Chapter question: A patient is in cardiac arrest and the compressions are not working. What exactly is it that ECPR offers — and why do three randomised trials of the same intervention read as though they studied three different things?
Evidence search date: 11 September 2026.
Primary sources: ECPR and Resuscitative ECMO (Shinar and Badulak), Chapter 2 — Optimizing the Pre-ECMO Resuscitation and Chapter 3, which supply the no-flow/low-flow framework and the low-flow outcome data; and ELSO Red Book 6th edition Chapter 32 — Extracorporeal Cardiopulmonary Resuscitation in Adult Patients, including Tables 32-1, 32-2 and the ECPR programme logistics table.
Retrieved in full from the publisher: the Bayesian meta-analysis of all three randomised trials (Critical Care 2024, DOI 10.1186/s13054-024-05008-9); the updated meta-analysis and trial sequential analysis (Critical Care 2024, DOI 10.1186/s13054-024-04830-5); and the trialists' own three-trial comparison (European Heart Journal: Acute Cardiovascular Care 2023, DOI 10.1093/ehjacc/zuad071).
The three randomised trials themselves — ARREST, PRAGUE-OHCA and INCEPTION — are published in journals this project cannot reach. Their designs, populations and results here come from structured abstracts and from the trialists' own comparative paper, which is written by authors from all three trials. Rows are marked Verified = No accordingly, and where the sources disagree with each other the disagreement is printed rather than resolved.
Currency warning. The ELSO ECPR interim guideline consensus statement (Richardson, Tonna, Nanjayya et al., ASAIO J 2021;67:221–228) predates none of the trials in publication but predates all of their synthesis. It remains the most recent society statement on ECPR, and the two randomised trials that reported after it — INCEPTION and the pooled individual-patient analysis — have not been incorporated into any updated guideline. This is the largest currency gap in this book.
What this chapter covers — and what it does not
This chapter owns | Deferred to |
What ECPR is and how it differs conceptually from every other ECMO indication in this book; the no-flow/low-flow framework; what conventional CPR can and cannot deliver, and the physiological reason compression pauses are expensive; why time is the indication rather than a criterion; the sixty-minute rule and what it actually is; reperfusion as an intervention; what the three randomised trials each tested and why their results diverge; how to read the meta-analyses; the programme as the unit of intervention; and what Part III hands forward | Who is a candidate — age, rhythm, no-flow, signs of life, aetiology, exclusions → Chapter 20
Cannulation under chest compressions → Chapter 21
The first hours after flow, coronary angiography, temperature, seizures → Chapter 22
Neurological assessment, imaging, biomarkers and prognostication → Chapter 23
Recognising futility, withdrawal and organ donation → Chapter 24
Resource allocation, cost and the ethics of offering ECPR at all → Chapter 77
Moving an ECPR patient between hospitals → Chapter 60 |
19.1 ECPR is not VA ECMO for a patient who happens to have arrested
Every other ECMO decision in this book is taken about a circulation that exists. In severe ARDS the heart is beating and the question is gas exchange. In cardiogenic shock the heart is ejecting badly and the question is how much of the output to take over. Even in the sickest patient in Chapter 11, there is a blood pressure, a lactate trend, a neurological examination and — usually — a conversation.
In ECPR none of that is true. There is no native output. The pump is not supporting the circulation; from the first revolution it is the circulation. Three consequences follow, and they organise the whole of Part IV.
Physiology — why ECPR is a different kind of decision
1. The indication is temporal, not physiological. Every ECPR candidate has the same physiology at the moment of decision: no cardiac output. What separates a candidate with a realistic chance from one without is not a measurement but elapsed time and what happened during it — how long there was no flow, how long there has been low flow, what rhythm started it, and whether anyone was compressing.
2. There is no pre-initiation baseline. Chapter 14 called the neurological examination before cannulation irreplaceable, because after cannulation the patient is sedated, anticoagulated and instrumented, and no later examination can be interpreted without it. In ECPR that baseline does not exist and never will. The single most important thing you would want to know about this patient is permanently unavailable. Chapters 23 and 24 exist because of this sentence.
3. The decision is made in minutes, by people with no history. There is no time to obtain records, no time to reach a family, and frequently no name. Chapter 20 is about making that decision reproducible in advance, because it cannot be made well in the moment.
The word that carries the entire indication is undefined
ECPR is offered for refractory cardiac arrest. The Red Book states the problem plainly: "Central to ECPR stands the definition of a refractory arrest. However, definitions differ (ranging between 10-60 mins) within the literature."
A sixfold range in the definition of the condition being treated is not a minor inconsistency. It means that two hospitals can both honestly report that they offer ECPR for refractory arrest and be treating populations with entirely different prior probabilities of benefit — and it means that when a study reports its results, the reader cannot assume that its patients resemble the patients in the next study. Much of the apparent disagreement in the ECPR literature is a disagreement about the word "refractory".
Danger — the bridge to nowhere
The ISCCM Manual names it directly: "A bridge to nowhere situation can be quite common."
ECPR restores circulation. It does not restore a brain, and it does not by itself treat the cause of the arrest. If the cause is not reversible and the brain has already been injured beyond recovery, the circuit converts a death in the emergency department into a death in the intensive care unit some days later, having consumed a cannulation, a bed, a team, and a family's hope.
Chapter 24 owns the exit. The decision to start must be taken by someone who already knows how it ends if it fails.
19.2 No-flow, low-flow, and what chest compressions can and cannot do
The vocabulary of Part IV comes from the ECPR literature and is used precisely.
Term | Definition | Why it matters for ECPR |
No-flow | The interval between collapse and the start of chest compressions — the period during which an arresting patient received no bystander CPR | Pure ischaemia at normothermia. Nothing the hospital does later recovers it. It is the strongest single argument for excluding a patient, and the hardest interval to estimate reliably |
Low-flow | The interval between the start of chest compressions and the return of spontaneous or extracorporeal circulation | This is the interval ECPR shortens. It is the only part of the timeline a system can modify, and every outcome gradient in this chapter is a gradient along it |
Arrest-to-flow | Collapse to established extracorporeal blood flow — no-flow plus low-flow | The number a programme should audit itself on. It is the number the sixty-minute rule refers to |
Conventional CPR is a holding measure, and it is explicit about being one
"CPR is not a definitive therapy; it rarely produces sufficient cardiac output and gas exchange to sustain normal metabolic activity. Rather, conventional CPR is a mitigating bridge... The primary goal of conventional CPR is prolonging the window of tolerable ischemic insult."
This is worth dwelling on, because it reframes what ECPR is competing with. Chest compressions do not restore physiology; they slow the rate at which it is lost. ECPR is therefore not an alternative to CPR — it is the endpoint that CPR is buying time to reach, in exactly the way that ROSC is.
What conventional CPR delivers | Target | Certainty |
Compression rate | 100–120 per minute | High — long-standing guideline standard |
Compression depth | At least 5 cm, with complete recoil and no leaning | High |
Compression fraction | At least 80% of total resuscitation time | High |
Coronary perfusion pressure | Human observational data record no instances of ROSC when coronary perfusion pressure failed to rise above 15 mmHg; large-animal models suggest a value closer to 25 mmHg is associated with ROSC, better cerebral perfusion pressure and higher brain oxygen tension | Low to moderate — the human figure was measured late in the resuscitation; the 25 mmHg figure is preclinical |
Practical surrogate | With a normal central venous pressure of 5–10 mmHg, a diastolic blood pressure of 25–30 mmHg approximates a coronary perfusion pressure of 20 mmHg. If central venous pressure rises during low flow, 35–40 mmHg is the more conservative aim | Low — arithmetic applied to preclinical targets |
Ventilation | Chest compressions alone are unlikely to generate significant ventilation in humans once anatomic dead space is accounted for | Moderate |
Physiology — why every pause costs more than the seconds it takes
Coronary perfusion pressure does not exist at the start of a compression sequence. It is built up over the first three or four compressions as an arteriovenous pressure gradient accumulates, and it dissipates within a few seconds of stopping.
The consequence is that a pause does not cost the duration of the pause. It costs the duration of the pause plus the compressions needed to rebuild the gradient — and during that rebuilding period, myocardial and cerebral perfusion are below what they were before the pause.
This is the physiological reason that Chapter 21 insists cannulation must be designed around uninterrupted compressions rather than accommodated to them. It is also why mechanical compression devices, which showed no clear benefit in unselected arrest populations, are near-universal in ECPR protocols: they were mandatory in ARREST and used in 85% and 90% of patients in the other two randomised trials. In ECPR their value is not that they compress better, but that they compress while somebody is putting a wire in a femoral artery.
End-tidal carbon dioxide during CPR — a real signal with a real confounder
During cardiac arrest, end-tidal CO₂ is proportional to cardiac output and pulmonary blood flow. Values above 20 mmHg are associated with successful resuscitation; values persistently below 10 mmHg after at least 20 minutes of resuscitation are associated with failure. An experimental ECPR model found that end-tidal CO₂ at the end of mechanical CPR was inversely associated with the extent of brainstem and renal injury after extracorporeal reperfusion, with a substantial relationship to carotid blood flow.
This is why an end-tidal CO₂ threshold — usually above 10 mmHg — appears in almost every published ECPR inclusion set, including the Red Book's Table 32-1.
Pitfall — the gate value is contaminated by the resuscitation that produced it
Epinephrine reduces end-tidal CO₂. Sodium bicarbonate transiently raises it.
An end-tidal CO₂ of 9 mmHg measured two minutes after a dose of adrenaline is not the same measurement as an end-tidal CO₂ of 9 mmHg measured in a patient who has had none — but the protocol treats them identically, and the patient is excluded.
This book's reading: where end-tidal CO₂ is used as an inclusion gate, record when the last dose of adrenaline and any bicarbonate were given alongside the value, and treat a single borderline reading immediately after a drug as uninterpretable rather than as a refusal. This is reasoning from the pharmacology, not a validated rule. Certainty: very low.
19.3 Time is not a criterion in ECPR — it is the indication
Every other chapter in this book selects patients by physiology. Part IV selects by the clock. The evidence for this is the most consistent finding in the entire ECPR literature, and it is worth seeing the numbers together.
Study and population | Finding | Certainty |
133 ECPR-treated in- and out-of-hospital refractory arrests (Critical Care 2017) | Survival 67%, 29%, 10% and 6% for low-flow durations of under 20, 20–45, 45–60 and 60–135 minutes | Low — single centre, retrospective; the most-reproduced figure in ECPR |
79 ECPR-treated out-of-hospital arrests, good neurological outcome in 14% | 40 minutes was the optimal cut point; at that point the likelihood of survival fell from approximately 30% to 15%. The longest low-flow interval in a case with a good neurological outcome was about 70 minutes | Low — single centre, retrospective |
190 in-hospital arrests, median low-flow 30 minutes | Favourable neurological outcome by initial rhythm: shockable 49%, PEA 34%, asystole 12%. The best low-flow cut-points differed by rhythm — 22 minutes for PEA, 46 minutes for shockable — and increasing no-flow duration harmed PEA patients more than shockable ones | Low — single centre; conceptually the most important of the group |
52 ECPR-treated out-of-hospital arrests | No survivor had a low-flow duration of 66 minutes or more | Very low — small single series |
Red Book Chapter 32 | "After 30 minutes of CPR, every additional minute without ECPR initiation adds ~2-2.5% mortality rate." The stated goal is ECPR on average within 60 minutes and ideally within 45 minutes of the arrest | Low — a synthesis statement, primary derivation not shown |
Red Book Chapter 32, on when to start considering ECPR | Survival declines after 10 minutes of CPR without ROSC; 10–20 minutes without ROSC is "often quoted as the point where ECPR should be considered" | Low — convention, not a derived threshold |
Population data on conventional resuscitation (Circulation 2016), reproduced in the ECPR textbook | The probability of survival with acceptable functional recovery declines steadily with CPR duration and asymptotically plateaus between 10 and 30 minutes, with the inflection point depending on shockable rhythm, witnessed collapse, bystander CPR and no-flow interval | Moderate — large multi-cohort analysis, read second-hand |
Clinical pearl — why the "right" time to activate is a subtraction, not a threshold
The textbook's argument is a subtraction and it is the cleanest piece of reasoning in the ECPR literature. The features that make a patient an attractive ECPR candidate — younger age, few comorbidities, shockable rhythm, witnessed collapse, bystander CPR — are the same features that make conventional CPR likely to work. Activating ECPR too early wastes the intervention on patients who were about to achieve ROSC and exposes them to its complications. Activating too late wastes it on patients whose brains have already gone.
The window therefore opens where the conventional-CPR survival curve flattens and closes where the low-flow curve reaches the floor. For ECPR candidates that opening is estimated at 10–20 minutes of professional resuscitation — nearer 20 minutes for shockable rhythms, nearer 10 for non-shockable.
Two arrests with the same elapsed time are not at the same point in this calculation if their rhythms differ.
Danger — a published cut-point is a property of the cohort, not of the brain
The numbers above give optimal low-flow cut-points of 22, 40, 46, 60 and 66 minutes, and they are all correctly derived.
They disagree because a cut-point extracted from a cohort describes where that cohort's own low-flow distribution happened to separate survivors from non-survivors. A centre whose median arrest-to-flow is 35 minutes will find its discriminating threshold at a different place from a centre whose median is 65 minutes, even if the underlying biology is identical — and both will publish their number as though it were biology.
Do not import another hospital's cut-point as a rule for your own. Audit your own arrest-to-flow distribution and find out where you actually sit on the curve.
One published figure in this literature runs backwards
The ECPR textbook, summarising the low-flow studies, reports that one study found 60 minutes to be the best cut-off "with survival 48% above and 12% below this interval." Every other finding in this section, and the study's own framing, requires survival to be higher below a low-flow cut-off and lower above it.
In accordance with this book's standing rule, the discrepancy is recorded and the number is not used. The conclusion the source draws — that 60 minutes discriminated well in that cohort — is retained; the two percentages are not. This is the fourth published numerical inconsistency flagged in this book and the first in Part IV.
The point of the diagram is the ownership column. No-flow belongs to the public and to the emergency medical dispatch system. Low-flow belongs partly to the ambulance service and partly to the hospital. The only interval a hospital can shorten by its own effort is the one from activation to flow — and §19.7 shows that this is precisely the interval where the largest randomised trial lost its advantage.
19.4 Controversy 1 — is sixty minutes a deadline?
Controversy 1 — Should a patient be refused ECPR because arrest-to-flow will exceed 60 minutes?
The question. Sixty minutes appears in the Red Book's example inclusion criteria, in the ELSO interim guidance as reported by the Red Book, in the ISCCM Manual ("good functional recovery is more likely if the time from collapse to start of extracorporeal blood flow is less than 60 minutes"), and in the inclusion criteria of the trials. Is it a deadline after which cannulation is futile, or a design target for a service?
The case that it is a deadline. The low-flow gradient in §19.3 is steep and consistent. Survival in the 133-patient series fell to 6% beyond 60 minutes. One 52-patient series had no survivor at or beyond 66 minutes. The Red Book's estimate of 2–2.5% additional mortality per minute beyond 30 minutes of CPR implies that a thirty-minute delay carries a 60–75 percentage-point penalty. Positive hospital-discharge outcomes are described as "rare when it exceeds 65-70 minutes."
The case that it is not. A prospective Stockholm programme applying a predefined 60-minute threshold across 95 patients met that threshold in 9% of cases. Survival to discharge was 25% overall and 39% among those reaching the intensive care unit on ECMO, with 78% of survivors at Cerebral Performance Category 1–2 within 12 months — and 8 of the 9 survivors had an arrest-to-ECMO time longer than 60 minutes. In the 79-patient cohort, the longest low-flow interval in a patient with a good neurological outcome was about 70 minutes. In PRAGUE-OHCA, among patients with neurologically intact survival after more than 45 minutes of CPR, 24 of 26 had received ECPR.
And the trials themselves did not meet it. Mean low-flow times were 59 minutes in ARREST, 62 minutes in PRAGUE-OHCA and 74.7 minutes in INCEPTION — all at or beyond the ceiling that the same protocols specified.
Resolution — the two uses of the number are different, and only one is supported.
— Used prospectively, to design a service, sixty minutes is legitimate and useful. It sets the standard that dispatch, transport, activation and cannulation must collectively meet, and it is the number a programme should audit itself against.
— Used retrospectively, at the bedside, to refuse a patient at minute 61, it is not supported. The relationship between low-flow time and outcome is a continuous decline, not a cliff, and every large series contains survivors beyond the threshold. What changes past 60 minutes is the prior probability of benefit, not its possibility.
The practical consequence: a programme that finds itself routinely exceeding 60 minutes has a system problem to fix, not a patient population to exclude. Stockholm's honest report of 9% adherence, published alongside acceptable outcomes, is the most useful paper in this section precisely because it declines to resolve the tension.
Certainty: low. No study has randomised patients by low-flow duration, and none will.
Clinical pearl — the modifiers that legitimately buy time
The Red Book's inclusion table carries a footnote that is easy to miss and does real work: the 60-minute low-flow limit applies "unless other favorable prognostic features are present: eg, periods of intermittent ROSC/hypothermia prearrest/young age/signs of life during CPR."
Each of these alters the ischaemic arithmetic rather than the clock. Intermittent ROSC means the brain was perfused during part of the interval the stopwatch counted as low-flow. Pre-arrest hypothermia reduces the metabolic cost of every minute of it, and accidental hypothermic arrest is the one setting where extraordinarily long intervals are compatible with intact recovery. Signs of life during CPR are direct evidence that cerebral perfusion has been adequate.
When you record a low-flow time for an ECPR decision, record these alongside it. The number without them is not the information.
19.5 Reperfusion is an intervention, not an event
Almost everything written about ECPR concerns how fast circulation is restored. Far less concerns how it is restored — and the moment of restoration is the most abrupt reperfusion in clinical medicine.
Physiology — what happens in the first sixty seconds of extracorporeal flow
The patient has spent between twenty minutes and over an hour at a fraction of normal cardiac output, with progressive lactic acidosis, hypercarbia, depleted adenosine triphosphate and an accumulating oxygen debt. Then, over a few seconds:
Flow goes from a small fraction of normal to full cardiac output. Not gradually — as fast as the pump can be turned up.
Oxygen tension goes from severe systemic hypoxaemia to whatever the sweep gas is delivering, which at the moment of cannulation is almost always an oxygen fraction of 1.0 through a fresh membrane. Post-membrane oxygen tensions in the hundreds of millimetres of mercury are normal, and they arrive at an ischaemic, acidotic brain.
Carbon dioxide is cleared at whatever rate the sweep gas dictates, which in a fresh circuit with a large membrane can be very fast, in a patient whose arterial carbon dioxide may be profoundly elevated.
Temperature falls, because the circuit is at ambient temperature until the heat exchanger is set.
Every one of these is a therapeutic decision being taken by default, in the middle of a resuscitation, by nobody in particular.
The oxygen problem has the largest dataset in this section
An analysis of 3,125 adult ECPR patients in the ELSO Registry (2009–2020) examined arterial gas values in the six hours before and the 24 hours after cannulation. Composite acute brain injury — ischaemic stroke, intracranial haemorrhage, seizures or brain death — occurred in 488 patients (16%), comprising 7% ischaemic stroke and 3% intracranial haemorrhage.
Exposure | Outcome | Adjusted odds ratio (95% CI) |
Moderate hyperoxia on ECMO (PaO₂ 200–299 mmHg) | Composite acute brain injury | 1.42 (1.02–1.97) |
Severe hyperoxia on ECMO (PaO₂ at or above 300 mmHg) | Composite acute brain injury | 1.59 (1.20–2.10) |
Severe hyperoxia on ECMO | Ischaemic stroke | 1.63 (1.11–2.40) |
Severe hyperoxia on ECMO | Intracranial haemorrhage | 1.92 (1.08–3.40) |
Severe hyperoxia on ECMO | In-hospital mortality | 1.58 (1.21–2.06) |
Mild hypercarbia before ECMO (PaCO₂ 45–54 mmHg) | Composite acute brain injury — protective | 0.61 (0.44–0.84) |
Mild hypercarbia before ECMO | Ischaemic stroke — protective | 0.56 (0.35–0.89) |
How much weight this carries. It is a registry analysis with two gas values per patient, multiple imputation for missing data, and an unavoidable confounding problem: a patient who is hyperoxic on ECMO may be hyperoxic because nobody has turned the sweep down yet, which is itself a marker of a chaotic resuscitation. The direction is biologically coherent and the effect sizes are modest and consistent across four separate outcomes, which is reassuring, but this is association. Certainty: low to moderate.
Set against it, the only randomised oxygen evidence on VA ECMO — examined in Chapter 14 — was null. That trial did not, however, study the first hours after extracorporeal reperfusion of a whole-body ischaemic insult, which is exactly the window this registry analysis implicates. The two findings are not in conflict; they concern different moments.
The Red Book's own immediate post-arrest table already carries the operational instruction: measure arterial oxygen tension from the right-sided arterial line, avoid hyperoxia if possible, and aim to avoid hypocarbia while titrating sweep gas.
Clinical pearl — give the sweep gas an owner and a clock
At the moment of cannulation the oxygen fraction is 1.0 and the sweep is high, and both are correct. Ten minutes later both are probably wrong, and nobody in the room has been given the job of changing them.
Name the person responsible for the first arterial blood gas from the right radial line, and name the time it will be taken. In practice this is the earliest, cheapest and most neglected intervention in ECPR, and it is the one the registry data point at.
Two cautions carried forward: the right-sided sampling site matters (Chapter 16), and the sweep gas is never simply switched off on VA ECMO (Chapters 10, 13 and 18).
The carbon dioxide problem is plausible and unproven
The mechanism is straightforward: a fresh, large-surface membrane at a high sweep gas rate can clear carbon dioxide from a severely hypercarbic patient within minutes, and an abrupt fall in arterial carbon dioxide causes cerebral vasoconstriction in a brain that has just been reperfused.
The one human study addressing it directly examined 84 ECPR patients and found that neither the initial arterial carbon dioxide tension (odds ratio 0.93, 95% CI 0.78–1.08) nor the maximum decrease in carbon dioxide over the first six hours (odds ratio 1.03, 95% CI 0.95–1.13) was associated with recovery of consciousness.
That is a null result and it is reported as one. The mechanism remains physiologically sound, the registry data above suggest that pre-cannulation carbon dioxide is not neutral, and the Red Book advises avoiding hypocarbia — but no study has shown that controlling the rate of carbon dioxide fall changes outcome. Certainty: very low. The reasonable practice is to start the sweep gas low and titrate up rather than the reverse, which costs nothing.
Pitfall — a second published figure in this section runs backwards
A three-centre prospective study of cerebral regional oxygen saturation in 26 ECPR patients concluded that higher mean cerebral saturation in the first 30 minutes after initiation was found in patients who regained consciousness (odds ratio 1.23, 95% CI 1.01–1.50), with no difference at baseline.
The abstract's accompanying percentages — quoted as 38% for those who regained consciousness versus 62% for those who did not — cannot both be as labelled and support the stated direction.
The conclusion and the odds ratio are used here; the two percentages are not. This is the fifth published numerical inconsistency flagged in this book.
Controlled reperfusion — the frontier, labelled as one
A line of work argues that the problem is not only when reperfusion happens but that conventional extracorporeal reperfusion is uncontrolled: full flow, high oxygen, non-pulsatile, at whatever temperature and composition the circuit happens to deliver.
The controlled automated reperfusion of the whole body approach modifies all of these deliberately — pulsatile high-pressure, high-flow perfusion, rapid hypothermia, a cytoprotective priming regimen, and continuous arterial blood-gas measurement to steer composition during the first reperfusion minutes. In porcine models it is reported to achieve up to 90% survival with intact neurological function after 20 minutes of arrest, and a multicentre European series of 69 patients reported 42% overall survival, with 79% of survivors at Cerebral Performance Category 1–2 at 90 days.
Danger — read the frontier correctly
These are single-group, uncontrolled clinical series reported by the developers of the device, in selected patients, with no randomised comparison against conventional ECPR. The survival figures cannot be compared with the trial figures in §19.6, which come from randomised populations with different selection.
Do not quote 42% as an ECPR survival rate. Certainty for any clinical effect: very low.
What to take from this work is the principle, not the number: experimental evidence indicates that tissue viability after prolonged ischaemia depends not only on how long the ischaemia lasted but on how the reperfusion is conducted — and conventional ECPR conducts it by default. That principle is actionable today, at zero cost, in the sweep gas, the oxygen fraction and the rate at which flow is increased.
19.6 The three randomised trials — what each one actually tested
Chapter 11 established how this book states VA ECMO evidence: read what the trial randomised, not what the intervention is called. Part IV needs the same discipline and its mirror image — three trials with very different systems must not be read as one answer.
Feature | ARREST | PRAGUE-OHCA | INCEPTION |
Setting | Single centre, Minneapolis | Single centre, Prague | 10 centres, Netherlands |
Design character | Predominantly explanatory | Hybrid — pragmatic eligibility, otherwise explanatory | Predominantly pragmatic |
Enrolment period | Aug 2019 – Jun 2020 | Mar 2013 – Oct 2020 | May 2017 – Feb 2021 |
Randomised | 30 | 256 | 134 analysed, of 160 randomised — 26 excluded for not meeting criteria at hospital admission |
Age | 18–75 | 18–65 | 18–70 |
Initial rhythm | VF/pulseless VT only | All rhythms — 61% VF/VT, 22% asystole, 17% PEA | Ventricular arrhythmia — 99% VF/VT |
Randomisation point | In hospital, on emergency department arrival | Pre-hospital, after 5 minutes of unsuccessful ALS | Pre-hospital, after 15 minutes of unsuccessful ALS |
Mechanical CPR | Mandatory | 85% of patients | 90% of patients |
Time ceiling in protocol | Estimated transfer under 30 min | Collapse to catheter laboratory under 60 min | Estimated cannulation within 60 min of arrest |
Mean on-scene time | 23 min | 13 min | 13 min |
Mean low-flow (arrest to ECMO) | 59 min | 62 min | 74.7 min |
Received ECPR in the ECPR arm | 80% (12/15) | 66% (82/124) | 74% (52/70) per the trialists' comparison; 66% (46/70) per the INCEPTION group's own secondary analysis — see the flag below |
Crossover in the control arm | 0 | 8% (10/132) | 4–5% (3/64 or 3/74, depending on the denominator reported) |
Primary outcome | Survival to hospital discharge | 180-day survival, CPC 1–2 | 30-day survival, CPC 1–2 |
Primary result | 6/14 (43%) vs 1/15 (7%); risk difference 36.2% (95% credible interval 3.7–59.2); posterior probability of ECMO superiority 0.9861 | 39 (31.5%) vs 29 (22.0%); odds ratio 1.63 (0.93–2.85); difference 9.5% (−1.3 to 20.1); P = 0.09 | 14/70 (20%) vs 10/64 (16%); odds ratio 1.4 (0.5–3.5); P = 0.52 |
30-day neurological recovery | Not the primary endpoint | 38 (30.6%) vs 24 (18.2%); odds ratio 1.99 (1.11–3.57); P = 0.02 | The primary endpoint above |
Why it stopped | Early for superiority at the first preplanned interim analysis, on the recommendation of the data safety monitoring board | Early for futility — 256 of a planned 285 | Completed enrolment |
ECPR caseload during the trial | 18 per year | 12 per year in the ECPR arm across 92 months | 0.29 attempted cannulations per centre per month |
Catchment population | 3.5 million | 1.25 million | About 800,000 per centre, with overlap across 10 sites |
Pitfall — the same trial group reports two different delivery figures
The trialists' comparative paper states that 52 of 70 patients randomised to ECPR in INCEPTION (74%) actually received it. A secondary analysis of the same trial, by overlapping authors, states that ECPR treatment was started in 46 (66%) of 70 patients in the ECPR arm.
Both figures are published; they are not reconciled in either source, and they may reflect different definitions — cannulation attempted versus extracorporeal flow achieved. Both are recorded here and neither is asserted. The point that matters is unaffected: in the pragmatic multicentre trial, a quarter to a third of patients assigned to ECPR never received it.
A second, smaller discrepancy: the same comparative paper gives PRAGUE-OHCA's mean low-flow time as 62 minutes in its table and 61 minutes in its text. The difference is immaterial and is noted only so that a reader comparing sources is not misled.
A third: the Red Book's summary table describes INCEPTION's resuscitation trigger as "No ROSC after 3 rounds of ALS CPR", while the trialists describe it as 15 minutes of unsuccessful advanced life support. These are not the same criterion. The trialists' description is used here, as they are the trial's own authors.
What each trial answered
Evidence — three different questions
ARREST asked: does ECPR work under ideal conditions? One centre, one high-volume team, one rhythm, mandatory mechanical compressions, a transfer time ceiling of 30 minutes, and in-hospital randomisation so that the comparison is purely the addition of ECPR to a patient already at the hospital door. Answer: yes, decisively — survival to discharge 43% versus 7%, stopped early at 30 patients because the prespecified monitoring boundary was crossed. This is an efficacy result, and the smallest trial in the group.
PRAGUE-OHCA asked: does an invasive bundle — mechanical compressions, intra-arrest transport, ECPR and immediate invasive assessment — beat continued on-scene advanced life support? Note that the randomised unit was the whole strategy, not the circuit. Answer: the primary 180-day endpoint missed significance (31.5% versus 22.0%, P = 0.09) and the trial stopped for futility, while 30-day neurological recovery was significantly better (30.6% versus 18.2%, P = 0.02). Two further findings matter more than the headline. First, the control arm did far better than designed — 22% against a presumed 10% — partly because building the invasive pathway improved the logistics of on-scene care as well. Second, widening entry to non-shockable rhythms diluted the effect: survival in that subgroup was about 5%, while a post-hoc analysis restricted to shockable rhythms showed about 40% CPC 1–2 survival.
INCEPTION asked: what happens when ECPR is implemented across an ordinary national health system? Ten centres, no imposed ECPR protocol, local practice variation permitted. Answer: no detectable benefit — 20% versus 16%, odds ratio 1.4 (0.5–3.5). This is an effectiveness result, and it is the one most hospitals should read first.
Two further INCEPTION analyses refine rather than reverse this. The prespecified per-protocol analysis (81 patients, excluding protocol violations, cannulation beyond 60 minutes, pre-hospital ROSC and crossovers) found 15% versus 9%, adjusted odds ratio 1.9 (0.4–9.3), P = 0.393, with a Bayesian posterior probability of 84% for any benefit and 61% for an absolute risk reduction of at least 5%. A separate Bayesian re-analysis of the intention-to-treat population estimated an absolute risk difference of 3.6% (95% credible interval −9.5 to 16.7) favouring ECPR, median relative risk 1.22 (0.59–2.51), with a 42% probability of a minimal clinically important difference at both 30 days and 6 months.
The pooled individual-patient analysis of the two invasive-strategy trials
ARREST and PRAGUE-OHCA shared a design feature INCEPTION did not: both compared intra-arrest transport with in-hospital ECPR initiation against continued standard advanced life support. Their individual patient data were pooled — 286 patients, 147 invasive and 139 standard.
Outcome | Invasive | Standard | Effect |
180-day survival with favourable neurological outcome | 45 (32.4%) | 29 (19.7%) | Absolute difference 12.7% (2.6–22.7), P = 0.015 |
180-day survival | 47 (33.8%) | 33 (22.4%) | Hazard ratio 0.59 (0.43–0.81), P = 0.0009 |
30-day favourable neurological outcome | 44 (31.7%) | 24 (16.3%) | Absolute difference 15.4% (5.6–25.1), P = 0.003 |
30-day cardiac recovery | 60 (43.2%) | 46 (31.3%) | Absolute difference 11.9% (0.7–23), P = 0.05 |
Subgroup — shockable rhythms | Larger effect | — | Absolute difference 18.8% (7.6–29.4), P = 0.01; hazard ratio 2.26 (1.23–4.15), P = 0.009 |
Subgroup — CPR longer than 45 minutes | Larger effect | — | Hazard ratio 3.99 (1.54–10.35), P = 0.005 |
Read this carefully. It pools the two trials run by dedicated single centres and excludes the multicentre pragmatic one, so it answers ARREST and PRAGUE-OHCA's question — is the invasive strategy better in centres that can deliver it — and not INCEPTION's. Median resuscitation duration was 58 minutes in the invasive arm and 49 in the standard arm, which makes the result more, not less, striking. The finding that the benefit was larger in patients with more than 45 minutes of CPR is the strongest published argument against a hard 60-minute cut-off. Certainty: moderate for the invasive strategy in comparable systems; it does not generalise to systems unlike these two.
19.7 Why the three trials are complementary rather than contradictory
The trialists — a group including authors from all three trials — published their own comparison and reached a conclusion this book adopts: "Three well-designed and well-performed randomized trials, addressing the same question, yielded apparently diverging conclusions... their results should be weighed separately, as they each highlight different important aspects of ECPR in refractory OHCA."
Their reasoning is that ARREST and PRAGUE-OHCA tested efficacy under near-ideal conditions, while INCEPTION tested effectiveness under ordinary ones, and that the pragmatic-versus-explanatory distinction explains the divergence without requiring any of them to be wrong.
Evidence — this book's reading, stated as reasoning rather than as a finding
Line the three trials up by how far each ECPR system was from ideal, and the effect size falls monotonically along that ordering.
ARREST — 18 ECPR cases per year, one centre, one team, one rhythm, mandatory mechanical compressions, a 30-minute transfer ceiling, randomisation at the hospital door. Largest effect.
PRAGUE-OHCA — 12 ECPR cases per year in the trial arm, one centre, one team, but entry widened to all rhythms and randomisation moved to the roadside. Intermediate effect, with the widened rhythm criterion visibly costing it: about 5% survival in non-shockable patients against about 40% in shockable ones on post-hoc analysis.
INCEPTION — 0.29 attempted cannulations per centre per month, ten centres, no imposed ECPR protocol, deliberate practice variation. Smallest effect.
The ordering is not by patient selection alone, and it is not by the technology, which was the same everywhere. It is by how much ECPR each team did and how tightly the pathway was specified.
The single most instructive number in Part IV
INCEPTION had a mean on-scene time of 13 minutes — identical to PRAGUE-OHCA's and ten minutes shorter than ARREST's. Its ambulance services were fast.
And it had the longest mean low-flow time of the three: 74.7 minutes, against 59 and 62.
Danger — the delay was inside the hospital
The pre-hospital phase was not INCEPTION's problem. The trialists' own figures show door-to-ECMO times of 12 minutes in ARREST and 12 minutes in PRAGUE-OHCA, against 16 minutes from hospital arrival to the start of cannulation plus a further 20 minutes of cannulation in INCEPTION.
Roughly a quarter of an hour of the excess low-flow time in the trial that found no benefit was generated by the hospitals that had agreed to deliver the intervention.
This is the finding that should change behaviour. A hospital considering an ECPR programme will naturally focus on cannulation skill and equipment. The trial evidence says the thing most likely to make the programme ineffective is the twenty minutes between the ambulance arriving and the wire going in.
The volume signal
The trialists cite an ELSO Registry analysis reporting that the adjusted odds ratio for hospital survival after ECPR was 0.36 in hospitals performing fewer than 10 ECPR procedures per year. A separate meta-regression across ECPR studies found that centre volume was associated with reduced odds of mortality, with a regression coefficient of −0.17 (95% CI −0.32 to −0.017, P = 0.030) per doubling of annual centre volume.
Neither is randomised, both are vulnerable to confounding by everything that correlates with volume, and the ELSO registry figure is used here second-hand — the underlying analysis was not retrieved. Certainty: low. But two independent observational approaches point the same way, and they agree with the ordering of the three trials.
Clinical pearl — the sentence to take from Part IV's evidence base
The trialists' own summary is the most useful thing written about ECPR:
"Excellent results can be achieved with a very high level of dedication, provided that strict selection criteria are applied. However, pragmatic implementation of extracorporeal cardiopulmonary resuscitation does not necessarily lead to improved outcome of refractory out-of-hospital cardiac arrest."
And their recommendation: "Centres that perform ECPR or aspire to do so should critically assess their setting, logistics, performance, and outcomes to assure that their resources are well spent."
The question "does ECPR work?" is not answerable in general. The answerable question is "does it work here?" — and it is answerable only by a unit that measures itself. §19.9 says what to measure.
19.8 Controversy 2 — four meta-analyses of three trials, four answers
Controversy 2 — Does the randomised evidence show that ECPR works?
The question. Four syntheses of essentially the same randomised evidence reached different verdicts within about a year of each other. A reader who finds one of them has the whole answer; a reader who finds all four has a problem.
What they found.
— A 2023 meta-analysis in Artificial Organs (four randomised trials, 433 patients): survival with favourable neurological outcome at longest follow-up 59/220 (27%) versus 39/213 (18%), odds ratio 1.72 (1.09–2.70), P = 0.02, I² = 26%, number needed to treat 9. Shockable rhythms only: odds ratio 1.90 (1.16–3.13), P = 0.01, number needed to treat 7. Positive.
— A 2023 European Heart Journal meta-analysis (three trials, n = 428, fixed effects): odds ratio 1.71 (1.08–2.71), P = 0.02, I² = 28. Positive.
— A 2023 meta-analysis in Internal and Emergency Medicine (three trials, 418 patients, random effects): favourable neurological survival at shortest follow-up 26.4% versus 17.2%, relative risk 1.47 (0.91–2.40), P = 0.12; at 6 months 28.3% versus 18.6%, relative risk 1.48 (0.88–2.49), P = 0.14. Negative.
— A 2024 Bayesian hierarchical meta-analysis in Critical Care (three trials, 209 versus 211 patients), written by authors of all three trials: see the table below. Neither positive nor negative — a probability distribution.
Why they disagree, in four parts.
1. The denominators are not the same. The totals are 418, 420, 428 and 433 for what is nominally the same three trials. The differences come from intention-to-treat versus modified intention-to-treat populations, from whether the ARREST patient who withdrew is counted, and — in the Artificial Organs analysis — from the inclusion of a fourth randomised trial that the other three syntheses did not include. That fourth trial is not identified in the material available to this project and is deliberately not named or characterised here.
2. Fixed versus random effects. A fixed-effects model on trials this heterogeneous narrows the confidence interval and will cross into significance where a random-effects model does not.
3. Odds ratios versus relative risks. With event rates near 20%, an odds ratio systematically exaggerates a relative risk. An odds ratio of 1.72 and a relative risk of 1.47 are not two different findings.
4. The dichotomy itself. Three of the four report a P value against 0.05 and convert it into a verdict, on a body of evidence comprising roughly 420 patients across trials of 30, 134 and 256.
The resolution, and it is a striking one: the point estimate barely moves. Across all four syntheses the central estimate of the effect sits between about 1.47 and 1.90 in favour of ECPR. What changes between papers is not the estimate of benefit but the verdict placed on it.
Certainty: moderate that the effect is positive; low for its magnitude.
What the Bayesian synthesis actually says
Quantity — 6-month neurologically favourable survival | All rhythms | Shockable rhythms |
Median relative risk | 1.47 (95% CrI 0.73–3.32) | 1.54 (95% CrI 0.79–3.71) |
Mean absolute risk difference | 8.7% (−5.0 to 42.7) | 10.8% (−4.2 to 73.9) |
Number needed to treat | 11 | 9 |
Assumed control-arm survival | 18.4% (11.8–27.7), I² = 33%, τ² = 0.077 | 19.9% (8.3–40.7), I² = 76%, τ² = 0.513 |
Probability the benefit exceeds 0% | 91.0% | 92.4% |
Probability it exceeds 5% absolute | 71.1% | 75.8% |
Probability it exceeds 10% absolute | 43.7% | 50.4% |
Probability it exceeds 15% absolute | 23.5% | 29.8% |
Probability it exceeds 20% absolute | 13.4% | 17.6% |
This is the most honest statement of the randomised ECPR evidence available, and it is a statement a clinician can actually use. It is about nine to one that ECPR helps. It is about seven to three that it helps by at least five absolute percentage points — the threshold the trialists themselves nominated as minimally clinically important. It is a coin-toss whether it helps by ten. And the between-trial heterogeneity in the shockable-rhythm analysis (τ² = 0.513, I² = 76%) is itself a finding: the trials disagree most about the population everyone agrees is the best candidate, which is what §19.7 predicts if the differences are systemic rather than biological.
Clinical pearl — this book has been here before
Chapter 4 tells the same story about EOLIA: a trial reported as negative on a frequentist primary endpoint, re-analysed under a range of priors, and found to carry a high posterior probability of a benefit whose magnitude remained uncertain.
The pattern is not a coincidence. Both trials asked whether an expensive, logistically demanding extracorporeal intervention helps a population with high mortality, and both were sized for an effect larger than the one they found. In both cases the dichotomous verdict destroyed information that the probability statement preserves.
When a reader tells you that ECPR "was negative in the randomised trials", the accurate reply is that the trials were individually underpowered and that the pooled probability of benefit is around 91%, with the magnitude genuinely unresolved.
And the observational literature says something different again
An updated meta-analysis combining 3 randomised trials and 10 propensity-score-matched studies — 14 pairwise comparisons, 6,336 ECPR and 7,712 conventional CPR patients — found:
- In-hospital mortality odds ratio 0.63 (0.50–0.79), graded high certainty, with the trial sequential analysis confirming the required information size was met.
- Out-of-hospital arrest subgroup odds ratio 0.62 (0.45–0.84) — newly significant with the addition of recent studies, where the same group's earlier analysis had found none.
- Short-term favourable neurological outcome odds ratio 1.57 (1.14–2.15); 30-day survival odds ratio 1.70 (1.29–2.26).
Pitfall — a GRADE rating of "high certainty" on a body dominated by matched observational studies
Ten of the thirteen included studies were propensity-score-matched cohorts. The randomised trials in this field do not individually reach significance; the matched observational studies do. That is the classic signature of residual confounding by indication — patients selected for ECPR are, by the act of being selected, different from those who were not, in ways propensity matching cannot fully capture because the selecting clinician saw things the dataset does not contain.
The authors themselves list residual confounding first among their limitations. This book grades the observational body moderate at best, and treats the concordance of its direction with the randomised body as the reassurance, not the certainty rating.
19.9 ECPR is a system, and its output is time
If §19.3 is right that time is the indication, and §19.7 is right that the trials are ordered by system maturity, then the unit of intervention in ECPR is not the circuit. It is the programme. Three reports make the case better than any argument.
Programme | What happened | What it demonstrates |
A UK tertiary centre that had performed ECPR for over a decade with poor results | 22 cases over three preceding years, 2 survivors (9.1%). A structured bundle was then introduced: systematic screening, strict exclusion criteria, assessment of resuscitation adequacy through physiological stop criteria, a drilled and standardised cannulation process, defined post-resuscitation care and neuroprognostication. In the following year, 13 patients, 9 (69.2%) surviving with good neurological outcome at 6 months; hazard ratio 4.56 (2.1–10.2), P less than 0.05 | Same hospital, same technology, different system. The intervention that changed was organisational |
A prospective Stockholm out-of-hospital programme | 95 patients met criteria; 23% achieved ROSC before initiation, 41% were excluded, 36% had ECMO started. The 60-minute target was met in 9%. Survival to discharge 25% overall and 39% (9/23) of those reaching intensive care on ECMO, with 78% of survivors at CPC 1–2 within 12 months | A programme can be safe, feasible and produce reasonable outcomes while missing its own principal target — and should say so |
A regional emergency medical services routing programme | 233 patients routed to ECPR-capable centres; 58 (27%) were cannulated; median arrest-to-cannulation 66 minutes. Survival to discharge 27% and CPC 1–2 21%, both for all routed patients and for those cannulated. On inverse-probability-weighted regression, ECPR was not associated with improved outcomes | Good absolute numbers and no measurable effect are compatible. Selection and system performance, not the intervention, are doing the work |
How much weight these carry. All three are before-after or observational. The UK series is small, and its improvement reflects changed selection as well as changed process — tighter exclusion criteria raise survival among those treated by definition. Certainty: low for any individual figure. What survives the caveats is the direction, which matches the volume–outcome data in §19.7 and the ordering of the trials.
What a programme is required to have
The Red Book's ECPR programme table, drawn from the ELSO interim guidance, sets out the requirements. Reproduced here because it is the closest thing to a standard that exists.
Domain | Requirement |
Programme development | Should be part of an established ECMO programme; multidisciplinary and multiprofessional engagement is mandatory; eligibility criteria must be clear and reproducible; in-house ECPR protocols must be robust and agreed through hospital governance; out-of-hospital ECPR requires integrated prehospital protocols covering patient identification and timely transport with high-quality CPR, embedded in ongoing training |
Training and education | Regular individual, system and team-based simulation is recommended |
Quality improvement | Data monitoring, audit and quality improvement embedded in the programme; quality-of-care and outcome metrics measured, reported and reviewed regularly; case review for every case |
Governance | A robust governance process; ideally every patient should be part of an ongoing clinical trial or registry to allow benchmarking |
Clinical pearl — the seven numbers a programme must be able to state about itself
The trialists' recommendation is to "critically assess their setting, logistics, performance, and outcomes." In practice that reduces to seven numbers, and a unit that cannot produce them is not in a position to say whether its ECPR programme helps anyone.
- Median arrest-to-flow time, and the proportion under 60 minutes. This is the programme's core performance measure, because it is the only interval the hospital controls.
- Median door-to-flow time. The trials' benchmark is 12 minutes; INCEPTION's excess sat here.
- Activations per cannulation. The Stockholm and regional programmes cannulated 36% and 27% of those screened. A very high ratio means the criteria are too loose; a very low one means they are too tight or the pathway is too slow.
- Cannulations per year. The observational signal turns at about 10 per year, and the trials that worked did 12 and 18.
- Cannulation success and vascular complication rate.
- Survival with CPC 1–2 at 6 months — not survival, and not survival to decannulation. Chapter 18's rule applies with more force here than anywhere: weaning is not surviving.
- The proportion of cases formally reviewed. The guidance says every one.
None of these requires research infrastructure. All of them require someone to own the numbers.
Addendum — 11 September 2026, added when Chapter 20 was written
The statement above that "the observational signal turns at about 10 per year" needs qualifying, and Chapter 20 §20.9 sets out why.
A secondary analysis of the SAVE-J II registry covering 1,740 patients across 36 centres grouped those centres into tertiles by annual ECPR volume — high (21 or more per year), medium (11–20), low (fewer than 11) — and reported survival at discharge of 33.4%, 24.1% and 26.8% respectively. The separation is at roughly 21 cases per year, not 10, and the relationship is not monotonic: the middle tertile did worse than the lowest. Two numerical problems in that analysis are flagged in Chapter 20 §20.9 and neither is corrected here.
The corrected form of the claim: more ECPR volume is associated with better outcomes, the direction is supported independently by the meta-regression in §19.7 and by the ordering of the three randomised trials, and the best-supported threshold lies somewhere between 10 and 21 cases per year. The figure of 10 in item 4 of the pearl above comes from the ELSO-registry odds ratio quoted in §19.7 and should be read as a floor rather than as a turning point.
19.10 What Part III hands to ECPR — and what it cannot hand over
Part IV is not a new ECMO. Most of the physiology, the circuit and the complications are the same, and this book does not restate them. What follows is the handover, including the items that do not transfer.
From | What transfers | What changes in ECPR |
Chapter 12 — cannulation | Arterial site selection, sizing by cannula-to-vessel ratio, the right radial arterial line | Chapter 12's rule that the distal perfusion wire goes in before the arterial cannula is harder under chest compressions, where the femoral pulse is generated by a machine and the field moves. Chapter 21 owns the answer |
Chapter 13 — VA haemodynamics | The three circulations, the moving mixing point, the afterload-sensitive pump, pulsatility as a readout of native ejection | At the moment of ECPR flow there is no native ejection at all, so there is no mixing point and no watershed. Both appear later — and their appearance is a sign of recovery, not a complication |
Chapter 14 — the first hours | The initiation sequence; the 4-hour distal perfusion deadline; the oxygen-target evidence | The pre-initiation neurological baseline does not exist. This is the single largest difference between ECPR and every other ECMO indication, and it is why Part IV needs two whole chapters on prognostication and withdrawal |
Chapter 15 — LV distension and unloading | The four inflows, the pressure–volume-area framing, the medical stage, the mechanical menu | §15.9 has already examined unloading in ECPR and concluded that the cardiogenic-shock unloading data must not be extrapolated to the arrested heart. Chapter 22 cites that conclusion; it does not re-argue it |
Chapter 16 — differential hypoxaemia | The three preconditions, the right-radial sampling rule, differential carbon dioxide, the four levers | Chapter 16 already contains the ECPR-specific material, including the recovery paradox and early post-arrest lung injury. Chapter 22 references it |
Chapter 17 — limb ischaemia | The six mechanisms, vasopressor dose as the dominant modifiable factor, NIRS over Doppler, the distal perfusion routes | ECPR cohorts carry a limb ischaemia incidence of about 10.6% — but a recent 42-patient ECPR registry reported distal limb ischaemia in 38.1% and cannulation-site bleeding in 31.0%, all managed conservatively. Chapter 17's definition problem explains the range: these are not the same outcome measured twice |
Chapter 18 — weaning | The three weaning techniques, the echo hierarchy, the aortic valve, measure the change not the value | Chapter 18's anchor systematic review deliberately excluded ECPR-only studies, because neurological prognosis changes the decision entirely. In one series only 20.3% of arrest patients were weaned. Weaning an ECPR patient is not a cardiac question alone |
Danger — the one thing that does not transfer at all
Every chapter before this one selects patients by a physiological state: a Murray score, a P/F ratio, a SCAI stage, a cardiac index, a lactate.
ECPR selects by the clock and by events that happened before anyone from the hospital was present — whether the arrest was witnessed, whether a bystander compressed, what the first rhythm was, how long it has been.
A clinician who reaches Part IV expecting to apply the reasoning of Chapter 11 will look for a number that says this patient needs ECMO, and there isn't one. The corresponding discipline in Part IV is to have decided in advance, in writing, and to let the protocol decide at three in the morning. That is Chapter 20.
19.11 The errors that recur
Error | Why it is wrong |
Treating ECPR as VA ECMO started slightly later | There is no native circulation, no baseline neurological examination and no time. The indication is temporal, not physiological |
Quoting another centre's low-flow cut-point as a rule | A cut-point describes where that cohort's low-flow distribution separated survivors. It is a property of the system, not of the brain |
Refusing a patient at minute 61 | The relationship is a continuous decline, not a cliff. The pooled trial analysis found the largest benefit in patients resuscitated for more than 45 minutes |
Excluding on a single low end-tidal CO₂ immediately after adrenaline | Adrenaline lowers end-tidal CO₂ and bicarbonate transiently raises it. Record the drug times alongside the value |
Leaving the oxygen fraction at 1.0 because nobody owns it | Severe hyperoxia on ECMO carries an adjusted odds ratio of 1.59 for acute brain injury and 1.58 for death in the largest ECPR registry analysis |
Running a high sweep gas into a profoundly hypercarbic patient at full rate | Mechanistically sound concern, one null human study. Start low and titrate up — it costs nothing |
Pausing compressions for the cannulation | Coronary perfusion pressure is built over 3–4 compressions and dissipates within seconds. A pause costs the pause plus the rebuild |
Reading ARREST as proof that ECPR works and INCEPTION as proof that it does not | They tested efficacy and effectiveness. Both results are true and they are about different things |
Quoting a single meta-analysis as the answer | Four syntheses of the same three trials give point estimates from 1.47 to 1.90 and opposite verdicts. Quote the probability, not the P value |
Investing in cannulation skill while ignoring door-to-flow time | INCEPTION's excess low-flow time was generated inside the hospital, despite the fastest on-scene times of the three trials |
Starting a programme without deciding how it stops | "A bridge to nowhere situation can be quite common." Chapters 23, 24 and 77 must exist before the first cannulation, not after it |
Reporting survival to decannulation as success | Chapter 18's rule, with more force here. The endpoint is CPC 1–2 at 6 months |
19.12 Key points
- ECPR is not VA ECMO for a patient who arrested. There is no native circulation, so the pump is the circulation from the first revolution; the indication is elapsed time rather than a physiological state; and the pre-initiation neurological baseline that Chapter 14 called irreplaceable does not exist.
- The word "refractory" carries the whole indication and is undefined — published definitions range from 10 to 60 minutes. Much of the apparent disagreement in this literature is disagreement about that word.
- No-flow is public; low-flow is shared; arrest-to-flow is what a hospital owns. Only the low-flow interval can be shortened by the hospital, and every outcome gradient in ECPR is a gradient along it.
- Conventional CPR is explicitly a holding measure. It does not restore physiology; it slows the rate at which physiology is lost. ECPR is not an alternative to CPR — it is one of the endpoints CPR is buying time to reach.
- A compression pause costs more than its duration. Coronary perfusion pressure accumulates over three or four compressions and dissipates within seconds of stopping. This is the physiological basis of everything Chapter 21 says about cannulating during CPR.
- The window for activation is a subtraction. The features that make a good ECPR candidate also make conventional CPR likely to succeed. The published balance point is 10–20 minutes of professional resuscitation — nearer 20 for shockable rhythms, nearer 10 for non-shockable.
- The low-flow gradient is the most reproduced finding in ECPR: survival 67%, 29%, 10%, 6% at under 20, 20–45, 45–60 and 60–135 minutes in the most-cited series. Optimal cut-points published from other cohorts range from 22 to 66 minutes and are not transferable.
- Sixty minutes is a service design target, not a bedside refusal. A Stockholm programme met it in 9% of cases and 8 of its 9 survivors exceeded it; the pooled trial analysis found the largest benefit in patients resuscitated beyond 45 minutes. A programme routinely exceeding 60 minutes has a system to fix, not a population to exclude.
- Reperfusion is an intervention, not an event. Flow, oxygen, carbon dioxide and temperature all change within seconds, by default, decided by nobody. In 3,125 registry patients, severe hyperoxia on ECMO carried adjusted odds ratios of 1.59 for acute brain injury and 1.58 for death. Give the first right-radial gas an owner and a time.
- The three randomised trials tested three different things — efficacy in an ideal single centre (ARREST, 43% versus 7%), a whole invasive strategy in a dedicated single centre (PRAGUE-OHCA, 31.5% versus 22.0%, P = 0.09), and pragmatic multicentre implementation (INCEPTION, 20% versus 16%). All three results are true.
- The effect size falls as the system moves away from ideal — 18, then 12, then the equivalent of about 3.5 cannulations per centre per year. The most instructive single number in Part IV is that INCEPTION had the shortest on-scene time and the longest low-flow time: the delay was inside the hospital.
- Four meta-analyses of three trials produce point estimates between 1.47 and 1.90 and opposite verdicts. The Bayesian synthesis is the usable statement: about 91% probability of any benefit, 71% probability of at least 5 absolute percentage points, 44% probability of at least 10. This is Chapter 4's EOLIA lesson repeated.
- The observational literature is more favourable than the randomised literature, in the direction that confounding by indication predicts. Take the concordance of direction as reassurance; do not take the GRADE rating as certainty.
- The unit of intervention in ECPR is the programme. One hospital moved from 9.1% to 69.2% good-outcome survival by changing its system and not its technology; the volume signal turns at about 10 cases a year. A unit that cannot state its seven numbers cannot say whether its programme helps anyone.
- Decide how the programme stops before the first cannulation. ECPR restores a circulation, not a brain, and it does not treat the cause. Chapters 23, 24 and 77 are not the end of Part IV — they are its precondition.
Cross-references
Backwards
- Chapter 4 — the Bayesian reinterpretation of EOLIA. §19.8 is the same argument applied to a different intervention, and the chapter's linked companion page on the site sets it out in full.
- Chapter 11 — how this book states VA ECMO evidence. Part IV applies the same discipline and adds its mirror image: three trials of very different systems must not be collapsed into one answer.
- Chapter 12 — arterial cannulation, sizing and the distal perfusion wire.
- Chapter 13 — the three circulations, the moving mixing point, the afterload-sensitive pump.
- Chapter 14 — the first hours on VA ECMO, the 4-hour distal perfusion deadline, the oxygen evidence, and the pre-initiation neurological baseline that ECPR does not have.
- Chapter 15 §15.9 — unloading in ECPR, already examined and already concluded.
- Chapter 16 — differential hypoxaemia, the right-radial sampling rule, and the ECPR-specific presentations.
- Chapter 17 — limb ischaemia, including the definition problem that explains the range of ECPR incidences in §19.10.
- Chapter 18 — weaning versus surviving, and why its anchor review excluded ECPR.
Forwards
- Chapter 20 — ECPR patient selection: age, rhythm, no-flow, signs of life, end-tidal CO₂, aetiology and exclusions. Everything §19.3 and §19.4 imply about writing criteria in advance is executed there.
- Chapter 21 — cannulation during chest compressions, where §19.2's coronary perfusion pressure physiology becomes a technique.
- Chapter 22 — the first hours after flow: coronary angiography, temperature, ventilation, and the reperfusion decisions §19.5 identifies as ownerless.
- Chapter 23 — neurological outcome and prognostication, made necessary by the missing baseline.
- Chapter 24 — failure, withdrawal and organ donation.
- Chapter 60 — transport of the ECPR patient.
- Chapter 77 — resource allocation and the ethics of offering ECPR.
- Chapter 30 — prognostic scoring, which should include RESCUE-IHCA (patient type, presenting rhythm, time of day, age, renal insufficiency and arrest duration; reported prediction ability 72%, external validation 68%) alongside the still-uncovered SAVE score.
References
Randomised trials — abstracts and secondary sources only; all such rows are marked Verified = No
- Yannopoulos D, Bartos J, Raveendran G, et al. Advanced reperfusion strategies for patients with out-of-hospital cardiac arrest and refractory ventricular fibrillation (ARREST): a phase 2, single centre, open-label, randomised controlled trial. Lancet. 2020;396:1807–1816. DOI 10.1016/S0140-6736(20)32338-2.
- Belohlavek J, Smalcova J, Rob D, et al. Effect of intra-arrest transport, extracorporeal cardiopulmonary resuscitation, and immediate invasive assessment and treatment on functional neurologic outcome in refractory out-of-hospital cardiac arrest: a randomized clinical trial (PRAGUE-OHCA). JAMA. 2022;327:737–747. DOI 10.1001/jama.2022.1025.
- Suverein MM, Delnoij TSR, Lorusso R, et al. Early extracorporeal CPR for refractory out-of-hospital cardiac arrest (INCEPTION). New England Journal of Medicine. 2023. DOI 10.1056/NEJMoa2204511. [VERIFICATION REQUIRED] — volume and pages not established; the author list beyond the first author is not confirmed from a retrieved source.
- Bol ME, Suverein MM, Lorusso R, et al. Early initiation of extracorporeal life support in refractory out-of-hospital cardiac arrest: design and rationale of the INCEPTION trial. American Heart Journal. 2019;210:58–68. DOI 10.1016/j.ahj.2018.12.008.
- Yannopoulos D, et al. Rationale and methods of the Advanced Reperfusion Strategies for Refractory Cardiac Arrest (ARREST) trial. American Heart Journal. 2020. DOI 10.1016/j.ahj.2020.07.006.
Pooled and secondary analyses
- Bělohlávek J, et al. Intraarrest transport, extracorporeal cardiopulmonary resuscitation, and early invasive management in refractory out-of-hospital cardiac arrest: an individual patient data pooled analysis of two randomised trials. eClinicalMedicine. 2023. DOI 10.1016/j.eclinm.2023.101988.
- Ubben JFH, et al. Early extracorporeal CPR for refractory out-of-hospital cardiac arrest — a pre-planned per-protocol analysis of the INCEPTION trial. Resuscitation. 2023. DOI 10.1016/j.resuscitation.2023.110033.
- Heuts S, et al. Extracorporeal life support in cardiac arrest: a post hoc Bayesian re-analysis of the INCEPTION trial. European Heart Journal: Acute Cardiovascular Care. 2023. DOI 10.1093/ehjacc/zuad130.
- Ubben JFH, et al. Favorable resuscitation characteristics in patients undergoing extracorporeal cardiopulmonary resuscitation: a secondary analysis of the INCEPTION trial. Resuscitation Plus. 2024. DOI 10.1016/j.resplu.2024.100657. Source of the 46/70 delivery figure flagged in §19.6.
The trialists' comparison — retrieved in full
- Ubben JFH, Heuts S, Delnoij TSR, Suverein MM, van de Koolwijk AF, van der Horst ICC, Maessen JG, Bartos J, Kavalkova P, Rob D, Yannopoulos D, Bělohlávek J, Lorusso R, van de Poll MCG. Extracorporeal cardiopulmonary resuscitation for refractory OHCA: lessons from three randomized controlled trials — the trialists' view. European Heart Journal: Acute Cardiovascular Care. 2023;12(8):540. DOI 10.1093/ehjacc/zuad071. Verified = Yes — full text retrieved. The structural backbone of §19.6 and §19.7.
Syntheses
- Scquizzato T, et al. Refractory out-of-hospital cardiac arrest and extracorporeal cardiopulmonary resuscitation: a meta-analysis of randomized trials. Artificial Organs. 2023. DOI 10.1111/aor.14516. Four randomised trials, 433 patients — the fourth trial is not identified in the abstract and is deliberately not named here.
- Gomes D, et al. Extracorporeal cardiopulmonary resuscitation for refractory out-of-hospital cardiac arrest: a systematic review and meta-analysis of randomized clinical trials. Internal and Emergency Medicine. 2023. DOI 10.1007/s11739-023-03357-x.
- Goel S, et al. Extracorporeal CPR versus conventional CPR for refractory out-of-hospital cardiac arrest: a meta-analysis of randomised clinical trials. European Heart Journal. 2023 (abstract supplement). DOI 10.1093/eurheartj/ehad655.1613.
- Heuts S, Ubben JFH, Kawczynski MJ, Gabrio A, Suverein MM, Delnoij TSR, Kavalkova P, Rob D, Komárek A, van der Horst ICC, Maessen JG, Yannopoulos D, Bělohlávek J, Lorusso R, van de Poll MCG. Extracorporeal cardiopulmonary resuscitation versus standard treatment for refractory out-of-hospital cardiac arrest: a Bayesian meta-analysis. Critical Care. 2024;28:217. DOI 10.1186/s13054-024-05008-9. Verified = Yes — full text retrieved.
- Low CJW, et al. Extracorporeal cardiopulmonary resuscitation versus conventional CPR in cardiac arrest: an updated meta-analysis and trial sequential analysis. Critical Care. 2024. DOI 10.1186/s13054-024-04830-5. Verified = Yes — full text retrieved.
- Low CJW, et al. Extracorporeal cardiopulmonary resuscitation versus conventional cardiopulmonary resuscitation in adults with cardiac arrest: a comparative meta-analysis and trial sequential analysis. Lancet Respiratory Medicine. 2023. DOI 10.1016/S2213-2600(23)00137-6. Source of the centre-volume meta-regression coefficient; abstract only.
Reperfusion physiology
- Shou BL, et al. Arterial oxygen and carbon dioxide tension and acute brain injury in extracorporeal cardiopulmonary resuscitation patients: analysis of the Extracorporeal Life Support Organization Registry. Journal of Heart and Lung Transplantation. 2022. [VERIFICATION REQUIRED] — the bibliographic index attaches a preprint identifier (DOI 10.1101/2022.03.10.22272203) to the journal record. The journal DOI was not established and is not guessed. This is the sixth bibliographic-metadata problem recorded in this book.
- Mandigers L, et al. Initial arterial pCO₂ and its course in the first hours of extracorporeal cardiopulmonary resuscitation show no association with recovery of consciousness in humans: a single-centre retrospective study. Membranes. 2021;11(3):208. DOI 10.3390/membranes11030208.
- Mandigers L, et al. Higher mean cerebral oxygen saturation shortly after extracorporeal cardiopulmonary resuscitation in patients who regain consciousness. Artificial Organs. 2023. DOI 10.1111/aor.14548. Numerical inconsistency flagged in §19.5.
- Ölander C-H, et al. End-tidal carbon dioxide impacts brain and kidney injury in experimental extracorporeal cardiopulmonary resuscitation. Shock. 2020. DOI 10.1097/SHK.0000000000001645.
- Beyersdorf F. Next-generation extracorporeal membrane oxygenation for multi-organ recovery after refractory cardiac arrest. Asian Cardiovascular and Thoracic Annals. 2026. DOI 10.1177/02184923261436888. Developer-reported uncontrolled series; used for the principle, not the numbers.
Programmes and implementation
- Akhtar W, et al. Improving quality and outcomes of extracorporeal cardiopulmonary resuscitation in refractory cardiac arrest: the Phoenix ECPR project. BMJ Open Quality. 2025. DOI 10.1136/bmjoq-2024-002934.
- Frykler Abazi L, et al. Implementation of an extracorporeal resuscitation (ECPR) program for out-of-hospital cardiac arrest in Stockholm, Sweden: feasibility, safety, and outcome. Resuscitation Plus. 2024. DOI 10.1016/j.resplu.2024.100596.
- Bosson N, et al. Implementation of a regional extracorporeal membrane oxygenation program for out-of-hospital cardiac arrest: results of the prospective observational study. Resuscitation Plus. 2025. DOI 10.1016/j.resplu.2025.101177.
- Kwinta A, et al. Feasibility and outcomes of anaesthesiology- and intensive care-led ECPR in a hospital without cardiac surgery: a 2.5-year prospective registry. BMC Anesthesiology. 2026. DOI 10.1186/s12871-026-03678-2. Source of the 38.1% distal limb ischaemia and 31.0% cannulation-site bleeding figures in §19.10.
Guidelines, nomenclature and background — all second-hand
- Richardson ASC, Tonna JE, Nanjayya V, et al. Extracorporeal cardiopulmonary resuscitation in adults: interim guideline consensus statement from the Extracorporeal Life Support Organization. ASAIO Journal. 2021;67:221–228. Not retrieved. Read entirely through Red Book Chapter 32, which reproduces its inclusion criteria and programme logistics tables. The highest-value outstanding target in Part IV.
- Conrad SA, Broman LM, Taccone FS, et al. The Extracorporeal Life Support Organization Maastricht Treaty for Nomenclature in Extracorporeal Life Support. American Journal of Respiratory and Critical Care Medicine. 2018;198(4):447–451. Not retrieved.
- Reynolds JC, Grunau BE, Rittenberger JC, et al. Association between duration of resuscitation and favorable outcome after out-of-hospital cardiac arrest. Circulation. 2016;134:2084–2094. Not retrieved; the figure is reproduced in the ECPR textbook.
- Wengenmayer T, Rombach S, Ramshorn F, et al. Influence of low-flow time on survival after extracorporeal cardiopulmonary resuscitation (eCPR). Critical Care. 2017;21(1):157. Not retrieved; the 67/29/10/6% figures are reproduced in the ECPR textbook.
- Ko R-E, Ryu J-A, Cho YH, et al. The differential neurologic prognosis of low-flow time according to the initial rhythm in patients who undergo extracorporeal cardiopulmonary resuscitation. Resuscitation. 2020;148:121–127. Not retrieved.
- Otani T, Sawano H, Natsukawa T, et al. Low-flow time is associated with a favorable neurological outcome in out-of-hospital cardiac arrest patients resuscitated with extracorporeal cardiopulmonary resuscitation. Journal of Critical Care. 2018;48:15–20. Not retrieved.
- Supady A, et al. Extracorporeal cardiopulmonary resuscitation for refractory cardiac arrest. Lancet Respiratory Medicine. 2025. DOI 10.1016/S2213-2600(25)00122-5. Abstract only — a current seminar, journal blocked.
- Scquizzato T, et al. Post-resuscitation care after adult extracorporeal cardiopulmonary resuscitation: a scoping review. Resuscitation. 2025. DOI 10.1016/j.resuscitation.2025.110880. 133 studies, 90% retrospective, only 17% with more than 500 patients. Held for Chapter 22.
- Kang JK, et al. Post-cardiac arrest care in extracorporeal cardiopulmonary resuscitation. Critical Care Medicine. 2023. DOI 10.1097/CCM.0000000000006102. Held for Chapter 22.
Textbooks
- ELSO Red Book, 6th edition, Chapter 32 — Extracorporeal Cardiopulmonary Resuscitation in Adult Patients: Table 32-1 example inclusion criteria, Table 32-2 key studies, Table 32-3 CPR-specific immediate post-arrest management, the ECPR programme logistics table, the RESCUE-IHCA score, and the timing statements. [VERIFICATION REQUIRED] — page numbers not confirmed.
- Shinar Z, Badulak J. ECPR and Resuscitative ECMO. Chapter 2, Optimizing the Pre-ECMO Resuscitation (Reynolds) for the no-flow/low-flow definitions, compression physiology, coronary perfusion pressure targets and end-tidal CO₂; and Chapter 3 for the low-flow outcome data, rhythm effects and aetiology. [VERIFICATION REQUIRED]. The remainder of this book is the primary source for Chapters 20–24.
- ISCCM Manual of RRT and ECMO in ICU, Chapter 40 (Extracorporeal Cardiopulmonary Resuscitation) and Chapter 28 (indications, including the bridge-to-nowhere warning). [VERIFICATION REQUIRED].
Chapter status
Drafted and audited 11 September 2026. Ten-pass quality control completed: clinical, physiology, evidence, citation, numerical, safety, contradiction, redundancy, bedside utility and literature-currency passes.
This chapter opens Part IV. It deliberately contains no selection criteria — no age limit, no rhythm rule, no no-flow ceiling — because Chapter 20 owns them, and stating them twice would guarantee that the two chapters drift apart. What this chapter establishes instead is why those criteria have to be written down in advance.
Two controversies were set out rather than smoothed. §19.4 sets the sixty-minute rule against a programme that met it in 9% of cases and a pooled analysis that found the largest benefit beyond 45 minutes, and resolves it by separating the prospective use of the number from the retrospective one. §19.8 sets four meta-analyses of the same three trials against each other, shows that their point estimates agree while their verdicts do not, and argues that the Bayesian statement is the only one a clinician can act on.
The numerical audit found three problems that are flagged and not corrected: a published low-flow survival comparison in the ECPR textbook whose two percentages run opposite to the direction of every other finding in the same section; a cerebral oximetry study whose abstract percentages contradict its own stated conclusion; and two different delivery figures — 52/70 and 46/70 — published by the same trial group for the same quantity in INCEPTION. In each case the conclusion was used and the numbers were not. A sixth bibliographic-metadata problem was recorded, a preprint identifier attached to a journal record, and no identifier was reconstructed.
The citation audit also recorded a criterion discrepancy: the Red Book and the INCEPTION trialists describe the trial's resuscitation trigger differently. The trialists' description is used, with both printed.
The chapter's organising argument — that ECPR is a system whose output is time, and that the three randomised trials are ordered by system maturity rather than by biology — is this book's reasoning built on the trialists' own efficacy-versus-effectiveness framing, and is labelled as such throughout. The single number it rests on is that INCEPTION had the shortest on-scene time of the three trials and the longest low-flow time.