Chapter 28 — Cerebral Monitoring

🧭

Chapter question: Every other organ on ECMO is watched through a number the circuit itself produces. The brain has no line. At the same moment, sedation removes the examination and the hardware removes the magnetic resonance scanner. What is left, what does each remaining instrument actually detect, and how should the bedside act on it?

📅

Evidence search date: 20 September 2026.

Primary sources: ELSO Red Book 6th edition — the general ECLS management chapter (neurological management, the Sutter systematic review figures, the EuroELSO neuromonitoring survey), the adult cardiac management chapter (the near-infrared spectroscopy trend argument, transcranial Doppler and electroencephalography on adult support), Red Book Chapter 33 (complications of venoarterial support: the autopsy series, the standardised-neuromonitoring cohort, neuron-specific enolase and haemolysis, the near-infrared spectroscopy operating characteristics), Red Book Chapter 9 and the paediatric cardiac chapter (carbon dioxide fluctuation and cerebral ischaemia, the nonconvulsive seizure range, the registry-as-minimum-estimate warning), and the bedside ultrasound chapter (continuous flow and pulsatility, brain-death considerations). ISCCM Manual of RRT and ECMO in ICU — Chapter 33 (neurological monitoring during ECMO; the absence of a minimum standard) and ISCCM Chapter 35 (sedation, the post-cannulation neurological assessment, delirium).

External evidence: the 2024 ELSO consensus guidelines on neurological monitoring and management for adult ECMO patients, read as full text; a cerebral pathophysiology and neuromonitoring review; a single-centre venoarterial stroke cohort; two prospective transcranial Doppler microembolic-signal cohorts and one earlier prospective series; a prospective multimodal neuromonitoring feasibility study; a before-and-after standardised-neuromonitoring cohort; a peri-cannulation carbon dioxide cohort; a near-infrared spectroscopy trend study; a cerebral autoregulation feasibility study; a retrospective electroencephalography cohort; a four-patient brain autopsy series; a quantitative pupillometry prognostication cohort; and a multidisciplinary review of brain injury on extracorporeal support.

Every external reference was read as an indexed abstract or full-text passage, except the 2024 ELSO consensus guidelines, which were read as complete full text. Digital object identifiers are reproduced from the search record and have not been independently resolved. Every database row seeded from this chapter is marked Verified = No, except the ELSO consensus guideline row.

What this chapter covers — and what it does not

Chapter 25 established that every haemodynamic instrument on a circuit was calibrated in a circulation that no longer exists. Chapter 26 established that every echocardiographic number is a measurement at a stated flow. Chapter 27 established that a blood gas is uninterpretable without its sampling site. This chapter is about the organ for which there is no instrument at all — and about what follows, at the bedside, from monitoring an organ through proxies that were all validated somewhere else.

This chapter owns	Owned elsewhere — cite, do not re-argue
The neurological examination on a sedated, cannulated patient — what survives, what does not, and how often	The choice and dosing of sedatives on a circuit, and their altered pharmacokinetics — Chapter 49; the sedation and delirium strategy — Chapter 57
Quantitative pupillometry, near-infrared spectroscopy, cerebral autoregulation indices, transcranial Doppler, electroencephalography, evoked potentials and serum biomarkers — what each detects and what each misses	Post-arrest prognostication and long-term neurological outcome after ECPR — Chapter 23; brain death determination, the apnoea test on a circuit, and organ donation — Chapter 24
The detection gap — why every published neurological complication rate is a detection rate rather than an incidence	Registry structure and reporting limitations in general have no owner in the current map. The nearest homes are Chapter 1 and Appendix 12. Recorded in §28.19
Peri-cannulation physiological targets that bear on the brain: the rate of carbon dioxide correction, early hyperoxaemia, pulse pressure and blood pressure	Gas targets as gas exchange — Chapter 27; the haemodynamic instruments themselves — Chapter 25; differential hypoxaemia physiology — Chapter 16
Assembling a cerebral monitoring protocol, and the escalation pathway when an instrument changes	Bleeding as a complication and the anticoagulation decision itself — Chapter 36, Part VII; circuit thrombosis — Chapter 31
Neurological prognostication specifically as a monitoring problem — which tests are confounded by the circuit	The prognostication algorithm after extracorporeal cardiopulmonary resuscitation — Chapter 23; withdrawal of life-sustaining therapy — Chapter 74, Chapter 75

⚠️

Structural gap in this book, recorded here because this chapter found it

A scope sweep of all 92 chapters was run before drafting. Part VI — Complications — contains eight chapters, and all eight are circuit complications: thrombosis, oxygenator failure, pump failure, cannula problems, air, bleeding, haemolysis and mechanical failure. Chapter 23 owns neurological outcomes and prognostication after extracorporeal cardiopulmonary resuscitation. Chapter 24 owns brain death and donation.

No chapter in the map owns ischaemic stroke, intracranial haemorrhage or seizures on ECMO as complications to be managed. This is the third scope gap the book has found, after brain death (closed into Chapter 24) and extracorporeal carbon dioxide removal (still open).

Two consequences follow, and both shape what you are about to read.

This chapter must own detection and surveillance completely, because nothing downstream will pick it up.
Section 28.15 carries the acute management recommendations as a stopgap, clearly labelled, pending a decision about where they permanently belong. That decision is owed before Part VI is drafted. This chapter's recommendation is that Part VI gain a chapter on neurological complications, or that Chapter 38 be redefined.

28.1 The organ the circuit does not monitor

Stand at the bedside of a patient on ECMO and count the instruments. The pump tells you revolutions and flow. The pressure transducers tell you access, internal and return pressures, and the difference across the membrane. The blood gas analyser tells you what the membrane is delivering, at four sites. The oximeter on the drainage limb tells you venous saturation continuously. The echo machine tells you what the heart and the cannulae are doing. Almost every organ system that matters is being watched, in real time, through numbers the circuit itself generates.

The brain is not on that list. There is no cerebral line, no cerebral pressure trace, no cerebral saturation that the console reports. And the two instruments that normally substitute for one in a critically ill patient are both removed at the same moment, by the treatment itself:

The neurological examination is removed by sedation and neuromuscular blockade, which are heaviest in exactly the window — the first 24 to 72 hours — when the risk of acute brain injury is highest.
Magnetic resonance imaging is removed by the hardware. This is not a relative contraindication or a logistical difficulty. The ELSO Red Book and the ISCCM manual both state it plainly: magnetic resonance imaging is not possible during ECMO because magnetic-resonance-compatible consoles and consumables do not exist. The single most sensitive test for the injuries in question is unavailable for the entire duration of the exposure.

🧠

Physiology · Why the brain is the organ most exposed by this arrangement

Every organ is at risk on a circuit. The brain is differently at risk, for four reasons that compound.

It has no reserve. Cerebral oxygen consumption is approximately 3.5 mL of oxygen per 100 g of brain tissue per minute, sustained, with essentially no anaerobic capacity. Minutes of inadequate delivery produce irreversible loss. Every other organ tolerates a deficit that the brain does not.
It sits at the top of the aorta, at the far end of the mixing zone. On peripheral venoarterial support the brain is perfused by whichever blood reaches the arch — which may be native, poorly oxygenated blood from a recovering heart (Chapter 16). The brain is the first organ to be injured by differential hypoxaemia and the last to show it.
It is downstream of the arterial return cannula on venoarterial support. Anything embolic that leaves the circuit — thrombus, fibrin, gas — enters a major artery and can be delivered directly to the cerebral circulation without passing a pulmonary filter.
It is simultaneously the organ most injured by bleeding. Full systemic anticoagulation sits on top of all of the above, and intracranial haemorrhage in this population carries a mortality that the Red Book gives as 80 to 90 percent.

No other organ combines zero reserve, last place in the oxygen queue, direct embolic exposure and catastrophic bleeding risk. That combination is why cerebral monitoring is a separate discipline on ECMO rather than a subsection of general monitoring.

💡

Clinical pearl · Cerebral monitoring on ECMO is not refinement, it is reconstruction

It is tempting to think of neuromonitoring as an optional extra — something added on top of good care in well-resourced centres. That framing is wrong, and it changes what you do.

In an ordinary ICU patient, the examination is the monitor, and near-infrared spectroscopy or electroencephalography would indeed be refinements laid on top of it. On ECMO the examination has been taken away. Everything in this chapter exists to rebuild a sense that the treatment removed, not to sharpen one that is already working.

The practical consequence: the question at the bedside is never "does this patient need neuromonitoring?" but "which proxy am I using in place of the examination I cannot perform, and what is that proxy blind to?"

28.2 What actually injures the brain on ECMO

The instruments in §§28.4 to 28.12 only make sense against the mechanisms they are trying to catch. Six recur across every source consulted.

Mechanism	Configuration mainly at risk	Timing	Which instrument can see it
Hypoxic-ischaemic injury sustained before or during cannulation — the low-flow or no-flow state that led to ECMO in the first place	Venoarterial, above all extracorporeal cardiopulmonary resuscitation	Already present at the moment the circuit starts	Electroencephalography background; evoked potentials; pupillometry; late computed tomography. Not preventable by anything done after cannulation
Thromboembolism from the circuit, the cannula, the left heart or the aortic root	Venoarterial predominantly; possible in both	Throughout the run	Transcranial Doppler microembolic signals (with the caveats of §28.8); new focal deficit; computed tomography
Intracranial haemorrhage on full anticoagulation, with acquired coagulopathy and platelet dysfunction	Both	Throughout; early in infants, less clearly time-bound in adults	Computed tomography only. No bedside monitor detects it reliably
Differential hypoxaemia delivering desaturated native blood to the arch while the femoral gas looks normal	Peripheral venoarterial, when the heart recovers before the lung	Any time cardiac ejection returns	Right radial arterial gas; near-infrared spectroscopy, especially right-sided
Abrupt correction of chronic hypercapnia, producing cerebral vasoconstriction on a bed already vasodilated and autoregulation-impaired	Both, venovenous best documented	The peri-cannulation window specifically	Nothing detects the injury in real time. The monitor is the blood gas trend itself (§28.13)
Loss of cerebral autoregulation under continuous, non-pulsatile flow, making perfusion pressure-passive	Venoarterial more than venovenous	Greatest on day 1, improving by day 2	Cerebral oximetry index and related autoregulation indices (§28.7)

💡

Clinical pearl · Half of the brain injury on ECMO happened before the pump was switched on

The single most consequential thing to hold in mind when interpreting any cerebral monitor. The Red Book states it for the venoarterial population: in the majority of cardiac ECMO patients, hypoxic-ischaemic brain injury occurs prior to cannulation, due to haemodynamic instability. The same chapter gives a range of 14 to 61 percent for hypoxic-ischaemic injury in venoarterial patients — the widest range in this chapter, and a range that says more about detection than about biology.

Two clinical consequences, and they point in opposite directions.

A grossly abnormal electroencephalogram on day 1 of extracorporeal cardiopulmonary resuscitation is very often describing the arrest, not a new event on the circuit. It is a baseline, not an alarm.
Which means the alarm is the change, not the value — the same principle Chapter 26 established for ejection fraction and Chapter 25 for pulsatility. This is the third time this book has arrived at it from a different direction, and it is the closest thing Part V has to a unifying rule.

🚩

Pitfall · Treating venovenous support as neurologically safe

Brain injury is more common on venoarterial than venovenous support, and the reasoning is sound: arterial return, embolic exposure, a mixing zone, often a preceding arrest. It does not follow that the venovenous patient is safe, and two specific mechanisms are easy to forget.

Abrupt carbon dioxide correction is best documented in the venovenous population, not the venoarterial one. The largest signal in the literature for peri-cannulation carbon dioxide and intracranial haemorrhage comes from registry analyses of venovenous patients with acute respiratory distress syndrome (§28.13). The venovenous patient is the one most likely to arrive with a profoundly elevated carbon dioxide tension — and therefore the one in whom rapid correction is most available.
The 2024 ELSO consensus specifically flags large dual-lumen venovenous cannulae as possibly associated with acute brain injury, through venous hypertension and cannula-related thrombosis, including cerebral venous sinus thrombosis (§28.15).

28.3 The detection gap: every published rate is a detection rate

This is the second structural claim of the chapter, and it should change how you read every number in it — including the numbers in this chapter's own tables.

Source and method	What it found	What it is actually measuring
ISCCM manual, citing the recent literature	Neurological complications (stroke and bleeding) 4 to 9 percent	Events detected and recorded in routine care
Sutter, Tisljar and Marsch systematic review of 44 studies of adult ECLS, cited in the Red Book	Median risk of acute neurological complications 13 percent: intracranial haemorrhage 5 percent, ischaemic stroke 5 percent, seizures 2 percent. Median mortality with a neurological complication 83 percent versus 42 percent without, p under 0.001	Pooled detection across studies with heterogeneous monitoring
ELSO Registry, adult cardiac, cited in the Red Book	Any neurologic event 9 percent in adult cardiac patients (24 percent in neonates, 20 percent in children)	Voluntary reporting. The Red Book says explicitly that registry CNS complications "should be viewed as minimal estimates"
2024 ELSO consensus guidelines, for the highest-risk group	Acute brain injury in 27 to 32 percent of extracorporeal cardiopulmonary resuscitation patients	Detection in a protocolised, neurology-engaged setting
Cho and colleagues, autopsy shortly after decannulation (cited in both the Red Book and the 2024 ELSO consensus; 90 percent venoarterial)	68 percent of ECMO non-survivors had acute brain injury, most commonly hypoxic-ischaemic injury (44 percent), then intracranial haemorrhage and ischaemic infarct. A second cohort: 9 of 10 brains showed acute brain injury at autopsy	Actual pathology, in the subgroup who died
Rastan and colleagues, autopsy series (cited in the Red Book)	30 percent rate of undiagnosed thromboembolic events	The discrepancy itself, measured directly
Long-term magnetic resonance imaging in survivors, cited in the 2024 ELSO consensus	Cerebral infarction or haemorrhage in 37 to 52 percent of adult ECMO survivors, higher in venoarterial than venovenous, and associated with cognitive impairment	Pathology in the subgroup who lived — found once the scanner became available again

⚠️

Danger · The reassuring number and the real one differ by a factor of five to ten

Set the two ends of that table against each other. Routine care detects neurological injury in something like 4 to 13 percent of adult patients. Autopsy finds it in 68 percent of those who die. Delayed magnetic resonance imaging finds it in 37 to 52 percent of those who live.

These are not contradictory findings and it is important not to read them as a scandal about registries. They are measurements of different things: one is a detection rate under conditions in which the examination is unavailable and the definitive scanner is prohibited; the other is the prevalence of pathology.

The clinical rule that follows is the most important one in this chapter. Because the gap between detection and pathology is this large, no cerebral monitoring result on ECMO can be used to reassure. Every instrument in §§28.4 to 28.12 is, in operating-characteristic terms, a rule-in test. A normal near-infrared spectroscopy trace, a normal transcranial Doppler, an unremarkable computed tomogram and a settled electroencephalogram, taken together, still leave a substantial probability of established injury. They do not license a lower index of suspicion. They license only the absence of a specific action right now.

🔬

Evidence · Does standardised neuromonitoring improve outcome, or improve diagnosis?

This distinction is repeatedly blurred in the literature, and it is worth being precise because it determines what you should expect from a protocol.

What was done. Ong and colleagues compared two eras at a single centre: era 1 (2016 to 2017, 70 patients) without standardised neuromonitoring, and era 2 (2017 to 2020, 145 patients) with a protocol of serial neurological examinations, electroencephalography, transcranial Doppler and somatosensory evoked potentials. Total 215 patients, mean age 54, 60 percent male, 71 percent venoarterial.

What was found. Acute brain injury was diagnosed in 23 percent in era 1 and 33 percent in era 2 (p equal to 0.12 on the unadjusted comparison). On multivariable logistic regression, standardised neuromonitoring was independently associated with a diagnosis of acute brain injury: odds ratio 2.24, 95 percent confidence interval 1.12 to 4.48, p equal to 0.02, alongside pre-ECMO cardiac arrest (odds ratio 2.17, 1.14 to 4.14). The paper also reports a greater proportion of patients with good neurological outcome in era 2.

How to read it. The finding that is solid is the first one: a protocol roughly doubles the odds that brain injury is diagnosed. That is a statement about the detection gap, and it is exactly what §28.3 predicts.

The outcome claim is weaker and should be held loosely. It is a before-and-after comparison at one centre over a period in which much else changed, including, by the authors' own account, the presence of neurocritical care consultation. Secular improvement in ECMO outcomes over 2016 to 2020 is well documented independently. This design cannot separate "we found more injury and acted on it" from "our programme matured."

Certainty: low for the outcome claim, moderate for the detection claim. The 2024 ELSO consensus cites this literature in the careful form — that protocolised monitoring "is shown to improve the detection of acute brain injury" — and then notes it is limited to a few centres. That is the accurate summary, and it is the one to repeat.

Sources: Ong CS, Etchill E, Dong J, Shou BL, Shelley L, et al. J Thorac Cardiovasc Surg 2023;165(6):2104–2110.e1. DOI 10.1016/j.jtcvs.2021.09.063. Cho SM, Hwang J, Chiarini G, Amer M, Antonini MV, Barrett N, et al. Crit Care 2024;28(1):296. DOI 10.1186/s13054-024-05082-z.

Note on the citation: this paper carries a 2021 online publication date and a 2023 print citation. Both are reproduced above as the record gives them. [VERIFICATION REQUIRED: print year and online year should be reconciled against the publisher record.]

28.4 The neurological examination: what survives sedation

Every source consulted, without exception, places the serial bedside examination first. The 2024 ELSO consensus opens its monitoring section with it: "Serial bedside examination remains the mainstay of neurological assessment in ECMO patients." It then immediately concedes the problem — that evaluation, especially early after cannulation, is frequently confounded by sedatives and paralytics, which is why multimodal monitoring exists at all.

Three practical specifications come out of the consensus document, and they are the ones to adopt.

Question	2024 ELSO consensus position	What this means at the bedside
When is the baseline taken?	Before and immediately after cannulation, then serially throughout support and after weaning	The pre-cannulation examination is often the only unsedated examination that will exist for days. It must be recorded, by someone, in the emergency. If it was not, say so explicitly in the notes rather than leaving a blank
How often?	The ideal frequency is not established. Daily assessment by a neurologist or neurointensivist, where available, can improve care. Nursing assessment every 1 to 4 hours, stratified by brain-injury risk, is reasonable	A defensible default: hourly in the first 24 hours after extracorporeal cardiopulmonary resuscitation or after any new neurological event; every 4 hours in a stable venovenous patient
What is examined?	Mental status; brainstem reflexes — pupillary light response, oculocephalic, corneal, cough and gag; motor examination. Standardised tools: Glasgow Coma Scale, Confusion Assessment Method	The brainstem reflexes are the part that survives light sedation. They are the examination on ECMO

🚩

Pitfall · Three parts of the examination are simply unavailable, and charting them is worse than omitting them

The consensus is unusually candid about the limits, and each limit has a charting consequence.

Motor response of the extremities is only helpful when analgosedation and paralytic are lightened or off. A motor score recorded through a paralytic is not a finding; it is a record of the paralytic.
Sensory examination is "mostly limited" in ECMO patients — in practice, absent.
Spinal cord injury, which the consensus calls rare but devastating, is described as very challenging to examine for. It is the injury that will be missed.

The harm of charting an unobtainable examination as normal is specific: it creates a documented normal baseline against which a later genuine deterioration appears as a change from normal, and it removes the prompt to obtain a real examination. Chart what was not assessable, and why.

💡

Clinical pearl · The single most actionable sedation statement in the guideline

The 2024 ELSO consensus makes a recommendation that is easy to skim past and changes daily practice: "Overall, intermittent (as-needed) analgo-sedation is preferred over continuous infusion." Short-acting, non-benzodiazepine agents could be considered. Daily reassessment of sedation goals, stepwise weaning and sedation interruptions can improve neurological examinations and the diagnosis of acute brain injury.

The ISCCM manual says the same thing more bluntly for the first hours: deep sedation to the level of light anaesthesia for the cannulation itself — to prevent air entrainment on spontaneous breathing, to avoid movement, and to reduce oxygen consumption — and then, once the patient is on the circuit, "it is prudent to stop all sedation and narcotics to assess the neurological status." After 48 hours, with a stable patient, minimal or no sedation is encouraged.

The Red Book adds the framing sentence that makes this possible to say on a ward round: "ECLS per se does not mandate sedation."

Read together, these are an argument that sedation strategy is a neuromonitoring intervention. The sedation interruption is not a comfort or a weaning issue on ECMO. It is the only way the primary monitor is switched back on.

⚠️

Danger · Fixed dilated pupils at the time of cannulation do not mean what they used to mean

The 2024 ELSO consensus corrects a piece of received practice, and the correction has life-or-death consequences at the point of decision.

The old teaching: absence of brainstem reflexes with fixed, dilated pupils before cannulation was equated to irreversible brain injury and treated as a contraindication to ECMO.

What the consensus states: during cardiopulmonary resuscitation, fixed and dilated pupils are frequently seen after adrenaline administration, and patients have achieved favourable outcomes despite these findings.

The rule that follows: a pupillary examination performed during or immediately after a resuscitation in which adrenaline was given is not a prognostic test. It is a pharmacological observation. Do not decline extracorporeal cardiopulmonary resuscitation on the pupils alone.

The consensus moves the useful pupillary information in the opposite direction, toward signs of life — gasping, pupillary light response, increased consciousness observed before or during resuscitation and while on support — which may be associated with improved neurological outcomes.

Chapter 20 owns signs of life as a selection criterion and has already quantified them, so this chapter cites rather than re-argues: favourable outcome 38.0 versus 8.1 percent in 1,395 registry patients (adjusted odds ratio 5.65), pooled odds ratio 9.63, negative predictive value 0.96 and positive predictive value 0.22 — that is, a strong rule-out and a weak rule-in, which is the same operating-characteristic shape as everything else in this chapter.

The monitoring consequence, which is this chapter's part: signs of life must be observed and recorded to exist. In a patient who is about to be sedated and cannulated, the window in which anyone can see gasping or a pupillary response is minutes long, and it closes. Record them in the cannulation note, positively or negatively.

28.5 Quantitative pupillometry: the one instrument with a published zero false-positive rate

Automated infrared pupillometry converts the bedside pupil from a binary observation into a number, the Neurological Pupil index. It deserves separate treatment because its published operating characteristics on venoarterial ECMO are unlike anything else in this chapter.

🔬

Evidence · Neurological Pupil index for early prognostication after venoarterial ECMO

What was done. An observational cohort of 100 consecutive sedated, mechanically ventilated venoarterial ECMO patients — 51 with refractory cardiogenic shock and 49 with refractory cardiac arrest — at a single European centre. Pupillary light reactivity was measured every 12 hours during the first 72 hours after ECMO insertion using an automated infrared pupillometer, generating the Neurological Pupil index. Trends were correlated with 90-day mortality, alone and combined with the 12-hour PREDICT VA-ECMO score.

What was found. Observed 90-day survival was 43 percent. Non-survivors (n equal to 57) had significantly lower index values than survivors at every time point (all p under 0.01). An abnormal index — below 3, at any time from 24 to 72 hours — was 100 percent specific for 90-day mortality, with 0 percent false positives. Adding the 12-hour PREDICT VA-ECMO score gave the best combined performance: specificity 100 percent (95 percent confidence interval 92 to 100 percent), sensitivity 60 percent (46 to 72 percent), area under the receiver operating characteristic curve 0.82.

What it means. This is the only bedside instrument in this chapter for which a published series reports no false positives, and the only one that is cheap, repeatable at the bedside, requires no neurophysiologist, and works through sedation. That combination is why it deserves a section.

What it does not mean, and the limits are serious.

A single centre, 100 patients, and a threshold derived in the same dataset in which it was tested. A 100 percent specificity in 57 non-survivors has an exact confidence interval whose lower bound is well below 100 percent; the paper reports the interval only for the combined score.
The outcome is 90-day mortality, not neurological outcome, in a population where mortality is heavily determined by cardiac and multi-organ factors.
Self-fulfilling prophecy is not excluded and cannot be excluded by this design. The 2024 ELSO consensus names this bias explicitly in its prognostication section: prognostic test results indicating poor outcome influence treatment decisions. If the treating team saw the index values, a 0 percent false-positive rate is partly a description of clinical behaviour.
Note the timing window: 24 to 72 hours, not 0 to 24. This is exactly consistent with §28.4's warning about adrenaline. The test is deliberately not applied at the moment the pupils are pharmacologically unreliable.

Certainty: moderate that the index carries real prognostic signal; low that the threshold of 3 can be transported to another unit and acted on.

Source: Miroz JP, Ben-Hamouda N, Bernini A, Romagnosi F, Bongiovanni F, Roumy A, Kirsch M, Liaudet L, Eckert P, Oddo M. Chest 2020;157(5):1167–1174. DOI 10.1016/j.chest.2019.11.037. A subsequent short report addressed the same question in cardiac arrest patients on ECMO: Menozzi M, Oddo M, Peluso L, Dessartaine G, Sandroni C, Citerio G, Payen JF, Taccone FS. ASAIO J 2022;68(7):e118–e120. DOI 10.1097/MAT.0000000000001569. This second reference was retrieved without an abstract and its findings are not reproduced here. [VERIFICATION REQUIRED: content of Menozzi 2022.]

💡

Clinical pearl · How to use pupillometry without over-reading it

Three rules make the instrument useful without making it dangerous.

Use it as a trend, twice daily, from the moment of cannulation. Its value in the cited cohort came from separating survivors and non-survivors at every time point, not from a single reading.
Do not act on a value obtained inside the first 24 hours, and above all not on one obtained during or shortly after a resuscitation in which adrenaline was given.
Treat a normal index as uninformative, not reassuring. Sensitivity was 60 percent even in the best combination. A normal pupil index in a patient with a large intracranial haemorrhage is an entirely expected finding.

28.6 Near-infrared spectroscopy: the signal is the asymmetry, not the number

Near-infrared spectroscopy is the most widely deployed cerebral monitor on ECMO. The EuroELSO survey cited in the Red Book found it was the most commonly used modality across both populations — 48 percent of adult centres and 80 percent of paediatric centres. It is also, on the evidence, the instrument most often misread.

🧠

Physiology · What the probe is actually measuring

Near-infrared light penetrates roughly 2 to 2.5 cm below the sensor, sampling a volume of frontal cortex that is predominantly venous — conventionally taken as around 70 to 75 percent venous and 25 to 30 percent arterial blood. The reported regional cerebral oxygen saturation is therefore a mixed tissue saturation in a small, superficial, frontal territory, used as a surrogate for the balance between cerebral oxygen delivery and consumption, and by extension for cerebral blood flow.

Four things follow, and each one explains a specific failure mode.

It is a regional monitor read as a global one. A posterior circulation infarct, a deep infarct, or a haemorrhage outside the frontal sampling volume can be entirely invisible.
Its absolute value is not individually validated. The Red Book's adult cardiac chapter states this precisely: the absolute regional saturation "varies significantly between individuals and has not sufficiently been validated" — but a drop of over 25 percent from baseline may be sensitive for acute brain injury, "which emphasizes the importance to monitor trends."
Extracranial contamination is real. Scalp and skull perfusion contribute; oedema, haematoma under the sensor and probe placement all alter the reading.
On venoarterial support it is the one instrument positioned to see differential hypoxaemia. The 2024 ELSO consensus says so directly: monitoring cerebral oxygenation using near-infrared spectroscopy may be useful in diagnosing differential oxygenation. This is its highest-value use and the one to build a protocol around.

Source for penetration depth and the surrogate relationship: Kazmi SO, Sivakumar S, Karakitsos D, Alharthy A, Lazaridis C, Leenen L. Crit Care Res Pract 2018;2018. DOI 10.1155/2018/3237810.

🔬

Evidence · The trend study that reframes how the number should be read

What was done. Sixty-nine adult venoarterial ECMO patients (mean age 56 years, 65 percent male) underwent cerebral near-infrared spectroscopy with left and right sensors, values collected every hour from cannulation. Acute brain injury was defined as intracranial haemorrhage, ischaemic stroke, hypoxic-ischaemic brain injury or brain death during ECMO. Two candidate signals were tested: cerebral desaturation (consecutive hours of regional saturation under 40 percent) and asymmetry between the two sides, expressed both as the mean left-right difference across the whole run and as consecutive hours of asymmetry.

What was found. Eighteen patients (26 percent) had acute brain injury.

Mean asymmetry over 8 percent: adjusted odds ratio 39.4 (95 percent confidence interval 4.1 to 381.4) for acute brain injury.
Asymmetric desaturation — regional saturation under 40 percent together with asymmetry over 10 percent, for over 10 consecutive hours: adjusted odds ratio 5.2 (1.2 to 22.2).
Neither criterion alone was significantly associated with injury.
Operating characteristics: mean asymmetry over 8 percent gave sensitivity 39 percent, specificity 98 percent, area under the curve 0.86. Asymmetric desaturation gave sensitivity 33 percent, specificity 88 percent, area under the curve 0.72.

What it means, and this is the organising point of the section. The number clinicians watch — the absolute regional saturation, with its familiar alarm threshold — was not, by itself, associated with brain injury in this cohort. The signal was the difference between the two hemispheres. Near-infrared spectroscopy on venoarterial ECMO is a comparison instrument, and a single-sensor installation discards the thing that carries the information.

And the operating characteristics tell you exactly how to use it: 98 percent specificity with 39 percent sensitivity is the profile of a rule-in test. Sustained asymmetry over 8 percent should trigger imaging. A symmetrical, normal trace excludes almost nothing — it missed 61 percent of the injuries in this series.

Limitations, stated plainly. Single centre, 69 patients, 18 events. The confidence interval of 4.1 to 381.4 is a direct signal of sparse data, and the point estimate of 39.4 should not be quoted as though it were a stable effect size. Both thresholds (8 percent and 10 percent, over 10 hours) were derived in this dataset and have not been prospectively validated.

Certainty: moderate that asymmetry outperforms the absolute value; low for the specific thresholds.

Source: Zhao D, Shou BL, Caturegli G, Whitman GJR, Kim BS, et al. ASAIO J 2023;69(12):1083–1089. DOI 10.1097/MAT.0000000000002032.

⚠️

Danger · The three conflicting thresholds in current use, and what to do about them

Three different action thresholds for near-infrared spectroscopy appear in the sources consulted for this chapter, and they are not reconcilable. A unit that has not chosen between them does not have a protocol.

Threshold	Source	What it is built on
A drop of over 25 percent from baseline	ELSO Red Book, adult cardiac chapter	Expert synthesis; explicitly framed as trend rather than absolute
Absolute saturation under 50, or a 20 percent drop from baseline	A paediatric congenital heart disease review	Paediatric practice, transported
Mean left-right asymmetry over 8 percent; or saturation under 40 percent with asymmetry over 10 percent for over 10 hours	Zhao and colleagues, adult venoarterial	The only adult ECMO dataset in which thresholds were tested against adjudicated brain injury

These differ in kind, not only in value. Two are relative-change thresholds, one is absolute, and one is a between-hemisphere comparison that the other two do not consider at all. They will not fire at the same time and will not fire on the same patients.

This book's position: in the adult venoarterial patient, use bilateral sensors and treat sustained asymmetry as the primary alarm, with the absolute value and the relative drop as secondary. Record which threshold your unit has adopted, and record that it is not validated.

💡

Clinical pearl · What to do when the cerebral saturation falls — the differential, in order

A usable action list, adapted from the paediatric review cited above and cross-checked against the 2024 ELSO consensus and this book's earlier chapters. Work through it before concluding that the brain is injured, because most causes of a falling cerebral saturation are not in the brain.

Check the sensor. Displacement, a haematoma or oedema beneath it, and ambient light are the commonest causes of a new abnormal value. Look before you act.
Differential hypoxaemia on peripheral venoarterial support — draw a right radial gas (Chapter 16, Chapter 27). A unilateral right-sided fall is this until proven otherwise.
Inadvertent hyperventilation or an abrupt fall in carbon dioxide tension — cerebral vasoconstriction. Check the blood gas and the sweep gas (§28.13).
Cannula malposition altering flow to the arch, or a fall in circuit flow.
Falling oxygen delivery for non-cerebral reasons — anaemia, falling venous saturation, falling cardiac output. Check haemoglobin and the drainage-limb saturation (Chapter 27).
Rising cerebral metabolic demand — subclinical seizures, hyperthermia, inadequate analgesia and sedation.
Only then: a new intracranial event. Which means a computed tomogram, not another monitor.

The order matters. Steps 1 to 6 are fixable at the bedside in minutes; step 7 requires a transfer that carries its own risk.

🚩

Pitfall · Treating near-infrared spectroscopy as a standard of care

It is the most deployed cerebral monitor on ECMO, and deployment is not the same as validation. A 2022 monitoring review states the position accurately: surrogates of tissue perfusion and near-infrared spectroscopy "are not standards of care but may provide useful information in select patients."

The Red Book's complications chapter reports the one adult operating-characteristic figure available from a prospective cohort — sensitivity 86 percent and specificity 55 percent for early detection of acute brain injury — which is a very different profile from the 39 percent sensitivity and 98 percent specificity found for asymmetry by Zhao and colleagues. These two sets of numbers are not compatible and this chapter cannot reconcile them. They may be measuring different signals (absolute versus asymmetry), in different populations, against differently adjudicated outcomes. Both are reproduced; neither should be quoted as the performance of "NIRS."

Sources: Castro DM, Morris I, Teijeiro-Paradis R, Fan E. Curr Opin Crit Care 2022;28(3):348–359. DOI 10.1097/MCC.0000000000000939. The 86 percent and 55 percent figures are attributed in the ELSO Red Book to a prospective cohort study by Hunt and colleagues; the primary paper was not retrieved. [VERIFICATION REQUIRED: Hunt et al., full citation and the cohort in which these operating characteristics were derived.]

28.7 Cerebral autoregulation: the index that turns blood pressure into a personal number

This is the newest of the instruments here, and the one most likely to change practice over the life of this book, because it addresses a question the guidelines admit they cannot answer: what mean arterial pressure does this particular brain need?

🧠

Physiology · Autoregulation under continuous flow

The intact cerebral circulation holds blood flow roughly constant across a range of perfusion pressures. Two things happen to that mechanism on ECMO.

Prolonged hypercapnia before cannulation — the rule rather than the exception in the venovenous patient — dilates the cerebral bed and impairs autoregulation, producing high cerebral blood flow and, in the 2024 ELSO consensus's phrase, "a narrow regulatory pressure window."
Continuous, non-pulsatile flow itself is associated with impaired autoregulation. The venoarterial stroke cohort states it directly: cerebral autoregulation is impaired in patients with non-pulsatile flow, and this contributes to brain injury.

When autoregulation fails, cerebral blood flow becomes pressure-passive — it tracks mean arterial pressure linearly. At that point the brain has no defence against either hypotension or hypertension, and the choice of a blood pressure target stops being a haemodynamic convenience and becomes a neurological intervention.

And pressure-passive does not mean pressure-sufficient. Chapter 25 handed this chapter a finding that belongs here: in a porcine model, a carotid flow deficit persisted after mean arterial pressure had been restored. Read alongside the cerebral oximetry index data, it makes a single point with two independent sources of support — restoring the blood pressure number does not demonstrate that cerebral flow has been restored, and on a circuit there is no reason to assume the two move together.

The cerebral oximetry index (COx) makes this measurable at the bedside. It is a moving correlation coefficient between mean arterial pressure and the slow waves of regional cerebral oxygen saturation from near-infrared spectroscopy. A value approaching 1 means the two move together — pressure-passive flow, impaired autoregulation. A value near 0 means flow is defended. Plotting the index against pressure identifies an optimal mean arterial pressure and the individual limits of autoregulation.

🔬

Evidence · Feasibility and what it showed

What was done. A single-centre prospective observational study of continuous autoregulation monitoring in adult ECMO patients: 15 patients (median age 57, interquartile range 47 to 69), 11 venoarterial and 4 venovenous, yielding 150 autoregulation measurements. Optimal mean arterial pressure and the individual limits of autoregulation were derived from the cerebral oximetry index and examined against modified Rankin Scale scores at 3 and 6 months.

What was found.

Autoregulation was impaired on day 1 and improved by day 2: mean index 0.2 on post-cannulation day 1 versus 0.09 on day 2, p under 0.01.
It was worse on venoarterial than venovenous support: 0.12 versus 0.06, p equal to 0.04.
Optimal mean arterial pressure was highly variable across the cohort, ranging from 55 to 110 mmHg.
Time spent with the observed pressure outside the limits of autoregulation was higher on day 1 than day 2 (18.2 versus 3.3 percent, p under 0.01).
Patients with a good functional outcome (modified Rankin Scale 0 to 3) spent less time outside optimal pressure than those with poor outcome (74 versus 82 percent, p equal to 0.01).

What it means. Two findings are worth carrying to the bedside even from 15 patients.

The optimal pressure for an individual brain ranged from 55 to 110 mmHg. There is no population mean arterial pressure target that serves this group. A protocol that targets 65 mmHg for everyone is, in roughly half of these patients, targeting the wrong number — and the direction of the error is unknowable without measuring.
Day 1 is the vulnerable day. Impairment was worse and time outside limits was five times greater on day 1 than day 2. This aligns with every other finding in this chapter about the peri-cannulation window.

Limitations. Fifteen patients. The outcome comparison — 74 versus 82 percent of time outside optimal pressure — is a narrow separation between two high numbers, which is as much a comment on how poorly pressure was controlled in everyone as on the difference between outcome groups. Feasibility was the stated primary aim and feasibility is what was demonstrated. No interventional study has tested whether targeting an individualised pressure changes anything.

Certainty: low. Reported here because it names a problem the guidelines concede they cannot solve, not because it solves it.

Source: Zhang LQ, Chang H, Kalra A, Humayun M, Rosenblatt KR, Shah VA, Geocadin RG, Brown CH, Kim BS, Whitman GJR, Rivera-Lara L, Cho SM. Neurocrit Care 2024;41(1):185–193. DOI 10.1007/s12028-023-01932-w. An earlier preprint of the same work carries DOI 10.21203/rs.3.rs-3300834/v1; the peer-reviewed version is cited here.

The 2024 ELSO consensus reaches the same conclusion from the guideline side: "individualized BP management tailored to dynamic cerebral autoregulation function is likely needed in this complex population", while stating that evidence and related therapeutic actions remain limited and are a mandatory objective for future research.

28.8 Transcranial Doppler: three things it measures, and only one of them is settled

Transcranial Doppler is the only bedside technique that images cerebral blood flow in real time, and the only one that can detect emboli as they pass. It is used on ECMO for three separate purposes, and they have very different evidential standing.

Use	What it measures	Standing on ECMO
Flow velocities	Mean, systolic and diastolic velocity in the middle cerebral and other basal arteries	Descriptive. Feasible; interpretation confounded by the flow pattern (below)
Pulsatility index	The shape of the waveform, conventionally a surrogate for distal vascular resistance and, indirectly, intracranial pressure	Confounded by design on venoarterial support (below). The most misread number in this section
Microembolic signals	High-intensity transient signals from solid or gaseous emboli crossing the insonated vessel	Unresolved. Three studies, three different conclusions — §28.8 controversy box

⚠️

Danger · Continuous flow destroys the meaning of the pulsatility index

This is the transcranial Doppler equivalent of Chapter 25's central warning about arterial waveform analysis, and the Red Book's bedside ultrasound chapter states it explicitly: "continuous flow confounds traditional measurements of pulsatility on VA ECMO." In adults, as in children, overall pulsatility is diminished by the proportion of cardiac output captured by the continuous flow.

Three specific failure modes follow.

A low pulsatility index on venoarterial support is a circuit finding, not a cerebral one. It tells you how much of the cardiac output the pump is carrying. Reading it as low distal resistance, or as reassurance about intracranial pressure, is a category error.
Vasospasm may be masked. The Red Book: arteriospasm may be hidden in situations where the heart is contributing little to cerebral blood flow.
The direction that may carry signal is the opposite one. Some series indicate that a rise in pulsatility index, not otherwise explained by improving cardiac function, may indicate acute neurological injury — and the Red Book is careful to call this an area for investigation rather than a rule.

The bedside consequence: a pulsatility index on ECMO is interpretable only alongside the circuit flow and the degree of native ejection at the time it was measured — which is the same rule Chapter 26 established for ejection fraction and Chapter 25 for every haemodynamic instrument. Record the flow with the velocity, every time.

The neuromonitoring review adds two observations from one centre's own practice that illustrate how unlike normal these tracings are: low mean middle cerebral artery velocities with a normal pulsatility index on venovenous support, and a low pulsatility index on venoarterial support. Source: Kazmi SO, Sivakumar S, Karakitsos D, Alharthy A, Lazaridis C, Leenen L. Crit Care Res Pract 2018;2018. DOI 10.1155/2018/3237810. These are single-centre descriptive values and are not normal ranges.

⚖️

Controversy 1 · What does a microembolic signal mean on ECMO?

The clinical question. A transcranial Doppler study shows showers of high-intensity transient signals in a patient on venoarterial ECMO. Should anything change — anticoagulation, circuit inspection, imaging, the decision to continue?

Position A: microembolic signals mark embolic load and correlate with arterial thrombotic events. A prospective multimodal neuromonitoring study of 20 ECMO patients (median age 60; 17 had transcranial Doppler, 13 electroencephalography, 7 somatosensory evoked potentials) found acute neurological complications in 17 of 20 (85 percent) and unfavourable outcome in 14 (70 percent). Eighteen thromboembolic events were observed, 14 of which had positive microembolic signals; all 10 patients with arterial-sided thrombotic events had positive signals. Source: Cho SM, Ziai W, Mayasi Y, Gusdon AM, Creed J, Sharrock M, et al. ASAIO J 2020;66(4):388–393. DOI 10.1097/MAT.0000000000001013.

Position B: microembolic signals are common, configuration-dependent and clinically inert. A prospective observational study at a single referral centre performed 248 transcranial Doppler studies in 42 venovenous and 11 venoarterial patients over 18 months. Signals were present in 26.2 percent of venovenous and 81.8 percent of venoarterial patients (p under 0.001), with severe grading confined entirely to the venoarterial group (5 of 11). No correlation with extracorporeal flow rates or activated partial thromboplastin time in either configuration; an inverse correlation with left ventricular ejection fraction in venoarterial patients (p equal to 0.037); and no clinical neurological impairment correlated with signal detection at 6-month follow-up. Source: Marinoni M, Migliaccio ML, Trapani S, Bonizzoli M, Gucci L, Cianchi G, et al. Acta Anaesthesiol Scand 2016;60(7):934–944. DOI 10.1111/aas.12736.

Position C, from the same group as Position A, in a larger cohort: no association with brain injury at all. A prospective cohort of 145 adult ECMO patients undergoing transcranial Doppler with embolic monitoring — 100 venoarterial (187 examinations) and 45 venovenous (65 examinations). Signals in 35.0 percent of venoarterial and 4.7 percent of venovenous patients (p under 0.001). Signal presence was not associated with cannulation mode (23.4 percent peripheral versus 25.8 percent central), not associated with the presence of clot or fibrin in the circuit, and not associated with any studied haemodynamic, laboratory or ECMO parameter at the time of the study. Acute brain injury occurred in 38 percent of venoarterial and 31.1 percent of venovenous patients — and on multivariable analysis neither acute brain injury nor a composite of arterial thromboembolic events was associated with the presence or number of signals. Source: Caturegli G, Kapoor S, Ponomarev V, Kim BS, Whitman GJR, Ziai W, et al. JTCVS Tech 2022;15:111–122. DOI 10.1016/j.xjtc.2022.07.026.

Where the evidence actually sits. One finding is robust and replicated across all three studies: microembolic signals are several times more common on venoarterial than venovenous support (81.8 versus 26.2 percent; 35.0 versus 4.7 percent). That is a real, consistent, mechanistically sensible observation — arterial return delivers what the circuit generates directly into the systemic circulation.

Everything beyond that is unresolved, and the disagreement is not between rival groups but within one. Positions A and C come from the same institution, and the larger, later, better-powered study found the association absent. A fourth data point deepens it: an earlier six-patient report found signals correlated with high circuit flow rates over 4 L/min, which both Marinoni and Caturegli explicitly failed to reproduce — and Marinoni's discussion argues the earlier study's 15-minute recording was too short to evaluate signals properly under European guidelines, with an unspecified decibel threshold.

What this book concludes. A microembolic signal count on ECMO is currently a research measurement, not a clinical one. It does not have an established action threshold; it does not predict brain injury in the largest cohort; it does not track circuit clot, anticoagulation intensity or flow. Do not alter anticoagulation on the basis of a signal count.

What would resolve it. A multicentre prospective study with standardised recording duration and decibel threshold, blinded adjudication of acute brain injury, and signal counts analysed as a time-varying exposure rather than a per-patient dichotomy. Marinoni's own conclusion — that multicentre studies with more extensive cases are still required — remains the correct summary a decade later.

Certainty: very low for any clinical use; moderate for the venoarterial-versus-venovenous difference in prevalence.

🚩

Pitfall · Using transcranial Doppler to support brain death determination on a circuit

Chapter 24 owns brain death determination and this chapter does not re-argue it. One monitoring-specific trap belongs here because it is a property of the instrument.

The Red Book sets out the conventional agreement for pulsatile circulations: oscillating arterial flow, absence of diastolic flow with short systolic spikes, and absence of flow compared with a previously documented flow by the same operator are consistent with brain death. It then states the problem: "The absence of flow is potentially problematic on VA ECMO in case of poor cardiac ejection, as continuous cerebral blood flow may be difficult to detect."

In plain terms: on venoarterial support the finding that signifies cerebral circulatory arrest and the finding produced by a poorly ejecting heart under continuous flow can look the same. Transcranial Doppler is not a reliable ancillary test in this setting. The 2024 ELSO consensus routes around it — where the apnoea test is not feasible, cerebral angiography or a radionuclide brain scan is preferred.

28.9 Electroencephalography: high yield, low deployment

The electroencephalogram is the only instrument that detects a specific, treatable, and otherwise completely invisible event: the nonconvulsive seizure in a paralysed patient.

🔬

Evidence · What the electroencephalogram shows on adult ECMO

What was done. Retrospective analysis of all patients undergoing venovenous or venoarterial ECMO with a concomitant electroencephalographic recording at a single centre over almost ten years (April 2009 to December 2018), recorded either intermittently or continuously. Background was classified as mild or moderate encephalopathy (largely defined by the presence of reactivity), severe encephalopathy (largely the absence of reactivity), burst-suppression, or suppressed. Outcome was unfavourable if the Glasgow Outcome Scale was under 4 at 3 months.

What was found. 139 patients out of 596 met the inclusion criteria — median age 54 (41 to 62), 60 male (43 percent), 98 (71 percent) venoarterial, continuous recording in 113 (81 percent). Unfavourable outcome in 99 (71 percent).

Severe encephalopathy in 29 (21 percent); burst-suppression in 4 (3 percent); suppressed background in 19 (14 percent).
Seizures or status epilepticus in 11 (8 percent); generalised periodic discharges in 10 (7 percent).

What it means, and the denominator is the finding. Only 139 of 596 ECMO patients at this centre had an electroencephalogram at all — under a quarter. The 8 percent seizure figure is therefore a yield among the patients someone thought worth recording, not a population incidence. In a cohort selected for clinical concern, roughly one patient in twelve had seizures or status epilepticus, and roughly one in three had a severely abnormal or suppressed background.

Source: Peluso L, Rechichi S, Franchi F, Pozzebon S, Scolletta S, Brasseur A, et al. Crit Care 2020;24(1):629. DOI 10.1186/s13054-020-03353-z.

The wider range. A neuromonitoring review reports that the electroencephalogram is abnormal in 50 to 80 percent of ECMO patients with electrographic seizures reported at 8 to 20 percent, and notes that continuous recording is a scarce resource, so many institutions check periodic 20-minute to 1-hour studies instead. Source: Kazmi and colleagues, DOI 10.1155/2018/3237810. In paediatric extracorporeal support the Red Book reports nonconvulsive seizures including status epilepticus ranging from 17 to 23 percent with continuous monitoring, and notes that seizures on ECLS correlate with lower intelligence quotient, cerebral palsy and increased mortality.

Certainty: moderate that clinically significant electrographic seizures occur in a meaningful minority; low for any specific incidence figure, because every figure is conditioned on who was recorded.

💡

Clinical pearl · Most seizures on extracorporeal support have no clinical sign — and there is a number for it

A large paediatric and neonatal series of 450 ECMO patients at a single children's hospital over seven years, presented in abstract, found electrographic seizures in 17 percent (33 of 199) of those who had continuous recording, of which 67 percent (22 of 33) were electrographic only — that is, invisible at the bedside. Seizures were seen in the setting of focal injury in 85 percent. The authors' conclusion is the one to carry: the electroencephalogram is necessary to detect seizures on ECMO because the majority are electroencephalogram-only.

Two further observations from the same series. Background classification predicted mortality better than seizures did: mortality with versus without electrographic seizures was 55 versus 50 percent and not significant (p equal to 0.6), while 90 percent (27 of 30) of those with a severe background classification died. And the timing is usable: median time from cannulation to first seizure was around 2 days, well after the median time from cannulation to electroencephalogram of around 15 to 17 hours.

Source is a conference abstract presented at the 47th Annual Meeting of the Child Neurology Society, Ann Neurol 2018;84(S22):S281–S428. DOI 10.1002/ana.25305. Abstract-only, paediatric, and a single centre. [VERIFICATION REQUIRED: whether this series was subsequently published in full, and by whom.] It is reproduced because the two findings — that most seizures are subclinical, and that background beats seizures for prognosis — are consistent with every adult source consulted.

🚩

Pitfall · Requesting an electroencephalogram for an isolated change in vital signs

A large paediatric cardiovascular intensive care series looked at what actually prompts a study and what that prompt yields: 605 continuous studies in 411 patients, median duration 26 hours, seizures detected in 57 of 605 (9 percent) overall.

Broken down by indication: 0 of 104 studies (0 percent) requested for isolated vital-sign changes detected a seizure (p under 0.001), against 10 of 101 (10 percent) for non-vital-sign symptoms, 34 of 356 (10 percent) for risk factors alone, and 13 of 44 (30 percent) when vital-sign changes were combined with other symptoms — roughly five times the odds of a study requested on risk factors alone (odds ratio 4.93, 2.45 to 9.77). Within this series, venoarterial ECMO as an indication yielded seizures in 14 of 116 (12 percent), and extracorporeal cardiopulmonary resuscitation in 8 of 47 (17 percent).

The practical point: an unexplained tachycardia, a blood pressure swing or a desaturation, on its own, is a poor reason to record. Combined with gaze deviation, abnormal limb movements or a change in mental status, it is a strong one. Source: Segal JB, Yang JK, Silverman A, Darji H, He Z, Campen CJ. Epilepsia 2025;66(4):1187–1198. DOI 10.1111/epi.18253. Paediatric and neonatal cardiovascular intensive care, not adult ECMO — the indication logic transfers; the numbers do not.

💡

Clinical pearl · Reactivity is the variable to ask the neurophysiologist about

Across the adult sources, the electroencephalographic feature that separates outcome groups is not the presence of epileptiform activity but whether the background is reactive. The Peluso classification is built on it: severe encephalopathy is defined largely by the absence of reactivity. The paediatric series found background classification, not seizure burden, predicted death.

When you telephone for a report, the question that carries the most information is "is the background reactive?" — not "are there seizures?"

And then §28.16's warning applies immediately: absent reactivity on a circuit has been shown, in autopsy-verified cases, not to guarantee structural injury. It is a prognostic variable, not a diagnosis.

One further consequence, inherited from Chapter 24. In some jurisdictions an inconclusive electroencephalogram can now block a death determination outright. That turns an investigation ordered for one purpose — finding treatable seizures — into a document with a second, unintended legal life. It is not a reason to avoid recording. It is a reason to record with a clear clinical question stated on the request, to ensure the report addresses that question, and to know your own jurisdiction's rule before the situation arises rather than during it. Chapter 24 owns the determination itself.

28.10 Serum biomarkers: the haemolysis problem

⚠️

Danger · Neuron-specific enolase is confounded by the circuit itself, and the size of the confounding is unknown

Neuron-specific enolase is the most used serum marker of neuronal injury and is part of the standard multimodal prognostication set after cardiac arrest. On ECMO it has a specific, mechanical problem: it is released by red cells as well as neurons, and ECMO haemolyses.

The Red Book's complications chapter notes that some have advocated regular enolase measurement as an adjunct in timing neuroimaging, "although ongoing hemolysis in ECMO may confound the interpretation." The 2024 ELSO consensus is more specific and more useful: "Neuron-specific enolase values are often higher in ECMO patients due to ongoing hemolysis. The most accurate neuron-specific enolase threshold for predicting an unfavourable neurological outcome in ECPR remains unknown, possibly exceeding 100 μg/L."

Set that against the general post-arrest threshold, which the same document reproduces: over 60 μg/L at 48 or 72 hours. The ECMO threshold may be above 100 — that is, the standard threshold may be exceeded by a substantial margin in a patient whose brain is intact, purely because of the circuit.

Three bedside rules.

Never act on an enolase value on ECMO without the concurrent haemolysis indices — plasma free haemoglobin, lactate dehydrogenase, haptoglobin (Chapter 37 owns haemolysis; Chapter 27 owns the carboxyhaemoglobin line as an additional haemolysis signal).
A single value is close to uninterpretable. The rising or falling trend, read against the haemolysis trend, carries what signal there is.
Do not apply the 60 μg/L threshold to an ECMO patient. The guideline says the correct number is unknown and probably higher. An unknown threshold cannot support a withdrawal decision.

Provenance of this obligation. Chapter 23 flagged the enolase-plus-haemolysis-index requirement and handed it to Chapter 29 (laboratory monitoring) and Chapter 37 (haemolysis), not to this chapter. This chapter supplies the number those chapters will need and hands it on: the number is that there is no validated number, and the 2024 ELSO consensus estimate is "possibly exceeding 100 μg/L". Chapters 29 and 37 retain ownership of the assay and the haemolysis indices; this chapter owns only the statement that the threshold is unvalidated on a circuit.

And Chapter 23 named the study that would settle it, which still does not exist: a trial measuring neuron-specific enolase and free haemoglobin simultaneously and serially in ECMO patients, against a neurological endpoint. Until it is done, the enolase threshold on a circuit is borrowed rather than derived — Chapter 23's phrasing, reproduced because it is exact.

Other markers. The neuromonitoring review lists glial fibrillary acidic protein, S100b, neuron-specific enolase, ICAM5, brain-derived neurotrophic factor and MCP-1/CCL2 as candidates, and notes debate over optimal cut-offs and over whether serial assessment adds meaning. The 2024 ELSO consensus states that data are sparse on ECMO for neurofilament light chain and tau. None has an established role on ECMO and none should be ordered expecting to act on it.

28.11 Somatosensory evoked potentials: the preserved N20 that is not reassurance

Bilateral absence of the N20 cortical response is one of the sturdiest predictors of poor outcome after cardiac arrest in patients not on a circuit, and it appears in the standard multimodal set the 2024 ELSO consensus reproduces (bilateral absence at 24 hours or more).

🔬

Evidence · What happened when evoked potentials and electroencephalography were recorded in the same patients

In the prospective multimodal neuromonitoring cohort of 20 ECMO patients, 7 patients had both somatosensory evoked potentials and electroencephalography. All 7 had intact N20 responses despite poor outcomes. Four of these 7 showed absent electroencephalographic reactivity despite the intact N20.

Two things follow, and they are opposite in direction.

A present N20 on ECMO carries very little reassurance. In the only series that looked, it was present in every patient who did badly.
The two electrophysiological tests disagreed in more than half of the patients in whom both were done. That is not a failure of either test; N20 tests the integrity of a specific somatosensory pathway, and electroencephalographic reactivity tests cortical responsiveness. But it means they cannot be summed into a single "electrophysiology" impression, and the multimodal prognostication rule of "at least two indicators of severe injury" becomes much harder to apply when the indicators routinely conflict.

The same study is the source of the broader feasibility claim: multimodal neuromonitoring caused no adverse effects, found neurological complications in 85 percent of patients, and the unfavourable outcome group was characterised by absent electroencephalographic reactivity, coma, central cannulation, higher transfusion requirement, and higher APACHE II and SOFA scores.

Certainty: very low — 7 patients. Reported because it is the only direct head-to-head observation available and because it argues against a specific false reassurance.

Source: Cho SM, Ziai W, Mayasi Y, Gusdon AM, Creed J, Sharrock M, et al. ASAIO J 2020;66(4):388–393. DOI 10.1097/MAT.0000000000001013.

28.12 Neuroimaging: one scanner, and it is the wrong one

Modality	Availability on ECMO	What it answers	What it misses
Non-contrast computed tomography of the head	Available, but requires intrahospital transport of a cannulated patient	Intracranial haemorrhage — the question that must be answered before anything else. Established large infarct; mass effect; hydrocephalus	Early ischaemia. The venoarterial stroke cohort states plainly that the sensitivity of computed tomography to detect brain ischaemia is limited
Computed tomographic angiography, with perfusion	Same transport requirement	Large vessel occlusion, and salvageable penumbra — the finding that makes thrombectomy possible (§28.15)	Small-vessel and embolic showers
Magnetic resonance imaging	Not possible during ECMO. Magnetic-resonance-compatible consoles and consumables do not exist	Would answer almost everything	Not applicable — it is unavailable
Cranial ultrasound	Only through an open fontanelle. Not applicable to adults	Fast, portable, no radiation; intraventricular and large parenchymal haemorrhage in neonates	Poor sensitivity even where usable: in one autopsy and imaging study, 19 of 74 patients had neurological or magnetic resonance sequelae despite a normal head ultrasound
Low-field portable magnetic resonance imaging	Emerging. The venoarterial stroke cohort reports it has recently been used safely in intensive care for ECMO patients	Potentially the missing test, at the bedside	Not established practice. [VERIFICATION REQUIRED: primary studies of portable low-field magnetic resonance imaging in ECMO patients, safety data and field-strength limits]

⚠️

Danger · The transport decision is itself a clinical decision with a risk

The Red Book is explicit that computed tomography on ECMO "requires intrahospital transport of a patient on ECLS, with all inherent risks and challenges", and that its use is in general limited to a clinical indication. The venoarterial stroke cohort says the same from the other side: "Doing a brain CT scan is sometimes difficult due to risky transfer of critically ill patients, and the sensitivity of CT scan to detect brain ischemia is limited."

So the decision to scan trades a real, immediate risk against a test with limited sensitivity for the commonest injury. That is an uncomfortable trade and it is why the low threshold recommended by every source must be paired with a clear question.

The question that justifies transport is almost always "is there blood?" Because the answer changes anticoagulation within the hour, and because computed tomography is genuinely good at that question. "Is there an infarct?" is a much weaker indication for transport, because a negative scan will not exclude one.

The 2024 ELSO consensus sets the standard plainly: non-contrast head computed tomography is imperative to rule out intracranial haemorrhage with any acute neurological examination change, and computed tomographic angiography is needed to assess for large vessel occlusion.

💡

Clinical pearl · Consider a pre-cannulation scan when the clock allows it

The Red Book notes that some centres perform cranial computed tomography before initiating extracorporeal support. In the transcranial Doppler series, a brain computed tomogram was performed at admission in the majority of patients and repeated when required — but in venoarterial patients it was not always possible to obtain one at admission.

That asymmetry is the practical teaching. The venovenous patient is usually being cannulated semi-electively and often can be scanned first; the venoarterial or extracorporeal cardiopulmonary resuscitation patient usually cannot. Which means the group with the highest risk of pre-existing brain injury is the group least likely to have a baseline image — and every subsequent scan in that patient will be read without one.

If the patient is stable enough for the scanner before cannulation, the image is worth more then than at any later point in the run.

28.13 The peri-cannulation window: the part that is actually modifiable

Sections 28.4 to 28.12 are about detecting injury. This section is about the small number of things that plausibly prevent it, and they are all concentrated in the same 24 hours. Every quantitative finding in this section comes from the 2024 ELSO consensus guidelines and the primary studies it cites.

🧠

Physiology · Why the first hours are different

At the moment the pump starts, a specific and unusual physiological transition occurs, and it occurs faster than anything in ordinary intensive care.

A patient who has been profoundly hypercapnic and acidotic for hours or days is corrected within minutes, by turning a sweep gas flowmeter.
A patient who has been severely hypoxaemic is, in the same minutes, at risk of hyperoxaemia.
A cerebral vascular bed that has been maximally dilated by chronic hypercapnia, with autoregulation impaired and, per §28.7, at its worst on day 1, meets both of those changes at once.

Carbon dioxide is a potent cerebral vasodilator that increases cerebral blood flow and neuronal metabolic demand. The 2024 ELSO consensus sets out the sequence: prolonged hypercapnia may impair cerebral autoregulation, producing high cerebral blood flow and a narrow regulatory pressure window; rapid correction of sustained high carbon dioxide soon after initiation sometimes leads to rapid hypocapnia, which may cause cerebral vasoconstriction and a decrease in cerebral oxygen delivery, resulting in cerebral ischaemia. And then the sentence that completes the mechanism: "Routine use of full-dose anticoagulation therapy at ECMO initiation and thereafter may cause hemorrhagic conversion of an ischemic injury."

That is a complete pathway from a flowmeter to an intracranial haemorrhage, and every step of it happens in the first day.

🔬

Evidence · The rate of carbon dioxide correction — a number to work to

This closes the obligation carried forward from Chapter 27, which owned the correction-rate argument for one patient and owed this chapter its general form.

The recommendation. The 2024 ELSO consensus: "avoiding a large ΔPaCO2 over 50 percent in the peri-cannulation period for both VA and VV ECMO is reasonable."

The evidence behind it, and it is not uniform.

Venovenous. An ELSO registry analysis found a rapid early decrease in carbon dioxide tension independently associated with increased risk of intracranial haemorrhage in patients with acute respiratory distress syndrome on venovenous support. A second ELSO retrospective study of 11,972 venovenous ECMO patients showed that those with a peri-cannulation carbon dioxide fall of over 50 percent were more likely to experience acute brain injury — both infarct and haemorrhage.
Venoarterial, single centre, granular data. A retrospective study of 129 venoarterial patients at a centre with standardised neuromonitoring (median age 60, 63 percent male), using the mean of all carbon dioxide values in the 12 hours before and after cannulation — a median of 11 values per patient (interquartile range 8 to 14). Acute brain injury in 43 (33 percent). Pre-cannulation (39 versus 42 mmHg, p equal to 0.38) and post-cannulation (37 versus 36 mmHg, p equal to 0.82) values did not differ between those with and without brain injury. But higher pre-cannulation carbon dioxide (odds ratio 2.10, 1.10 to 4.00, p equal to 0.02) and a larger fall (odds ratio 2.69, 1.18 to 6.13, p equal to 0.02) were associated with intracranial haemorrhage specifically. Source: Shou BL, Ong CS, Zhou AL, Al-Kawaz MN, Etchill E, et al. ASAIO J 2022;68(12):1501–1507. DOI 10.1097/MAT.0000000000001699.
The disconfirming study. An ELSO retrospective study of 3,125 extracorporeal cardiopulmonary resuscitation patients found the carbon dioxide fall was higher in those with brain injury than without, but was not significantly associated with acute brain injury.

How to read the disagreement. Note what changes between the studies that find an effect and the one that does not: the outcome. The venoarterial single-centre study found the association with intracranial haemorrhage, not with brain injury overall. The extracorporeal cardiopulmonary resuscitation study looked at brain injury overall — in a population where the dominant injury is hypoxic-ischaemic damage sustained before cannulation, which no post-cannulation gas management could have prevented. A real effect on haemorrhage can easily be diluted to nothing inside a composite dominated by pre-existing anoxic injury. That is a coherent reading, but it is this book's reading and not a finding.

The consensus is candid about why certainty is low: a lack of sensitive, reliable and readily available diagnostic markers of brain injury; retrospective observation; and inconsistent arterial blood gas sampling.

Certainty: moderate that a large, rapid fall in carbon dioxide is harmful, strongest for intracranial haemorrhage in venovenous patients; low for the 50 percent figure as a precise threshold.

⚠️

Contradiction found during this chapter's audit — between two chapters of this book

Chapter 22 §22.3 already owns the rate of carbon dioxide correction after extracorporeal cardiopulmonary resuscitation, and states the physiology with numbers this chapter does not repeat: between 20 and 60 mmHg, 1 mmHg changes cerebral blood flow by 2 to 3 percent, beginning within 2 minutes and equilibrating within 12; three ELSO Registry populations (11,972 adults, 7,270 children, 3,583 neonates) associate a large early fall with neurological complications; the two best-measured human studies are null; and a porcine extracorporeal resuscitation experiment in which sweep gas was set as a fraction of blood flow found that rapid correction impaired cerebral autoregulation at every phase and produced histological neuronal injury.

Two things must be recorded rather than smoothed over.

First, an apparent conflict that is really a refinement. Chapter 22 reports that the best-measured human studies are null. This chapter's granular venoarterial cohort is null for acute brain injury overall and positive for intracranial haemorrhage specifically (odds ratio 2.69 for the fall; 2.10 for the pre-cannulation value). These are the same finding stated at two levels of resolution, and the 2024 ELSO consensus reaches the same shape: the signal is strongest for haemorrhage, weakest for brain injury as a composite. Chapter 22's null is correct for the endpoint it names.

Second, a real conflict in the bedside rule. Chapter 22's operational instruction is to set the initial sweep low, leave it ten minutes, then titrate toward a carbon dioxide tension of about 40 mmHg against right-radial gases. This chapter's rule, taken from the 2024 ELSO consensus, is to avoid a fall of over 50 percent in the peri-cannulation period. In a patient whose pre-cannulation carbon dioxide tension is 100 mmHg, those two instructions cannot both be followed — reaching 40 mmHg is a 60 percent fall.

The resolution this book adopts: the percentage rule binds, and the absolute target is the destination rather than the first stop. In that patient, the peri-cannulation floor is roughly 50 mmHg, and 40 mmHg is reached later, over hours. Chapter 22 should be amended to state the ceiling on the fall alongside its 40 mmHg target. This correction is recorded in the tracker and in the book status.

💡

Clinical pearl · How to make the sweep gas obey a carbon dioxide target

The recommendation is easy to state and easy to violate in the first ten minutes of a cannulation, when the instinct is to normalise the gas. A practical translation, building on Chapter 27's levers:

Record the pre-cannulation carbon dioxide tension before the pump starts. You cannot honour a 50 percent rule without the denominator. In the study above, the pre-cannulation value was the mean of everything in the preceding 12 hours — a single spot value is a weaker anchor but is better than none.
Calculate the floor. If the pre-cannulation value is 90 mmHg, do not go below roughly 45 mmHg in the peri-cannulation period, whatever the pH suggests.
Start the sweep low and climb. Chapter 27's rule — the sweep is the on-off switch — cuts both ways: a sweep opened wide on a patient with a carbon dioxide of 100 mmHg will overshoot, and the overshoot is the harm.
Titrate the sweep to the pH and to the rate of change, not to the carbon dioxide alone, and recheck within 30 to 60 minutes rather than at the next routine gas.
Write the target on the chart, as a number, with the pre-cannulation value beside it. This is a handover-critical piece of information and it is the one most often lost at the change of shift immediately after cannulation.

Physiological variable	2024 ELSO consensus position	The evidence behind it	Certainty
Carbon dioxide fall after cannulation	Avoid a fall of over 50 percent in the peri-cannulation period, both configurations	ELSO registry analyses in venovenous patients; a single-centre venoarterial cohort with granular gases; one large extracorporeal cardiopulmonary resuscitation study finding no significant association	Moderate for the direction; low for the threshold
Hyperoxaemia, venoarterial — Chapter 19 §19.5 owns the reperfusion case and has quantified it: severe hyperoxia carried an adjusted odds ratio of 1.59 for acute brain injury and 1.58 for death in the same 3,125 ELSO Registry extracorporeal cardiopulmonary resuscitation patients whose carbon dioxide analysis appears in the row above	Avoid severe hyperoxaemia (oxygen tension over 300 mmHg) within 24 hours of cannulation. Definitions given: mild hyperoxaemia over 100 or 120 mmHg; severe over 300 mmHg; hypoxaemia under 60 or 70 mmHg	Associations with brain injury and poor neurological outcome in observational data; optimal targets explicitly unknown	Low to moderate
Hypoxaemia, venovenous	Oxygen tension under 70 mmHg was associated with brain injury, especially intracranial haemorrhage, in a single-centre observational cohort. No data on hyperoxaemia in venovenous support, as it is rarely a clinical issue	Single-centre observational	Low
Pulse pressure, venoarterial	Pulse pressure under 20 mmHg in the first 24 hours was associated with acute brain injury. But the data on improving pulse pressure with inotropes or left ventricular venting are described as weak	Observational association; no interventional evidence	Low. An association without a demonstrated intervention
Blood pressure, no brain injury	No data exist on early and optimal blood pressure goals for preventing brain injury. Individualised management tailored to autoregulation is "likely needed"; avoid hypotension	§28.7; the optimal pressure ranged 55 to 110 mmHg across 15 patients	Very low. The guideline states the absence of evidence directly
Blood pressure, established ischaemic stroke	Permissive hypertension (up to 220/120 mmHg) per national stroke guidance is reasonable if the heart tolerates it — allowing the patient to autoregulate	Extrapolated from non-ECMO stroke guidance	Low. Extrapolation, with an explicit ECMO-specific caveat below
Blood pressure, established intracranial haemorrhage	Lower: systolic under 140 mmHg and mean under 90 mmHg, because the haemorrhage is anticoagulation-associated	Extrapolated from non-ECMO haemorrhage guidance	Low
Temperature	Unresolved. Hypothermia during ischaemia prolongs organ tolerance and the preclinical case is strong; one randomised controlled trial in venoarterial patients with cardiogenic shock comparing 33 to 34 °C with 36 to 37 °C found no mortality difference at 30 days	A meta-analysis of 2,643 extracorporeal cardiopulmonary resuscitation patients across 35 studies is described as severely heterogeneous, low-quality evidence. No data at all in venovenous support	Very low. See the controversy in §28.17

⚠️

Danger · The blood pressure target on venoarterial ECMO is a direct conflict between two organs, and the guideline says so

Almost every blood pressure recommendation in critical care can be followed without harming something else. On venoarterial ECMO with an established ischaemic stroke, it cannot, and the 2024 ELSO consensus states the conflict in one sentence: "Higher BPs lead to increased afterload, which may hinder myocardial recovery (VA ECMO only), particularly when the left ventricle is not vented."

So the brain wants permissive hypertension and the recovering ventricle wants the afterload down, and there is no evidence that resolves it. The consensus's position — allow the patient to autoregulate "if the heart can tolerate it" — is an honest statement that the decision has been handed back to the bedside.

What to do with that. Make the trade explicit and revisit it daily. Record in the notes which organ you are currently prioritising and why, so that the next team knows a choice was made rather than inheriting a number. And note that left ventricular venting changes the terms of the trade (Chapter 15 owns left ventricular distension and unloading) — a vented ventricle tolerates a higher pressure, which is an argument that the venting decision and the neurological blood pressure target should be made together rather than separately.

💡

Clinical pearl · The oxygen blender is the hyperoxaemia knob even though it is not the hypoxaemia knob

Chapter 27 established, firmly, that raising the sweep gas oxygen fraction does essentially nothing for a hypoxaemic patient, because a working membrane already saturates the outlet blood. It does not follow that the blender is useless — and the 2024 ELSO consensus uses it for precisely the opposite task, recommending that early severe hyperoxaemia and hypoxaemia be avoided "by manipulating the fraction of delivered oxygen from the ECMO sweep gas."

The asymmetry is real and worth holding as a rule: turning the blender down always works; turning it up usually does not. Lowering the delivered oxygen fraction reliably lowers the post-membrane oxygen content, because the membrane will only deliver what it is given. Raising it adds nothing once the outlet blood is saturated.

Practical version: if the arterial oxygen tension is over 300 mmHg in the first 24 hours of venoarterial support, turn the blender down. If the patient is hypoxaemic, the blender is the wrong control and Chapter 9's differential is the right response.

28.14 Assembling a protocol

There is no minimum standard. Two of this book's primary sources say so in as many words. The ISCCM manual: "There is no guideline for the minimum neurological monitoring standard during ECMO support." The neuromonitoring review: "the optimal neuromonitoring protocol has not been well established with current practices dependent on physician preference or device availability." The Red Book adds that although standardised neuromonitoring on ECLS may improve outcomes, there is no widely accepted standard.

What exists instead is a survey of what centres actually do, and it is worth knowing because it tells you what your colleagues elsewhere will and will not have.

Modality	Proportion of centres using it	Population and source
Any neuromonitoring beyond the examination	79 percent	EuroELSO Neurologic Monitoring and Outcome Working Group survey, reported in the Red Book's neonatal chapter
Near-infrared spectroscopy	66 percent in that survey; 48 percent of adult and 80 percent of paediatric centres in the European centres survey	As above; Cvetkovic M, Chiarini G, Belliato M, et al. Perfusion 2021 (cited in the Red Book)
Intermittent electroencephalography	35 percent	EuroELSO survey
Transcranial Doppler	29 percent	EuroELSO survey
Brain biomarkers	25 percent	EuroELSO survey
Cranial ultrasound, paediatric centres	73 percent routinely; frequency daily 25 percent, twice weekly 20 percent, three times weekly 17 percent, only on concern 15 percent	European centres survey, paediatric
Would obtain brain computed tomography if clinically concerned	74 percent	European centres survey
Electroencephalography routinely applied in extracorporeal cardiopulmonary resuscitation institutions	13.9 percent	Practice survey, carried forward from Chapter 22 and recorded by Chapter 23

💡

Clinical pearl · A defensible default protocol, built from what this chapter found

No source endorses a specific schedule, so this is this book's synthesis, offered as a starting point to be adapted and audited, not as a guideline. It is built on three principles established above: the examination is the primary monitor and sedation strategy determines whether it works; the instruments are rule-in tests; and day 1 is the vulnerable day.

Before cannulation, if the clock allows

Neurological examination, recorded. Note explicitly if it was not obtainable.
Non-contrast head computed tomography in the semi-elective venovenous patient.
Record the carbon dioxide tension and the blood pressure. These are the denominators for §28.13.

Immediately after cannulation

Repeat examination, per the consensus.
Stop sedation to assess neurological status once the patient is stable on the circuit, per the ISCCM manual, unless there is a specific reason not to.
Bilateral near-infrared spectroscopy sensors from this point, so an asymmetry baseline exists (§28.6).
Set and chart the carbon dioxide floor and the delivered oxygen fraction ceiling.

First 24 hours — the high-risk window

Nursing neurological assessment hourly after extracorporeal cardiopulmonary resuscitation or with any risk factor; up to 4-hourly in a stable venovenous patient.
Quantitative pupillometry 12-hourly, with the results not acted upon before 24 hours (§28.5).
Blood gases frequently enough to honour the carbon dioxide and oxygen targets — every 30 to 60 minutes initially, not at routine intervals.
Right radial arterial sampling on peripheral venoarterial support (Chapters 16 and 27).

Day 2 onward

Daily neurological assessment by a neurologist or neurointensivist where available.
Daily sedation interruption, treated as a monitoring procedure, with the examination performed during it and documented.
Continue pupillometry and bilateral near-infrared spectroscopy trends.
Electroencephalography on indication, not routinely — and the indications that yield are unexplained coma off sedation, any focal or abnormal movement, gaze deviation, or a vital-sign change combined with another sign (§28.9). Prefer continuous recording; accept a 20-minute to 1-hour study if that is what exists.
Consider baseline and interval neuron-specific enolase only if haemolysis indices are drawn with it (§28.10).

At any new neurological finding

Non-contrast head computed tomography, to answer the question "is there blood?" Add angiography if a focal deficit suggests large vessel occlusion.

Before discharge

Magnetic resonance imaging once off the circuit. Long-term series find infarction or haemorrhage in 37 to 52 percent of adult survivors — most of it not diagnosed during the run.
Modified Rankin Scale at discharge and at every follow-up, per the 2024 ELSO consensus, with structured long-term follow-up including neurological and psychiatric assessment.

🚩

Pitfall · Buying the monitor without buying the response

The commonest failure of a neuromonitoring programme is not a missing device. It is a device generating a signal that nobody has agreed how to act on.

Before installing any instrument in this chapter, write down three things. What value triggers a response; what that response is; and who is authorised to initiate it at three in the morning. A near-infrared spectroscopy monitor with no agreed asymmetry threshold, no documented escalation pathway and no out-of-hours neurology contact is not monitoring. It produces a number in the chart that will, in retrospect, be shown to have been abnormal for eight hours.

28.15 When something changes: the escalation pathway

⚠️

Recorded as a stopgap — see the scope note at the head of this chapter

The material in this section is acute management, not monitoring, and it does not properly belong to a monitoring chapter. It is reproduced here because the scope sweep found that no chapter in this book's 92-chapter map owns neurological complications as complications, and because a detection chapter that stops at detection would leave the reader with a diagnosis and no next step.

It is a faithful summary of the 2024 ELSO consensus recommendations and nothing more. It is not this book's synthesis, it is not complete, and it must be replaced. When Part VI is drafted, this section should be reduced to a cross-reference.

The first decision is always the same. The consensus is unambiguous: non-contrast head computed tomography is imperative to rule out intracranial haemorrhage with any acute neurological examination change. Everything downstream branches on that result.

Finding	2024 ELSO consensus position	The ECMO-specific complication
Acute ischaemic stroke — thrombolysis	Intravenous alteplase is generally not indicated in ECMO patients. The expert consensus is to avoid it	High bleeding risk on systemic anticoagulation with platelet dysfunction. This is one of the clearest single statements in the whole document
Acute ischaemic stroke — thrombectomy	Mechanical thrombectomy should be pursued for large vessel occlusion detected by computed tomographic angiography, with perfusion imaging to assess salvageable penumbra, by consulting stroke specialists — precisely because thrombolysis is not available	Requires an interventional service willing to take a cannulated patient. The logistics must be arranged before they are needed
Malignant hemispheric infarction	Decompressive craniectomy may be indicated; hyperosmolar therapy for cerebral oedema. Successful craniectomy on ECMO has been reported. Anticoagulation monitoring and resumption are necessary postoperatively	Evidence limited; risks and benefits to be discussed in a multidisciplinary manner
Intracranial haemorrhage — anticoagulation	Prevent expansion by blood pressure control and discontinuing systemic anticoagulation. Venovenous support may allow discontinuation until decannulation, on multiple reports of heparin-free running with a heparin-coated circuit. Venoarterial carries a much higher thromboembolic risk when anticoagulation is held. Early cessation without reversal, then judicious resumption with repeated neuroimaging, appeared feasible in a reported cohort	The configuration changes the answer. This is the single most useful distinction in the section
Intracranial haemorrhage — surgery	Surgical or minimally invasive haematoma evacuation may be considered. Data are limited; multidisciplinary decision-making involving neurosurgery and neurology	Coagulopathy
Hydrocephalus and intracranial pressure	External ventricular drainage may be considered in selected patients at risk of imminent death from intraventricular haemorrhage and hydrocephalus. It is explicitly described as a high-risk procedure with intra- and post-procedural bleeding	Invasive intracranial pressure and brain tissue oxygen monitoring have not been shown to improve long-term outcomes and may increase the risk of parenchymal haemorrhage in ECMO patients — a rare instance of a guideline actively discouraging a monitor
Cerebral venous sinus thrombosis	Requires a high index of suspicion with risk factors including internal jugular cannulation. Large dual-lumen venovenous cannulae may be associated with brain injury through venous hypertension and cannula-related thrombosis. Diagnosis made with brain computed tomography. Systemic anticoagulation is the primary treatment; endovascular thrombectomy in advanced centres for deteriorating patients; decompressive craniectomy for hemispheric oedema	Presentation is non-specific — headache, seizure, encephalopathy — that is, indistinguishable from everything else on this list in a sedated patient
Positioning	Head of bed at 30 degrees may benefit raised intracranial pressure; but brain oxygenation and circulation improve supine, which benefits perfusion-dependent ischaemic stroke. Position may be guided by transcranial Doppler velocity and near-infrared spectroscopy regional saturation	A genuine use for the monitors in this chapter as real-time feedback on a bedside manoeuvre

💡

Clinical pearl · The one place in this chapter where a monitor directly steers a bedside action

Almost everything in §§28.4 to 28.12 detects rather than guides. The head-of-bed recommendation is the exception, and it is worth noticing because it shows what these instruments are for when they work properly.

The two standard positions conflict: head up for intracranial pressure, supine for cerebral perfusion in ischaemic stroke. A patient with both problems has no protocol answer. The consensus's suggestion is to let the monitors arbitrate — adjust the position and watch the transcranial Doppler velocity and the regional saturation.

This is the model for how every instrument in this chapter should ideally be used: not as a screening test generating alarms, but as feedback on a manoeuvre you are performing deliberately. It is the same logic Chapter 26 reached for echocardiography during a deliberate flow change — the information is in the response, not in the reading.

28.16 Prognostication as a monitoring problem

Chapter 23 owns prognostication after extracorporeal cardiopulmonary resuscitation, and this chapter does not re-argue it. What belongs here is narrower and more uncomfortable: every test in the standard prognostication set is, on ECMO, an instrument with a known confounder — and the set has never been validated in this population.

⚠️

Danger · The multimodal prognostication rule has not been validated on ECMO, and the guideline says so

The 2024 ELSO consensus reproduces the standard post-arrest rule: an unfavourable neurological outcome in patients without ECMO is strongly suggested by at least two indicators of severe brain injury, namely

absence of pupillary and corneal reflexes at 72 hours or more;
bilateral absence of N20 cortical waves on somatosensory evoked potentials at 24 hours or more;
highly malignant electroencephalographic patterns after 24 hours;
neuron-specific enolase over 60 μg/L at 48 or 72 hours;
status myoclonus within 72 hours;
extensive diffuse anoxic injury on brain computed tomography or magnetic resonance imaging.

Then, in the next sentence: "This approach has not been validated in ECMO patients and has limited evidence."

Work through what that means, item by item, using this chapter's own findings.

The pupils are unreliable in the early window because of adrenaline (§28.4) — though quantitative pupillometry from 24 to 72 hours performed well in one cohort (§28.5).
The N20 was intact in all 7 patients who had it recorded alongside electroencephalography in the only series that looked, every one of whom did badly (§28.11).
Enolase is confounded by haemolysis and the correct threshold is unknown and probably above 100 μg/L, not 60 (§28.10).
Magnetic resonance imaging, the most informative item on the list, is unavailable for the entire duration of support (§28.12).

Four of six indicators are compromised. The rule requires two concordant indicators; on ECMO it is difficult to assemble two that are not individually confounded.

The clinical consequence is not paralysis but timing and humility: prognostication on ECMO should be slower, more multidisciplinary, and more explicitly provisional than off it.

🔬

Evidence · Four brains, and the reason this section exists

What was done. All patients who underwent brain autopsy at a single centre from November 2017 to December 2018 were reviewed. Four patients with neuromonitoring had post-mortem brain examination. Median age 56; all four had venoarterial cannulation; none had a known acute neurological injury before ECMO. All four were persistently comatose off sedation on support. Continuous electroencephalography was performed on all four, and all four showed delta frequency with absent reactivity.

What was found at autopsy.

Case 1: small infarcts in the cerebral cortices.
Case 2: cortical microhaemorrhages and chronic microinfarcts in the basal ganglia.
Case 3 and Case 4: diffuse hypoxic-ischaemic brain injury.

The authors' conclusion, reproduced exactly because the phrasing matters: "The clinical presentation was consistent with the widespread neuropathologic injury in two cases (Cases 3 and 4) but not in the other two (Cases 1 and 2). While 'poor neurological status' was provided in the end-of-life discussion in all four, only two had significant brain injuries."

What it means. Persistent coma off sedation, with an unreactive delta electroencephalogram, was not explained by structural brain injury in half of these four patients. Something else — the circuit, residual sedation pharmacokinetics, metabolic derangement, critical illness — produced the same clinical and electrophysiological picture as devastating injury.

And the finding that should genuinely disturb: in all four, "poor neurological status" was used in the end-of-life conversation. In two of them, that statement was not supported by the pathology.

Limitations: four patients, one centre, all of whom died — the design cannot tell you how often this happens. It is the smallest study cited in this chapter and one of the most important, because it is the only one that checked.

Certainty: very low as an estimate of frequency; high as a demonstration that the failure mode exists.

The authors' own summary is the right one to carry: "Extrapolations related to prognostication from experience in other brain injury needs to be approached cautiously."

Source: Cho SM, Geocadin RG, White B, Chen LL. Int J Artif Organs 2020;43(9):614–619. DOI 10.1177/0391398820901829.

💡

Clinical pearl · Name the self-fulfilling prophecy out loud in the meeting

The 2024 ELSO consensus instructs clinicians to "exercise caution to mitigate the 'self-fulfilling prophecy' bias, which occurs when prognostic test results indicating poor outcomes influence treatment decisions." Two additional practical points from the same section are worth adopting verbatim.

Rule out confounders first — sedatives, significant electrolyte disturbance, hypothermia — "to prevent an overly pessimistic prognosis." On ECMO, add ongoing haemolysis to that list whenever enolase is being quoted.
In patients undergoing targeted temperature management, the most crucial assessment is the one performed after rewarming.

The consensus also recommends routine ethics consultation within 72 hours of cannulation where the resource exists, to mitigate ethical conflicts by setting clear expectations. Chapter 74 and Chapter 75 own the ethics; it is noted here because 72 hours from cannulation is inside the window this chapter covers, and because arranging it is a monitoring-team action.

28.17 Controversy 2 · Should the venoarterial patient be cooled?

⚖️

The clinical question. A patient is cannulated onto venoarterial ECMO after a prolonged low-flow state. The circuit's heat exchanger makes temperature a free parameter — the operator chooses the patient's temperature by turning a dial. Should that dial be set to 33 to 34 °C, or to normothermia?

Position A: cool, because the biology is strong and this is the population it was designed for. The 2024 ELSO consensus makes the argument itself: "Inducing hypothermia during ischemia prolongs the tolerance of organs to ischemia, improving neurological outcomes. Thus, it could be reasonable to use hypothermia in VA ECMO patients where cerebral ischemic and hypoperfusion time is prolonged. This rationale is even more important in patients who have already suffered severe hypoxic-ischemic brain injury, as in ECPR." It adds that the basic and preclinical science on hypothermia in ischaemia is strong, that venoarterial patients have a high incidence of brain injury and prolonged absent or low cerebral perfusion, and that in a meta-analysis of extracorporeal cardiopulmonary resuscitation patients bleeding complications and coagulopathy were similar with and without hypothermia — removing the main clinical objection.

Position B: do not cool, because the one randomised trial was negative. A randomised controlled trial in patients with cardiogenic shock requiring venoarterial ECMO compared moderate hypothermia (33 to 34 °C) with normothermia (36 to 37 °C) and found no mortality difference at 30 days.

Where the evidence actually sits. The consensus lists three specific limitations of that trial, and they are not rhetorical.

Insufficient sample size, because the effect size was estimated from non-ECMO studies.
No formal neurological assessment was performed.
The primary outcome was mortality at 30 days rather than neurological outcome at 90 or 180 days — that is, the trial did not measure the thing hypothermia is hypothesised to protect.

The wider evidence base is worse: the meta-analysis of 2,643 extracorporeal cardiopulmonary resuscitation patients across 35 studies is described by the consensus as "severely heterogeneous and limited to low-quality evidence." And there is no data at all on hypothermia in venovenous support.

What this book concludes. This is not a resolved negative trial; it is an underpowered trial of the wrong endpoint. It should not be cited as evidence that hypothermia does not work on venoarterial ECMO, and equally it is not evidence that it does. The honest position is that a temperature is being chosen for every venoarterial patient, by someone, with no evidence either way — which is an unusual situation, because on ECMO the choice cannot be avoided. The heat exchanger has no neutral setting.

What would resolve it. The consensus says it: a robust multicentre prospective study testing hypothermia strategically in each major venoarterial cohort separately — postcardiotomy shock, extracorporeal cardiopulmonary resuscitation, post-myocardial-infarction and non-ischaemic cardiogenic shock — with neurological outcome at 90 to 180 days as the primary endpoint and formal neurological assessment built in.

Certainty: very low in both directions.

[VERIFICATION REQUIRED: the identity, name, sample size and full citation of the randomised trial of moderate hypothermia versus normothermia in venoarterial ECMO for cardiogenic shock, and of the 35-study meta-analysis of 2,643 extracorporeal cardiopulmonary resuscitation patients. Both are described in the 2024 ELSO consensus text but their reference numbers were not resolvable from the full text as retrieved.]

28.18 A worked case: the patient whose monitors all agreed, and were all wrong

🧮

Worked example · Reasoning through the detection gap at the bedside

The patient. A 58-year-old man, day 3 of peripheral venoarterial ECMO after refractory cardiogenic shock following an anterior myocardial infarction. No preceding cardiac arrest. Circuit flow 3.8 L/min, sweep 3 L/min, delivered oxygen fraction 0.5. Sedation on a continuous infusion of midazolam and fentanyl, no paralytic since day 1.

The monitors, all recorded within the same hour.

Bilateral near-infrared spectroscopy: left 64 percent, right 62 percent. Both within 5 percent of the post-cannulation baseline; asymmetry 2 percent.
Right radial arterial gas: oxygen tension 96 mmHg, carbon dioxide tension 39 mmHg, pH 7.41.
Quantitative pupil index 4.2 bilaterally, unchanged over four measurements.
Transcranial Doppler: mean middle cerebral artery velocity low-normal bilaterally, pulsatility index low and symmetrical.
Nursing neurological assessment four-hourly: "sedated, pupils equal and reactive, no focal deficit." Recorded as unchanged for 48 hours.

The question asked on the ward round. The team is considering whether the anticoagulation intensity can be reduced, given a falling platelet count. Someone says: "Neurologically he's been rock solid — all the monitoring is normal."

Why that sentence is the error, worked through.

The four-hourly examination has not been performed for 48 hours. What has been performed is an observation of a sedated patient. Motor response through continuous sedation is not a motor examination (§28.4). The one sentence that would change this — "sedation was interrupted and he localised" — is absent from the chart. That absence is the finding.
Symmetrical near-infrared spectroscopy excludes very little. Its sensitivity for brain injury as an asymmetry measure was 39 percent in the study that defined the threshold (§28.6). Sixty-one percent of injuries were missed. And it samples a small frontal volume: a posterior or deep infarct is outside it entirely.
A normal pupil index has a sensitivity of 60 percent at best (§28.5), and the instrument was validated against 90-day mortality, not against the presence of a lesion. A large haemorrhage with a normal pupil index is an ordinary finding.
The low, symmetrical pulsatility index is a circuit measurement, not a cerebral one — it reflects the proportion of cardiac output the pump is carrying (§28.8). Reading it as reassurance is a category error.
The right radial gas is the right sample and it is genuinely reassuring about one specific thing: differential hypoxaemia is not currently occurring. That is all it says.
And the base rate: in a venoarterial patient on day 3, autopsy series find brain injury in 68 percent of those who die and long-term imaging finds it in 37 to 52 percent of survivors (§28.3). A set of rule-in tests, all negative, moves that prior very little.

What should be said instead. "No monitor has alarmed, and no examination has been performed. Before we change anticoagulation, interrupt the sedation and examine him. If he localises symmetrically and follows commands, we have information. If he does not, we need a scan before we touch the heparin, not after."

The teaching point, which is the chapter in one line. Absence of an alarm from a panel of rule-in tests is not the same as evidence of an intact brain — and on ECMO the only test with a useful negative predictive value is the neurological examination, which is the one the treatment has switched off. Switching it back on is a deliberate act that someone has to order.

This case is constructed to illustrate the reasoning in §§28.3 to 28.8 and the operating characteristics cited there. It is not drawn from a published report.

28.19 What this chapter could not close

🚩

Open questions, recorded rather than concealed

Temperature correction of blood gases on a circuit — alpha-stat versus pH-stat — remains unresolved for the second consecutive chapter. Chapter 27 flagged it and could not close it. A targeted search for alpha-stat or pH-stat management, temperature correction and blood gases in extracorporeal membrane oxygenation returned 95 indexed records, of which none was ECMO-specific; the literature is entirely cardiopulmonary bypass and therapeutic hypothermia. This is itself a finding and is recorded as such: there appears to be no ECMO-specific evidence on how blood gases should be temperature-corrected on a circuit whose heat exchanger sets the patient's temperature. Given §28.17, where the temperature is a chosen parameter and the carbon dioxide tension is a cerebral vasomotor drug, this gap matters more on ECMO than on bypass, not less.
The near-infrared spectroscopy operating characteristics cannot be reconciled. Sensitivity 86 percent and specificity 55 percent (Red Book, attributed to Hunt and colleagues) against sensitivity 39 percent and specificity 98 percent (Zhao and colleagues, asymmetry). The primary Hunt paper was not retrieved.
The randomised trial of hypothermia in venoarterial ECMO is described but not identified, nor is the 35-study meta-analysis. Both are cited within the ELSO consensus text without resolvable reference numbers in the retrieved full text.
The three near-infrared spectroscopy thresholds in §28.6 have no common basis and no study compares them.
The Ong and colleagues citation carries a 2021 online and a 2023 print year, reproduced as found.
The paediatric electroencephalography series in §28.9 is a conference abstract and may or may not have been published in full.
The Menozzi 2022 pupillometry report was retrieved without an abstract and its content is not reproduced.
Author order on the venoarterial stroke cohort (DOI 10.1155/ccrp/9058296) appears irregular in the retrieved citation record, with the corresponding author apparently placed second-to-last. The citation is reproduced as the record gives it. [VERIFICATION REQUIRED.]
The duplicate-identifier pattern flagged in Chapter 27 recurred: the neuromonitoring review was returned under both a plain digital object identifier and a compound form with identical text. The plain form is cited.
No chapter in the map owns registry structure and reporting limitations, although the argument in §28.3 depends on them. The nearest homes are Chapter 1 and Appendix 12.

Corrections owed to other chapters, arising from this chapter's audit

Chapter 22 §22.3 should state a ceiling on the rate of carbon dioxide fall alongside its target of about 40 mmHg, because in a patient starting at 100 mmHg the two instructions conflict (§28.13).
Chapters 29 and 37 inherit the enolase threshold statement from this chapter: the standard 60 μg/L figure does not apply on a circuit and the correct figure is unknown (§28.10).
Part VI should acquire a chapter on neurological complications, or Chapter 38 should be redefined, so that §28.15 can be reduced to a cross-reference.

Carried forward from earlier chapters and still open: a home for extracorporeal carbon dioxide removal; the ELSO extracorporeal cardiopulmonary resuscitation interim guidance; and now, newly, a home for neurological complications as complications.

28.20 Evidence and certainty summary

Claim	Best evidence	Certainty
Magnetic resonance imaging is unavailable during ECMO	ELSO Red Book; ISCCM manual; consistent across all sources	High — a hardware fact
Detected neurological complication rates are several-fold lower than pathological prevalence	Autopsy series (68 percent of non-survivors; 9 of 10 brains; 30 percent undiagnosed thromboembolism) against registry and review rates of 4 to 13 percent; long-term imaging 37 to 52 percent in survivors	High for the direction and rough magnitude of the gap
Acute brain injury roughly doubles to triples in-hospital mortality	2024 ELSO consensus (factor of 2 to 3); Sutter systematic review (83 versus 42 percent, p under 0.001)	High
Brain injury is more common on venoarterial than venovenous support, and highest in extracorporeal cardiopulmonary resuscitation (27 to 32 percent)	2024 ELSO consensus; registry data; consistent across sources	High
Most hypoxic-ischaemic injury in cardiac ECMO patients occurred before cannulation	ELSO Red Book Chapter 33	Moderate to high
Standardised neuromonitoring increases the detection of acute brain injury	Ong and colleagues, before-and-after, odds ratio 2.24 (1.12 to 4.48)	Moderate
Standardised neuromonitoring improves neurological outcome	Same study; before-and-after design at one centre over a period of secular improvement	Low
Sedation interruption improves the neurological examination and the diagnosis of brain injury; intermittent analgo-sedation is preferred to infusion	2024 ELSO consensus recommendation; ISCCM manual; Red Book	Moderate — consensus recommendation, mechanistically compelling, not trial-tested
Fixed dilated pupils during resuscitation with adrenaline do not predict a poor outcome	2024 ELSO consensus, explicit statement	Moderate
An abnormal quantitative pupil index from 24 to 72 hours predicts 90-day mortality with no false positives	Miroz and colleagues, 100 patients, single centre, threshold derived in the same dataset	Moderate for signal; low for the transportable threshold
Near-infrared spectroscopy asymmetry outperforms the absolute value for detecting brain injury on venoarterial support	Zhao and colleagues, 69 patients, 18 events; specificity 98 percent, sensitivity 39 percent	Moderate for the principle; low for the 8 percent threshold
Near-infrared spectroscopy is not a standard of care	Castro and colleagues; Red Book ("has not sufficiently been validated")	High
Cerebral autoregulation is impaired under continuous flow, worse on venoarterial than venovenous, and worst on day 1	Zhang and colleagues, 15 patients, 150 measurements	Low — small, single centre, feasibility design
There is no population blood pressure target that serves this group; optimal pressure ranged 55 to 110 mmHg	Zhang and colleagues; 2024 ELSO consensus states no data exist on optimal early blood pressure goals	Low for the range; high that the evidence for a universal target is absent
Continuous flow confounds the transcranial Doppler pulsatility index	ELSO Red Book bedside ultrasound chapter, explicit; paediatric literature	High
Microembolic signals are far more common on venoarterial than venovenous support	Marinoni (81.8 versus 26.2 percent); Caturegli (35.0 versus 4.7 percent)	Moderate to high
Microembolic signals predict brain injury or thromboembolic events	Three studies, three conclusions; the largest (145 patients) found no association	Very low — do not act on it
Most electrographic seizures on extracorporeal support are subclinical	Paediatric series, 67 percent electroencephalogram-only (abstract); adult data consistent	Moderate for the principle; low for the proportion in adults
Electroencephalographic background and reactivity predict outcome better than seizure burden	Peluso and colleagues; paediatric series (90 percent mortality with severe background)	Moderate
Absent electroencephalographic reactivity off sedation does not guarantee structural brain injury	Cho and colleagues, four brain autopsies, two discordant	Very low as frequency; high that the failure mode is real
Neuron-specific enolase is confounded by haemolysis; the ECMO threshold is unknown and possibly above 100 μg/L	2024 ELSO consensus, explicit; Red Book	High that it is confounded; very low for any threshold
A preserved N20 does not reassure on ECMO	Seven patients, all with intact N20 and poor outcome	Very low — but it is the only direct observation available
A peri-cannulation carbon dioxide fall of over 50 percent should be avoided	2024 ELSO consensus recommendation; ELSO registry analysis of 11,972 venovenous patients; single-centre venoarterial cohort (odds ratio 2.69 for intracranial haemorrhage); one extracorporeal cardiopulmonary resuscitation study of 3,125 patients finding no significant association	Moderate for the direction; low for the threshold
Severe hyperoxaemia (over 300 mmHg) within 24 hours of venoarterial cannulation should be avoided	2024 ELSO consensus; observational associations. Optimal targets stated to be unknown	Low to moderate
Thrombolysis is generally not indicated for ischaemic stroke on ECMO; thrombectomy should be pursued for large vessel occlusion	2024 ELSO consensus, expert consensus with limited literature	Moderate as a consensus position; low in underlying evidence
Invasive intracranial pressure and brain tissue oxygen monitoring have not improved outcomes and may increase parenchymal haemorrhage risk on ECMO	2024 ELSO consensus, explicit	Low to moderate
Hypothermia improves neurological outcome on venoarterial ECMO	Strong preclinical rationale; one underpowered randomised trial with a mortality endpoint and no neurological assessment; a heterogeneous meta-analysis	Very low in both directions — genuinely unresolved
The standard multimodal prognostication set is valid on ECMO	2024 ELSO consensus: "has not been validated in ECMO patients and has limited evidence"	Very low — this is a statement that validation is absent

28.21 Key points

The brain is the only organ on ECMO that the circuit does not monitor, and the treatment simultaneously removes the two instruments that normally substitute — the examination, through sedation, and magnetic resonance imaging, through the hardware. Cerebral monitoring on ECMO is reconstruction of a lost sense, not refinement of a working one.
Every published neurological complication rate is a detection rate. Routine care finds injury in 4 to 13 percent of adults; autopsy finds it in 68 percent of non-survivors; delayed imaging finds it in 37 to 52 percent of survivors. The gap is a factor of five to ten.
Consequently, every instrument in this chapter is a rule-in test. None can be used to reassure. A normal panel licenses the absence of an action right now; it does not license a lower index of suspicion.
The serial examination is the primary monitor, and sedation strategy determines whether it exists. Intermittent analgo-sedation over infusion; daily interruption treated as a monitoring procedure; "ECLS per se does not mandate sedation."
Fixed, dilated pupils at the time of a resuscitation in which adrenaline was given are a pharmacological observation, not a prognosis — and are not a reason to decline extracorporeal cardiopulmonary resuscitation.
Quantitative pupillometry from 24 to 72 hours is the one instrument with a published zero false-positive rate for 90-day mortality, at a threshold of 3, in a single cohort of 100. Sensitivity was 60 percent at best: a normal index is uninformative.
On near-infrared spectroscopy, the signal is the asymmetry between hemispheres, not the absolute saturation. Use bilateral sensors. A single sensor discards the information. And three incompatible thresholds are in circulation — choose one and write it down.
Continuous flow destroys the meaning of the transcranial Doppler pulsatility index. A low index reflects the pump, not the brain. Record the circuit flow alongside every velocity — the third time this book has reached that rule from a different direction.
Microembolic signals are a research measurement. They are several times more common on venoarterial support, and in the largest cohort they predicted nothing. Do not change anticoagulation on a signal count.
Most seizures on extracorporeal support are subclinical, and the background matters more than the seizures. Ask the neurophysiologist whether the background is reactive.
Neuron-specific enolase is confounded by haemolysis. The standard 60 μg/L threshold does not apply; the correct one is unknown and possibly above 100. Never quote a value without the concurrent haemolysis indices.
The peri-cannulation window is where prevention lives. Avoid a carbon dioxide fall of over 50 percent; avoid an oxygen tension over 300 mmHg in the first 24 hours of venoarterial support; record the pre-cannulation carbon dioxide so the target has a denominator.
The oxygen blender is the hyperoxaemia knob even though Chapter 27 showed it is not the hypoxaemia knob. Turning it down always works; turning it up usually does not.
On venoarterial support with an ischaemic stroke, the brain and the recovering ventricle want opposite blood pressures, and no evidence resolves it. Make the trade explicit, revisit it daily, and make it together with the venting decision.
Absent electroencephalographic reactivity off sedation has been shown, in autopsy-verified cases, not to guarantee structural injury — in two of the only four patients in whom anyone checked, while "poor neurological status" was nevertheless used in the end-of-life conversation. Prognosticate slowly, multidisciplinarily, and after rewarming.
Four of the six standard prognostication indicators are individually confounded on ECMO, and the consensus states the set has never been validated in this population.
Chapter 22's carbon dioxide rule and the ELSO 50 percent rule conflict in the severely hypercapnic patient. The percentage ceiling binds; 40 mmHg is the destination, not the first stop. Chapter 22 is owed a correction.
This book has no owner for neurological complications as complications. Chapter 28 carries detection; §28.15 carries management as a labelled stopgap. That decision is owed before Part VI is drafted.

📌

Chapter answer: What is left is a panel of rule-in tests and one switched-off examination. The near-infrared spectroscopy sensors detect asymmetry, not adequacy; the pupil index detects a trajectory, not a lesion; the Doppler detects the pump as much as the brain; the electroencephalogram detects what the paralytic hides; and computed tomography detects blood. None of them detects a normal brain. The single most powerful thing anyone does for cerebral monitoring on ECMO is not an instrument at all — it is turning the sedation off and looking.