The Bleeding You Caused

01

The pearl

Andexanet alfa achieved haemostasis in 67.0% of factor Xa-associated intracerebral haemorrhages against 53.1% with usual care — and thrombotic events ran 10.3% against 5.6%, with ischaemic stroke 6.5% against 1.5%. That is the shape of every reversal decision in miniature: a real gain on the bleeding, paid for in clot, in a patient whose original indication has not gone away. And then the clot side of that ledger closed the drug down. On longer postmarketing follow-up the thrombosis signal grew — 14.6% against 6.9% at 30 days, with thrombosis-related death 2.5% against 0.9% — and AstraZeneca withdrew andexanet alfa from the United States market on 22 December 2025. Do not go looking for it in an American pharmacy in 2026; four-factor prothrombin complex concentrate is what is on the shelf. The reversal is rarely the hard part. Knowing what you have bought is — and this time the bill arrived after the trial was published.

02

Why it is true

mechanism

An intracranial bleed in an anticoagulated patient presents as an emergency with an obvious answer, and the obviousness is the problem. Reversal is the right thing to do and it is not a free action: every agent on the table buys clot at the price of clot somewhere else, in a patient who was anticoagulated for a reason that has not disappeared.

Warfarin

The most established row. Intravenous vitamin K 5 to 10 mg should be given along with either FFP 20 mL/kg or PCC 20 to 40 IU/kg to normalize the prothrombin time. Both parts matter: the factor product corrects the patient in the next half hour, and the vitamin K prevents them drifting back as it clears.

Four-factor PCC contains factors II, VII, IX and X plus proteins C and S, and it has demonstrated superiority to plasma in rapid reversal of warfarin-associated coagulopathy. The trial that established this is worth reading for what the two endpoints actually say. Effective haemostasis was 72.4% versus 65.4% — a modest difference. Rapid INR correction was 62.2% versus 9.6% — an enormous one. Safety, including thromboembolic events, was similar between groups.

So the case for PCC is not mainly that fewer patients bleed. It is speed and volume. About 1 L of plasma is required to normalize coagulation, because 10 to 15 mL/kg raises factor levels by only 10 to 20%. In heart failure, or in a patient who is already oedematous, that litre is a second problem arriving to solve the first.

Dabigatran

Idarucizumab 5 g is a specific antibody fragment and can be given for life-threatening bleeding or before an emergent procedure. PCC is the fallback where it is unavailable.

The useful thing to know about this row is logistical rather than pharmacological: find out now whether your institution stocks it. That is not a question to be researching while a patient is being wheeled in.

Factor Xa inhibitors

This is where the textbook and the literature have come apart, and it is worth seeing the gap rather than being told the answer.

The source text says patients on rivaroxaban or apixaban may be treated with PCC. In 2024, ANNEXA-I randomized 530 patients with factor Xa-associated intracerebral haemorrhage to andexanet alfa or usual care — where usual care included PCC in 85.5% of controls.

Haemostatic efficacy, defined as haematoma expansion of 35% or less at 12 hours with an NIHSS increase under 7 points and no rescue therapy, was achieved in 67.0% with andexanet against 53.1% with usual care, an adjusted difference of 13.4 percentage points (95% CI 4.6–22.2, P = 0.003).

Thrombotic events occurred in 10.3% against 5.6%. Ischaemic stroke occurred in 6.5% against 1.5%.

Read those together. The gain is real and it is on a surrogate — haematoma expansion, an endpoint that has repeatedly failed to translate into functional outcome in intracerebral haemorrhage. The cost is roughly one additional ischaemic stroke for every twenty patients treated, and it is not a surrogate at all.

Antiplatelet agents

The weakest row, and the one most often acted on with confidence. In patients on antiplatelet medication — particularly dual therapy combining aspirin with a P2Y12-ADP inhibitor such as clopidogrel, prasugrel or ticagrelor — the effect of platelet transfusion on outcome is not clear, and it carries a class IIb recommendation.

Knowing that a recommendation is class IIb is a different kind of knowledge from knowing a dose, and it is the kind that changes how you talk about a decision rather than what you do.

Two things not to do

Recombinant factor VIIa prophylactically. The data are inconsistent on mortality benefit, transfusion requirement and complications, and the thromboembolic risk is serious enough that it cannot be recommended in patients without acute bleeding. PCC offers significant cost savings and is a viable alternative for coagulopathy in traumatic brain injury.

Tranexamic acid late, or in the wrong patient. CRASH-3 randomized 12,737 patients with traumatic brain injury. The primary outcome — head injury-related death within 28 days among those treated within 3 hours — was 18.5% versus 19.8%, risk ratio 0.94 (95% CI 0.86–1.02). It crossed 1.

The interpretable signal is in severity. Risk ratio 0.78 in mild to moderate injury against 0.99 in severe, with early treatment more effective than late in the mild and moderate group and timing making no obvious difference in severe injury. Give it early in the patient whose brain is not already destroyed.

The sentence to say out loud

Reversal ends one problem and starts another, and the second one belongs to somebody else — the intensivist tomorrow, the cardiologist on day three, whoever has to decide when this patient’s mechanical valve gets its anticoagulation back.

That decision is made from your record. “PCC given” is close to useless. What was given, how much, at what time, and what the patient was on — that is the handover, and it is the part of this day that costs nothing and is skipped most often.

Specific reversal available Reversal with a measured cost Weakest evidence

Figure 1. The doses are those given in the source text; the andexanet row reflects a 2024 trial that postdates the chapter's advice, which is why that row reads differently from the others. This is a reference table, not a protocol — follow local policy and confirm any dose before giving it. The line at the bottom is the part worth memorizing: the trade is present in every row, and it is least visible in the row where the evidence is weakest.

03

The evidence

3 RCTs · n = 13,469 · two positive on the surrogate

ANNEXA-I

Connolly SJ, Sharma M, Cohen AT, et al. Andexanet for factor Xa inhibitor-associated acute intracerebral hemorrhage. N Engl J Med. 2024;390(19):1745-1755. Benefit

Design: Randomized, controlled, open-label trial
Population: 530 patients with acute intracerebral haemorrhage while taking a factor Xa inhibitor; 263 assigned andexanet, 267 usual care
Intervention: Andexanet alfa
Control: Usual care, which included prothrombin complex concentrate in 85.5% of control patients
Primary: Haemostatic efficacy — haematoma expansion of 35% or less at 12 hours, an NIHSS increase of less than 7 points, and no rescue therapy between 3 and 12 hours
Result: 150/224 (67.0%) with andexanet vs 121/228 (53.1%) with usual care · adjusted difference 13.4 percentage points (95% CI 4.6–22.2) · P = 0.003
Harm signal: Thrombotic events 27/263 (10.3%) vs 15/267 (5.6%); ischaemic stroke 17 (6.5%) vs 4 (1.5%)

A won primary outcome that does not settle the question The primary endpoint is a composite built largely around haematoma expansion — a surrogate, and one that has repeatedly failed to translate into functional outcome in intracerebral haemorrhage. Set the 13.4-point haemostatic gain beside a near-doubling of thrombosis and a more-than-fourfold increase in ischaemic stroke and the net benefit is not obvious from these data. The trial also compared against usual care rather than against PCC specifically, though most controls received PCC.

4F-PCC versus plasma for vitamin K antagonist reversal

Sarode R, Milling TJ Jr, Refaai MA, et al. Efficacy and safety of a 4-factor prothrombin complex concentrate in patients on vitamin K antagonists presenting with major bleeding: a randomized, plasma-controlled, phase IIIb study. Circulation. 2013;128(11):1234-1243. Benefit

Design: Randomized, plasma-controlled, phase IIIb trial
Population: 202 patients on vitamin K antagonists with acute major bleeding
Intervention: Four-factor prothrombin complex concentrate
Control: Plasma
Primary: Co-primary — 24-hour haemostatic efficacy, and INR correction to 1.3 or below at 0.5 hour after the end of infusion
Result: Effective haemostasis 72.4% vs 65.4% · rapid INR reduction 62.2% vs 9.6%

What the second endpoint is actually telling you The haemostasis difference is modest; the INR difference is enormous, and it is a laboratory value rather than an outcome. What that gap really describes is speed and volume — PCC corrects the number in half an hour where plasma mostly does not, and it does so without the litre of fluid plasma requires. Safety, including thromboembolic events, was similar between groups. This is not a trial about who bleeds less; it is a trial about how fast and at what volume cost.

CRASH-3

CRASH-3 trial collaborators. Effects of tranexamic acid on death, disability, vascular occlusive events and other morbidities in patients with acute traumatic brain injury (CRASH-3): a randomised, placebo-controlled trial. Lancet. 2019;394(10210):1713-1723. Benefit

Design: International randomized placebo-controlled trial
Population: 12,737 patients with traumatic brain injury
Intervention: Tranexamic acid, 1 g loading dose over 10 minutes then 1 g over 8 hours
Control: Placebo
Primary: Head injury-related death in hospital within 28 days, in patients treated within 3 hours of injury
Result: 18.5% vs 19.8% · risk ratio 0.94 (95% CI 0.86–1.02) · benefit in mild to moderate injury (RR 0.78) with no benefit in severe injury (RR 0.99)

A primary outcome that did not reach significance The headline comparison crosses 1. The signal lives in the severity subgroup — mild and moderate injury benefited, severe injury did not, and early treatment was more effective than late in the mild and moderate group while timing made no obvious difference in severe injury. That is a coherent story rather than a fishing expedition, and it is still a subgroup. Read it as: worth giving early in the patient whose brain is not already destroyed, and not a rescue for the one whose is.

Synthesis. Three trials, and the honest summary is that reversal is well established as a practice and thinly established as a benefit.

What is solid: four-factor PCC corrects the INR far faster than plasma and without a litre of fluid, with a similar safety profile. That is a real and practical advantage, and it is the reason PCC has displaced plasma for vitamin K antagonist reversal in most places.

What is newer and more uncomfortable: andexanet alfa achieved better haemostatic efficacy than usual care in factor Xa-associated intracerebral haemorrhage, and roughly doubled thrombotic events while quadrupling ischaemic stroke. The primary outcome was won on a composite built around haematoma expansion. Nobody has yet shown that the patients did better.

And CRASH-3 is worth holding beside CRASH from day 09, because the same collaboration produced both. Steroids in head injury: clear harm. Tranexamic acid in head injury: a primary outcome that crossed 1, with benefit concentrated in mild to moderate injury treated early. Two large, honest trials of plausible drugs, and neither delivered the clean answer the physiology predicted.

The practical position that survives all three: reverse promptly, choose the agent for speed and volume rather than for elegance, and say out loud that you have just made the patient prothrombotic — because the person who has to decide when to restart anticoagulation is going to need to know exactly what was given and when.

04

What you actually do

in the room

Normalize coagulation as soon as possible when an oral anticoagulant is implicated. This is the one instruction with no serious dissent. The debate is about which agent and what it costs, not about whether to act.
For warfarin, give vitamin K and a factor product — not vitamin K alone — and prefer PCC to plasma when volume matters. Intravenous vitamin K 5 to 10 mg with either 4F-PCC 20 to 40 IU/kg or FFP 20 mL/kg. Vitamin K takes hours to work; the factor product is what corrects the patient now, and the vitamin K is what stops them re-anticoagulating as it wears off. About 1 L of plasma is needed to normalize coagulation — 10 to 15 mL/kg raises factor levels by only 10 to 20% — so in heart failure, or in a patient already oedematous, that litre is its own problem.
Know before the emergency whether your pharmacy stocks idarucizumab. Idarucizumab 5 g reverses dabigatran in life-threatening bleeding or before an emergent procedure. It is a specific antibody fragment, and the time to discover it is not held locally is not at 2 a.m.
For factor Xa inhibitors, treat the choice between PCC and andexanet as genuinely open. Andexanet improved haemostatic efficacy by 13.4 percentage points and roughly doubled thrombotic events. The source text still lists PCC. Whichever your institution uses, be able to state the trade rather than the protocol.
Be honest about platelet transfusion for antiplatelet-associated bleeding. The effect on outcome is not clear, including for dual therapy combining aspirin with a P2Y12 inhibitor, and it carries a class IIb recommendation. That is the weakest evidence on the table and it is often given with the most confidence.
Give tranexamic acid early in traumatic brain injury, or not at all. Benefit in CRASH-3 concentrated in mild to moderate injury, and early treatment beat late in that group. Timing made no obvious difference in severe injury, where no benefit was seen.
Do not give recombinant factor VIIa prophylactically. The data are inconsistent on mortality and transfusion requirement, and the thromboembolic risk is serious enough that it cannot be recommended in patients who are not actively bleeding.
Document exactly what was given and when. Someone will have to decide when to restart anticoagulation in a patient who has a reason to be on it. That decision is made on your record, and a vague one makes it worse.

05

The pitfall

Reversing the drug and forgetting the indication

The reversal is the easy decision and it feels like the whole problem. The bleed is in front of you, the mechanism is obvious, the agent is on the shelf. What is not in front of you is the reason the patient was anticoagulated — the mechanical valve, the recent pulmonary embolism, the atrial fibrillation with a high stroke score — and that reason is still true ten minutes after the drug has been given.

ANNEXA-I put a number on it. Andexanet won its haemostatic endpoint, 67.0% against 53.1%, an adjusted difference of 13.4 percentage points. In the same trial, thrombotic events occurred in 10.3% against 5.6%, and ischaemic stroke in 6.5% against 1.5%. Those are not signal-of-concern rates. That is roughly one extra ischaemic stroke for every twenty patients treated, in a trial whose primary outcome was a composite built around how much the haematoma grew.

Two habits follow. Say the trade out loud when you give the drug, so that the person managing the patient tomorrow inherits a decision rather than an event. And write down exactly what was given and at what time, because the question that arrives on day three — when can we restart anticoagulation — is answered from your notes, and a reversal recorded as "PCC given" is worth very little.

The version of this error that reaches an anaesthetist most often is quieter: the patient who comes to theatre reversed, whose chart says the bleeding has stopped, and for whom nobody has thought about the fact that they are now sitting in a prothrombotic state for the next several days.

06

Sources

primary first

01 Connolly SJ, Sharma M, Cohen AT, et al. Andexanet for factor Xa inhibitor-associated acute intracerebral hemorrhage. N Engl J Med. 2024;390(19):1745-1755. Full text · Preston doi 10.1056/NEJMoa2313040
02 Sarode R, Milling TJ Jr, Refaai MA, et al. Efficacy and safety of a 4-factor prothrombin complex concentrate in patients on vitamin K antagonists presenting with major bleeding: a randomized, plasma-controlled, phase IIIb study. Circulation. 2013;128(11):1234-1243. Full text · Preston doi 10.1161/CIRCULATIONAHA.113.002283
03 CRASH-3 trial collaborators. Effects of tranexamic acid on death, disability, vascular occlusive events and other morbidities in patients with acute traumatic brain injury (CRASH-3): a randomised, placebo-controlled trial. Lancet. 2019;394(10210):1713-1723. Full text · Preston doi 10.1016/S0140-6736(19)32233-0

Full text · Preston goes through LibKey to whatever Preston holds — the PDF where there is one — and works on a phone. The doi link always reaches the publisher's record, whether or not we hold a subscription. If you are reading on a computer, LibKey Nomad adds the same one-click PDF button to PubMed and publisher pages everywhere else; it is a desktop browser extension, so it does nothing on a phone or tablet.

07

Check yourself

3 items · one attempt each

Item 1 Analysis Neuro 1 · Day 13

In ANNEXA-I, andexanet alfa achieved haemostatic efficacy in 67.0% of patients with factor Xa-associated intracerebral haemorrhage versus 53.1% with usual care, adjusted difference 13.4 percentage points (95% CI 4.6–22.2). Which reading is best supported?

Item 2 Application Neuro 1 · Day 13

A 78-year-old on warfarin for a mechanical mitral valve presents with an intracerebral haemorrhage and an INR of 3.8. She has moderate heart failure. What is the best reversal strategy?

Item 3 Recall Neuro 1 · Day 13

CRASH-3 randomized 12,737 patients with traumatic brain injury to tranexamic acid or placebo. What did it show?