Evidence Database

28,767 adults on V-A ECLS, ELSO Registry 2010-2020. Artif Organs 2022;46(11):2266-2283.

Internal validation only; weak discrimination; no laboratory haemostasis variables; abstract only read.

36, 42, 45

29.0% bleeding complications. AUC 0.66. Predictors: sex, BMI, surgical cannulation, pre-ECLS respiratory and haemodynamic variables, pre-ECLS support and interventions, diagnosis. Authors: external validation warranted.

Internal validation

Multivariable logistic prediction model for bleeding

Use individual risk factors rather than a score to inform anticoagulation intensity.

DOI 10.1111/aor.14340; PMID 35712783

132 adult ECLS patients, single centre (University of Maryland), about 3 years. Ann Thorac Surg 2016;101(2):682-689.

Single centre, small, retrospective; wide confidence intervals; abstract only read; transfusion lies on the causal path of bleeding.

36, 46

Serious bleeding 56.1% (74/132), 10 events per 100 days. Crude OR death with bleeding 2.22 (1.00-4.94); adjusted 0.90 (0.37-2.19). Per RBC unit crude 1.03 (1.01-1.04), adjusted 1.03 (1.00-1.06, p=0.04). 90-day survival trend lower with bleeding (46.7% vs 64.9%, p=0.08).

No bleeding; per-unit transfusion

Exposure: serious bleeding events and red cell transfusion

Track red cell units transfused as a quantitative marker; the adjusted association with mortality attached to transfusion volume rather than to bleeding per se in this cohort.

DOI 10.1016/j.athoracsur.2015.07.046; PMID 26443879

VA ECMO patients: analysis cohort 349 (Nanjing, 2017-2023); external validation 83 (Suqian, 2022-2024). BMC Gastroenterol 2026;26(1).

Two centres, small external cohort (10 events), abstract only read.

36, 55

GI bleeding 17.8% (62/349) and 12.0% (10/83). After SIPTW, GI bleeding associated with higher 28-day mortality HR 2.25 (1.67-3.05). Independent predictors: CRRT, baseline INR, baseline calcium. AUC 0.797 training, 0.803 internal validation, 0.753 (0.531-0.976) external.

No GI bleeding (stabilised IPTW)

Exposure: GI bleeding; LASSO/logistic prediction model

GI bleeding on VA ECMO is common and associated with death; patients on CRRT or with raised baseline INR are at higher risk.

DOI 10.1186/s12876-026-04991-3; PMID 42351017; PMC13343996

18,658 adults on V-V ECLS, ELSO Registry 2010-2020. Artif Organs 2022;46(5):932-952.

Internal validation only; weak discrimination; no laboratory haemostasis variables; abstract only read.

36, 42, 45

3,933 (21.1%) bleeding complications. AUC 0.63. Predictors: pre-ECLS arrest, surgical cannulation, lactate, pO2, HCO3, ventilation rate, mean airway pressure, pre-ECLS CPB or RRT, pre-ECLS surgical interventions, diagnosis.

Internal validation

Multivariable logistic prediction model for bleeding

Registry-based bleeding prediction is only modestly better than chance; should not set anticoagulation intensity.

DOI 10.1111/aor.14148; PMID 34904241

7,579 adults on VV-ECMO in the ELSO Registry 2010-2017 (of 11,880); exclusions as companion VA study. Mean age 46.5; median 8 days; single-site cannulation 36.7%. Intensive Care Med 2022;48(2):213-224 (online Dec 2021).

Registry; missing-data exclusions; no VTE, DVT, PE or upper respiratory bleeding collected; no anticoagulation or laboratory data; no event timing; ICD-coded comorbidities; in-hospital outcomes only; observational.

36, 31, 38a, 45, 9

40.2% had >=1 BTE; of these 41.7% thrombosis only, 37% bleeding only, 21.2% both. Of 4,840 BTEs, 54.9% thrombotic: circuit clotting 31.8%, oxygenator/pump failure 12.7%; cannulation bleeding 15.5%, surgical site 9.6%; medical bleeding 18.7%; ICH 4.5%, ischaemic stroke 1.9% of events. Mortality 34.9%. Adjusted OR death: any bleeding 1.69, any thrombosis 1.23; 1 BTE 1.66, 2 BTEs 1.68, >=3 2.16; ICH 5.71, ischaemic stroke 4.50, pulmonary 2.02, GI 1.54; no surgical bleeding event significant. Mortality with event: ischaemic stroke 73.9%, ICH 73.2%, pulmonary 53.6%, GI 48.6%, tamponade 46.6%, oxygenator/pump failure 45.7%, haemolysis 42.2%, surgical site 39.6%, cannula site 39.3%, circuit clotting 34.7%. Bleeding risk: pre-ECMO AKI, vasopressors, longer duration, younger age, higher pH. Thrombosis risk: higher weight, multisite cannulation, pre-ECMO arrest, higher PaCO2. Ischaemic stroke with delta PaO2 >80 mmHg at 24 h (15% higher odds per 10 mmHg); ICH with delta PaO2 <80.

Patients without BTEs

Not applicable (observational; exposure = bleeding and thrombotic events)

On VV-ECMO thrombotic events are more common but bleeding carries more mortality, especially medical bleeding; AKI and vasopressors before ECMO flag bleeding risk; avoid large early PaO2 rises; minimise duration.

DOI 10.1007/s00134-021-06593-x; PMID 34921625; PMC9178906 (correction PMID 35041017)

11,984 adults on VA-ECLS in the ELSO Registry 2010-2017 (of 17,715); excluded pre-ECLS coagulopathy/DIC, GI or pulmonary bleeding, stroke, multiple runs, missing core covariates. Mean age 54.5; 44% pre-ECLS arrest; 79.3% peripheral; median 4 days.

Self-reported registry; threshold-based definitions (e.g. >3 U PRBC/24 h) miss minor events and under-report; 26.1% excluded for missing data; no anticoagulation or laboratory data; no timing of events; ICD-coded comorbidities; in-hospital outcomes only; observational. Abstract lists female sex as bleeding risk factor but results describe only a trend.

36, 31, 38a, 45, 46

8,457 HRAEs in 5,285 patients (44.1%); 62.1% bleeding, 37.9% thrombotic. Bleeding 32.8% of patients, thrombosis 17.2%. Commonest: cannulation-site 23.5%, surgical-site 20.8% of events; medical bleeding 12.6% of events; circuit clots 18.4%; ischaemic stroke 4.3% of patients, haemorrhagic stroke 2%. Crude mortality 58.6%. Adjusted OR death: 1 HRAE 1.43, 2 HRAEs 1.86, >=3 3.27. Haemorrhagic stroke 7.71, pulmonary 3.08, GI 1.95, tamponade 1.40, surgical site 1.31, cannulation site 1.05 (0.91-1.22, NS), ischaemic stroke 2.31, haemolysis 1.58. Bleeding risk: central cannulation 1.89, each day 1.06, renal failure 1.3, PMP oxygenator 1.15, older age 1.05/10y; lower with later year 0.93, higher pH 0.93/0.1, Asian race 0.8. Medical bleeding: time 1.07/day, respiratory failure 1.3, lower pH. Time on ECLS not independently associated with mortality. HRAE rates fell over time but medical bleeding (p trend 0.08) and ischaemic stroke did not.

Patients without HRAEs

Not applicable (observational; exposure = hemocompatibility-related adverse events)

Distinguish surgical from medical bleeding: cannulation-site bleeding is common and not associated with death; ICH, pulmonary and GI haemorrhage carry the mortality signal. Minimise ECLS duration; watch renal failure, central cannulation and acidosis as bleeding risk markers.

DOI 10.1016/j.jchf.2020.09.004; PMID 33121701; PMC8427553

Three patients on ECMO. Perfusion 2021;36(1):97-99 (online May 2020).

Three cases, abstract only read, configuration not stated in abstract.

35, 38a

Air entered from a right jugular venous sheath during percutaneous septal repair, a pulmonary artery catheter in ICU, and sewing holes on the atrial wall during surgery. Cases 2 and 3 managed successfully; case 1 suspected cerebral air embolism through transseptal right-to-left shunt.

None

Not applicable (air entry events)

Venous access open to air near the drainage cannula, especially with an intracardiac right-to-left communication, is a route to systemic air embolism.

DOI 10.1177/0267659120918471; PMID 32423312

In vitro venovenous model: Cardiohelp with HLS set, 23 Fr 55 cm multistage drainage and 17 Fr return cannula in latex vessels; 40% glycerol at 36 C; CVP 10 cm H2O; 1500/2500/3500/4500 rpm (1.61/3.15/4.56/5.91 L/min); each experiment x3.

In vitro, blood substitute, VV only, one CVC size, only distal CVC lumen tested, no respiration (authors expect vigorous inspiration to increase entry), fixed CVP, one console; descriptive only, three repeats.

35, 34, 33, 32

Protected hub (stopcock or clave): no air entry at any speed or distance. Open hub near distal hole: air at 2500, 3500, 4500 rpm with tips touching, 3500 and 4500 only at 1-2 cm; near proximal side hole: 3500 and 4500 rpm at all distances, none at 1500/2500. Pump sound at 1 mL and above = first warning sign. Air reached return-side detector at 1 mL (4500 rpm), 5 mL and 25 mL (3500, 4500). 25 or 50 mL stopped flow completely at every speed; smaller volumes transient flow drop with slow oxygenator venting. Authors: return-side detector inappropriate for monitoring circuit air because membrane retains most air; raising speed after airlock risks forcing air through the membrane, so clamp. Cites (secondary) AE incidence up to 4.6% adults, 4.9% neonates; animal lethal dose 200-300 mL or 3-5 mL/kg.

Open hub vs protected hub; pump speeds

7 Fr triple-lumen CVC tip 0/1/2 cm from drainage cannula distal or proximal hole with hub open vs closed 3-way stopcock vs clave; separately 1/5/25/50 mL air injected into drainage cannula.

Keep every central line lumen closed (stopcock or needle-free valve) in ECMO patients; treat a new hiss from the pump head as air until proven otherwise; do not rely on a post-membrane bubble detector for venous air; never raise pump speed to clear an airlock - clamp.

DOI 10.3390/medicina60111883; PMID 39597068; PMC11596159

In vitro venoarterial ECLS circuit with centrifugal pump, membrane oxygenator and occluder; 44% glycerin prime; 2500/3000/3400 rpm. Artif Organs 2020;44(3):268-277.

In vitro, abstract only read, blood substitute.

35, 33, 34

Microbubble generation highly dependent on negative pressure at the inflow line; increasing speed and inflow restriction increased microbubbles; centrifugal pump identified as a main source; oxygenator's ability to withhold larger microbubbles limited; dynamic bubble traps significantly reduced count and gas volume.

With vs without dynamic bubble traps

Static or dynamic inflow occlusion to generate microbubbles; oxygenator and dynamic bubble traps assessed for elimination.

Avoid excessive negative inlet pressure and inflow restriction; the membrane lung is not a reliable microbubble filter.

DOI 10.1111/aor.13557; PMID 31408537

116 ECMO patients on Cardiohelp HLS, single centre, October 2009 to July 2020; 4 with circuit gas bubbles or systemic gas embolism.

Four patients, one centre, one console; abstract only read; backflush technique not described in abstract.

35

Source located in all 4; venous side in 3; centrifugal pump airlock with flow cessation in 2. All 4 weaned from ECMO; 3 discharged alive. Authors: systemic gas embolism avoidable with effective management; backflush safe and effective for airlock in Cardiohelp.

None

Circuit change or de-airing with backflush technique

Circuit air source is usually findable and usually venous; an airlock can be recovered by de-airing or circuit change.

DOI 10.1097/MAT.0000000000001494; PMID 34081051

391 responses from 45 countries (43% response rate; invitations to directors/coordinators of 396 ELSO-member centres in 51 countries); US 181, UK 32; neonatal, paediatric and adult peripheral ECMO; some double enrolment acknowledged.

Self-report, recall and social-desirability bias, no patient denominator (outcome = centre recalling any event), possible double enrolment, US-dominated. Never-suture group very small (upper CI 28% implies roughly a dozen centres, book's arithmetic). Volume and suturing associations confounded by exposure.

34, 58, 57, 59, 60

Suturing at insertion site 93% (Americas and Europe 96%, Australia/NZ 68%). Primary securement along line: skin suture 45%, bed/fixed object 19%, sutureless device 16% (42% in Australia/NZ), adhesive tape 10%. Femoral fixation points: two 48%, more than two 26%, one 21%. 34% reported a line malposition, dislodgement or decannulation with adverse outcome in previous 5 years. Causes (71 events, 61 respondents): inadequate securement 28%, insertion/manipulation 14%, patient removed 13%, turning/bathing 11%, transport 4.2%, material failure 2.8%, ambulation 1.4%, not stated 25%. Always-suture centres 35% (95% CI 30-41) vs never-suture 0% (0-28), chi-square 4.40, p=0.04, not significant at Bonferroni-adjusted 0.01. Adult centres over 30 cases/year 56% vs 1-6 cases 22% (p under 0.001). Dressings: transparent 75%; CHG-alcohol commonest antiseptic (53% insertion, 54% maintenance); 78% want an evidence-based guideline.

Exploratory comparisons between practice groups

Not applicable (survey of line practices: antisepsis, dressings, securement, circuit access)

Securement fails often and inadequate securement is the commonest named cause; secure at multiple points, standardise, and do not change securement method on the basis of the non-significant suturing association.

DOI 10.1371/journal.pone.0227248; PMID 31887197; PMC6936833

Single case: 70-year-old on femoral venoarterial ECMO for refractory post-infarction cardiogenic shock.

Single case report; full text not available in PMC; outcome details not in abstract.

34, 26, 12

Malposition produced no disturbance of ECMO venous return. Moving or replacing the cannula judged high-risk for hepatic laceration and haemoperitoneum; repositioning deferred until day 10 when the ventricular defect was repaired and ECMO weaned. At insertion, TEE showed the tip in the IVC but not the right atrium; authors state wire visualised in IVC but not RA is insufficient. Indexed with MeSH Fatal Outcome; abstract does not relate death to the cannula.

None

Venous cannula found in an accessory hepatic vein on day 8 by ultrasound; repositioning deliberately deferred.

A malposition can be clinically silent and found only by imaging; leaving a functioning malpositioned cannula can be a deliberate risk-balance decision when manipulation carries greater hazard and support is short.

PMID 30250343; PMC6146273 (abstract only in PMC)

Single case: 17-year-old on venovenous ECMO with a 27 Fr bicaval dual-lumen cannula in the left hepatic vein. Perfusion 2023;38(3):645-650 (online December 2021).

Single case, abstract only read, adolescent.

34, 5, 26

Successful repositioning from the left hepatic vein. Authors identify confirming wire position in the infrahepatic IVC as ensuring safe position and avoiding hepatic vein cannulation.

None

Retraction of the cannula, reinsertion of wire and introducer together, manipulation under serial chest radiographs and transthoracic echocardiography.

For dual-lumen cannula repositioning out of a hepatic vein, re-establish a wire in the infrahepatic IVC before re-advancing.

DOI 10.1177/02676591211063829; PMID 34927476

Two adults on venovenous ECMO with a jugular dual-lumen bicaval cannula that migrated — one into a hepatic vein, one into the right ventricle.

Two cases, single centre; abstract only read; presentation of each case not described in the abstract.

34, 26, 5

Both cannulas repositioned; the right ventricle case without antegrade wire placement into the cannula, without discontinuation of ECMO, and without bleeding. Authors state malposition can cause inadequate ECMO flow, hypoxia and structural injury.

None

Hepatic vein case: repositioning under echocardiographic guidance without a guidewire. Right ventricle case: repositioning using a guidewire from the femoral vein under fluoroscopy.

A migrated dual-lumen cannula can be corrected under imaging without stopping ECMO flow; a femoral wire under fluoroscopy is an option when the tip is in the right ventricle.

DOI 10.1111/jocs.12629; PMID 26358888

Single case: 16-year-old with traumatic brain injury and ARDS on venovenous ECMO via bicaval dual-lumen cannula. Am Surg 2025;91(7):1132-1135.

Single case, abstract only read; durability and safety of adhesive repair unknown.

34, 36

Cannula site oozed from day 9 for 13 days despite dressing changes and attempted suture repair before the crack was found. After adhesive application the site was haemostatic and cannula position stable; ECMO continued 49 more days to respiratory recovery and decannulation; discharged to rehabilitation then home.

None

Crack on the return side of the cannula sealed with skin adhesive (Dermabond) instead of cannula exchange.

Persistent cannula-site oozing unresponsive to local measures should prompt inspection of the cannula itself for a crack.

DOI 10.1177/00031348251346544; PMID 40455563; PMC12419875

178 ECMO-related accidents reported to the Japan Council for Quality Health Care patient-safety accident database, January 2010 to December 2021 (reporting mandatory for tertiary teaching hospitals, voluntary for others; about 20% of Japanese hospitals). Free-text narratives reviewed by two authors. Events limited to intraoperative cardiopulmonary support excluded.

No denominator; authors state incidence cannot be calculated. Voluntary reporting dependent on individuals; system designed around errors, so bleeding likely under-reported. About 20% of hospitals. Categorisation subjective. 12-year window spanning equipment change. Japanese-language database, free-text search.

34, 5, 12, 21, 35, 36

Cannula malposition 28% (50), accidental decannulation 19% (34), bleeding 15% (26), air in circuit 7% (12), thromboembolism 5% (9), limb ischaemia, gas supply and power supply 4% each, foreign body 3%. Harm: death 23%, high potential residual disability 26%. Malposition fatal in 44% (22/50), thromboembolism 56% (5/9). MALPOSITION: ECPR 46%, VA 28%, VV 26%; 74% over 60 years; retroperitoneal or intraperitoneal injury 63% (27/50), all via femoral vessels; right internal jugular commonest single insertion site (20%) with mediastinal/thoracic injury; 38% (19/50) had neither fluoroscopic nor ultrasound guidance and combined guidance was used in none; some malpositions despite fluoroscopic wire confirmation; surgery 54%, trans-arterial embolisation 18%, death before treatment 26%. DECANNULATION: all 34 complete removals requiring reinsertion; arterial cannula 85% (29/34); during another procedure (surgery, PCI) 41% (14/34); insufficient fixing reported as 44% with numerator 23/34 (internal inconsistency); blinded insertion position 41%; inadequate communication 21%. BLEEDING: 9/26 due to circuit damage including fissure at arterial cannula-circuit connection; 11/26 needed surgical repair. AIR: incorrect clamp release 33%, stopcock error 33%, machine breakage 17%. Authors cite Kim et al for a 1.3% decannulation rate.

None

Not applicable (descriptive analysis of reported accidents)

Cannula events are the commonest and among the most lethal ECMO accidents that reach an incident system. Use real-time imaging (ideally combined) for femoral cannulation; treat a flow that will not rise at initiation as possible extravascular cannula; assign one person to the cannulas whenever another procedure is performed on an ECMO patient; secure and keep insertion sites in view.

DOI 10.1007/s10047-023-01386-z; PMID 36795227; PMC9933024

612 neonates (28 days or under) supported with venovenous ECMO for primary pulmonary failure, cannulated via the right internal jugular vein with dual-lumen venovenous cannulas and polymethylpentene oxygenators, reported to the ELSO Registry 2016-2020. Centrifugal 340, roller 272. Deliberately homogeneous: from 1,742 raw paediatric venovenous runs, patients with multiple ECMO runs, transplants, missing diagnosis, still on ECMO, cardiac indication and ECPR were all excluded. Leading primary diagnosis neonatal and meconium aspiration.

Registry, retrospective, no randomisation. THE AUTHORS' OWN FOUR LIMITATIONS: centre volume could not be determined; centres recently switched to centrifugal may still be on a learning curve and the registry does not record centre experience; pump, cannula and oxygenator BRAND NAMES were not released, and the authors note different dual-lumen cannulas alone can differ by 100 percent in pressure drop; and the groups differed in pre-ECLS diagnoses and treatments. THEIR OWN LEADING EXPLANATION FOR THE SURVIVAL FINDING IS CONFOUNDING BY CENTRE: roller pumps have been used for years in established high-volume centres while centrifugal pumps were recently introduced in new low-volume ones. METHODOLOGICAL POINT ADDED BY THIS BOOK AND NOT RAISED BY THE AUTHORS: the multivariable model adjusts for haemolysis, mechanical complications and circuit thrombosis, which are not confounders but the MECHANISMS through which pump type could act at all; adjusting for variables on the causal path removes part of the effect being estimated and can move it unpredictably. The estimate moved from a non-significant 0.77 to a significant 0.53 on that adjustment, which is the direction that should prompt most caution. Entirely neonatal — nothing here transfers to adults, and the mechanism most implicated (long residence time in the pump head at low flow) is a small-patient problem by construction.

33, 31, 37, 60

THE AUTHORS' PRESPECIFIED HYPOTHESIS WAS THAT CENTRIFUGAL PUMPS WOULD BE ASSOCIATED WITH GREATER SURVIVAL; the result was the opposite. UNIVARIATE: no survival difference (272/340 versus 228/272; OR 0.77, 0.51-1.17, p = 0.22). MULTIVARIABLE (pump type, haemolysis, mechanical complications, thrombosis/clots in circuit components, clots in haemofilter, meconium aspiration): centrifugal pump use associated with LESSER odds of survival, OR 0.53 (0.33-0.84), p under 0.008. Thrombosis and clots in circuit components: OR 0.28 (0.16-0.60), p under 0.001 for survival. Haemolysis NOT independently associated with survival: OR 0.60 (0.31-1.19), p = 0.14, and the authors note the upper limit does not exclude a 19 percent increase in odds. Meconium or neonatal aspiration as primary diagnosis: OR 7.57 (4.02-15.74), p under 0.001. COMPLICATIONS, UNIVARIATE, AND THIS IS THE COHERENT PART: centrifugal pumps had FEWER mechanical complications overall (85/340 versus 95/272; OR 0.62, 0.44-0.88, p = 0.008) and FEWER circuit thromboses (30/340 versus 48/272; OR 0.45, 0.28-0.73, p = 0.001), but MORE haemolysis defined as plasma free haemoglobin over 50 mg/dL (52/340 versus 12/272; OR 3.79, 2.07-7.49, p under 0.001). CO-TREATMENT DIFFERED MARKEDLY: roller patients received more neuromuscular blockade (OR 1.73), more systemic steroids (OR 1.98), more dopamine, dobutamine, alprostadil and surfactant, and far less high-frequency ventilation (OR 0.25, 0.14-0.44); centrifugal patients received more adrenaline, noradrenaline, sildenafil and epoprostenol. CONTEXT THE AUTHORS SUPPLY: two earlier ELSO analyses found the same direction — 8,366 patients under 10 kg (2011-2016, propensity-matched) with higher mortality and complications on centrifugal, and 14,026 patients under 18 (2010-2015) with more kidney injury, hyperbilirubinaemia, circuit thrombosis and haemolysis — but both pooled indications and support modes. Several high-volume neonatal and paediatric centres reverted to roller pumps within months of transitioning to centrifugal. ENGINEERING STATEMENT USED THROUGHOUT CHAPTER 33: the roller pump is occlusive and overcomes downstream resistance to maintain the same flow regardless of downstream pressure; the centrifugal pump is non-occlusive and its flow depends on rotation speed AND downstream resistance.

Conventional roller blood pump

Centrifugal blood pump

Carry the complication trade-off, which is mechanistically secure and matches what Chapter 31 found in a single pump head: a centrifugal pump clots less and grinds more. That should change what is monitored on a centrifugal circuit — expect haemolysis, watch the free haemoglobin, and be most vigilant at low flow where residence time in the head is longest — rather than what is purchased. Do NOT change adult practice on the survival signal: it is unadjusted-null, adjusted-significant, probably over-adjusted, confounded by centre on the authors' own account, and entirely neonatal. In neonatal and small-infant practice treat the question as genuinely unresolved in both directions.

DOI 10.1097/PCC.0000000000003251; PMID 37102713; PMC11069414

53 explanted ECMO circuits from 46 paediatric patients, single centre. Tubing, pump and oxygenator examined for visible thrombi, with representative samples taken for histological, immunofluorescence and immunohistochemical analysis.

ABSTRACT READ ONLY; full text not retrieved. Forty-six patients, all paediatric, single centre. Proportions are of circuits examined rather than of patients supported, so they are not incidence figures. No exposure denominators are given for the antifibrinolytic association, which is therefore hypothesis-generating only. No adult equivalent series was retrieved.

33, 31, 32, 37, 42

THROMBI WERE FOUND IN 81 PERCENT OF CIRCUITS. By site, as proportions of circuits: pump axle thrombi with fragments trapped in the oxygenator 45 percent (classed by the authors among the LESS clinically significant); arterial tubing 30 percent; venous tubing 26 percent; connectors 26 percent; deep oxygenator membrane 15 percent; INFLOW OXYGENATOR MEMBRANE SURFACE 11 percent. The most clinically significant were the inflow oxygenator membrane surface thrombi, the arterial tubing thrombi, and the venous tubing and connector thrombi. INFLOW SURFACE THROMBI PRODUCED A RAPIDLY INCREASING PRESSURE DROP ACROSS THE OXYGENATOR OVER 1 TO 2 DAYS, OXYGENATOR FAILURE AND CIRCUIT REPLACEMENT, and were associated with intravascular venous thrombosis and bacterial infection present before ECMO began. IN 11 PERCENT OF CASES A VENOUS TUBING THROMBUS GREW LARGE ENOUGH TO BREAK OFF AND EMBOLISE TO THE PUMP, RESULTING IN INCREASED HAEMOLYSIS. In one case an arterial cannula or tubing thrombus led to aortic and mesenteric artery thrombosis followed by bowel infarction. SAFETY SIGNAL: antifibrinolytic therapy during ECMO was associated with an increased risk of pump thromboembolism. The authors conclude that examining circuits after removal is a useful quality-improvement tool that can elucidate the cause of circuit problems, identify patients at increased thrombotic risk, and suggest areas for improvement.

Not applicable; descriptive with within-series associations

Systematic post-removal examination of the complete circuit — the examination Chapter 31 section 31.8 asks programmes to perform routinely

The pump is not only a place where clot forms but the place clot ARRIVES: a thrombus that grew in the venous limb or at a connector can break off and land in the head, where it is ground rather than passed to the patient. That reframes a rising plasma free haemoglobin — it may be reporting a clot nowhere near the pump. A RAPIDLY rising transmembrane pressure drop over 1 to 2 days points specifically at inflow membrane surface thrombus, which distinguishes it from the slower climb of diffuse fibre deposition. And the antifibrinolytic signal sits beside Chapter 31's finding that 2 of 4 patients given activated factor VII developed acute oxygenator thrombosis, and the protamine hazard: three prohaemostatic interventions, three circuit-thrombosis signals, none quantified. Finally, this series is the demonstration of what Chapter 31 section 31.8 asks for — examining the explanted circuit produces data that exist nowhere else.

DOI 10.1097/MAT.0000000000001605; PMID 34860711

282 adults receiving venovenous or venoarterial ECMO at a single centre between January 2020 and December 2023

ABSTRACT READ ONLY; full text not retrieved. Single centre and retrospective; propensity weighting does not remove confounding by indication in device selection. THE AUTHORS STATE THEY HAD NO DIRECT MEASUREMENT OF GAS TRANSFER and call for prospective multicentre work incorporating direct measurements — so this is an absence of a detectable difference at the level of the patient, not a demonstration of equivalence at the level of the membrane. NOTE ADDED BY THIS BOOK AND NOT STATED BY THE AUTHORS: measured oxygen transfer across a membrane falls when the inlet blood arrives better oxygenated, so both the decline over the course of support (patients recover and their venous saturation rises) and the venovenous-versus-venoarterial difference are partly predicted by that arithmetic alone and cannot be read straightforwardly as device or configuration effects.

32, 3

No statistically significant difference in oxygen transfer across the membrane between oxygenator types after adjustment for covariates. OXYGEN TRANSFER DECLINED OVER TIME DURING SUPPORT (p under 0.001) and was higher on venovenous than on venoarterial ECMO (p equal to 0.002). A higher Charlson Comorbidity Index was associated with lower oxygen transfer (p equal to 0.046). Coagulation and haemolysis markers were comparable across all devices. The authors conclude that current-generation oxygenators provide similar gas exchange efficiency and haemocompatibility in clinical practice.

Head-to-head between the four devices, with propensity weighting and adjustment for APACHE II and baseline Charlson Comorbidity Index

Four contemporary adult oxygenators: Medtronic Nautilus and Nautilus Smart, Getinge Heart Lung Support Module Advanced 7.0, Getinge Quadrox-iD Adult, and Fresenius X-Lung

A decline in measured oxygen transfer over a run is the EXPECTED trajectory rather than a finding, so it is not on its own a trigger for exchange. And within the current generation of adult devices there is no evidence that the choice of oxygenator changes gas exchange or haemocompatibility at the level of the patient, which means device selection should turn on local familiarity, circuit compatibility and supply rather than on claimed performance differences.

DOI 10.1097/MAT.0000000000002805; PMID 42574647

Ex vivo only — no patients. A closed blood-primed circuit with two adult Maquet Quadrox-iD oxygenators in series on a Maquet PLS Rotaflow pump; the first oxygenator used to set the carbon dioxide tension delivered to the second, sweep flows of 100 percent oxygen varied to give different ratios of gas flow to blood flow.

ABSTRACT READ ONLY; full text not retrieved. Ex vivo, a single device, two oxygenators in series in a closed circuit — no patients and no clinical endpoint. The ratio of 4.0 is a property of this oxygenator in this configuration and should not be transferred to another device as a number; what transfers is the existence of a ceiling and the mechanism (rising dead space fraction) behind it. Published in a journal supplement.

32, 84, 10, 26

Carbon dioxide transfer rose with gas flow and PLATEAUED above a gas-to-blood flow ratio of about 4.0. Exhaust carbon dioxide rose linearly with the carbon dioxide tension delivered to the device, with the slope of that line falling at high ratios. OXYGENATOR DEAD SPACE FRACTION WAS 0.3 TO 0.4 AT LOWER RATIOS AND ROSE TO 0.8 TO 0.9 AT RATIOS ABOVE 4.0. The authors conclude that carbon dioxide clearance within this oxygenator is limited at high ratios and that the limitation correlates with a rising dead space fraction, with implications for patients requiring high levels of support.

Within-circuit comparison across ratios

Systematic variation of the gas-flow-to-blood-flow ratio and of the pre-oxygenator carbon dioxide tension

There is a ceiling on what the sweep can buy, and it arrives earliest at low blood flow — at 2 L/min of blood flow a gas-to-blood ratio of four is a sweep of 8 L/min, which many units reach during weaning or in a carbon dioxide removal configuration. Past that ceiling the knob still turns and the flowmeter still rises while nothing further is cleared, and the team discovers this at the worst possible moment. The ratio is what matters and it is displayed nowhere, so it has to be calculated from the sweep and the blood flow recorded on the same line. When the ratio approaches four, escalate the enquiry rather than the knob, and consider raising the blood flow instead where the patient allows it.

DOI 10.1177/0267659120906767; PMID 32397880

Two components: an unsystematic literature review of ECMO circuit monitoring (no inclusion criteria or quality assessment framework applied, by the authors' own statement), and a customer-discovery interview study of 45 stakeholders in the United States and Canada conducted September to October 2023

The literature component applied no formal inclusion or exclusion criteria and no quality assessment framework, by the authors' own statement. THE INTERVIEW COMPONENT IS A MARKET-DISCOVERY INSTRUMENT RATHER THAN A CLINICAL PRACTICE SURVEY: of 45 stakeholders, 15 were from sales and business and a further 10 were scientists or biomedical engineers, leaving 20 clinicians of any kind; recruitment was largely by web search and referral with 58 percent from a single US state. The ELSO recommendations are reproduced here at second hand and the guideline text itself was not retrieved. Its pain points should be read as hypotheses about what clinicians find difficult rather than as evidence of what they do.

32, 30, 31, 87, 88, 89, 92

Summarises ELSO gas-exchange monitoring recommendations as: continual arterial and venous blood gas monitoring; pre-membrane and post-membrane blood gas analysis daily; regular checking of the membrane lung pressure gradient; monitoring of blood flow; maintaining adequate haemoglobin; and monitoring lactate. THE KEY STATEMENT: no unified protocol exists specifying how these parameters should be integrated or recorded to characterise overall circuit gas-exchange efficiency. Measurement of oxygen and carbon dioxide concentrations within the sweep gas helps determine uptake and clearance and is available on some devices, but MANY MEMBRANE LUNGS LACK INTEGRATED SENSORS FOR POST-MEMBRANE OXYGEN PARTIAL PRESSURE AND SATURATION. Two published continuous gas-exchange monitoring approaches — an algorithmic one and a sensor-based one using two oximetry probes, an ultrasonic flow meter and infrared exhaust carbon dioxide — are described as not widely adopted because of modelling complexity, extra sensors and workflow interruption. Console flow and pressure data are typically recorded manually only once or twice an hour into the record. OPERATIONAL FIGURE: clinicians described preparation for a membrane lung swap taking about 15 to 30 minutes or longer, while the swap itself typically takes under 2 minutes, during which the patient loses ECMO support. Haemolysis and debris in the membrane lung are currently assessed visually with a flashlight, which interviewees described as not always accurate. The authors note it is not conclusive whether current clinical decision support systems improve patient outcomes, and list skill atrophy, mode confusion and automation bias as risks.

Not applicable

Not applicable — appraisal of current monitoring practice and of the case for an automated clinical decision support system

The field has a list of what to measure and no protocol for what to do with it, so any integrated reading of membrane lung function is assembled locally and should be written down as such. The measurement most likely to give early warning — continuous exhaust carbon dioxide — exists on some devices and has not been prospectively evaluated. And the asymmetry between a 15 to 30 minute preparation and a sub-2-minute swap is the practical reason surveillance matters: it does not shorten the procedure, it moves it into working hours.

DOI 10.1051/ject/2026022; PMID 42743428; PMC13577623

Five domestic female pigs (80 kg) on venovenous ECMO with a polymethylpentene membrane (PLS Quadrox D, Bioline coating), studied before and after induction of multiple organ failure by saline lung lavage and faecal peritonitis. 45 paired measurements in each condition. Predefined sequence of blood flows (0.5, 1.5, 3.0 L/min) with sweep flows giving gas-to-blood ratios of 1:2, 1:1 and 2:1. Carbon dioxide transfer measured by the reference method, from the gas exhaled by the membrane.

Five animals, one device, one coating; the coefficients are not transferable as numbers to a human circuit and only the rank order and direction of the determinants should be carried across. A higher measured oxygen transfer in the device does not mean more oxygen reaching tissues, as the authors state. Low animal numbers could produce type II errors for the variables that dropped out. THE AUTHORS PRINT ONE COEFFICIENT THEY CANNOT EXPLAIN: higher oxygen transfer at LOWER pre-membrane carbon dioxide runs against the expected direction because a low carbon dioxide raises haemoglobin's oxygen affinity; they say plainly they were unable to bring forth a physiological explanation, offer a speculative one, and concede they cannot exclude residual confounding with blood flow.

32, 27, 6, 9, 84

OXYGEN TRANSFER (mixed generalised linear model, backward elimination, pseudo-R squared 0.81) had three independent determinants: blood flow (beta plus 0.066 mL/min per mL/min, p under 0.001), pre-membrane oxygen saturation (beta MINUS 1.669 mL/min per percent, p under 0.001 — better blood arriving means LESS measured transfer), and pre-membrane carbon dioxide tension (beta minus 0.964 per mmHg, p equal to 0.001). SWEEP GAS FLOW, TEMPERATURE AND HAEMOGLOBIN ALL DROPPED OUT OF THE OXYGEN MODEL, and stratified by sweep-to-blood ratio the oxygen transmembrane gradient and the oxygen transfer were the same at every ratio tested — no effect of gas flow on oxygenation within the range studied. CARBON DIOXIDE TRANSFER (pseudo-R squared 0.89) had four independent determinants: sweep gas flow (plus 32.910 mL/min per L/min, p under 0.001), blood flow (plus 0.017 per mL/min, p under 0.001), pre-membrane carbon dioxide tension (plus 1.192 per mmHg, p under 0.001) and haemoglobin (minus 3.478 per g/dL, p equal to 0.042); temperature and oxygen saturation dropped out. THE EFFECT OF BLOOD FLOW ON CARBON DIOXIDE TRANSFER WAS ROUGHLY 1.9 TIMES THAT OF GAS FLOW. A 10 mmHg rise in pre-membrane carbon dioxide raised transfer by about as much as an extra 0.5 L/min of sweep, which the authors read as support for permissive hypercapnia. Across the blood flow range oxygen transfer rose about six-fold from 0.5 to 3.0 L/min while carbon dioxide transfer rose about two and a half fold. POST-MEMBRANE VALUES IN A CLEAN DEVICE: oxygen tension 422 (341-463) mmHg at baseline and 354 (188-441) mmHg in multiple organ failure; pH rose 0.09 (0.07-0.13) at baseline and 0.21 (0.13-0.25) in MOF; carbon dioxide fell 10 and 28 mmHg respectively; ionised calcium fell 0.03 mEq/L and chloride rose 2 mEq/L, both statistically significant and clinically trivial; potassium and lactate essentially unchanged.

Within-animal comparison across settings, and healthy versus multiple organ failure state

Systematic variation of blood flow and sweep gas flow

Blood flow is the oxygen knob and sweep gas flow is the carbon dioxide knob — but blood flow is ALSO a carbon dioxide knob, with about 1.9 times the effect of gas flow over the ranges studied, which means every intervention that lowers blood flow lowers carbon dioxide clearance too (a weaning trial, a drainage problem, a reduction made for another reason). Sweep gas flow does nothing at all for oxygen transfer, which is the same conclusion Chapter 27 reached from the saturation side. And the negative coefficient on pre-membrane saturation is the arithmetic behind the denominator problem: a measured oxygen transfer falls when the patient improves, in a device that has not changed.

DOI 10.1371/journal.pone.0054954; PMID 23383011; PMC3558498

All 45 consecutive adults supported with venovenous ECMO for at least 48 hours, single Dutch tertiary centre, October 2018 to March 2022. Median ECMO duration 16 days; 28 of 45 (62 percent) discharged alive; 23 of 45 (51 percent) COVID-19 ARDS. Cardiohelp with the HLS advanced 7.0 set throughout; unfractionated heparin to aPTT 50-70 s; oxygenator inspected twice daily by a perfusionist and an intensivist; weaning by lowering the sweep rather than the blood flow; no strict exchange protocol.

Forty-five patients and 29 events; retrospective and single centre; a single oxygenator brand throughout, so results may not transfer to other devices; over half the cohort had COVID-19 ARDS with its own prothrombotic phenotype. THE AUTHORS' OWN CENTRAL LIMITATION: the variables tested were the same variables clinicians used to decide on an exchange, so there is a risk of a self-fulfilling prophecy; they note there were no emergency exchanges due to an acute shutdown of the system, and that the indication for exchange was not generally recorded so emergency exchanges during office hours could be missed. The emergency-exchange finding rests on 10 events with a confidence interval touching 1.00. No threshold is offered for any parameter and the authors explicitly call for research to define cut-off values.

32, 31, 33, 37, 89

19 of 45 patients (42 percent) had at least one oxygenator exchange, on median day 12; 10 (22 percent) had a second on median day 9 of that circuit; 29 exchanges in total. 10 of 29 (34 percent) were emergency exchanges. THE MOST PREVALENT INDICATION FOR BOTH EMERGENCY AND ELECTIVE EXCHANGE WAS A RISING TRANSMEMBRANE PRESSURE DROP. Independent risk factors for exchange (backward stepwise Cox regression): higher arterial carbon dioxide tension with correspondingly lower pH, higher transmembrane pressure drop, and higher haemoglobin; Harrell c-statistic 0.77 for the combined linear predictor. Univariately associated: higher Hb, plasma free haemoglobin, platelets, pressure drop, PaCO2, lower pH, SOFA score and COVID-19 as indication. EMERGENCY EXCHANGE: lower lactate dehydrogenase was the only independent risk factor, OR 0.90 (0.82-1.00), borderline, on 10 events; both LDH and CRP were LOWER before emergency than before elective exchanges. NEGATIVE FINDINGS THAT MATTER: the percentage change in pressure drop adjusted for blood flow over the 12 and 24 hours before the exchange showed NO correlation with the exchange, and a sensitivity analysis substituting the pressure-drop-to-blood-flow ratio for the raw pressure drop gave similar results throughout. D-dimer was higher in the exchange group but was not associated with exchange in the regression. Plasma free haemoglobin was associated univariately but did not survive, and the authors attribute haemolysis more plausibly to the pump than to the membrane. Oxygenator exchange had no impact on ICU survival. The authors describe deteriorating gas exchange as a LATE marker of functional decline, and cite an in-vitro study finding alteration of the carbon dioxide concentration at the oxygenator OUTLET to be an early indicator, a measurement they note is not generally performed.

Patients with no oxygenator exchange; and elective versus emergency exchanges

Oxygenator exchange, classified as elective or emergency by TIMING (outside the 10 a.m. to 5 p.m. window) rather than by indication

Expect an oxygenator exchange in roughly four in ten venovenous runs and expect about a third of them to fall outside working hours — which is the entire argument for surveillance, since the swap itself takes under two minutes but its preparation takes 15 to 30 minutes or more. The carbon dioxide channel carries independent information about membrane decline, so record the sweep, the blood flow and the arterial carbon dioxide on the same line. Do NOT expect the pressure drop to give a day's notice: the one study that tested the short window found nothing. And treat every operating characteristic quoted for these parameters as partly a description of local practice until a study evaluates them against an outcome determined independently of them.

DOI 10.1177/02676591231168642; PMID 36995209; PMC11191661

265 consecutive adults on venovenous ECMO, single centre (Regensburg), January 2009 to December 2013. Retrospective analysis of prospectively collected data. Age 50 (36-60); SOFA 11.0 (8.0-14.3); Murray score 3.33 (3.33-3.67); primary lung failure 68 percent. Cumulative support 3,411 days, median 9 (6-15) days.

Single centre, venovenous only, 2009-2013, retrospective. Six different ECMO systems with differing pumps and coatings but an identical polymethylpentene gas-exchange membrane throughout, so the study cannot speak to membrane material. The 31 percent exchange rate reflects that centre's thresholds, anticoagulation protocol and case mix and is not transferable. D-dimer reported in mg/dL by a local assay, so the absolute numbers do not transfer between laboratories. The mortality comparison is not randomised and not powered — p = 0.113 with a 10-point absolute difference is not evidence of equivalence. Before-and-after design for the reversal findings is uncontrolled. No routine circuit-infection surveillance was performed, so 4 of 83 is a detection rate rather than an incidence. The activated factor VII signal rests on 4 patients. No cut-off is offered for any parameter.

31, 32, 36, 37, 39, 42, 58

83 of 265 (31 percent) required at least one system exchange: 37 acute, 46 elective. Reasons summing to 83 — mechanical failure 8 (pump 5, membrane lung 2, cannula 1); pump head thrombosis 13; acute oxygenator thrombosis 16; worsening gas transfer 10; device-related coagulation disorder 32 (local fibrinolysis 24, device-induced hyperfibrinolysis 8); suspected circuit infection 4. PUMP HEAD THROMBOSIS: plasma free haemoglobin rose about 10-fold before and fell about 6-fold after exchange (p under 0.001 both); clot documented in every pump head removed; a warning rise on the preceding day in only 6 of 13 (55 [49-60] to 143 [117-266] mg/dL, p = 0.031). ACUTE OXYGENATOR THROMBOSIS: dpMO divided by blood flow rose 2.6-fold (2.1-4.0) before and fell 3.2-fold (2.0-5.2) after exchange (p under 0.001 both), and did not rise in any other acutely exchanged patient; D-dimer elevated beforehand in 11 of 16, with 5 of 16 missed by routine D-dimer testing; gas transfer already worsened in 8 of 11 (53 percent [45-59] fall in outlet pO2; outlet pCO2 40 [32-48] mmHg at gas flows of 10 L/min or more). DEVICE-INDUCED HYPERFIBRINOLYSIS (n = 8): fibrinogen fell below 200 mg/dL with rising D-dimer over 4 days, diffuse unaccountable bleeding, reversible by system exchange, and gas exchange of the respective membrane lungs only minimally affected. LOCAL FIBRINOLYSIS (n = 24): rising D-dimer and falling platelets over 3-4 days with fibrinogen unchanged, yet 40 percent bled. OUTCOME: exchanged patients received a median of 2 (2-3) membrane lungs, range up to 5, with longer support (20 [14-27] versus 8 [6-10] days) and more renal replacement (28 versus 15 percent, p = 0.028), but NO significant mortality difference — weaning 69 versus 79 percent, died on ECMO 31 versus 21 percent, p = 0.113; acute versus elective died on ECMO 17 versus 29 percent, p = 0.320. Acutely exchanged circuits had a shorter first-membrane-lung life (7.0 [5.0-10.3] versus 9.5 [7.0-13.0] days, p = 0.023). SAFETY SIGNAL: 2 of the 4 patients given activated coagulation factor VII developed acute oxygenator thrombosis. INFECTION: 4 exchanges for suspected circuit infection; mycobacteria identified microscopically in one membrane lung, Enterococcus grown from another on long-term ex-vivo incubation; noradrenaline requirement fell 15-fold in one patient after exchange.

Patients not requiring a system exchange, at comparable support times

ECMO system exchange (whole circuit except cannulae), acute or elective

Run a five-channel surveillance system (visual circuit check, dpMO normalised to blood flow, post-membrane gases, the haemolysis panel, the coagulation panel), read for trend rather than value, and accept that it converts about half of acute clot events into planned procedures rather than abolishing them. Read the pressure drop as a ratio to blood flow, not as a raw number. Put the circuit near the top of the differential for a diffusely bleeding ECMO patient with a rising D-dimer and falling platelets over 3-4 days — a normal fibrinogen does not exclude it and normal gas exchange does not exclude it. Do not require the circuit to be failing before suspecting it. A programme that exchanged circuits in a third of its patients, sometimes up to five times, did not visibly pay for it in survival.

DOI 10.1371/journal.pone.0112316; PMID 25464516; PMC4251903

470 ECMO-treated intensive care patients in the Tokushukai Medical Database (2018-2022) covering 71 hospitals; 357 included for model development

The outcome is a treatment decision, not a physiological event, so the model learns the transfusion practice of the contributing hospitals — and its top features are precisely the variables a transfusion protocol uses. Retrospective; no external validation. Abstract only.

30, 46, 36

LightGBM and random forest performed best, with ROC AUC above 0.7 for both (accuracy 70.5 percent, ROC AUC 0.703, recall 0.784; and ROC AUC 0.705). SHAP identified circulating blood volume, haemoglobin and weight as the most important predictive factors.

Between algorithms

Machine learning models (LightGBM, random forest, extreme gradient boosting, support vector classification) using 47 variables, with red cell transfusion events as a surrogate marker for bleeding; SHAP analysis for feature importance

For Chapter 30: if a model's outcome is a treatment decision, the model has learned local practice and will not transfer. Chapter 46 owns transfusion strategy and should carry this caution.

Kamio T, Ikegami M, Mizuno M, Ishii S, Tajima H, Machida Y, Fukaguchi K. Transfusion 2025;65(6):1051-1060. DOI 10.1111/trf.18261

Participants identified from the International ECMO Network and ELSO, including consumers, multidisciplinary clinicians, researchers and industry partners, from 15 countries across five continents

Abstract only; the substantive content — which instruments, measured when — was not retrievable. Consensus rather than outcome evidence.

30, 23, 88

Response rates 46 of 65 (71 percent) and 40 of 46 (87 percent). Seven measurement tools were identified for the core outcome set of patients on ECMO. [VERIFICATION REQUIRED: the seven tools and their measurement timings are NOT named in the abstract and are not reproduced in this book.]

Candidate tools identified from a systematic review of the literature and clinical trial registrations

Two-round international web-based modified Delphi to select measurement tools and timing of measurement for the previously published core outcome set

A standardised instrument set exists, so a programme recording functional outcome at discharge does not need to invent one. Its absence is a named limitation in the ECPR registry literature. Recorded here as a retrieval target rather than as usable content.

Hodgson CL, Fulcher B, Mariajoseph FP, Burrell AJC, Pellegrino V, Brodie D, Fan E. Crit Care Med 2021;49(12):e1252-e1254. DOI 10.1097/CCM.0000000000005110

652 adults on venovenous ECMO in 41 centres, December 2018 to February 2021; 8,471 ECMO days analysed

Observational; practice varied by centre and was not protocolised. Abstract only.

30, 42, 36

Unfractionated heparin was the initial anticoagulant in 77 percent and the most frequently used during the course (6,221 days; 73 percent). aPTT was the monitoring test on 86 percent of days; median 52 s (IQR 39-61), falling by 5.3 s (95 percent CI -7.4 to -3.2, p under 0.01) after the first bleeding event. Bleeding occurred on 1,202 days (16.5 percent); 342 patients (52.5 percent) had at least one bleeding event, one episode every 215 hours on ECMO, of which 10 (1.6 percent) were fatal. Higher aPTT was a potentially modifiable risk factor for a first bleed (HR 1.07 per 20-s increase). The authors describe anticoagulation as a dynamic process, frequently stopped for bleeding and restarted according to the clinical picture.

Within-cohort; before and after a first bleeding event

Anticoagulation recorded DAILY by type, dosage and monitoring strategy, with bleeding events reported by site, severity and impact on mortality

For Chapter 30, this is the demonstration of what granularity buys: a registry field recording which anticoagulant was used cannot represent a practice that changes daily, and only daily data showed that it does. Chapter 42 owns the anticoagulation content.

Martucci G, Giani M, Schmidt M, Tanaka K, Tabatabai A, Tuzzolino F, Agerstrand C, Riera J, Ramanan R, Grasselli G, Ait Hssain A, Gannon WD, Buabbas S, Gorjup V, Trethowan B, Rizzo M, Fanelli V, Jeon K, De Pascale G, Combes A, Ranieri MV, Duburcq T, Foti G, Chico JI, Balik M, Broman LM, Schellongowski P, Buscher H, Lorusso R, Brodie D, Arcadipane A. Am J Respir Crit Care Med 2024;209(4):417-426. DOI 10.1164/rccm.202305-0896OC

Neonates (0-28 days), children (29 days to 17 years) and adults (18 years and over) supported with ECMO from 2009 through 2022 and reported to the ELSO Registry. 154,568 ECMO runs; 780 centres submitted over the period, 557 in 2022.

Voluntary reporting by fee-paying member centres. Descriptive statistics are complete-case, so every reported percentage carries its own unstated denominator and denominators differ between variables in the same table. Abstract only.

30, 1

Median annual adult ECMO runs per centre rose from 4 to 15 while paediatric and neonatal runs per centre fell from 12 to 7. Over 50 percent of patients were transferred to the reporting centre, and 20 percent of those were transported on ECMO. Prone positioning before respiratory ECMO in adults rose from 15 percent (2019) to 44 percent (2021). CRITICAL METHODS LINE for Chapter 30: 'Missing values were excluded before calculating descriptive statistics.'

Year-on-year and by age group

Descriptive report of programme characteristics, processes of ECMO care, adjunctive therapies, support modes, treatments, complications and survival

Registry descriptive percentages must be read as complete-case proportions; derived figures must not be computed by multiplying two of them together. This is the fourth level of Chapter 30's measurement pyramid.

Tonna JE, Boonstra PS, MacLaren G, Paden M, Brodie D, Anders M, Hoskote A, Ramanathan K, Hyslop R, Fanning JJ, Rycus P, Stead C, Barrett NA, Mueller T, Gomez RD, Malhotra Kapoor P, Fraser JF, Bartlett RH, Alexander PMA, Barbaro RP. ASAIO J 2024;70(2):131-143. DOI 10.1097/MAT.0000000000002128

119 adults (18 years and over) on venovenous ECMO at Johns Hopkins Hospital 2016-2024; 59.6 percent female; 62 (52.1 percent) mortality at discharge

119 patients, 62 events, 80/20 split, four algorithm families compared — roughly 12 events in the test set. No external validation. The abstract presents a WEANING model but reports only mortality at discharge as its outcome statistic, so the endpoint definition is not recoverable [VERIFICATION REQUIRED]. The abstract also prints 'median age = 48%'. Abstract only.

30, 18, 4

AUROC by model: RESP alone 0.815; pre-ECMO 0.715; first 24 hours 0.515; last 24 hours before decannulation 0.881; full ECMO run 0.904. The authors conclude high-granularity models outperform RESP-based models. READ BY DECISION WINDOW the conclusion reverses: the two best-performing models require the outcome to have already occurred, while the two windows in which a prediction would be wanted gave 0.715 and 0.515 against a 4-variable pre-cannulation score at 0.815.

A model built on the RESP score alone

Machine learning models (Random Forest, CatBoost, AdaBoost, XGBoost) built on EMR time-series across four windows: pre-ECMO, first 24 h, last 24 h before decannulation, and the full run. 80/20 train-test splits; SHAP feature importance.

Judge a prediction model by its performance in the window in which the decision is taken, not by its best reported figure. A model needing the last 24 hours before decannulation to predict decannulation is describing, not predicting.

Sivakumar S, Fan L, Cao M, Brown P, Kim BS, Whitman GJ, Cho S, et al. Artif Organs 2026. DOI 10.1111/aor.70198

Adults receiving ECPR for in-hospital or out-of-hospital cardiac arrest reported to the ELSO Registry 1 Jan 2020 to 31 Dec 2023, first run only. 7,911 ECPR runs reported; 3,736 (47.2 percent) had a valid ECPR addendum; 3,215 analysed. Median centre volume 7 (IQR 3-13); median age 57 (45-66); 68.4 percent male; 44.1 percent shockable; median conventional CPR 44 min (22-58); median ECMO 67 h (21-136).

Observational; residual unmeasured confounding. The registry does NOT record the degree of cardiac dysfunction, the indication for unloading, the timing or duration of unloading, or the timing and indication for PCI. Very few centres sufficiently recorded neurological outcomes at discharge. Immortal time bias is named by the authors: patients must survive long enough to receive the device. Two internal inconsistencies as printed: '1,876 patients (74 percent) had an IHCA' in a cohort of 3,215 (division gives 58.4 percent), and the INCEPTION CI is identical to the primary analysis CI while the point estimate differs. Both reproduced, neither corrected. [VERIFICATION REQUIRED for both.]

30, 15, 22, 19, 20, 21

No difference in 90-day survival time: HR 0.92 (0.79-1.08). Concordant across IPTW 0.92 (0.82-1.03), complete case 0.99 (0.86-1.15), re-matched 0.92 (0.80-1.06), double-adjusted 0.94 (0.81-1.09), excluding contraindications 0.93 (0.78-1.11), adding PCI 0.94 (0.81-1.09); and across emulated ARREST 0.89 (0.75-1.07), PRAGUE-OHCA 0.92 (0.77-1.10) and INCEPTION 1.05 (0.79-1.08 as printed) eligibility. No difference in favourable functional outcome, OR 1.15 (0.67-1.99). HARMS: renal OR 1.55 (1.16-2.07); cardiovascular 1.60 (1.14-2.26); CNS bleeding 1.75 (1.03-2.96); arrhythmias 1.56 (1.04-2.36); haemolysis 1.85 (1.10-3.09). ECMO duration longer by 23.3 h (8.3-38.4). POST-HOC META-ANALYSIS in the same paper, pooling five other propensity-matched studies (six cohorts, no RCTs) with this one, gave 30-day mortality OR 0.55 (0.35-0.87), graded LOW certainty; IABP 0.66 (0.37-1.17) and mLVAD 0.51 (0.25-1.04) both very low.

ECPR without mechanical LV unloading. 621 matched pairs.

Mechanical LV unloading during the ECPR run (intra-aortic balloon pump or microaxial LVAD), identified by CPT codes. Target trial emulation with confounders selected by a causal directed acyclic graph, multiple imputation with chained equations for missing data, and propensity-score matching.

Unloading during ECPR has not been shown to improve survival and is associated with more complications. Methodologically this is Chapter 30's anchor: where two analyses of the same data disagree, ask which one modelled time. And the 47.2 percent addendum completion rate is the single most consequential denominator in the book's ECPR evidence.

Ling RR, Chen Y, Low CJW, Agerstrand C, Jung JS, Lim SL, Lorusso R, Mueller T, Okada Y, Tonna JE, Shekar K, Brodie D, MacLaren G, Ramanathan K. Crit Care 2025;29(1):186. DOI 10.1186/s13054-025-05345-3. PMC12063446. READ AS COMPLETE FULL TEXT.

Patients with acquired von Willebrand syndrome from high-shear states — extracorporeal support, left ventricular assist devices, aortic stenosis

Narrative review; not ECMO-specific; no ECMO-specific thresholds or management data

29, 36, 39

Places extracorporeal support in the same mechanistic class as aortic stenosis and ventricular assist devices: increased proteolysis through shear stress followed by ADAMTS13, with frequent and recurrent mucosal bleeding. Under high shear the multimer unfolds and exposes the ADAMTS13 cleavage site, so the largest and most haemostatically potent multimers are removed first.

Congenital von Willebrand disease and other acquired bleeding disorders

Review of mechanism, diagnosis and management

Explains why the quantity of von Willebrand factor can be normal or high while function is severely impaired: what is lost is a size distribution, not a concentration — and an antigen level cannot see a size distribution.

Tiede A, Susen S, Lisman T. Haemophilia 2024;30(S3):29-38. DOI 10.1111/hae.14995

23 children on ECMO; 191 samples taken before support, daily to day 14, and 12 to 24 hours after decannulation

Conference abstract only; paediatric; 23 patients; single centre. [VERIFICATION REQUIRED: whether published in full.]

29, 36, 39

18 of 23 (78 percent) had at least one collagen-binding-to-antigen ratio below 0.6; 8 (35 percent) below 0.4; 3 (13 percent) below 0.2. The ratio fell over the course of support and NORMALISED 12 to 24 hours after decannulation. Multimer loss was seen during the first 3 days. The abstract records that bleeding complications occur in 50 to 70 percent of paediatric ECMO patients and that the aetiology is poorly understood.

Each child's own pre-ECMO and post-decannulation values

Serial von Willebrand collagen-binding-to-antigen ratio, with multimer patterns in three patients

Acquired von Willebrand syndrome develops over days and reverses fast after decannulation — worth knowing when planning a procedure, and the strongest available evidence on its time course.

Drop J, Wildschut E, de Hoog M, de Maat M, van Ommen CH. Abstract, Res Pract Thromb Haemost 2021;5(S2). DOI 10.1002/rth2.12591

Neonatal and paediatric ECMO

Narrative review; neonatal and paediatric; the pump-type direction is reported rather than independently analysed here

29, 31, 68

ELSO data indicate a significant INCREASE in intravascular haemolysis in neonates and children with centrifugal pumps compared with their use in adults — the opposite of the adult roller-versus-centrifugal direction. Stated risk factors: small cannula size, high negative inlet pressure in the pump head, thrombosis in the pump head and oxygenator. Haemolysis increases renal injury, direct hyperbilirubinaemia and thrombosis WITHOUT an increase in mortality, and is described as an unavoidable side effect of current ECMO technology for which there are no approved treatments or treatment guidelines.

Adult ECMO experience

Narrative review of haemolysis mechanism, risk factors and consequences

The haemolysis-generating variable is not the pump's name or price but the pressure and geometry it works against. And there is no treatment for the finding — what a rising free haemoglobin changes is surveillance and circuit inspection, not a prescription.

Garg M. NeoReviews 2020;21(11):e728-e740. DOI 10.1542/neo.21-11-e728

28 specimens from children on ECMO, plus controls

Conference abstract only; paediatric; single centre; 28 specimens. [VERIFICATION REQUIRED: whether published in full.]

29, 36

24 of 28 specimens were classified as acquired von Willebrand syndrome on the gel. ALL 24 had a high-molecular-weight multimer fraction below 40 percent, but only 46 percent had an activity-to-antigen ratio below 0.7. Among controls, 92 percent had a normal multimer fraction and 89 percent a normal ratio. Multimers below 40 percent gave sensitivity 100 percent, specificity 75 percent, overall accuracy 96 percent. Authors: the activity-to-antigen ratio cannot be a screening test for acquired von Willebrand syndrome.

The conventional von Willebrand activity-to-antigen ratio (cut-off 0.7)

Direct multimer analysis by immunofixation electrophoresis, adjudicated independently by two specialists from the gel

A normal activity-to-antigen ratio does NOT exclude acquired von Willebrand syndrome on ECMO. If the question is being asked because a patient is bleeding, request multimer analysis by name. The direction is supported by mechanism: a ratio cannot see a size distribution.

Abstract, International Journal of Laboratory Hematology 2024;46(S2):3-121 supplement. DOI 10.1111/ijlh.14396

Method-comparison study of a point-of-care free haemoglobin device against laboratory reference methods

Analytical study, not a clinical outcome study; device-specific

29

Within-run imprecision under 3 percent, with acceptable concordance with laboratory methods established FROM 50 mg/dL upward. Agreement below the ELSO threshold was not established.

Established laboratory methods

Point-of-care plasma free haemoglobin measurement

The interesting clinical range is 11 to 50 mg/dL and the assay concordance was established only above it. Before a unit builds a protocol around sub-threshold haemolysis it must know its own assay's performance in that band — a written question for the laboratory, not the perfusion team.

Baud B, Dupuy AM, Zozor S, Badiou S, Bargnoux AS, Mathieu O, Cristol JP. Bioanalysis 2023;16(2):65-74. DOI 10.4155/bio-2023-0116

20 patients undergoing lung transplantation with short-term perioperative ECMO

20 patients, lung transplantation, short-run support — not a general ECMO population

29, 39

The ristocetin aggregometry channel fell from 104.7 plus or minus 44.2 to 39.6 plus or minus 24.6 (p under 0.0001), suggesting severe von Willebrand deficiency. The formal assays said the opposite: antigen ROSE 136.4 to 181.8 percent (p=0.0068), ristocetin cofactor 129.4 to 168.8 (p=0.0088), collagen binding 117.5 to 160.5 (p=0.01); ratios unchanged; multimer index fell only 98.4 to 94.8 percent, which the authors called not clinically significant. No significant correlation between the ristocetin channel and any laboratory von Willebrand test. The ADP channel also fell (71.4 to 45.7, p=0.0004).

Paired pre-support versus post-explantation values

Multiple-electrode aggregometry alongside formal von Willebrand assays (antigen, ristocetin cofactor, collagen binding, multimer index), before support and after explantation

A ristocetin-based point-of-care channel is named after its reagent, not after the abnormality it detects; here it was detecting acquired platelet dysfunction. And short perioperative support did not produce clinically significant multimer loss, which qualifies the prevalence figures from longer runs.

Garaj M, Durila M, Vajter J, Solcova M, Marecek F, Hrachovinova I. Artif Organs 2021;46(5):899-907. DOI 10.1111/aor.14142

35 consecutive adult venovenous ECMO patients at a single centre over almost six years

35 patients, retrospective, single centre. The flow finding is most easily read as confounding by indication: flow falls because something is wrong, and that something is haemolysing the blood.

29, 55, 32

A reduction in ECMO blood flow of 1 L/min was associated with a daily haptoglobin consumption of 93.371 mg/dL (p=0.001) — lower flow tracked with MORE haemolysis. Membrane oxygenator dead space was associated with haptoglobin reduction (B = minus 215.307, p=0.004). CRRT connected to the ECMO circuit was associated with both higher daily free haemoglobin and lower haptoglobin.

Within-cohort regression on circuit and patient variables

Daily plasma free haemoglobin and haptoglobin examined against patient characteristics, daily laboratory findings and circuit variables

Do not lower blood flow to treat a haemolysis number. Look at inlet pressure, drainage, transmembrane gradient and any circuit-connected CRRT instead. The CRRT association is actionable and easily overlooked.

Costantini S, Belliato M, Ferrari F, Gazzaniga G, Ravasi M, Manera M, De Piero ME, Curcelli A, Cardinale A, Lorusso R. Perfusion 2022;38(3):609-621. DOI 10.1177/02676591211073768

501 children on ECMO

Retrospective; paediatric, so extrapolation to adults is indirect

29, 55, 37

Free haemoglobin — described by the authors as the most specific of the haemolysis markers — carried an adjusted HR of 2.9 (1.4 to 5.6) for a composite of acute kidney injury or renal replacement therapy, rising to 6.6 (3.3 to 13.2) for stage 2 or worse injury. LDH gave 3.1 and 2.8 respectively.

Children without elevated haemolysis markers

Elevated plasma free haemoglobin and elevated lactate dehydrogenase as time-varying exposures, with the pre-ECMO creatinine as baseline

Haemolysis identifies real organ-specific harm beyond the neurological signal, and the renal endpoint is the better-powered one. Supports treating sub-threshold haemolysis as a finding rather than noise.

Strong AE, Zee J, Fulchiero R, Kilbaugh TJ, Connelly J, Makeneni S, Campos D, Laskin BL, Denburg MR. Kidney360 2023;4(11):1536-1544. DOI 10.34067/KID.0000000000000253

80 adult ECMO patients in otherwise identical circuits

Retrospective, single centre, non-randomised pump allocation

29, 31

No significant difference in daily plasma haemoglobin (5.7 plus or minus 3.6 versus 5.7 plus or minus 4.2 mg/dL), lactate (2.8 versus 3.0) or LDH (2656 plus or minus 1607 versus 2689 plus or minus 1726) over the first 10 days. Pre-ECMO LDH was 4004 plus or minus 3583 and 3604 plus or minus 3354 — HIGHER than anything recorded on support, in both arms. Authors: the difference in cost of the devices does not correlate with performance and outcomes.

Each other; and on-support values against pre-ECMO values

Two centrifugal pumps differing 20 to 30-fold in cost

Pump price does not predict haemolysis. And the pre-cannulation LDH was the highest value in the whole dataset, which is the second pillar of the pre-cannulation-baseline argument.

Palanzo DA, El-Banayosy A, Stephenson E, Brehm C, Kunselman A, Pae WE. Artif Organs 2013;37(9):E162-E166. DOI 10.1111/aor.12158

21 continuous-flow ventricular assist device patients, 20 ECMO patients, 12 heart failure patients without a device, 7 with aortic stenosis, and healthy donors

Single centre; 20 ECMO patients; cross-sectional

29, 36, 39

Loss of high-molecular-weight multimers, consistent with acquired von Willebrand syndrome, in 14 of 20 ECMO patients and 18 of 19 device patients. Elevated soluble glycoprotein VI with significantly reduced surface glycoprotein Ib-alpha and glycoprotein VI versus healthy donors. Critically, platelet receptor levels were ALSO significantly reduced in the heart failure patients who had no device.

Heart failure patients WITHOUT a device, and healthy donors — the design feature that makes this study useful

Von Willebrand multimer analysis and platelet receptor measurement (soluble and surface glycoprotein VI, surface glycoprotein Ib-alpha)

Most patients on support have acquired von Willebrand syndrome. But an unknown share of the platelet abnormality belongs to the disease that made the circuit necessary, not to the circuit — the same attribution correction the free-haemoglobin data force.

Lukito P, Wong A, Jing J, Arthur JF, Marasco SF, Murphy DA, Bergin PJ, Shaw JA, Collecutt M, Andrews RK, Gardiner EE, Davis AK. J Thromb Haemost 2016;14(11):2253-2260. DOI 10.1111/jth.13497

Adult and paediatric ECLS patients, with in vitro parameter-variation experiments alongside patient samples

Single centre; mixed population; the multi-parameter condition is modelled rather than a clinical outcome study

29, 42

Anti-Xa correlated with heparin r=0.97 and with antithrombin r=0.98, unaffected by other parameters. PT correlated with coagulation factors r=0.88. ACT correlated with coagulation factors r=0.99, was insensitive to heparin below 0.5 U/mL, and correlated with NEITHER heparin nor factors when multiple parameters changed together. Authors describe ACT as too insensitive to heparin, sensitive to too many other factors, and too imprecise to be useful.

Isolated single-parameter change versus simultaneous multi-parameter change

Head-to-head evaluation of ACT, aPTT, anti-Xa and PT against directly measured heparin, antithrombin and coagulation factor levels

The anchor evidence for Chapter 29. Specific assays (anti-Xa, PT) retain their meaning on a circuit; global assays (ACT, aPTT) do not. Never act on an isolated ACT.

Saifee NH, Brogan TV, McMullan DM, Yalon L, Matthews DC, Burke CR, Chandler WL. ASAIO J 2020;66(2):230-237. DOI 10.1097/MAT.0000000000000993

150 adults on venoarterial ECMO at a single centre over almost six years; mean age 55 plus or minus 14; 33 percent women

Single centre, retrospective, VA only. Causation not established: haemolysis may mark a thrombogenic circuit rather than cause the strokes. No interventional data.

29, 37, 28

62 of 150 (41 percent) met the definition. Non-haemorrhagic stroke in 20 of 62 (32 percent) versus 4 of 88 (5 percent); adjusted HR 7.6 (2.2 to 25.9, p=0.001). Graded: 26-50 mg/dL HR 11.3 (3.6 to 35.1); 11-25 mg/dL HR 4.4 (1.36 to 14.85). Two-fold greater rise in mean platelet volume (0.98 plus or minus 1.1 versus 0.49 plus or minus 0.96 fL, p=0.03). Operative mortality with stroke 83 versus 45 percent.

No low-level haemolysis (free haemoglobin below 11 mg/dL)

Low-level haemolysis defined as 48-hour plasma free haemoglobin of 11 to 50 mg/dL

The ELSO 50 mg/dL threshold is a definition of an event, not a level below which nothing is happening. A rising free haemoglobin inside the 11-50 mg/dL band with concordant markers justifies looking for a cause and raising neurological surveillance, not a circuit change.

Saeed O, Jakobleff WA, Forest SJ, et al. Ann Thorac Surg 2019;108(3):756-763. DOI 10.1016/j.athoracsur.2019.03.030

215 adults with cardiac failure on extracorporeal life support, in three groups: cannulated during ongoing CPR (n=110), after ROSC with sustained shock (n=45), severe shock without preceding resuscitation (n=60)

Retrospective, single centre, cardiac population; not designed to test attribution

29, 14, 22

Pre-cannulation free haemoglobin 318 (138-586) versus 212 (107-439) versus 79 (53-232) mg/L across the three groups, p under 0.001. Free haemoglobin declined significantly within 24 hours ON support; implantation did not aggravate haemolysis but decreased it. Pre-cannulation free haemoglobin had no prognostic value on multivariable analysis; only a low pre-cannulation lactate predicted successful weaning (p under 0.0001) and hospital discharge (p=0.0028).

The three pre-cannulation clinical states compared with each other and with values on support

Daily plasma free haemoglobin, lactate and pH before and during support

The circuit is routinely blamed for haemolysis the illness and the resuscitation caused. Take a full laboratory panel BEFORE the pump starts; without it every later value is read against the wrong comparison.

Lehle K, Lubnow M, Philipp A, Foltan M, Zeman F, Zausig Y, Lunz D, Schmid C, Muller T. Eur J Heart Fail 2017;19(S2):110-116. DOI 10.1002/ejhf.854

605 consecutive continuous video-electroencephalography studies in 411 patients aged 21 years or younger in one paediatric and neonatal cardiovascular intensive care unit over 38 months; median study duration 26 hours (IQR 20 to 41)

Paediatric and neonatal cardiovascular intensive care, not adult ECMO — the indication logic transfers but the numbers do not. Single centre, retrospective; does not establish the risk of electrographic seizures by individual clinical risk factor.

Ch 28 (s28.9, s28.14)

Seizures detected in 57 of 605 (9 percent) overall. Zero of 104 studies (0 percent) requested for isolated vital-sign changes detected a seizure (p under 0.001), versus 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 — odds ratio 4.93 (2.45 to 9.77) against risk factors alone. Venoarterial ECMO as an indication yielded seizures in 14 of 116 (12 percent) and extracorporeal cardiopulmonary resuscitation in 8 of 47 (17 percent).

Seizure yield by indication category

Continuous electroencephalography, with indications categorised as index symptoms of concern (vital-sign or non-vital-sign) and/or clinical risk factors

An unexplained tachycardia, blood pressure swing or 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. Supports recording electroencephalography on indication rather than routinely on ECMO.

DOI 10.1111/epi.18253

Adults on venovenous and venoarterial ECMO; narrative review with single-centre descriptive observations

Narrative review; the transcranial Doppler figures are single-centre descriptive values, not normal ranges; the seizure range is wide and conditioned on who was recorded. The article was returned in the search record under both a plain digital object identifier and a compound form with identical text; the plain form is cited.

Ch 28 (s28.6, s28.8, s28.9, s28.10, s28.12, s28.14)

States that the optimal neuromonitoring protocol has not been well established, with current practices dependent on physician preference or device availability. Near-infrared light penetrates 2 to 2.5 cm, with regional saturation a surrogate for cerebral blood flow and the cerebral oximetry index approaching 1 indicating pressure-passive flow. Transcranial Doppler pulsatility index is lower during ECMO initiation; rising pulsatility may indicate cerebral pathology; the centre's own figures show low mean middle cerebral artery velocities with normal pulsatility index on venovenous and low pulsatility index on venoarterial support. Magnetic resonance imaging cannot be used while on ECMO due to hardware incompatibility. Electroencephalogram abnormal in 50 to 80 percent of ECMO patients with electrographic seizures reported at 8 to 20 percent; continuous recording is a scarce resource so many institutions check periodic 20-minute to 1-hour studies. Biomarkers GFAP, S100b, neuron-specific enolase, ICAM5, BDNF and MCP-1/CCL2 are candidates with debate over optimal cut-offs and whether serial assessment adds meaning.

Not applicable

Review of near-infrared spectroscopy, transcranial Doppler, electroencephalography, evoked potentials, optic nerve sheath diameter, diffuse correlation spectroscopy and serum biomarkers

The source of the framing statement that there is no established neuromonitoring protocol on ECMO, and of the physiological detail needed to read a near-infrared spectroscopy or transcranial Doppler trace correctly — particularly that the sampling volume is small, superficial and frontal, so posterior or deep lesions are invisible.

DOI 10.1155/2018/3237810

20 adult ECMO patients (median age 60) at a single centre: 17 had transcranial Doppler, 13 electroencephalography and 7 somatosensory evoked potentials

Twenty patients, seven with paired electrophysiology; single centre; the microembolic finding is directly contradicted by the same group's larger 145-patient cohort (Caturegli 2022).

Ch 28 (s28.8, s28.11, s28.16); Ch 23

Neurological complications found in 17 of 20 (85 percent); unfavourable outcome in 14 (70 percent), associated with absent electroencephalographic reactivity, coma, central cannulation, higher transfusion requirement and higher APACHE II and SOFA scores. Eighteen thromboembolic events observed, 14 with positive microembolic signals; all 10 patients with arterial-sided thrombotic events had positive signals. Of the 7 patients with both evoked potentials and electroencephalography, ALL 7 had intact N20 responses despite poor outcomes, and 4 of those 7 showed absent electroencephalographic reactivity despite the intact N20. No adverse effects from monitoring.

Favourable (modified Rankin Scale 0 to 3) versus unfavourable (over 3) outcome at discharge

Noninvasive multimodal neuromonitoring — neurological examinations, transcranial Doppler, electroencephalography and somatosensory evoked potentials at prespecified intervals

A present N20 carries very little reassurance on ECMO — in the only series that looked, it was intact in every patient who did badly. And the two electrophysiological tests disagreed in more than half the patients in whom both were done, so they cannot be summed into a single 'electrophysiology' impression, which makes the multimodal rule of 'at least two indicators of severe injury' much harder to apply.

DOI 10.1097/MAT.0000000000001013

Monocentric prospective observational study at a regional ECMO referral centre, December 2013 to August 2015: 248 transcranial Doppler monitoring sessions in 42 venovenous and 11 venoarterial adult patients, plus 2 on extracorporeal carbon dioxide removal. Patients with bilateral inadequate temporal windows, right-to-left shunt or haemodynamic carotid stenosis excluded; all assessed by transoesophageal echocardiography.

Single centre; 11 venoarterial patients only; no control of recording duration against outcome; the authors themselves conclude that multicentre studies with more extensive cases are required. Explicitly failed to reproduce an earlier six-patient report correlating signals with circuit flow over 4 L/min, and criticises that report's 15-minute recording and unspecified decibel threshold.

Ch 28 (s28.8, Controversy 1)

Signals detected in 26.2 percent of venovenous and 81.8 percent of venoarterial patients (p under 0.001); severe grading occurred only in the venoarterial group (5 of 11). No correlation with extracorporeal flow rates or activated partial thromboplastin time in either configuration. Inverse correlation between left ventricular ejection fraction and signal grading in venoarterial patients (p=0.037). No clinical neurological impairment correlated with signal detection at 6-month follow-up; 6-month survival 50 percent.

Venovenous versus venoarterial configuration; correlation with extracorporeal flow rate, activated partial thromboplastin time, left ventricular ejection fraction; and 6-month Glasgow Outcome Scale

Multigated transcranial Doppler detection and grading of microembolic signals, repeated during treatment

Establishes the one replicated finding in this literature — microembolic signals are several times more common on venoarterial than venovenous support, consistent with arterial return delivering circuit-generated emboli directly into the systemic circulation — while finding no relationship to flow, anticoagulation intensity or 6-month neurological outcome.

DOI 10.1111/aas.12736

15 adult ECMO patients at a single centre (median age 57, IQR 47 to 69) — 11 venoarterial and 4 venovenous — yielding 150 autoregulation measurements

Fifteen patients; feasibility was the stated primary aim and feasibility is what was demonstrated. The outcome comparison is a narrow separation between two very high numbers, which says as much about how poorly pressure was controlled in everyone as about the difference between outcome groups. No interventional study has tested whether targeting an individualised pressure changes anything. An earlier preprint carries DOI 10.21203/rs.3.rs-3300834/v1.

Ch 28 (s28.7, s28.13); Ch 25; Ch 54

Mean index higher (worse autoregulation) on day 1 than day 2 (0.2 versus 0.09, p under 0.01) and higher on venoarterial than venovenous support (0.12 versus 0.06, p=0.04). Median optimal mean arterial pressure for the cohort was highly variable, ranging 55 to 110 mmHg. Time with 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 good functional outcome spent less time outside optimal pressure than those with poor outcome (74 versus 82 percent, p=0.01).

Post-cannulation day 1 versus day 2; venoarterial versus venovenous; modified Rankin Scale 0 to 3 versus 4 to 6 at 3 and 6 months

Continuous cerebral oximetry index monitoring, a moving correlation between mean arterial pressure and slow waves of regional cerebral oxygen saturation from near-infrared spectroscopy, used to derive an optimal mean arterial pressure and individual limits of autoregulation

The optimal pressure for an individual brain ranged from 55 to 110 mmHg, so there is no population mean arterial pressure target that serves this group — a protocol targeting 65 mmHg for everyone is targeting the wrong number in roughly half of these patients, in an unknowable direction. And day 1 is the vulnerable day: impairment was worse and time outside limits five times greater than on day 2. Read alongside Chapter 25's porcine finding that a carotid flow deficit persisted after mean arterial pressure was restored: restoring the pressure number does not demonstrate that cerebral flow has been restored.

DOI 10.1007/s12028-023-01932-w

139 of 596 adult venovenous or venoarterial ECMO patients at a single centre (April 2009 to December 2018) who had a concomitant electroencephalographic recording: median age 54 (41 to 62), 60 male (43 percent), 98 (71 percent) venoarterial, continuous recording in 113 (81 percent)

Only 139 of 596 ECMO patients at this centre had an electroencephalogram at all — under a quarter. Every figure is therefore a yield among patients someone thought worth recording, not a population incidence. Retrospective, single centre, mixed intermittent and continuous recording.

Ch 28 (s28.9, s28.16); Ch 23; Ch 51

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).

Unfavourable 3-month neurological outcome, defined as Glasgow Outcome Scale under 4

Intermittent or continuous electroencephalography, with background classified as mild or moderate encephalopathy (defined largely by the presence of reactivity), severe encephalopathy (absence of reactivity), burst-suppression or suppressed

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. The feature that separates outcome groups is whether the background is REACTIVE, not whether epileptiform activity is present — so the question that carries most information when telephoning for a report is 'is the background reactive?', not 'are there seizures?'

DOI 10.1186/s13054-020-03353-z

Four adult patients (median age 56) who underwent brain autopsy at a single centre between November 2017 and December 2018. All four had venoarterial cannulation, none had a known acute neurological injury before ECMO, all were persistently comatose off sedation on support, and all four had continuous electroencephalography showing delta frequency with absent reactivity.

Four patients, one centre, all of whom died — the design cannot estimate how often this occurs. The smallest study cited in Chapter 28.

Ch 28 (s28.9, s28.16); Ch 23; Ch 24; Ch 74; Ch 75

Case 1 showed small cortical infarcts; Case 2 cortical microhaemorrhages and chronic basal ganglia microinfarcts; Cases 3 and 4 diffuse hypoxic-ischaemic brain injury. The authors conclude that the clinical presentation was consistent with widespread neuropathologic injury in two cases but not in the other two, and that while 'poor neurological status' was provided in the end-of-life discussion in all four, only two had significant brain injuries.

Clinical and electrophysiological impression versus neuropathology

Post-mortem brain examination compared against ante-mortem clinical and electroencephalographic findings

Persistent coma off sedation with an unreactive delta electroencephalogram was not explained by structural brain injury in half of these patients. Absent electroencephalographic reactivity on a circuit is a prognostic variable, not a diagnosis. Prognosticate slowly, multidisciplinarily, after rewarming, and having excluded sedation, electrolyte and haemolysis confounders.

DOI 10.1177/0391398820901829