Appraising a new oral anticoagulant trial for the P&T committee

Statistical: p = 0.03 is below 0.05, but the upper confidence limit of 0.98 almost touches 1. The data are therefore compatible with a relative reduction as large as 28% or as small as 2% - a borderline result from a single trial, not a robust one.

Absolute:

• Event rates 9.6% vs 8.1% → ARR = 1.5%, RRR ≈ 16%.
• NNT = 1/0.015 ≈ 67, i.e. about 67 patients treated for the ~2.8-year trial duration to prevent one stroke or systemic embolism.
• Using the confidence limits, the ARR spans roughly 0.2% to 2.8%, so the NNT ranges from about 36 to about 500 - that width is the honest message for the committee.

Bleeding: no significant difference, but the confidence interval is compatible with anything from a 20% relative reduction to a 10% relative increase in major bleeding. The safety claim is 'no worse than warfarin demonstrated to a limited degree', not 'safer'.

Presentation: show both the relative and the absolute numbers on one slide, state the NNT range rather than the point estimate alone, and explicitly flag that warfarin's comparator performance depends on the time in therapeutic range achieved in your own clinic. A drug that beats poorly-managed warfarin may not beat well-managed warfarin.

Limited weight - it is hypothesis-generating, not decision-making.

• It is underpowered: 1,400 patients with a wide CI (0.85 to 1.48) is compatible with a meaningful benefit or a substantial harm.
• Multiplicity: with many prespecified subgroups, one containing the null is expected by chance; a subgroup effect should be supported by a significant interaction test, and none is reported here.
• Direction of the point estimate still matters: HR 1.12 in the group at highest absolute stroke risk is not reassuring, and it deserves comment even if it does not change policy.
• Check whether the absolute event rates were higher in the elderly subgroup, which would make the same relative effect clinically more valuable, and whether the bleeding signal moved in the same direction.

How to use it: do not restrict the drug to under-80s on the basis of this alone, and do not ignore it. Use it to justify a specific caution for frail elderly patients - dose criteria, renal function monitoring, falls assessment, review of interacting drugs - and to ask for real-world registry data in over-80s at the 12-month review.

Recommendation: restricted formulary addition, not open listing.

Approved for: stroke prevention in non-valvular atrial fibrillation where the patient has labile INR control (TTR <65%) despite adherence, documented warfarin intolerance, or inability to attend INR monitoring - subject to renal function criteria and exclusion of mechanical heart valves or moderate-to-severe mitral stenosis.

Conditions

1. Prescriber criteria: initiation by cardiology, internal medicine or the anticoagulation service against a written checklist, including weight, age and creatinine for dose selection.
2. Monitoring: renal function at baseline and at least annually (more often if CrCl <60 mL/min), full blood count, and a documented bleeding-risk assessment and falls review in patients over 80.
3. Capacity answer: this is the argument the committee actually needs. Redirect stable, well-controlled patients to continued warfarin and use the freed clinic capacity for the unstable group - the drug does not solve a capacity problem by itself.
4. Economic framing: at NNT ~67 over ~3 years, cost per event avoided should be calculated against the local acquisition price and presented alongside the option of investing in INR control quality instead.
5. Audit: bleeding events, thromboembolic events, off-criteria prescribing, and a formal reassessment at 12 months, at which point the elderly subgroup data can be revisited with real-world evidence.
6. Explicitly note the fragility of the primary result (CI upper limit 0.98) in the minutes so that the decision is transparent and revisitable.