Formulary Decision for Apixaban Versus Warfarin Under a Cost-Neutrality Mandate

Absolute risk reductions (per patient-year):

• Stroke: 2.3% − 1.4% = 0.9% → NNT ≈ 111
• Major bleeding: 3.1% − 2.5% = 0.6% → NNH ≈ 167

Incremental drug cost per patient-year: $5,600 − $1,150 = $4,450

Event offsets per patient-year:

• Strokes avoided: 0.009 × $38,000 = $342
• Bleeding avoided: 0.006 × $14,000 = $84
• Total offset: $426

Net incremental cost: $4,450 − $426 = $4,024 per patient-year → Cost per stroke avoided = $4,024 / 0.009 = ≈ $447,000 → Plan-wide budget impact for 12,000 members = 12,000 × $4,024 = ≈ $48.3 million per year

Is cost neutrality achievable? No. For the drug premium to be offset by events, the stroke cost would have to be roughly $494,000 per stroke (4,450/0.009), which is an order of magnitude above the first-year figure used. Older-generation stroke cost estimates and multi-year accumulations (rehabilitation, nursing home, lost productivity, caregiver costs) can push per-stroke cost into six figures and materially reduce the ratio - at $100,000 per stroke the net falls to ≈ $3,550 per patient-year, or ≈ $394,000 per stroke avoided - but never to zero.

Value framing: $447,000 per stroke avoided is expensive against a $100,000/QALY benchmark if strokes averted carry only a few QALYs each, but that calculation depends entirely on the utility assumptions and the multi-year cost horizon. State those assumptions explicitly rather than defending $447,000 as if it were a QALY figure - it is not.

Abandonment breaks the model in a specific direction. The 30% who abandon do not generate savings - they revert (or default) to warfarin, so the plan captures neither the drug cost nor the benefit; it just fails to realize the clinical gain. The committee should model the realized population: 8,400 effectively treated, 3,600 effectively untreated, and probably an excess of untreated patients with poor INR control. The incremental cost of the formulary action is therefore ~$4,024 × 8,400 ≈ $33.8 million, with the clinical benefit proportionally reduced - a worse buy than the naive calculation suggests.

Design recommendations:

1. Reduce patient cost share to keep copay at or below $75/month. The plan pays more per fill but recovers the intended event reduction; abandonment is the dominant leak, and closing it is cheaper than the drug premium.
2. Prior authorization for indication and dose only (AF, appropriate renal dosing), not step therapy through warfarin - step therapy in a population that will simply not persist on warfarin mostly manufactures abandonment.
3. Prefer apixaban over other DOACs where efficacy and bleeding profiles are distinguishable and cost is comparable - the committee is choosing among DOACs, not only between DOAC and warfarin.
4. 90-day fills and adherence outreach for the first two fills, where abandonment and early discontinuation cluster.
5. Outcome and persistence metrics (proportion of days covered, stroke hospitalization rate) reported at 12 months, with a contractual trigger to revisit.

Message to the CFO: cost neutrality is not achievable; the honest question is whether the plan will buy a 39% relative reduction in stroke for about $4,000 per patient-year, and how much of that it will actually capture given abandonment.

Assumption 1 - the stroke cost figure. $38,000 first-year cost is the single most influential input. If the plan accounts for multi-year post-stroke costs (rehabilitation, long-term care, readmissions, caregiver burden) or if a substantial share of strokes occur in members who cross into Medicare (where the plan's exposure ends), the offset changes direction. Test: run a one-way sensitivity from $38,000 to $200,000 and a two-way sensitivity with bleeding cost; report the threshold value at which the program becomes budget-neutral. Also disaggregate by member age band, since cost exposure is age-dependent.

Assumption 2 - adherence and persistence. The 30% abandonment figure applies to fills; persistence at 12 months for DOACs commonly runs 65–80%, and the event rates in the trial-derived inputs assume continuous therapy. If real-world persistence is 70%, the effective ARR falls from 0.9% to roughly 0.6%, and NNT rises to about 167 - the cost per stroke avoided exceeds $650,000. Test: model 60%, 70%, and 85% persistence; pair with a 90-day persistence pilot and use the observed PDC in the next cycle.

Other inputs worth a tornado diagram: the warfarin monitoring cost (assumes tight INR control; poor control in the real world worsens warfarin's outcomes and favors apixaban), the bleeding costs, and the proportion of members with prior stroke or renal impairment (higher-risk subgroups show a larger absolute benefit and a better ratio - a targeted program confined to CHA₂DS₂-VASc ≥3 could be defensible even where a whole-population mandate is not). Recommendation structure: approve with the copay mitigation and a 12-month evidence review, not an unconditional open formulary.