What this is

The ANNEXA-I trial showed that andexanet alfa reduces hematoma expansion (HE) in factor Xa inhibitor-associated intracerebral hemorrhage but increased thrombotic events. This secondary analysis proposes a practical bedside rule — prescan hematoma growth rate, diastolic blood pressure, and prior arterial thromboembolism — to triage which patients should receive andexanet.

1. Shadow Audit

The paper is structured as an exploratory secondary analysis of the parent ANNEXA-I RCT. The authors hedge the headline: “due to the exploratory nature of our analyses and insufficient power, we are unable to demonstrate statistically significant treatment interactions for greater absolute benefit in patients at the highest risk of HE, and the numerical trends suggesting this should be considered hypothesis-generating.” That hedge is doing more work than the conclusions admit.

What is absent from the headline framing: The primary trial was stopped early for hemostatic efficacy, but the disability and mortality outcomes were unaffected. The discussion concedes only that “the 30-day duration of follow-up in the trial was too brief to detect differences in disability between the treatment arms, and the study was not powered for death or disability.” Yet the central message — use the “11-to-1 rule” to select patients for andexanet — is built on a surrogate (HE) that the parent trial already showed does not translate into better function at 30 days. The whole clinical question is whether preventing HE translates to better mRS, and the parent trial did not answer that.

No head-to-head with 4-factor PCC in the high-risk quartile. 86.4% of the usual-care arm received PCC. The NNT=4 for HE in the highest-risk quartile is for an unvalidated surrogate against a “PCC + best supportive care” comparator — exactly the comparison that would matter clinically.

Excluded populations are exactly the ones where the rule is most wanted. Exclusion criteria: “a Glasgow Coma Scale score of <7 at the time of randomization, planned surgery to evacuate the hematoma within 12 hours, or a history of TE within the past 2 weeks.” The very patients excluded — comatose, surgical, recent TE — are the ones clinicians most fear to treat or not to treat. The rule cannot be applied to them.

Claims stated without evidence: “DBP is a stronger predictor than SBP” is presented as exploratory, but the rule depends on it; the authors themselves note “the biological underpinning of our observation that DBP is a stronger predictor of HE in FXa inhibitor–associated ICH is uncertain.” The actionability statement — “our estimates favor treatment with andexanet over usual care… except in patients with prior arterial TE who present with low-to-intermediate prescan hematoma growth rates (<11 mL/h)” — rests on NNT/NNH calculations from a 30-day TE count that “the small total number of TE negates further stratification.”

2. Inversion Engine

The headline conclusion is: Use prescan hematoma growth rate, DBP, and prior arterial TE to individualize andexanet use in FXa-inhibitor ICH; treat high-risk (rate >11 mL/h) patients, withhold from low-risk with prior TE. For the opposite to be right — andexanet should not be used routinely in any of these patients because the surrogate target does not translate to function and the thrombotic harm is real — the following would have to hold:

(1) The ANNEXA-I functional outcome data remain null at 12-month follow-up. The mechanism assumption — “the requisite thresholds of hematoma growth of ≥35% and ≥12.5 mL that were used in our definition of HE have been determined to independently predict poor outcome in patients with acute ICH, with positive predictive values of 70% and 80%, respectively, for a modified Rankin Scale score of 4 to 6 at 90 days” — depends on observational data, not on the within-trial experiment showing that preventing HE reduces mRS.

(2) The TE risk is not just statistically significant but clinically undercounted. The NNH=26 is from 30-day TE. If the TE risk extends to 90 days — and post-ICH VTE, stroke, and MI commonly present later — the NNH likely falls well below the NNT.

(3) PCC has a smaller hemostatic effect than reported because dosing was inadequate. Most patients got 4F-PCC at site discretion; non-uniform dosing in the usual-care arm could be inflating andexanet’s apparent benefit. The paper does not report PCC dose or timing.

(4) The prescan hematoma growth rate is a poor operationalization for emergent decision-making. It requires knowing the time of symptom onset (often unknown) and the baseline hematoma volume (requires Quantomo software, “performed at a dedicated imaging core laboratory by trained physicians”). In real-world ED workflow, the precise volume is unknown for the first ~30 minutes and the time-from-onset is regularly estimated.

One assumption that breaks first: the assumption that preventing HE in this trial translates to better function. The parent trial’s primary endpoint was hemostatic efficacy (a composite), not functional outcome. The secondary analyses build HE risk stratification on top of a surrogate that the parent trial already showed doesn’t deliver what patients care about. If the 12-month ANNEXA-I data stay null, the 11-to-1 rule is purely a surrogate-marker exercise.

3. Second-Order Catalyst

First to change practice (0–6 months): Comprehensive stroke centers (CSCs) with andexanet on formulary and Quantomo-style volumetric analysis. Pharma: AstraZeneca/Alexion sales will use the NNT=4 datapoint to push andexanet into the highest-risk quartile explicitly. With list prices ~$20,000–50,000 per dose, even a small shift in the eligible denominator has substantial revenue implications.

Last to change practice (12–24 months): Community hospitals where andexanet is not stocked and PCC is the default. The rule requires volumetric analysis that may not be available. Payers (CMS, UnitedHealthcare) will be slow to reimburse andexanet specifically for the high-risk quartile unless AHA/ASA guidelines endorse it. Expect a draft guideline amendment by mid-2026, formal endorsement by 2027.

What the field does in 12 months: A prospective registry of patients treated under the “11-to-1 rule” with 90-day mRS as the primary outcome. The authors themselves recognize this: “Additional planned secondary analyses of ANNEXA-I will further explore the relationship between systolic blood pressure, DBP, and HE, as well as potential treatment effect modification between acute blood pressure control and andexanet treatment in achieving hemostatic efficacy.” Competition: PCC + tranexamic acid combination studies, and emerging minimally invasive surgical evacuation trials, will be framed as alternatives to andexanet selection.

A second-order catalyst that nobody is talking about: if HE is the right target, this paper implicitly argues that early aggressive BP lowering in FXa-inhibitor ICH (where DBP >95 mmHg predicts HE) matters at least as much as reversal agent choice. The bundle-of-care implication (citing INTERACT3, Code ICH, the expert consensus statement) reorders the priority: BP control first, reversal agent second.

4. Asymmetric Leverage

The largest claim: “One could apply andexanet preferentially to the high-greater-rate quartile (NNT=4) and avoid it in patients with prior arterial TE with low-intermediate growth rates (NNH=12).” If true, this single decision rule changes hospital pharmacy budgets (saves $15k–40k per avoided dose), bed-day utilization (earlier hematoma control → earlier ICU discharge), and litigation patterns (a documented “rule” creates a standard of care). The leverage multiplier is enormous because the underlying FXa-inhibitor population is large and growing.

The smallest claim that costs the most if wrong: “DBP predicts HE in FXa-inhibitor ICH.” If DBP turns out to be a confounder (e.g., correlated with severity of underlying cerebral small vessel disease, which is the actual causal driver), clinicians would inappropriately target DBP in the ED without addressing the underlying vasculopathy. The cost of the wrong claim: hundreds of ICH patients over-treated with IV antihypertensives to a DBP target that doesn’t matter, with attendant risk of worsening perihematomal hypoperfusion.

Where the leverage is concentrated but the evidence is thinnest: the “11-to-1 rule” for individual treatment decisions. The NNT and NNH numbers are derived from a 459-patient cohort with only ~30-day TE follow-up and no functional outcome validation. The clinical decision threshold being proposed is a single number (prescan hematoma growth rate >11 mL/h) that is being asked to carry enormous operational weight.

5. Paradigm Destroyer

What this paper kills:

(1) The “andexanet for everyone with FXa-inhibitor ICH” reflex. The discussion explicitly nudges against blanket use: “our estimates favor treatment with andexanet over usual care in patients presenting with FXa inhibitor–associated ICH who fulfill the ANNEXA-I trial eligibility, except in patients with prior arterial TE who present with low-to-intermediate prescan hematoma growth rates (<11 mL/h).” This is a meaningful shift from the original ANNEXA-I implication.

(2) The “PCC is universally adequate” consensus for FXa-inhibitor ICH. The paper shows that in the highest-risk quartile, usual care (86.4% of which got PCC) still produces a 50–60% HE rate. PCC is not enough in this group; that is a paradigm shift that the discussion makes explicit.

(3) The “SBP is the only BP that matters in ICH” framing. The paper identifies DBP as the predictive variable in anticoagulated ICH (counter to the general ICH literature where SBP dominates). This is presented as exploratory but, if validated, reverses a decade of SBP-centric trial design (ATACH-2, INTERACT3).

What this paper does NOT kill: the role of andexanet in life-threatening FXa-inhibitor bleeds outside the CNS; the general principle that reversal agent choice should be individualized; the role of PCC in lower-risk FXa-inhibitor ICH. And nothing is killed in the disability domain — the paper does not (and cannot) overturn the null functional outcome from the parent ANNEXA-I trial. The surrogate-vs-function tension remains unresolved.

MVP — Minimum Viable Proof

The MVP: A prospective, single-arm registry of 200–300 FXa-inhibitor ICH patients at 5–10 high-volume CSCs, treated under the “11-to-1 rule” (andexanet for growth rate >11 mL/h and no prior arterial TE; PCC otherwise), with the primary outcome being 90-day mRS 4–6, compared to the original ANNEXA-I usual-care arm (PCC) as a benchmark. Cost: ~$1–2M (registry infrastructure, not a trial). Time: 18–24 months. Required because the parent trial’s primary endpoint was hemostatic efficacy, not function, and the paper’s clinical implications are entirely about functional significance.

A cheaper MVP (#2): A prespecified retrospective analysis of the 12-month ANNEXA-I follow-up data, comparing 90-day and 12-month mRS stratified by the quartiles used in this paper. If the parent trial sponsors have already collected this (likely), the cost is near zero and the answer is achievable in 6 months.

Best Combination

Adopt the “11-to-1 rule” as a triage tool, not a definitive treatment decision — but bundle it with aggressive BP control and the ICH bundle of care. The asymmetric payoff is:

If HE prevention does translate to function (parent trial’s 12-month follow-up confirms efficacy): you get the full NNT=4 benefit at the highest-yield subgroup.If HE prevention does NOT translate to function (the more conservative reading): you haven’t lost anything because aggressive BP control and bundle-of-care are themselves evidence-based (INTERACT3, Class I in many AHA statements).If andexanet is overused: the rule prevents that. NNH=26 in the full sample becomes NNH=12 in patients with prior stroke/MI; the rule explicitly excludes the latter group from andexanet.The single move: Operationalize the prescan hematoma growth rate at the bedside using a simple volumetric approximation (ABC/2 or similar) divided by the time-from-onset, and use it to drive a triple-tier decision: high rate + no prior TE → andexanet; high rate + prior TE → PCC; low rate → PCC. Bundle with target DBP <95 mmHg. This is a rule that is mostly right under both scenarios, and the cost of being wrong is bounded (you’ve still done bundle-of-care).

Overvalue Warning

Surrogate-endpoint inflation. HE is the primary outcome; mRS is not measured. The paper depends on the assumption that HE prevention → better function. The parent trial showed equivalent function at 30 days; the paper inflates this surrogate into a clinical-action rule.

Industry sponsorship bias. “Initially funded by Portola, which was purchased by Alexion AstraZeneca Rare Disease. Portola, followed by Alexion, was the trial sponsor and provided the andexanet free of charge.” The sponsor did not approve submission, but the entire framing (a “rule” that prioritizes andexanet in the highest-risk subgroup) is commercially aligned.

Subgroup analysis inflation. Quartile-based subgroup analyses are notoriously prone to spurious findings. The trial was not powered for treatment-by-quartile interactions; the authors explicitly say “due to the exploratory nature of our analyses and insufficient power, we are unable to demonstrate statistically significant treatment interactions.” Yet the rule is built on these interaction-looking results.

Sample size and event counts. Only 459 patients, only ~150 HE events, and only ~30-day TE. The number needed to harm calculations are derived from very small TE counts.

Operational feasibility of the predictor. Prescan hematoma growth rate requires both an accurate symptom-onset time AND a precise volumetric measurement. In real-world ICH, both are typically noisy. The rule may select the wrong patients in routine practice.

Novelty and recency bias. ANNEXA-I (2024) is fresh; the secondary analyses are even fresher. The temptation to “do something” with andexanet in high-risk FXa-inhibitor ICH is real, and the rule provides a defensible rationale — but defensibility is not the same as correct.

Mechanism plausibility gap for DBP. The authors themselves admit “the biological underpinning of our observation that DBP is a stronger predictor of HE in FXa inhibitor–associated ICH is uncertain.” They invoke ROCKET-AF as a parallel but the analogy is loose. The single most actionable BP-related claim in the paper rests on a soft observation.