis and more anticoagulation complexity; see the Impella page.
• Percutaneous pulmonary artery or left atrial drainage, a cannula placed into the PA or transseptally into the left atrium and Y-connected into the venous limb, so the ECMO pump does the venting.
• Atrial septostomy, a balloon-created interatrial communication decompressing the left atrium into the right, where the circuit can drain it.
• Surgical vent, a cannula placed directly into the LV apex, left atrium or pulmonary artery in the operating room and spliced into the venous limb. Definitive, and it requires a sternotomy.

Which one you use is a program decision. What is not negotiable is that somebody has to look for distension every day, and the resident who charts the pulse pressure and asks whether the aortic valve opened on the echo is usually the one who finds it first.

Weaning

The two modes wean by completely different mechanisms, because they support completely different things.

VV: turn the sweep down

On VV the machine's job is gas exchange, and gas exchange is what the sweep controls. So the trial is a sweep trial. Leave the blood flow exactly where it is, so the circuit keeps moving and does not clot, and turn the sweep down and eventually to zero. The membrane is now just plastic with blood running through it, and whatever the blood gas shows is the patient's own lungs.

• Confirm they are a plausible candidate first: improving compliance, improving film, resolving underlying process, modest ventilator requirement.
• Set the ventilator to what you would be willing to leave them on after decannulation, usually a lung protective mode rather than rest settings, and let it stabilize.
• Sweep to zero for a meaningful period, often several hours. A five minute trial proves nothing.
• Watch respiratory rate, work of breathing, tidal volumes and comfort as closely as the gas. A patient holding an acceptable PaO2 and PaCO2 by breathing at 40 has failed regardless of the numbers.
• Serial gases through the trial, not one at the end. If they pass, decannulate. If they fail, turn the sweep back up and try another day. There is no penalty for a failed trial and a substantial penalty for a premature decannulation.

The trap: do not turn the blood flow down to wean VV. You gain nothing, since sweep determines gas exchange, and you add stasis and clotting risk for no reason.

VA: turn the flow down, with echo at every step

On VA the machine provides cardiac output, so the trial has to test whether the heart can provide it instead. Reduce blood flow in steps and watch what the heart does, with an echocardiogram at each step performed by someone who can do it properly, not a glance at the screen.

Preconditions: minimal or no vasopressor and inotrope requirement, a normal or normalizing lactate, adequate urine output, a resolved or resolving primary insult, and pulsatility on the arterial trace showing the ventricle is actually ejecting. A patient with a flat waveform is not ready to wean, they are a patient who needs a vent. Come down stepwise, pausing long enough at each level for the hemodynamics to settle, typically toward a minimum flow your program considers safe from a thrombosis standpoint. Many units raise the anticoagulation target during a low-flow trial precisely because low flow is thrombogenic.

At each step, what you are looking for:

• LVOT VTI, the velocity time integral across the left ventricular outflow tract and the direct echo surrogate for stroke volume. Maintained or improving as circuit flow falls is a ventricle taking over. Collapsing is a ventricle that is not ready. This is the single most informative measurement in a VA wean.
• Aortic valve opening, on every beat and well, as support comes down.
• Left ventricular size and function, and specifically whether it dilates as you load it.
• Right ventricular size and function, which people forget. Plenty of these patients fail on the right ventricle, not the left, and an RV that dilates and goes flat as flow comes down is your answer.
• Filling pressures: rising wedge or PA diastolic, rising CVP, mitral inflow patterns of elevated filling pressure.
• The clinical variables regardless of the pictures: mean arterial pressure, pulse pressure, drainage saturation, lactate, urine output, and whether the vasopressor requirement climbed during the trial.

If the patient tolerates minimal flow for a sustained period with good ventricular performance, decannulation is a surgical event. Arterial cannulas usually need a cutdown or a closure device, and the leg needs assessment immediately afterwards and repeatedly for hours, because a limb can declare itself ischemic after decannulation as well as during the run.

Complications

The exit strategy, honestly

ECMO has a failure mode no other therapy has quite so completely. A ventilator can be withdrawn in a way that families broadly understand as letting a disease take its course. A machine circulating a patient's entire blood volume through a plastic box in the corner of the room is different. Turning it off looks and feels like causing a death rather than permitting one, even though it is the same act. That perception is the whole problem.

The term for what goes wrong is bridge to nowhere. Someone is cannulated in a crisis, often at 2 a.m., often by a team who correctly judged there was no time for the full conversation. The heart does not recover. The patient turns out not to be a transplant candidate, or the neurologic injury declares itself, or the malignancy is found. Now they are awake, or awake enough, stable on the circuit, not dying and not living, and the only remaining decision is one that everybody in the room experiences as killing them. Families end up consenting to their relative's death weeks after they consented to a procedure that was explained to them as a chance. That outcome is foreseeable, and it is avoided at the moment of cannulation, not later.

Three things make those conversations go better. Use the word "trial" every single time and never use the word "permanent." Define the milestone in terms the family can see for themselves, such as the echo or coming off support, rather than in numbers only you understand. And hold the scheduled meeting even when nothing has changed, especially when nothing has changed, because a family who hears from you weekly during a stable stretch will trust you when the news is bad. The mechanics are on the running a family meeting page, and the framework for the underlying decisions, including how to document code status on a patient who is on mechanical support, is on the goals of care and code status page.

One last thing, plainly. There will be a night when you are asked to cannulate someone who is going to die either way, and the honest answer is that the machine will not change that, only lengthen it. Saying so, respectfully, with a plan for comfort instead, is not withholding care. It is the correct clinical recommendation, and the fact that it is harder to say than "yes" is not a reason to say "yes."

Related: choosing between the devices, the intra-aortic balloon pump, the Impella, PA catheter numbers, arterial line and CVP waveforms, and ventilator settings and ABG interpretation for the lung half of the problem.