Forecasting the plan-level financial consequence of adding a new therapy to formulary — population sizing, uptake curves, PMPM, and net budget impact for payers
Every budget impact analysis (BIA) begins with the same question: how many of the payer's own members will actually receive this therapy? Unlike cost-effectiveness analysis, which asks whether a treatment is worth its price for an average patient, BIA asks a plan-specific accounting question — and that means the population estimate must be built from the payer's own membership, not from a national trial population.
A defensible eligible-population estimate is built as a cascading funnel, each stage multiplying the prior:
1. Plan membership: total covered lives (commercial, Medicaid, Medicare Advantage segments modeled separately — cost-sharing and rebate structures differ materially) 2. Disease prevalence: sourced from published epidemiology (SEER, CDC surveillance, claims-based algorithms in IQVIA or Optum Clinformatics) — typically expressed per 100,000 covered lives, adjusted for the plan's demographic mix (age/sex distribution differs from national averages) 3. Diagnosis rate: fraction of prevalent cases formally diagnosed and coded (ICD-10-CM) in claims — often 60–85% depending on disease awareness and screening intensity 4. Treatment-eligible rate: fraction of diagnosed patients meeting label/guideline criteria (line of therapy, biomarker status, contraindications) — pulled from NCCN or specialty-society guidelines 5. Plan-specific adjustment: member turnover (average 15–20%/year commercial disenrollment), seasonal enrollment, and benefit design restrictions (step therapy, site-of-care mandates)
Each multiplier compounds uncertainty, so ISPOR's Budget Impact Analysis Good Practice II Task Force (Sullivan SD, Mauskopf JA, Augustovski F, et al., Value Health 2014;17(1):5–14) recommends reporting the population estimate with an explicit derivation table, not a single opaque number, so the P&T committee can audit every assumption independently.
A 2019 ICER budget impact reassessment of a gene therapy launch found that manufacturer-submitted eligible-population estimates were 30–40% higher than claims-derived plan estimates — almost entirely because the manufacturer used trial-eligibility criteria instead of real-world diagnosis and treatment-initiation rates. Payers now routinely request claims-anchored population tables as a formulary-submission requirement.
The Academy of Managed Care Pharmacy's AMCP Format for Formulary Submissions (v4.1) formally requires manufacturers to submit a budget impact model alongside clinical and economic evidence when petitioning for formulary inclusion. Population inputs are expected to be traceable to named sources:
• Administrative claims databases: IQVIA PharMetrics Plus, Optum Clinformatics Data Mart, Merative (formerly IBM) MarketScan — used to validate prevalence and diagnosis rates against real-world coding patterns rather than trial populations • CDC/NIH surveillance data: for population-level prevalence and incidence benchmarks • Plan-specific eligibility files: enrollment counts by product line, since Medicaid, commercial, and MA populations have different age structures and disease burden • Published epidemiologic literature: peer-reviewed incidence/prevalence studies, weighted toward the most geographically and demographically relevant cohort available
A critical, frequently underestimated component is population growth and disease-trend adjustment: if incidence is rising (as with many autoimmune and metabolic conditions) or the plan's membership is growing through group acquisitions, the Year-3 eligible population can differ materially from Year-1, and static single-year estimates understate the multi-year budget trajectory.
A budget impact model is fundamentally a comparison of two futures for the same population — "world without the new drug" versus "world with the new drug." Characterizing the current standard-of-care mix accurately, including the sizable share of patients who remain untreated, is what makes that counterfactual credible to a P&T committee.
The counterfactual scenario is not simply "the single leading competitor" — it is the actual weighted mix of everything eligible patients are currently receiving, including:
• Named pharmacologic comparators: each existing agent used in the eligible population, weighted by its current claims-derived market share within the plan (not national market share, which can diverge substantially by region and formulary tier) • Best supportive care / no active treatment: often the largest single "comparator" for undertreated conditions — omitting it systematically overstates the new drug's incremental budget impact by attributing displaced cost to therapies that were never actually being used • Off-label and compounded use: relevant in specialty and pediatric indications, priced and utilized differently than on-label comparators • Site-of-care mix: infused/injectable comparators split across physician office, outpatient hospital, and home infusion, each carrying a different site-of-care fee schedule (outpatient hospital typically 2–3× the office rate under commercial contracts)
Mix shares are typically extracted from the plan's own 12–24 month pharmacy and medical claims lookback, coded by NDC/HCPCS and current formulary tier, and validated against national market-share syndicated data (IQVIA National Prescription Audit) to check for plan-specific outliers.
Comparator costs must be priced on a consistent basis with the new drug, or the comparison is invalid:
• WAC (Wholesale Acquisition Cost): list price before any rebates or discounts — the "sticker price" manufacturers report, used as the default in early-stage models • ASP (Average Sales Price): CMS-published quarterly figure reflecting actual market transaction prices net of most rebates and discounts, used for Medicare Part B reimbursement (ASP+6%) and increasingly as the realistic cost basis for medical-benefit drugs • Net price: plan-specific price after negotiated rebates, which for branded specialty products in the US commercial market can run 20–50% below WAC depending on therapeutic category and competitive intensity • 340B and Medicaid Best Price interactions: safety-net providers and state Medicaid programs access substantially lower effective prices (Medicaid Drug Rebate Program mandates a minimum 23.1% AMP rebate for branded drugs), which must be excluded or separately modeled when the BIA is built for a commercial book of business
Mixing WAC-based new-drug costs against net-price comparator costs (or vice versa) is one of the most common — and most consequential — modeling errors flagged in P&T committee review, because it can flip the sign of the projected budget impact.
A new therapy rarely captures its full addressable market on day one. Uptake is modeled as a diffusion curve — analogous to Bass diffusion models used in technology adoption forecasting — shaped by formulary tier placement, prior-authorization stringency, prescriber familiarity, and competitive response.
Uptake curves are typically parameterized as an S-shaped (logistic or Bass) function rather than linear growth, reflecting three empirically observed adoption phases:
• Early phase (Months 0–6): slow uptake among early-adopter prescribers, gated by formulary review timing — most commercial plans review new molecular entities at the next quarterly P&T cycle, creating a built-in 1–3 month lag after FDA approval before any coverage exists at all • Acceleration phase (Months 6–18): uptake steepens as favorable coverage policy, published real-world evidence, and peer prescriber experience compound; this is the phase most sensitive to formulary tier (tier 2 preferred vs. tier 3 non-preferred can swing realized uptake by 2×) and prior-authorization criteria stringency • Plateau phase (Year 2–4): uptake approaches a ceiling set by clinical differentiation, competitive entrants (including biosimilars or generics eroding the category), and the practical limit of eligible patients willing/able to switch from stable existing therapy
Prior authorization (PA) and step-therapy requirements are the dominant utilization-management levers payers use to shape this curve directly: a PA approval rate of 60% versus 90% can reduce realized Year-1 uptake by 25–30 percentage points relative to the "unmanaged" clinical-adoption curve a manufacturer typically submits.
The 2014 hepatitis C launches (Sovaldi/Harvoni) triggered the most-cited real-world uptake shock in payer history: uptake vastly exceeded pre-launch forecasts because of pent-up demand from a large previously-untreated prevalent pool. A 2015 CMS letter to state Medicaid directors explicitly addressed the resulting Medicaid budget strain, and the episode became the reference case cited in nearly every subsequent BIA methodology paper on why "diagnosed pool depletion" must be modeled as a separate dynamic from steady-state incident uptake.
A frequent structural error is modeling uptake as a constant percentage of a static eligible population every year. In reality, two distinct patient flows must be modeled separately:
• Prevalent pool: patients already diagnosed and treatment-eligible at model start — this pool is drawn down (depleted) as they switch to the new therapy, and does not replenish at the same rate once early adopters have moved • Incident flow: newly diagnosed patients entering the eligible population each year, who choose among all available therapies fresh, unaffected by switching costs or existing therapy inertia
Models that conflate these two flows systematically overstate later-year budget impact (the prevalent pool is treated as if it continually regenerates) or understate it (steady-state incident demand is missed after the prevalent pool is exhausted). ISPOR guidance recommends explicit dual-flow modeling for any chronic condition with a substantial existing prevalent pool at product launch.
Per-member-per-month (PMPM) is the currency payers actually budget in — it converts a large, abstract total dollar figure into a small, comparable number that can be benchmarked against premium trend, medical loss ratio targets, and the plan's overall per-member cost trajectory.
The per-patient annual net cost that feeds the PMPM calculation is assembled from several additive and subtractive components:
Additive (cost to the plan): • Net drug acquisition cost: WAC minus negotiated rebate — the largest single line item for most specialty products • Administration cost: infusion/injection facility and professional fees (site-of-care dependent) for non-oral therapies • Monitoring cost: required labs, imaging, or REMS-program monitoring specified in the label • Adverse-event management: incremental cost of treating drug-specific toxicities not seen (or seen less often) with the comparator, weighted by incidence rate from the pivotal trial safety population
Subtractive (cost offset to the plan): • Averted comparator drug cost: the cost of whatever therapy the switching patient would otherwise have remained on • Averted downstream medical cost: reduced hospitalizations, ER visits, or disease-progression events, sourced from the clinical trial's or real-world evidence's effect on the relevant utilization endpoint • Averted comparator-related adverse-event cost: if the new therapy has a materially better tolerability profile
Net cost per patient = (net drug cost + administration + monitoring + AE management) − (averted comparator cost + averted downstream cost). This figure, multiplied by the number of patients switching in a given year and divided by total plan membership and 12 months, produces the PMPM.
Manufacturers and payers routinely disagree most sharply not on drug price, but on the size of the cost-offset assumptions:
• Offset magnitude: derived from surrogate endpoints (e.g., relapse-rate reduction, HbA1c improvement) translated into avoided medical events using published cost-per-event literature — this translation step introduces substantial uncertainty and is a common target of payer pushback • Offset timing: many offsets (e.g., avoided long-term complications in diabetes or cardiovascular disease) accrue years beyond the typical 3-year BIA horizon, meaning a short-horizon model captures the cost but not the corresponding benefit — a structural bias toward showing negative (unfavorable) short-term budget impact even for therapies that are cost-saving over a longer horizon • Population applicability: offset estimates from clinical trial populations (often healthier, more adherent, more closely monitored) may not generalize to the real-world plan population, where adherence to specialty therapies averages only 60–75% at one year
Because of this, sophisticated payer models frequently run the analysis both with and without downstream medical offsets, presenting the "gross drug-cost-only" impact alongside the "net-of-offsets" impact so the P&T committee can see the sensitivity explicitly rather than accepting a single blended number.
The core deliverable of the entire exercise is a year-by-year projection of total plan spending with versus without the new therapy on formulary, typically spanning a 3-year horizon and presented undiscounted per prevailing ISPOR guidance for short-horizon budget (as opposed to cost-effectiveness) models.
For each year t in the model horizon:
Budget Impact(t) = [Eligible Population(t) × Uptake(t) × Net Cost per Patient(t)] − [Eligible Population(t) × Uptake(t) × Comparator Cost per Patient(t) that would otherwise have been incurred]
Equivalently and more transparently, the model is usually built as two full parallel scenarios computed independently, then subtracted:
• Scenario A ("world without"): total plan pharmacy + medical cost for the eligible population under the current comparator mix, projected forward with no new entrant • Scenario B ("world with"): total plan cost for the same population under the projected uptake curve of the new therapy, with remaining patients still on the prior comparator mix • Budget Impact(t) = Scenario B(t) − Scenario A(t)
This two-scenario architecture — rather than a single netted formula — is strongly preferred by P&T committees and by AMCP dossier reviewers because every intermediate total (total drug spend, total medical spend, total membership) is independently checkable against the plan's own claims experience, catching modeling errors that a single collapsed formula would hide.
ICER's 2023 budget impact analysis of a novel Alzheimer's therapy modeled national annual budget impact exceeding $2 billion at moderate uptake assumptions even before accounting for required amyloid-PET and MRI monitoring costs — prompting several regional Medicare Advantage and Medicaid plans to implement restrictive prior-authorization criteria specifically to manage projected budget exposure ahead of finalized coverage determinations.
Once the annual Budget Impact($) is computed, it is converted to PMPM by dividing by total plan membership (not just the eligible sub-population) and by 12 months — this is a critical distinction: PMPM is always expressed against the whole insured population, because that is the base against which premiums are set and medical loss ratio is calculated.
Most commercial plans use an internal PMPM threshold — commonly a fraction of a percent of total per-member trend — above which a new therapy triggers escalated financial review, tiered formulary placement, or a requirement for a risk-sharing/rebate agreement before unrestricted coverage. Because pharmacy trend budgets for large commercial plans typically run in the 6–9% annual growth range against a base PMPM often exceeding $120–150 for total drug spend, a single new specialty product PMPM addition of even $0.10–$0.30 can represent a meaningful, board-visible share of that year's allowable trend growth — which is precisely why payers scrutinize every input in the model rather than accepting the manufacturer-submitted headline number.
No single-point budget impact estimate survives first contact with a P&T committee. One-way and scenario sensitivity analyses quantify which assumptions matter most, and that ranked list becomes the actual agenda for price and contract negotiation between payer and manufacturer.
Deterministic one-way sensitivity analysis varies each model input individually, holding all others at base case, and ranks the resulting swing in total budget impact — visualized as a "tornado diagram" with the widest-swinging variable at top:
• Uptake rate (peak share and speed of diffusion) is the dominant driver in the overwhelming majority of published BIAs, frequently accounting for 40–60% of total output variance • Net price / rebate percentage is typically the second-largest driver, and the only one directly controllable through negotiation rather than clinical/behavioral forecasting • Cost-offset magnitude and eligible population size round out the top four in most models
More sophisticated submissions add probabilistic sensitivity analysis (PSA): Monte Carlo simulation sampling every input simultaneously from a specified distribution (beta for rates, gamma for costs) across 1,000–10,000 iterations, producing a full distribution of possible budget-impact outcomes rather than a single point estimate — increasingly requested by larger payers and by ICER in its own independent reassessments of manufacturer-submitted models.
The ranked sensitivity output directly shapes the negotiation:
• Standard rebates: a flat percentage discount off WAC, the most common structure, typically negotiated within a therapeutic-class competitive band; PBM aggregators (Express Scripts, CVS Caremark, OptumRx formulary committees) negotiate these centrally across large client books • Utilization/volume caps: rebate percentage increases if actual utilization exceeds the modeled uptake forecast, directly hedging the payer against the dominant sensitivity driver (uptake risk) rather than static price alone • Outcomes-based (value-based) agreements: rebate or refund tied to a measured clinical outcome in the payer's own population — Harvard Pilgrim's 2017 agreement with Amgen for the PCSK9 inhibitor Repatha, tying additional rebates to real-world LDL-lowering performance, remains a widely cited template; CAR-T therapies (Kymriah, Yescarta) have similarly used response-based payment structures with major academic medical centers and payers • Budget caps / risk corridors: total annual payer spend on the product is capped, with the manufacturer refunding amounts above the cap — used selectively for very high per-patient-cost one-time therapies (gene therapies, CAR-T) where population-level financial exposure, not just per-patient price, is the payer's primary concern
The final negotiated net price is then fed back into the budget impact model as the new base case, closing the loop between analytical model and commercial contract.
When the first gene therapies (e.g., for spinal muscular atrophy) launched with list prices above $2 million per one-time dose, ICER and multiple state Medicaid programs explicitly modeled outcomes-based, multi-year installment payment structures — rather than a single lump-sum rebate — as a direct response to sensitivity analysis showing that a single catastrophic-cost patient could materially move an entire state Medicaid pharmacy budget in one fiscal year. Several states subsequently adopted supplemental rebate and installment-payment authority specifically to manage this class of budget-impact risk.