Decoding Medicare's Cost Projections: Why the Leqembi Alzheimer's Drug Estimate Missed the Mark

Overview

When Medicare actuaries issued their 2025 cost projection for the Alzheimer's drug Leqembi (lecanemab), they anticipated a staggering $3.5 billion expense for the program. Two years later, that forecast appears wildly inaccurate—a classic case of predictive hubris. This tutorial unpacks the mechanics behind Medicare's drug cost estimation, using Leqembi as a case study to reveal why even sophisticated actuarial models can fail. You'll learn the step-by-step process actuaries use, the data assumptions that lead to errors, and how to avoid common pitfalls when analyzing new drug spending.

Prerequisites

Before diving into the estimation process, ensure you have a baseline understanding of:

• Medicare Part B vs. Part D: Leqembi is administered via infusion, placing it under Part B (medical insurance), which covers outpatient drugs administered in a clinical setting.
• Alzheimer's Disease Pathology: Leqembi targets amyloid-beta plaques, but its clinical benefit is modest, affecting adoption rates.
• Actuarial Modeling: Basics of population forecasting, incidence rates, and cost modeling using tabular data.
• Drug Pricing and Reimbursement: Medicare pays for Part B drugs under the Average Sales Price (ASP) plus 6%, which influences total spending.

Step-by-Step Instructions

Step 1: Define the Population at Risk

Start with Medicare beneficiaries diagnosed with mild cognitive impairment (MCI) or early-stage Alzheimer's—the indicated population for Leqembi. Use CMS claims data or prevalence estimates. For 2025, actuaries likely used:

• Total Medicare beneficiaries (≈66 million in 2025 projection)
• Alzheimer's prevalence (≈10% of those aged 65+; higher for 75+)
• Proportion with early-stage disease (≈30-40% of diagnosed)
• Probability of seeking treatment (initially high due to media buzz)

Code example (pseudo Python):

total_beneficiaries = 66_000_000
alz_prevalence = 0.10
diagnosed = total_beneficiaries * alz_prevalence
early_stage_frac = 0.35
treatment_seek_rate = 0.80  # optimistic assumption
estimated_patients = diagnosed * early_stage_frac * treatment_seek_rate
print(f"Estimated patients: {estimated_patients:,.0f}")

This yields roughly 1.85 million potential patients—a number that seems unrealistic given real-world limitations.

Step 2: Estimate Annual Drug Cost per Patient

Leqembi is dosed every two weeks (26 infusions per year). The initial list price was $26,500 per year. Medicare's ASP reimbursement averages around list price minus mandatory discounts (e.g., 23% for Part B drugs with a rebate rule, though Leqembi may have exemptions). For modeling, use:

• Net annual cost per patient: $26,500 × (1 - average discount factor, say 0.77) ≈ $20,405
• Plus infusion and administration costs: often an additional 30% of drug cost, so ~$6,121, for total ~$26,526 per patient per year.

However, many patients discontinue treatment due to side effects (e.g., ARIA) or lack of efficacy, reducing real-world cost.

Step 3: Multiply by Adoption Rate and Adjust for Attrition

Not all eligible patients ultimately receive treatment. Actuaries apply an adoption curve. The original $3.5 billion figure likely assumed:

• Adoption rate of 5-7% of eligible patients in the first year, ramping to 15-20% by 2025.
• Monthly discontinuation rate: 2-5%.

Simplified annual cost projection:

avg_patients_per_year = 1_000_000  # from step 1 after adoption/attrition
cost_per_patient = 26_526
total_cost = avg_patients_per_year * cost_per_patient
print(f"Projected annual cost: ${total_cost:,.0f}")

That yields $26.5 billion—way above $3.5B. So clearly the actuaries used more conservative assumptions, e.g., slower adoption and lower attrition.

Step 4: Incorporate Real-World Constraints

Medicare's actuaries layer on constraints:

• Diagnostic hurdles: Requiring PET scans or spinal taps for amyloid confirmation; only ~20% of eligible patients are diagnosed early enough.
• Infusion capacity: Limited infusion clinic slots restrict how many can start treatment.
• Coverage policies: CMS's coverage with evidence development (CED) for Alzheimer's drugs until July 2023 likely lowered uptake.
• Provider reluctance: Neurologists often hesitant due to risk.

Applying a 10% downstream factor (from estimated 1.85M to 185,000 treated) brings total cost to $4.9B, closer to $3.5B after further discounting for price negotiations.

Step 5: Run Monte Carlo Simulations

Actuaries run thousands of simulations varying assumptions. The $3.5B median outcome assumed moderate adoption and frequent discontinuation. The actual 2025 spending is likely much lower due to:

• Lower-than-expected diagnosis rate (insurance barriers for PET scans)
• Negative media from ARIA deaths
• Competition from donanemab (Eli Lilly) splitting the market
• Price reductions after Medicare negotiation under Inflation Reduction Act

Common Mistakes

Ignoring Real-World Eligibility Criteria

Many forecasters use broad prevalence without accounting for the narrow indicated population (MCI stage only). Leqembi is not for moderate or severe Alzheimer's. Overestimating eligible patients is the #1 error.

Overlooking Infusion Capacity Constraints

Infusion clinics cannot handle millions of new patients overnight. Medicare's own data shows only ~2,000 centers equipped for Leqembi's monitoring requirements (multiple MRIs). This caps monthly starts.

Assuming Linear Adoption

Adoption rarely follows a straight line. Early hype (like aducanumab) creates a spike, then a plateau. The $3.5B forecast assumed continued growth, but real-world data shows flat or declining starts after initial surge.

Misjudging Discontinuation Rates

Clinical trials show ~10% discontinuation due to ARIA, but real-world rates may be higher (20-30%). Persistent brain swelling requires stopping treatment, turning a chronic cost into a one-time expense.

Neglecting Medicare's Negotiation Power

The Inflation Reduction Act grants Medicare the ability to negotiate prices for high-spend drugs from 2026 onward. Leqembi may be selected, slashing costs by 30-60%, which the original $3.5B model predated.

Summary

Medicare's $3.5 billion Leqembi cost projection for 2025 was a classic forecasting failure—not because the model was technically flawed, but because it relied on optimistic assumptions about diagnosis rates, adoption, and durability of use. By breaking down the estimation process step by step, we see how small changes in input parameters cascade into huge spending differences. For anyone involved in healthcare analytics, this case emphasizes the need to ground assumptions in real-world constraints, run sensitivity analyses, and update models with early market data. The real 2025 spending will likely land between $500 million and $1.2 billion, a far cry from the initial prediction.