Industry: Healthcare and Epidemiology Product: ModelRisk Application: Health outcomes — cost-effectiveness analysis
Health-technology assessment bodies — NICE in the UK, ICER in the US, CADTH in Canada — set willingness-to-pay thresholds around $50,000–$150,000 per quality-adjusted life year (QALY). A new biologic for moderate-to-severe psoriasis with an annual acquisition cost near $28,000 sits in the zone where reimbursement is decided by the probability the intervention is cost-effective at the threshold, not by a single deterministic ICER. The cost-effectiveness acceptability curve (CEAC) below is the chart NICE and ICER actually want to see — the entire probability-vs-threshold relationship — and on this biologic it tells a blunt story: at every conventional threshold the probability of cost-effectiveness is low, and only above roughly $274,000 per QALY does it cross 50%.
A health economics team built the probabilistic sensitivity analysis (PSA) for the new biologic in ModelRisk over a 10-year horizon, with parameter uncertainty on response rates, utility weights, costs and adverse-event burden. The mean ICER came out at $306,000 per QALY, with a median of $273,000 — far above the $50k–$150k decision band. The Monte Carlo PSA showed the probability of cost-effectiveness was effectively 0% at $50k, 0% at $100k, and just 5% at $150k. That is the verdict a point estimate would have obscured: at its current list price the biologic is not cost-effective at any conventional threshold, and the CEAC quantifies exactly how large a price concession the math demands.
PASI 75 response rate, treatment arm: Beta(76, 24) — mean 76%, 90% CI 67%–83%, calibrated from a Phase III pivotal trial. PASI 75 response rate, control (standard care): Beta(28, 72) — mean 28%, 90% CI 21%–35%, from real-world comparator data.
The Beta distribution is the only defensible choice for a response rate — bounded on [0, 1], conjugate to Binomial trial data, and allows the asymmetry that small-sample estimates always carry. A Normal here would let the simulator draw negative response rates.
Utility weight for responders: Normal(0.85, 0.04) clipped to [0, 1] — the standard EQ-5D-derived utility for a PASI 75 responder. Utility weight for non-responders: Normal(0.65, 0.06) clipped to [0, 1].
Annual treatment cost: LogNormal centered on $28,000 with σ_log = 0.15 — captures wholesale acquisition cost variation plus contract rebates. Annual standard-care cost: LogNormal centered on $3,200 with σ_log = 0.25 — wider relative uncertainty because comparator care is heterogeneous. Annual adverse-event cost (treatment only): LogNormal centered on $900 with σ_log = 0.5 — heavy-tailed because biologics carry rare severe-AE risk.
All costs and QALYs were discounted at 3% per year over the 10-year horizon, per UK and US health-economic guideline practice.
Twenty thousand PSA draws produced the joint distribution of incremental cost and incremental QALY per patient.
Every point is one PSA draw. The x-axis is incremental QALY (positive in 99.7% of draws — the treatment is more effective); the y-axis is incremental cost (positive in 100% of draws — the treatment is more expensive). The three dashed rays are the willingness-to-pay thresholds: any point below a given ray is cost-effective at that threshold. The cloud sits high above all three rays: only 0.3% of draws fall below the $50k ray, and 95% fall above the $150k ray — not cost-effective at any conventional threshold. The treatment clearly works (mean incremental gain 0.85 QALY), but at this price the cost per QALY lands well outside the reimbursement band.
The simulated ICER has a median of $273,000/QALY and a mean of $306,000/QALY — the mean sits above the median because the right tail dominates, a right-skew that comes from the LogNormal cost priors and the bounded-above utility weights. The distribution is decisive rather than borderline: there is a 78% probability the ICER exceeds $200,000/QALY and a 42% probability it exceeds $300,000/QALY. A deterministic point estimate near the median would have made the answer look merely "high"; the full distribution shows it is high with very little chance of being otherwise.
The CEAC at the top of this article reads: at $50k/QALY, probability of cost-effectiveness is effectively 0%; at $100k/QALY, still effectively 0%; at $150k/QALY, 5%. The probability does not reach 50% until a willingness-to-pay of roughly $274,000/QALY. The recommendation calculus follows directly: at the list price, no conventional threshold supports reimbursement — the question is not whether a price concession is needed but how large it must be to pull the curve into the decision band.
The biologic's manufacturer used the CEAC to frame the negotiation around price: because incremental cost is almost entirely drug-acquisition cost, the ICER scales nearly linearly with price, so closing the gap from a $273k median ICER to a $150k target implies a list-price reduction on the order of 45%. The manufacturer paired the concession analysis with a value-based contract structure that tied a portion of revenue to real-world PASI 75 rates relative to the Beta(76, 24) prior.
The responder utility weight has the biggest single effect on the ICER — a ±0.05 shift moves the mean ICER by roughly ±$72k/QALY, because utility scales the entire QALY denominator. Drug acquisition cost is second: a ±15% price change moves the ICER by roughly ±$44k/QALY. The treatment and control response rates come next, each worth about ±$28k/QALY for a ±5-percentage-point shift.
The implication for the manufacturer: drug price is the lever it actually controls, and a ±15% move on price is worth ±$44k/QALY — enough that a substantial concession is the only realistic route into the threshold band. The implication for the payer and the HTA committee: the utility weight is the single most influential evidence parameter, so tightening the EQ-5D utility estimate for responders is the highest-value analytic investment after the price negotiation.
A single ICER answers "is it cost-effective on the central estimate?" — but reimbursement committees today ask "with what probability is it cost-effective at our threshold?" Monte Carlo simulation in ModelRisk is what makes the CEAC computable, the question askable, and the price-and-evidence negotiation defensible.