A Monte Carlo simulation framework for the actuarial risk analysis of semaglutide coverage
Cardiovascular disease is the leading cause of death in the United States, yet broader insurance coverage of GLP-1 receptor agonists such as semaglutide—priced near $12,000 per year—poses a material financial risk to payers. We constructed a three-stage Monte Carlo simulation drawing on three primary data sources: a nationally representative NHANES cohort (n = 4,429) for baseline PREVENT inputs, 835 MEPS inpatient events for a log-normal hospitalization-cost distribution, and clinical-trial literature for stochastic treatment-effect modeling. Across 1,000 ten-year simulations per patient, avoided ASCVD hospitalizations alone (≤ $500 per patient over ten years) were insufficient to offset drug cost at any risk level. Incorporating a scale factor for broader GLP-1 benefits, we identify two clinically meaningful subgroups—high-risk (PREVENT ≥ 20%, ~2%) and moderate-risk (10–20%, ~13%)—where total savings plausibly justify coverage. We recommend a tiered framework with a dynamic threshold that expands eligibility as drug prices decline.
GLP-1 medications demonstrably reduce cardiovascular risk, but at roughly $12,000 per year the expected savings from avoided hospitalizations alone cannot justify universal coverage. 2 This project builds the actuarial framework needed to determine who should be covered, and why—using Monte Carlo simulation, bootstrap uncertainty estimation, and a tiered PREVENT-score recommendation that concentrates expenditure on the highest-return patients.
The central financial risk is structural: our PREVENT-score distribution shows that roughly 85% of BMI-eligible patients fall into the low-risk category, generating minimal return on a $12,000 annual investment. The gap between the high cost of the drug and the modest projected ASCVD savings of $200–$500 per patient over ten years is the primary driver of insurer financial exposure. The three quantities below frame that trade-off.
Three primary data sources, each chosen for reliability, sample size, and relevance, integrate into a single risk model.
n = 4,429 valid patients (from 6,615; excluded missing values, ages outside 30–79, and pregnant individuals).
BMI, SBP, HDL-C, total cholesterol, eGFR (computed from serum creatinine), smoking, statin use, diabetes history, antihypertensive use.
Baseline PREVENT inputs for a nationally representative cohort.
n = 835 ASCVD inpatient events; pooled across five years, adjusted to 2026 dollars via CPI-M.
ICD-10 I20–I25, I60–I69, I70–I74. Log-normal fit (MLE): μ = 9.518, σ = 1.124; mean $25,676, median $16,633.
Cost distribution for any single ASCVD event.
STEP 1 (NEJM, n~2,000), SUSTAIN 6 & PIONEER 6 pooled (eGFR), BMI study (n=8,857 vs 35,428), SBP meta-analysis (27,000 participants, 29 trials).
BMI effects stratified across four class bins (<29.9, 30–34.9, 35–39.9, 40+).
Normal distributions for treatment-effect modification.
| Data Source | Characterize Outcomes | Severity / Range | Frequency / Likelihood |
|---|---|---|---|
| NHANES (Table 1) | — | — | ✓ |
| MEPS (Table 2) | ✓ | ✓ | — |
| Research papers (Table 3) | — | — | ✓ |
Mean changes and standard deviations derived from clinical-trial distributions, applied stochastically across all 4,429 patients.
| Variable | Mean Change | Std. Dev. | Direction |
|---|---|---|---|
| Blood Pressure | −2.31 mmHg | ±0.41 | ↓ protective |
| Total Cholesterol | −3.3 mg/dL | ±0.31 | ↓ protective |
| HDL Cholesterol | +4.9 mg/dL | ±0.31 | ↑ beneficial |
| eGFR | −0.97 mL/min | ±0.11 | ≈ neutral* |
| BMI (<29.9) | −12.2 kg/m² | ±8.4 | ↓ protective |
| BMI (30–34.9) | −16.5 kg/m² | ±9.2 | ↓ protective |
| BMI (35–39.9) | −16.5 kg/m² | ±8.5 | ↓ protective |
| BMI (40+) | −14.1 kg/m² | ±8.4 | ↓ protective |
* Modest eGFR decline is expected and within acceptable physiological range; it does not represent kidney harm at this magnitude.
A three-stage Monte Carlo procedure: 1,000 independent ten-year simulations per patient, across 4,429 patients, integrating treatment uncertainty, event timing, and hospitalization cost.
For each patient, sample 1,000 treatment effects from the Table 2 normal distributions, apply to baseline variables, and compute 1,000 post-treatment PREVENT scores.
Convert PREVENT to a Poisson baseline hazard. Simulate the ten-year timeline at Δt = 0.01 yr, applying a spike–decay recurrence model after each event and a mortality hazard that terminates accumulation.
Each simulated event draws a cost from LogNormal(μ=9.518, σ=1.124), fitted to 835 MEPS records. Costs are summed per patient per simulation.
The baseline instantaneous hazard is derived from the ten-year cumulative PREVENT probability R; the probability of an event over an interval x follows from the survival function; and post-event risk is elevated by a decaying spike summed over prior events.
| Parameter | Spike Magnitude S | Decay Constant k |
|---|---|---|
| Event risk | 0.2409 | 0.4851 |
| Death risk | 0.2157 | 0.4111 |
Mean squared error: 6.718 × 10⁻⁵ (event), 7.833 × 10⁻⁵ (death).
Each event cost is an independent draw from the MEPS-fitted log-normal distribution.
Log-normal fit to 835 inpatient events from MEPS 2019–2023, adjusted to 2026 dollars. μ = 9.518, σ = 1.124.
Mean ten-year ASCVD costs by risk tier, before and after semaglutide therapy. Bootstrap standard errors estimated across 1,000 resamples per PREVENT percentile range.
Additional GLP-1 benefits plausibly close the remaining gap. A scale factor of 24 is within a defensible range.
Require a one-year lifestyle intervention (CDC DPP, ~$553) before authorization; reassess PREVENT and, if still ≥10%, authorize coverage.
At current pricing the scale factor exceeds 60. ASCVD savings represent under 1.7% of the benefit required to break even.
| PREVENT Score | 10-yr ASCVD Savings | Scale Factor | Recommendation | Population |
|---|---|---|---|---|
| ≥ 20% (High) | > $500 | ≤ 24 | Cover immediately | 2% |
| 10–20% (Moderate) | $200–500 | 24–60 | Cover with step therapy | 13% |
| < 10% (Low) | < $200 | > 60 | Do not cover | 85% |
As drug prices fall under market competition or legislative pressure, the same framework re-identifies cost-effective populations. Adjust the annual cost to observe the effect.
A replicable, actuarially sound decision rule for insurers evaluating semaglutide coverage.
Semaglutide generates savings across disease pathways beyond cardiovascular hospitalization—the empirical basis for the scale factor.
The framework's reach, and why resolving insurer uncertainty matters.
Lives are lost not only when treatments fail, but when proven treatments are never authorized. The primary barrier to semaglutide access in the non-diabetic population is insurer financial uncertainty. A tiered PREVENT-score framework resolves that uncertainty with a replicable, actuarially sound rule—and as prices fall under market and legislative pressure, the dynamic scale factor automatically expands eligibility, saving more lives each year without a new analysis.
The principal simplifying assumptions of the model, with justifications.