The question this chapter answers
A free haemoglobin result comes back high. Is the circuit breaking red cells, was the patient already breaking them, or is the number wrong? And if the circuit is responsible, which part of it — and does it need changing today?
Haemolysis is the complication where the laboratory and the circuit have to be read together. A number on its own tells the team almost nothing; the same number with the pre-cannulation value, the trend, the inlet pressure and the sound of the pump usually tells them where the red cells are breaking.
Chapter 29 owns how the markers are measured and what confounds them. This chapter owns what the team does next: where the haemolysis comes from, which sources need an exchange, and when.
Evidence search date: 24 September 2026. Two sources are read as complete full text and are the Verified anchors, both from the Regensburg ECMO registry: Lehle and colleagues on haemolysis in 318 adults on venovenous support, and Appelt and colleagues comparing 606 venoarterial with 457 venovenous adults. Primary sources are the ELSO Red Book 6th edition (the circuit chapter, the adverse-effects chapter and the adult respiratory complications chapter), ECMO in the Adult Patient (the haematology chapter), ECMO: A Practical Guide to Management (venovenous physiology and patient–circuit monitoring), the ISCCM Manual of RRT and ECMO in ICU (monitoring, complications and the renal replacement troubleshooting chapter), and Comprehensive Healthcare Simulation: ECMO Simulation (pump engineering). Other studies are graded as abstract-only where that is all that was read.
What this chapter covers — and what it does not
This chapter owns | This chapter does not own |
Haemolysis as a complication — how often it happens, what the thresholds mean, and why the published figures disagree | The markers themselves. Chapter 29 owns them — how free haemoglobin, LDH, bilirubin, haptoglobin and phosphate are measured, and what confounds each. This chapter uses them and does not re-derive them |
Where red cells break in a circuit — shear, cavitation, residence time, clot and flow through small cannulas | The pump as a machine. Chapter 33 owns it, including why a centrifugal pump grinds rather than bursts |
Localising a rising free haemoglobin to the pump, the drainage, the cannula, a connected renal circuit, the patient, or the sample | Pump head thrombosis as a circuit event. Chapter 31 owns it; this chapter owns it as a haemolysis diagnosis |
The haemolysis trigger for pump or circuit exchange — the decision Chapter 29 declined to make on a number alone | The exchange procedure. Chapter 31 owns it |
What haemolysis is associated with — kidneys, stroke, death — and how it distorts other laboratory tests | Acute kidney injury and renal replacement on ECMO. Chapter 55 owns them. Stroke is Chapter 38a's |
Neuron-specific enolase in the presence of haemolysis — the obligation Chapters 23 and 28 left open | Neuroprognostication as a whole. Chapter 24 owns it |
37.1 Two kinds of haemolysis
The structural claim of this chapter · Much of the haemolysis measured on ECMO arrived with the patient. The circuit's haemolysis usually announces itself as a change
Appelt and colleagues measured free haemoglobin before cannulation in 1,063 adults, and the finding changes how every later result should be read.
- Before venoarterial ECMO, the median free haemoglobin was 225 mg/L (22.5 mg/dL); before venovenous ECMO it was 72 mg/L (7.2 mg/dL). After cardiopulmonary resuscitation it was 289 mg/L, and after failure to wean from cardiopulmonary bypass 282 mg/L — both already above the level many units would call haemolysis.
- After cannulation, it fell. Across every pump system in the venoarterial group, the median first-day value was lower than the pre-cannulation value. In resuscitated patients it fell by 34 percent with a 15 Fr cannula and 22 percent with a 17 Fr cannula within one day. The authors' reading: restoring flow to ischaemic tissue "normalized RBC damage."
- Only 13 percent of the high values were explained by pump head thrombosis. The rest came from the indication, the resuscitation, the surgery, transfusion and the illness.
The authors conclude: "Hemolysis was dominant in VA ECMO patients, mainly caused by different indications and not by the ECMO support itself."
Two kinds of haemolysis follow, and they need different responses.
- Background haemolysis — present before cannulation or produced by the illness, and usually stable or falling on support. It needs recognising, not an exchange.
- Circuit haemolysis — a rise from the patient's own baseline, often a step, with a mechanical source that can be found. It needs localising, and sometimes an exchange.
So the first question about any high free haemoglobin is not how high it is, but what it is being compared with. The comparison is the patient's pre-cannulation value and their own trend — Chapter 29's rule, applied here to a management decision.
Certainty: moderate — one large single-centre registry; the direction is consistent with the pre-cannulation data Chapter 29 recorded from the same group.
Clinical pearl · A high first-day value in a resuscitated patient is expected
After extracorporeal resuscitation or failure to wean from bypass, a free haemoglobin above the unit's alarm threshold on day one is the rule, not a complication. If the pre-cannulation sample was taken, compare with it. If it was not, the day-one value becomes the baseline — and the trend over the next 48 hours is what matters.
A value that falls is the circuit restoring perfusion. A value that rises from an already high baseline is a new problem stacked on an old one.
37.2 How often the circuit breaks red cells
Evidence · The two Regensburg anchors
Venovenous. Lehle K, Philipp A, Zeman F, Lunz D, Lubnow M, Wendel HP, Göbölös L, Schmid C, Müller T. "Technical-Induced Hemolysis in Patients with Respiratory Failure Supported with Veno-Venous ECMO – Prevalence and Risk Factors." PLoS One 2015;10(11):e0143527. DOI 10.1371/journal.pone.0143527, PMC4659553. Full text read.
Venoarterial and venovenous compared. Appelt H, Philipp A, Mueller T, Foltan M, Lubnow M, Lunz D, Zeman F, Lehle K. "Factors associated with hemolysis during extracorporeal membrane oxygenation (ECMO) — Comparison of VA- versus VV ECMO." PLoS One 2020;15(1):e0227793. DOI 10.1371/journal.pone.0227793, PMC6984694. Full text read.
Lehle 2015 — venovenous | Appelt 2020 — venoarterial | Appelt 2020 — venovenous | |
Patients; years | 318; 2009 to 2014 | 606; 2012 to 2018 | 457; 2012 to 2018 |
Median time on ECMO | — | 4 days (2 to 7) | 8 days (6 to 14) |
Free haemoglobin measurements | 4,142 | 2,864 | 4,832 |
Measurements above 500 mg/L (50 mg/dL) | 1.7 percent | 4 percent | 2 percent |
Patients with at least one value above 500 mg/L | — | 12 percent | 10 percent |
Pump head thrombosis | 8 patients (2.5 percent) | 2 percent (15 patients) | 9 percent (41 patients) |
Median day of pump head thrombosis | — | Day 5 | Day 9 |
The reading this chapter takes. Severe haemolysis with modern equipment is uncommon, and when it is severe it is usually one event — pump head thrombosis. Lehle's severe cases "always indicated PHT." Appelt adds the configuration difference: venoarterial patients have more high values, mostly brought with them; venovenous patients have more pump head thrombosis, probably because they stay on longer. The authors also note lower heparin targets on venovenous support (aPTT 45 to 50 against 50 to 60 seconds) and more inflammation, and say plainly that "reasons for the different incidences of PHT cannot be clarified."
Certainty: moderate — large, prospective data collection, but one centre, one laboratory and its own pump mix.
Pitfall · "4 percent versus 2 percent" is a share of measurements, not of patients
Appelt's abstract says severe haemolysis "occurred in 4% (VA) and 2% (VV)." The Red Book repeats it as haemolysis being "significantly more frequent in VA ECMO when compared with VV ECMO (4% vs. 2%)." In the paper's own Table 6, those figures are the proportion of free haemoglobin measurements above 500 mg/L. The proportion of patients with at least one such value was 12 and 10 percent.
One further inconsistency to record. The legend of Appelt's Figure 2 gives values above 500 mg/L as 1 percent (venoarterial) and 0.4 percent (venovenous), which does not match Table 6. The book uses the table.
Why it matters at the bedside. A unit auditing its own haemolysis rate needs to know which denominator it is comparing with. Per patient, roughly one in ten adults crossed 500 mg/L at least once in this series.
Why published haemolysis rates run from about 2 to 18 percent
Source | Figure | What it counts |
Red Book adverse-effects chapter; Lehle and Appelt citing earlier work | 5 to 18 percent | A range quoted without a single definition behind it |
Lehle 2015 | 1.7 percent of measurements above 500 mg/L | Measurements, venovenous only |
Appelt 2020 | 12 and 10 percent of patients | At least one value above 500 mg/L, including pre-existing haemolysis |
ELSO Registry, venovenous (Nunez 2022), as the Red Book reports it | 5.3 percent | Registry-reported haemolysis |
Pan 2016, one Melbourne centre, 184 runs (abstract read) | 13 percent of runs (24 of 184) peaked above 0.5 g/L; 54 percent (99) peaked between 0.1 and 0.5 g/L | Peak value per run |
The spread is the denominator and the threshold. Measurements or patients; any value or a sustained one; 30, 50 or 100 mg/dL; with or without the haemolysis the patient arrived with. A haemolysis rate means nothing without all three stated.
37.3 What counts as haemolysis — and the units problem
The thresholds the sources give — including two from the same book
Source | Threshold | Meaning attached |
Red Book, adult respiratory complications chapter | Above 0.5 g/L | "Diagnostic"; plasma free haemoglobin is "the gold standard for identifying intravascular hemolysis on ECLS" |
Red Book, adverse-effects chapter | Above 30 mg/dL diagnostic; above 50 mg/dL most commonly reported; above 100 mg/dL severe | Above 100 mg/dL "should prompt immediate action" |
Lehle 2015, following the ELSO Registry | Above 500 mg/L; normal subclinical range 100 mg/L or less | Haemolysis complicating ECMO |
Appelt 2020 | Above 500 mg/L critical; above 100 mg/L "suspected of hemolysis" | Pump head thrombosis defined separately as a rapid rise of more than 300 mg/L, mostly with falling platelets and pump noise or vibration |
Lehle 2015 | A sudden rise to 1,000 mg/L or more | Can indicate pump head thrombosis |
Omar 2015 (abstract read) | Above 50 mg/dL at 24 hours | Severe haemolysis |
Saeed 2019 (Chapter 29) | 11 to 50 mg/dL at 48 hours | "Low-level haemolysis" |
The two Red Book chapters disagree. One calls 30 mg/dL diagnostic; the other calls 50 mg/dL diagnostic. Neither is wrong about what the literature reports; they are reporting different studies. The book uses 50 mg/dL as the threshold for calling an event and 100 mg/dL as the level at which the Red Book asks for immediate action, and treats both as conventions rather than safety margins (Chapter 29, Controversy 1).
Survey evidence confirms there is no common practice. Strong and colleagues (Blood Purif 2025, abstract read) surveyed 26 ECMO programmes: 68 percent had a haemolysis protocol, but the protocols and the definitions of significant haemolysis "varied widely", and significant haemolysis was identified mainly by red urine and a raised plasma haemoglobin.
Danger · The same number in three units
500 mg/L = 50 mg/dL = 0.5 g/L. Different laboratories, papers and chapters of the same book report free haemoglobin in each of the three, and a factor-of-ten error is one misplaced unit away.
mg/L | mg/dL | g/L | What the sources attach to it |
100 | 10 | 0.1 | Upper limit of the "normal subclinical range" (Lehle); lower limit of "low-level" (Pan) |
300 | 30 | 0.3 | "Diagnostic" in one Red Book chapter; the size of the rise that defined pump head thrombosis (Appelt) |
500 | 50 | 0.5 | The commonest definition of haemolysis; "diagnostic" in the other Red Book chapter |
1,000 | 100 | 1.0 | Severe; immediate action (Red Book). A sudden rise to this level can indicate pump head thrombosis (Lehle) |
The rule: write the unit on every free haemoglobin result that is handed over, and set the unit's alarm thresholds in the unit the laboratory reports.
Evidence · Two cheaper markers from the blood gas — promising, not validated
Chapter 29 covers the standard markers. Two further signals come from the same analyser that runs every arterial gas.
Bünger and colleagues (Anesth Analg 2025, abstract read), in 435 adults on venovenous ECMO for ARDS, found methaemoglobin independently associated with haemolytic events (free haemoglobin above 50 mg/dL): adjusted OR 2.99 (2.19 to 4.10). A methaemoglobin of 1.55 percent had 90 percent specificity for a haemolytic event. Mean carboxyhaemoglobin (OR 2.03) and methaemoglobin (OR 2.78) were both associated with ICU mortality, with cut-offs of 2 percent and 1.25 percent.
The same group (Shock 2024, abstract read), in the same cohort, found haptoglobin depletion below 0.39 g/L in the first week in 269 of 435 patients, with 28-day mortality 43.5 against 25.3 percent, and a starting haptoglobin below 0.66 g/L predicting depletion.
What the book takes from them. A rising methaemoglobin on the routine gas is a reason to send a free haemoglobin, not a replacement for one: the cut-off comes from one centre and has not been validated elsewhere. And haptoglobin's two-directional confounding (Chapter 29) applies to the depletion finding.
Certainty: low — single-centre, retrospective, abstracts only.
37.4 Where red cells break in a circuit
Physiology · Five places, one mechanism
The Red Book gives the mechanism in two sentences: "Normal shear stress in the circulatory system is between 1-10 Pa, but this is largely exceeded during extracorporeal circulation." When flow becomes turbulent, "the erythrocytes first deform to an ellipsoidal shape and finally progress to cell wall destruction." Every source of circuit haemolysis is a place where shear rises further, or where red cells are exposed to it for longer.
Site | What happens | What the sources say |
1. The drainage side — suction and cavitation | The centrifugal pump draws rather than being fed. When preload is inadequate, pressure at the pump inlet falls below atmospheric and dissolved gas comes out of solution | Red Book: inlet pressures of minus 250 to minus 300 mmHg risk cavitation; the micro-bubbles "cause additional shear stress," and "when these implode in the positive pressure zone (pump exit), serious blood cell damage may occur." Chattering drainage produces pressure spikes of minus 600 to minus 650 mmHg and haemolysis from high shear at the cannula tip |
2. The cannula — flow through a small bore | For a fixed flow, a narrower cannula means a higher velocity and more shear | Red Book: haemolysis occurs with "high flow velocity due to small cannulas." Lehle and Appelt measured it (§37.6) |
3. The pump head — clot | A thrombus on the impeller or axle rotates with it and grinds red cells at every turn | Lehle: "A rotating clot in the pumphead will cause excessive shear forces and mechanical destruction of erythrocytes." §37.7 |
4. The pump head — low flow | At low flow, blood spends longer in the head and is exposed to shear for longer | ECMO simulation text: at low flow "the blood spends more time in the pump head, and severe hemolysis can occur." Modern magnetically levitated heads reduce this; Chapter 33 treats it as structural, not historical |
5. The oxygenator | A rising resistance across the membrane raises the pressure gradient | The Red Book lists "increased membrane pressure gradients" and "fibrin deposition" among causes. Lehle's data disagree for acute oxygenator thrombosis — §37.6 |
ECMO: A Practical Guide to Management puts the pump-side causes together: "high rotation speed with inadequate pump preload, small cannula, and hypovolemia, or increased afterload and membrane thrombosis."
Certainty: high for the mechanisms, which are engineering; moderate for their relative clinical importance.
Physiology · What free haemoglobin does once it is out
The sources agree on the pathway.
- Scavengers are overwhelmed. Free haemoglobin and haem are normally bound by haptoglobin and hemopexin. When release exceeds their capacity, the unbound fraction does the damage (Red Book).
- Nitric oxide is consumed. Free haemoglobin "depletes endothelial-derived nitric oxide," causing vasoconstriction and a "procoagulant state secondary to abnormal coagulation and platelet activation" (Red Book). This is the proposed link to the thrombotic strokes Chapter 29 recorded (Saeed 2019).
- Haem and iron generate free radicals that "disrupt cell membranes, damage proteins, DNA and lipids, induce SIRS" (Red Book).
- The kidney takes a direct hit. ECMO: A Practical Guide to Management: "direct renal tubular toxicity and nitric oxide depletion." Lehle adds tubular obstruction and reactive oxygen species from free iron.
The practical point. Low-level haemolysis consumes haptoglobin before free haemoglobin rises far, which is why a falling haptoglobin can come first (Chapter 29). By the time the free haemoglobin is high, the scavengers have already been exceeded. Lehle and Appelt both call free haemoglobin a late marker of blood trauma.
37.5 A rising free haemoglobin: where to look
The localisation table — this book's synthesis, building on Chapter 31's five-item differential
Chapter 31 listed five causes of a rising free haemoglobin: sampling, excessive suction from a malpositioned cannula, circuit-connected renal replacement, thrombosis elsewhere, and large haematomas. This table adds what the anchors found and gives each source the bedside finding that points to it.
Source | Pattern | The finding that points to it | Where it is covered |
The sample | An isolated high value, often with no other marker moving | A repeat from a different site is normal. Lipaemia can produce spurious haemolysis (§37.5 pitfall) | Chapter 29 |
The patient's illness | High before cannulation; stable or falling on support | Cardiac arrest, bypass surgery, trauma, sepsis, a large haematoma, many transfusions | §37.1; §37.6 |
The drainage | A gradual rise, sometimes fluctuating | A more negative inlet pressure; chatter; flow limited by volume or cannula position | Chapter 34 §34.2 |
Flow through too small a cannula | A rise that tracks flow | Flow of 3 L/min or more through a small return cannula, or a cannula run above its designed flow | §37.6 |
The pump head | A step — often to several times baseline within a day | Platelets falling, a change in pump sound or vibration, dark urine or effluent, clot seen on the axle | §37.7; Chapter 31 |
A renal replacement circuit connected to the ECMO circuit | A rise after the connection | Timing with the renal circuit; access pressures on the renal machine | Chapter 55; Chapter 29 (Costantini) |
Something outside the circuit | Anything the circuit does not explain | A new drug, a transfusion reaction, microangiopathy, a second device such as a microaxial pump | §37.5 pearl |
The order matters. The sample and the baseline are checked first because they cost nothing and they account for many high results. The pump head is checked early because it is the one source that can progress to circuit failure within hours.
Pitfall · Lipaemia can produce spurious haemolysis
Venado and colleagues (Perfusion 2014, abstract read) reported two ECMO patients with spurious haemolysis caused by hypertriglyceridaemia — lipid interfering with the measurement of plasma free haemoglobin. The keywords name propofol and sirolimus as the drugs involved. The authors' advice: investigate hypertriglyceridaemia as part of the assessment of unexplained haemolysis on ECMO.
Why it catches teams out. Propofol is a common sedative on ECMO, and the sample may look turbid rather than red. An unexplained free haemoglobin in a patient on propofol needs a triglyceride level before it needs a pump change.
Certainty: low — two cases; the mechanism is analytical interference.
Clinical pearl · "Causes other than ECMO should be suspected"
ECMO in the Adult Patient: "Severe intravascular haemolysis should be investigated, and causes other than ECMO should be suspected." It also warns that "significant haemolysis may lead to anaemia followed by inappropriate investigations if not readily recognized" — the anaemia is chased while its cause is in the circuit.
Both directions of error happen. A circuit is changed for haemolysis that came from a transfusion reaction or a drug; and a patient is investigated for anaemia while the pump head grinds.
A second device matters too. Nakashima and colleagues (Eur Heart J Acute Cardiovasc Care 2026, abstract read) found higher microaxial pump levels associated with higher free haemoglobin when the microaxial pump was used alone. When a patient has both a microaxial pump and venoarterial ECMO, the haemolysis may come from either device; Chapter 80 owns combined support.
37.6 What the anchors found about cause
Evidence · Flow, cannulas, pumps, the oxygenator and the patient
Factor | Lehle 2015 — venovenous | Appelt 2020 — venoarterial and venovenous |
Pump type | No effect (p = 0.242). A pump running at about double the rotational speed of the others did not cause more haemolysis at the same flow: "rotational speed per se does not cause hemolysis" | No effect in either configuration (first day p = 0.077 venoarterial; p = 0.072 venovenous). Pump head thrombosis equally distributed across pump types, but more frequent with one system on venovenous than venoarterial support (10 against 2 percent) |
Cannula type | No effect overall (p = 0.906) | No effect of cannulation technique (central, surgical or percutaneous; p = 0.973), and dual-lumen cannulas did not cause more haemolysis than single-lumen (p = 0.358) |
Flow | 3.0 L/min or more gave a moderate rise against low flow (p = 0.004), but about three-quarters of values were still under 100 mg/L | In patients without pre-existing haemolysis or pump thrombosis, flow of 3 L/min or more through a 15 Fr return cannula produced significantly more haemolysis than 2.5 L/min or less |
Cannula run above its design | A 24 Fr dual-lumen cannula designed for carbon dioxide removal at up to 2 L/min gave free haemoglobin of 116 mg/L at 3 L/min or more, against 53 to 63 mg/L for the other cannulas (p 0.001 or less) | — |
Drainage pressure | A 27 Fr dual-lumen cannula ran at more negative pressure (median minus 29 mmHg) than single-lumen drainage cannulas, without more haemolysis within the flows used. Brief pressure excursions to minus 700 mmHg would not be recorded by routine measurement | — |
Acute oxygenator thrombosis | 15 episodes: the pressure drop rose sharply but free haemoglobin stayed stable (62 before, 86 mg/L on the day of replacement) | — |
Duration on the first system | No effect (p = 0.363) | Pump head thrombosis came later on venovenous support (day 9 against day 5) |
The patient | Multiple trauma (p = 0.003), more red cell units per day (p = 0.017) and higher flow velocity (p = 0.013) were independent; renal replacement before ECMO was associated in univariable analysis only | Resuscitation and failed weaning from bypass gave the highest pre-cannulation values |
Three readings, and each changes a bedside decision.
- Size the cannula for the flow the patient will need, not the flow they need now. Both anchors found haemolysis where flow exceeded what a small cannula carries comfortably. Appelt notes that exchanging for a larger cannula is usually avoided because of its own risk, so the decision belongs at insertion (Chapters 5 and 12).
- The oxygenator clots without grinding. Lehle's explanation: in acute oxygenator thrombosis "no disruption of erythrocytes occurred." Chapter 32 recorded the same thing from the other side: free haemoglobin did not predict oxygenator exchange. The Red Book's listing of membrane pressure gradients as a cause is not supported by the adult clinical data in the book's sources; the book records the disagreement and treats a rising free haemoglobin as a pump or drainage question first.
- A high pump speed is not itself the problem. Speed without adequate preload is. Lehle's comparison of pumps at different speeds separates the two.
One contradiction to record. Appelt's results paragraph states that high flow through 15 Fr cannulas produced more haemolysis than through 17 Fr cannulas, and in the next sentence that at high flow "the fHb levels of 15 Fr cannulas were significantly lower compared to 17 Fr cannulas (p = 0.027)." The discussion supports the first reading. The book uses the discussion's conclusion and flags the sentence.
Certainty: moderate for the pump, cannula and oxygenator findings; low for any single threshold, which neither anchor could define — Lehle: "it was not possible to define a cut-off value for flow velocity."
37.7 Pump head thrombosis: the haemolysis that cannot wait
Danger · A step in free haemoglobin with falling platelets is a clotted pump head until it is proven otherwise
What it looks like. Appelt's definition: "a rapid and substantial increase in fHb (> 300 mg/l) accompanied mostly by a decrease in platelets and an abnormal noise/vibration of pump head." Lehle: a marked rise within 24 hours in all 8 venovenous cases, and a rise into four figures (1,000 mg/L or more) "should immediately raise suspicion that the pumphead may be partially clotted." The ISCCM manual: "sudden change sound from pump head with vibrations in the circuit, progressive pump-induced hemolysis," with high free haemoglobin and D-dimers.
What it does not look like. The pressures may not change. Chapter 31 found that pump head thrombosis in its anchor cohort was found by free haemoglobin and was invisible to the pressure channel. Lehle adds that it "may cause a noisy pump run, but may be undetectable by inspection alone."
What happens after exchange. In Lehle's 8 patients free haemoglobin "normalized within 2 days after pumphead replacement," and clots were seen inside every removed pump head. Appelt: "The levels of fHb decreased and platelets normalized after immediate change of the circuit." Chapter 31's anchor found free haemoglobin about 10-fold higher before and 6-fold lower after exchange.
Why it cannot wait. A rotating clot grinds at every revolution, and the free haemoglobin it releases is itself prothrombotic (§37.4). Chapter 33's pump-failure pathway begins here: a thrombus can unbalance the magnetic coupling, and the pump stops moving blood while still switched on.
What to do.
- Look and listen. The ISCCM manual suggests slow-motion smartphone video of the pump head to see the thrombus.
- Send the confirming set — free haemoglobin, platelets, D-dimer, potassium — and look at the urine or renal effluent.
- Prepare the exchange while the results return. Appelt calls pump head thrombosis "an acute emergency event" in which "the pump head has to be changed." Chapter 31 owns the procedure.
- If platelets are falling, think of heparin-induced thrombocytopenia. Lehle: "a heparin-induced thrombocytopenia must be ruled out; a change of anticoagulation is advisable until receipt of laboratory values" (Chapter 44).
Certainty: moderate — consistent across two anchors, Chapter 31's anchor and the ISCCM manual; the thresholds differ between sources.