DOAC Reversal: What the ANNEXA-I Trial Actually Showed About Andexanet Alfa

ANNEXA-I, published in the New England Journal of Medicine in 2024, was the first randomized trial of a factor Xa-inhibitor reversal agent in intracranial bleeding. Andexanet alfa controlled hematoma expansion better than usual care, but roughly one in ten patients had a thromboembolic event, mostly ischemic stroke. That split, a bleeding endpoint that improved while a clotting signal worsened, is the whole story of this trial and the reason it settled less than many people hoped.

Why this trial mattered

Direct oral anticoagulants such as apixaban and rivaroxaban block factor Xa to prevent strokes and clots. When someone taking one of these drugs suffers a brain hemorrhage, that same anticoagulation can let the bleed keep growing. Andexanet alfa is a modified, inactive form of factor Xa that acts as a decoy, soaking up the inhibitor so the body's own clotting can resume.

The drug reached the United States market in 2018 under an accelerated pathway, based on a surrogate marker: it lowered anti-factor Xa activity in blood. What it did not yet have was evidence that it changed outcomes patients actually feel. As a condition of that approval, the manufacturer was required to run a randomized trial. ANNEXA-I (ClinicalTrials.gov NCT03661528) is that trial, and it deserves a careful read because it is the first randomized comparison of any DOAC reversal agent against standard treatment.

What ANNEXA-I measured

ANNEXA-I enrolled patients with acute intracerebral hemorrhage who had taken a factor Xa inhibitor within roughly 15 hours before presenting. Participants were randomized to andexanet alfa or to usual care, which in most cases meant prothrombin complex concentrate, a pooled clotting-factor product used off-label for this purpose. In all, 263 patients were assigned to andexanet and 267 to usual care.

The primary endpoint was a composite of hemostatic efficacy at 12 hours: hematoma expansion held to no more than 35 percent, no large worsening on the NIHSS stroke scale, and no need for rescue therapy. According to the NEJM report and the American College of Cardiology's trial summary, hemostatic efficacy was achieved in 67.0 percent of the andexanet group versus 53.1 percent with usual care, an adjusted difference of about 13 percentage points that was statistically significant. The trial was stopped early once this signal on bleeding control crossed a prespecified threshold at an interim analysis.

The number that complicates the picture

Stopping early for a positive bleeding result sounds like a clean win. The safety data are what make ANNEXA-I a genuinely hard trial to interpret.

Thromboembolic events occurred in 10.3 percent of patients who received andexanet alfa, compared with 5.6 percent on usual care. The gap was driven largely by ischemic stroke, which struck 6.5 percent of the andexanet group versus 1.5 percent of the usual-care group. In other words, the same mechanism that helped stop one clot-related problem, the expanding bleed, appears to have pushed some patients toward the opposite problem, a clot in the brain's arteries.

This is not a paradox once you consider what the drug does. Rapidly switching off anticoagulation in an older population that was on a blood thinner for a reason, often atrial fibrillation or prior clots, removes protection those patients still needed. A decoy that restores clotting does not distinguish between the clot you want to prevent and the clot you do not.

Better bleeding control, but did patients do better?

The most important caveat is what happened to outcomes that matter to a person and their family. Reduced hematoma expansion is a radiographic finding, a smaller blot on a CT scan. It is a reasonable stand-in for benefit, but it is not the same as surviving, thinking clearly, or walking.

On those harder endpoints, ANNEXA-I did not show that andexanet came out ahead. Mortality at 30 days was similar between the groups, in the mid-to-upper 20 percent range in both arms, and measures of disability did not favor the drug. A 2024 editorial in Research and Practice in Thrombosis and Haemostasis by Richard Buka captured the tension bluntly, noting that the trial improved a surrogate marker while showing no benefit for death or disability and an increased risk of thrombosis, predominantly stroke. When a treatment tightens control of the target lesion but the clinical scoreboard does not move, the honest conclusion is that we have learned something real about mechanism and something humbling about how loosely surrogate endpoints can track outcomes.

How to read this as evidence

A few appraisal points are worth holding onto. First, the comparator matters: usual care here was largely prothrombin complex concentrate, itself unproven in randomized trials for this setting, so ANNEXA-I compares one imperfectly evidenced approach against another. Second, early stopping for benefit tends to inflate the apparent size of an effect, so the efficacy difference may look larger on paper than it would prove over a full enrollment. Third, a roughly 5 percentage-point absolute increase in thrombosis is not a footnote; in a condition where the baseline risk of death is already high, a treatment must earn its place against that harm.

None of this makes andexanet alfa useless. It makes the decision genuinely situational, weighed case by case by the treating team against a specific patient's bleeding severity, timing, and clotting risk. This article is educational and is not medical advice. What ANNEXA-I offers is not a verdict but a clearer map of the tradeoff: real control of hematoma growth, a real thromboembolic cost, and no demonstrated gain yet in survival or function. That is a more useful thing to know than a false sense of a settled answer.