Reading a noninferiority trial of a new anticoagulant and applying it to an older patient
A colleague forwards a phase 3 randomized trial of a new oral factor XIa inhibitor versus apixaban in 21,000 patients with nonvalvular atrial fibrillation and CHA2DS2-VASc of 3 or more. Mean age is 71 years; patients older than 80 were eligible only if creatinine clearance exceeded 50 mL/min. Median follow-up is 2.4 years. Results: stroke or systemic embolism 1.6% per year with the new drug versus 1.8% per year with apixaban (hazard ratio 0.89, 95% CI 0.72–1.10). Major bleeding 3.1% versus 3.6% per year (HR 0.86, 95% CI 0.74–0.99). Intracranial hemorrhage 0.4% versus 0.8% per year. Permanent discontinuation 21% versus 16%. The prespecified noninferiority margin was an upper confidence limit of 1.38 for the primary efficacy outcome. The primary analysis excluded 4% of randomized patients who never received a dose. Your patient is an 82-year-old man, 68 kg, creatinine clearance 38 mL/min, CHA2DS2-VASc 5, HAS-BLED 4, with a prior gastrointestinal bleed, currently tolerating apixaban.
1. Has noninferiority been established for the primary efficacy outcome? Explain what the confidence interval does and does not tell you.
Read the worked answer
Yes, noninferiority is met. The upper bound of the 95% CI is 1.10, which is below the prespecified margin of 1.38. The trial has therefore excluded a clinically important excess of stroke with the new drug - that is all noninferiority means. It does not establish superiority: the point estimate of 0.89 favors the new drug and the CI includes 1.00, so a true benefit is not demonstrated, and the result is compatible with anything from a 28% reduction to a 10% increase in stroke.
Several features limit how far this can be pushed:
- The primary analysis excluded 4% of randomized patients who never received a dose. That is a modified intention-to-treat approach; excluding randomized patients can bias a noninferiority conclusion if the exclusions are unbalanced or related to outcome.
- Discontinuation was substantially higher with the new drug (21% vs 16%). Noninferiority is fragile when adherence differs - the efficacy conclusion depends on patients actually taking the drug.
- The margin itself (1.38) is generous. A trial designed to exclude a 38% relative increase cannot rule out a difference that many clinicians would consider clinically meaningful.
- Bleeding results are direct, not inferred: major bleeding and intracranial hemorrhage show CIs that exclude 1.00 (for major bleeding, 0.74–0.99), so those benefits are more robust than the stroke comparison.
In short: the drug is a plausible alternative on efficacy, and its case rests largely on less bleeding.
2. Calculate the absolute differences and numbers needed to treat or harm for stroke, major bleeding, intracranial hemorrhage and discontinuation, and say what they mean for decision-making.
Read the worked answer
All rates are per year of treatment, so the numbers are per patient-year.
| Outcome | New drug | Apixaban | Absolute difference | NNT/NNH (per patient-year) |
|---|---|---|---|---|
| Stroke/systemic embolism | 1.6% | 1.8% | 0.2% lower | NNT 500 |
| Major bleeding | 3.1% | 3.6% | 0.5% lower | NNT 200 |
| Intracranial hemorrhage | 0.4% | 0.8% | 0.4% lower | NNT 250 |
| Permanent discontinuation | 21% | 16% | 5% higher | NNH 20 |
Interpretation. A 0.89 hazard ratio sounds impressive, but the stroke difference is 2 events per 1,000 patient-years - 500 patients treated for a year to prevent one stroke. Meanwhile the bleeding and intracranial hemorrhage advantages are of similar or larger absolute magnitude, and the excess discontinuation (1 in 20 patients per year) is by far the largest absolute difference in the trial. That asymmetry matters: the drug's realistic selling point is bleeding avoidance, and its realistic cost is that one in five patients stops it within a year, so real-world persistence will be worse than the trial.
This is the classic place where relative risk and absolute risk diverge, and where a patient with a high baseline stroke rate (and therefore higher absolute benefit) is precisely the one who also has the highest bleeding risk.
3. Would you recommend switching your 82-year-old patient? Show how you weigh applicability, and what you would say to him.
Read the worked answer
Applicability is the problem, not the statistics. Consider the domains of applicability one by one:
- Age: the mean was 71, and patients over 80 were only enrolled with creatinine clearance above 50 mL/min. He is 82 with a clearance of 38 - he falls outside the population that generated the evidence.
- Renal function: clearance of 38 mL/min is moderate impairment; the trial's pharmacokinetic subgroup will be small, and the drug's renal elimination and dosing in impairment must be established before it is used here.
- Bleeding risk: HAS-BLED 4 with a prior GI bleed is a group likely underrepresented, and also the group that stands to gain most from less intracranial hemorrhage. Both statements are true at once.
- Current tolerability: he is already tolerating apixaban without bleeding. Switching a stable patient for a small absolute bleeding gain, using drugs and formulations with which the team has less experience, is not a neutral act.
Reasonable position: continue apixaban for now, note the new evidence, and revisit if a trial in older patients with renal impairment reports, or if he develops bleeding on apixaban. This is a preference-sensitive decision, so present it as a choice rather than a directive.
What to tell him: the new drug prevented about the same number of strokes as apixaban in the trial, caused somewhat less serious bleeding and less bleeding into the brain, but the people studied were younger with better kidney function than him. His current drug is working, and if he starts bleeding we will reconsider. Ask what matters most to him - avoiding stroke, avoiding a bleed, or avoiding a change - and document the answer.