<aside> 🧭
Chapter question: The machine was started on a patient who could not be asked. It is still running. The heart has not recovered, the brain may or may not have, and nobody has scheduled a moment at which anyone decides anything. How does an ECPR run end well — and what does the circuit become when it fails?
</aside>
<aside> 📅
Evidence search date: 15 September 2026.
Primary sources: ELSO Red Book 6th edition — Chapter 32 and the ethics and end-of-life material. ECPR and Resuscitative ECMO (Shinar and Badulak). The 2023 AAN/AAP/CNS/SCCM consensus guideline on death by neurologic criteria, which for the first time addresses determination during extracorporeal support.
External evidence: the mode-of-death literature (four cohorts, one registry analysis, one SAVE-J II secondary analysis totalling 1,660 patients), the apnoea-testing literature on ECMO (one systematic review, one multicentre series, one Harlequin-specific study, one 2026 national-directive analysis), the ECMO decision-making ethnography, the ECPR organ-donation literature (one 307-patient cohort, one systematic review of 254 donors, one integrated ECPR-and-donation programme), and the normothermic regional perfusion literature including its ethical scoping review.
Everything external in this chapter comes from a structured abstract. Nothing was retrieved in full text. Every database row seeded from this chapter is marked Verified = No.
</aside>
This is the last chapter of Part IV, and it is the only one about ending. Chapter 23 produced a probability. This chapter is about what is then done with it — and about three things that happen in no other ECMO chapter, because they happen almost only after ECPR.
| This chapter owns | Owned elsewhere — cite, do not re-argue |
|---|---|
| How ECPR runs actually end: the four modes of death and their four different clocks | Futility and non-beneficial ECMO as a concept — Chapter 74 |
| Why a withdrawal decision is structurally hard to make on a circuit — the escalation trap | The procedure of withdrawing ECMO, and dying well on a circuit — Chapter 75 |
| The ECPR consent asymmetry: started without consent, stopped only with it | Symptom control, family support and bereavement — Chapter 76 |
| Determination of death by neurologic criteria on ECMO — the apnoea test the circuit sabotages | Resource allocation, and whether a unit's criteria are a dial — Chapter 77; Chapter 20 §20.9 |
| Organ donation after failed ECPR, and normothermic regional perfusion | ECMO as a bridge to transplantation for the patient on it — Chapter 72, Chapter 82 |
| Recognising the ECPR run that cannot be maintained at all | Weaning and decannulation of a recovering patient — Chapter 18; the checklist — Chapter 91 |
| Prognostic input to a decision | How that prognosis is generated — Chapter 23; cerebral monitoring — Chapter 28 |
| The ECPR bridge to nowhere specifically | Bridge-to-decision as a strategy — Chapter 70; goals of ECMO — Chapter 69 |
<aside> ⚠️
Pitfall · Three subjects in this chapter have no other home in this book, and that is deliberate
A tracker query run while preparing this chapter established that no chapter in Understanding ECMO owns determination of death by neurologic criteria on a circuit, organ donation after ECMO, or normothermic regional perfusion. They are placed here rather than in Part XI for a physiological reason, not an editorial one:
Chapters 74 to 77 own the ethics. This chapter owns the physiology and the process.
</aside>
Almost every clinician's mental model of ECPR failure is neurological: the patient survives the arrest, the brain does not, and the family is asked to agree to stop. That model describes a minority of deaths, and it describes the slowest of them.
<aside> 📊
Evidence · Low certainty · What actually kills ECPR patients
Four cohorts, adjudicated independently, converge.
Two conclusions hold across all four. Refractory shock kills at least as many ECPR patients as the brain does. And the majority of ECPR deaths occur with the circuit still running, which means somebody must decide to stop it.
</aside>
The mode of death is not merely a label. Each mode has its own clock, and the single most useful table in this literature is the one that separates them.
| Mode of death | Median survival after cannulation | What it looks like at the bedside |
|---|---|---|
| Refractory shock | 0 days (IQR 0–2) | Flow cannot be maintained, lactate will not clear, vasopressor requirement climbs against full support. The decision is often made by the physiology rather than by anyone in the room |
| Brain death | 2 days (IQR 1–4) | Brainstem reflexes disappear early and completely. The question becomes one of determination, not prediction (§24.5) |
| Non-neurological withdrawal | 3 days (IQR 1–8) | Multiorgan failure, uncontrollable bleeding, an unreconstructable heart with no destination |
| Neurological withdrawal without brain death | 9 days (IQR 2–14) | The commonest single modality. The patient is not dead, will not recover, and the circuit is working perfectly |
<aside> 💡
Clinical pearl · The four clocks tell you which conversation to prepare
In the 95-patient series that produced those figures, 51 (53.7%) had died by day 28, and neurological withdrawal without brain death was both the commonest modality and the latest. Refractory shock was the earliest, with a median survival of zero days.
The practical consequence is a triage of conversations rather than of patients:
A unit that runs only one kind of end-of-life conversation is prepared for the least common and latest of the four.
</aside>
<aside> 🚨
Danger · Most withdrawal decisions are made before Chapter 23's assessment can begin
A secondary analysis of 1,660 out-of-hospital ECPR patients found that 510 (30.7%) had a decision to withhold or withdraw life-sustaining therapy. The number of such decisions was highest on the first day, and the median was two days after intensive care admission.
The reasons given were: perceived unfavourable neurological prognosis 300/510 (58.8%); perceived unfavourable cardiac or pulmonary prognosis 105/510 (20.5%); inability to maintain extracorporeal support 71/510 (13.9%); complications 10 (1.9%); exacerbation of pre-existing comorbidity 7 (1.3%). Thirty-day survival was 36/506 (7.1%) with a withdrawal decision against 386/1140 (33.8%) without.
Now set that beside Chapter 23 §23.5 and §23.9, whose central practical conclusion was that the prognostic clock starts when sedation clears, which on a circuit is later than 72 hours — and that at 72 hours in the largest prospective algorithm cohort, half of all patients were still indeterminate.
The commonest single reason for stopping ECPR is a neurological prognosis, and the commonest day on which it is invoked is the first. Nothing in Chapter 23 supports a neurological prognosis on day one in a sedated patient on a circuit. Either these decisions are being made on grounds other than the ones recorded, or they are being made on grounds the evidence does not support. Both possibilities are serious, and the registry cannot distinguish them.
</aside>
<aside> 🧠
Physiology · Why refractory shock, not the brain, is the early killer — and why the machine hides it
Venoarterial ECMO delivers flow. It does not deliver a working heart, and it does not reverse the metabolic debt already incurred. After a prolonged low-flow interval, the patient arrives on the circuit with a vasoplegic, acidotic, capillary-leaking circulation, a stunned myocardium and, frequently, an ischaemic gut. The pump conceals the first consequence of that state — a falling blood pressure — while doing nothing about the state itself.
This is why the early trajectory must be read from variables the pump does not set: lactate clearance, vasopressor requirement at a fixed flow, mixed venous saturation trend, and the ability to maintain flow without progressive volume loading. A mean arterial pressure of 70 mmHg on a centrifugal pump at maximal vasopressor support is not reassurance; it is the machine answering a question nobody asked.
Chapter 13 owns the haemodynamics and Chapter 22 owns the first 24 hours. The point here is narrower: the mode of death that arrives first is the one the circuit is best at masking.
</aside>
This is the organising idea of the chapter, and like Chapter 23's it is structural rather than clinical.
Every account of how ECMO should be stopped assumes that there is a moment at which stopping is decided. Somebody weighs the prognosis, somebody speaks to the family, a decision is reached. The one study that went and watched found that this moment does not reliably exist.
<aside> 📊
Evidence · Moderate certainty · What actually happens is escalation, not decision
A focused ethnography in two academic cardiothoracic intensive care units: 380 hours of direct observation, 34 weekly interviews with the families of 20 ECMO patients, and 13 interviews with unit clinicians.
The finding, in the authors' own words: "Following ECMO initiation, treatment was escalated as complications mounted until the patient either could be decannulated or interventional options were exhausted. Families were well-informed about treatment and prognosis but played minimal roles in shaping the trajectory of care."
And the conclusion that matters most: "Discussion between clinicians and families about prognosis and goals was frequent but did not occasion decision-making moments."
The authors offer this to explain something otherwise puzzling — why communication interventions designed to strengthen surrogate participation have had so little measurable effect on ECMO trajectories. They have little effect because they are aimed at a decision point that the structure of care does not produce.
</aside>