Outcomes-based reimbursement between manufacturer and payer — real-world response adjudication drives rebates, not list price
A value-based contract (VBC) ties some or all of a therapy's reimbursement to a measurable clinical or economic outcome achieved in routine practice, rather than to list price or volume alone. Before a single patient is dosed, the manufacturer and payer must agree on an endpoint precise enough to adjudicate without dispute, a follow-up window short enough to be commercially workable, and a rebate schedule that meaningfully shifts risk without making the deal unadministerable.
The single hardest design decision in a VBC is picking an endpoint that is (a) clinically meaningful, (b) already captured somewhere in routine care so it does not require a bespoke trial infrastructure, and (c) binary or near-binary enough to adjudicate without endless dispute.
Common endpoint families used in live US contracts: • Biomarker/lab threshold — LDL-C reduction ≥50% for PCSK9 inhibitors (Amgen Repatha, Harvard Pilgrim 2015); HbA1c <7% for diabetes agents • Composite clinical response index — PASI75 (≥75% Psoriasis Area and Severity Index improvement) at week 12–16 for biologics • Hospitalization/utilization endpoint — reduction in heart-failure hospitalization rate at the population level (Novartis Entresto, Cigna 2017) • Time-boxed response — "no response by day 28, no charge" for CAR-T (Novartis Kymriah, CMS 2017-era outcomes arrangement) • Durability endpoint — sustained biomarker response at 2 and 5 years for one-time gene therapies, paid as staged installments (Novartis Zolgensma, 2019)
Endpoints that reference a validated, guideline-anchored threshold (NCCN response criteria, ACR20/50/70 for rheumatoid arthritis, MRD-negativity <10⁻⁴ in oncology) are strongly preferred by both sides because a third party has already defined what "response" means — reducing adjudication disputes to a data-quality question rather than a definitional one.
In 2017, Novartis and CMS structured a value-based arrangement for the CAR-T therapy Kymriah (tisagenlecleucel) in pediatric acute lymphoblastic leukemia: if the patient did not respond by day 28 post-infusion, Novartis would not bill for the drug at all. It was among the first "no pay for non-response" agreements applied to a curative-intent cell therapy in the United States.
The financial mechanics sit on top of the clinical endpoint:
• Flat rebate — a fixed percentage of WAC (Wholesale Acquisition Cost) refunded per non-responder, typically 15–30% • Tiered/staged rebate — larger rebate the further the patient falls below threshold, or staged installment payments released only as durability checkpoints are met (Zolgensma's 5-year, up to $2.125M installment model) • Population-level corridor — rebate triggers only if the cohort-wide response rate falls below a pre-agreed corridor (e.g. below 70% aggregate response), insulating both sides from single-patient noise • Money-back guarantee — full refund if a hard clinical event occurs despite adherence (Amgen's Repatha guarantee: refund if a fully adherent patient still has an MI or stroke)
Two US regulatory frameworks make these deals administrable: • Medicaid Best Price — CMS's 2022 final rule (implementing the Bipartisan Budget Act of 2018) allows manufacturers to report multiple, outcome-contingent "best prices" for value-based purchasing arrangements instead of one contract collapsing the government's statutory minimum rebate nationwide • Anti-Kickback Statute safe harbor — the OIG's November 2020 final rule (42 CFR §1001.952) created a value-based enterprise safe harbor, permitting risk-sharing arrangements between manufacturers and payers that would otherwise raise remuneration concerns under the AKS
Without these two carve-outs, a single underperforming VBC rebate could legally reset a manufacturer's Medicaid rebate obligation across its entire national book of business — which is why VBC activity was rare before 2019 and accelerated sharply afterward.
A value-based contract is only as good as its ability to find the same patient again at follow-up and prove what their condition looked like before treatment. That requires deterministic or probabilistic linkage across data sources that were never designed to talk to each other: medical claims, pharmacy claims, EHR flowsheets, and increasingly a manufacturer-sponsored patient registry.
Enrollment in a VBC is an operational exercise in health-data plumbing, not just eligibility screening:
• Index event definition — the "index date" (typically first fill or first infusion) must be unambiguous in claims data (NDC/HCPCS code + date of service) so that baseline and follow-up windows anchor consistently across every patient • Deterministic linkage — matches on member ID, date of birth, and NPI when the payer and the specialty pharmacy or hub-services vendor share a common identifier • Probabilistic linkage — when identifiers are inconsistent across systems (common when a manufacturer hub program sits outside the payer's data warehouse), a Fellegi-Sunter-style probabilistic match on name, DOB, and ZIP is used, typically achieving 85–96% match rates versus deterministic linkage alone • Baseline extraction — the pre-index biomarker/severity score is pulled from the EHR structured field (lab result, PASI score entered by the dermatologist) or, where structured data is unavailable, abstracted by a chart-review vendor against a pre-specified case-report-form definition analogous to a CDISC SDTM baseline domain
Missing or ambiguous baseline data is the single most common reason a patient is excluded from adjudication — most contracts specify that a patient without a valid, timely baseline measurement is treated as a protocol deviation and removed from the denominator rather than defaulted to either responder or non-responder status, since defaulting either way would bias the rebate calculation.
Every live VBC is underpinned by a data-sharing agreement (DSA) that predates enrollment, specifying:
• Which party (payer, manufacturer, or an independent third-party vendor) hosts the linked, de-identified analytic dataset • The minimum data completeness threshold for a patient to be adjudicable — commonly requiring both a valid baseline and at least one valid follow-up value within the measurement window • Audit rights — typically a random 10–20% sample of adjudicated patients is available for source-document verification by either party • Alignment with the FDA's Real-World Evidence framework (established under the 21st Century Cures Act, 2016) where the same longitudinal dataset doubles as post-marketing evidence generation, giving the manufacturer a second use for the infrastructure it built to run the contract
HIPAA-compliant de-identification (Safe Harbor or Expert Determination method under 45 CFR §164.514) is applied before data crosses from the payer's claims warehouse into the shared adjudication dataset, so the adjudication vendor typically works with a coded patient identifier rather than direct identifiers.
Once treatment begins, the contract's entire financial logic is suspended until the follow-up window closes and an outcome value exists for each patient. This stage is where the largest source of contract friction actually lives: not disagreement about what counts as success, but incomplete or late-arriving real-world data.
Three capture mechanisms dominate current VBCs, in ascending order of data quality and descending order of scalability:
1. Manufacturer or society-run patient registry — a purpose-built case report form completed by the treating clinician at the follow-up visit (e.g. dermatologist enters PASI score in a psoriasis biologic registry). Highest data quality, but requires clinician participation and site-level data-entry burden, so completeness typically plateaus around 90% even in well-run programs. 2. Structured EHR extraction — an automated feed (via HL7 FHIR interfaces increasingly required by CMS interoperability rules) pulls the relevant lab value or score directly from the EHR at the visit closest to the follow-up window. Scales well but depends on the value being entered as structured data rather than buried in a clinical note. 3. Claims-based proxy — where no direct clinical measure is feasible (e.g. large chronic-disease populations), a validated claims algorithm proxies the outcome, such as counting heart-failure-related hospitalization claims (ICD-10 I50.x with inpatient revenue codes) in the follow-up period as the Entresto-style outcome measure. Lowest cost, but introduces coding-accuracy uncertainty into the adjudication.
Most mature contracts blend mechanisms: registry or EHR data as primary source, claims-based proxy as the fallback for patients missing a structured clinical value, with an explicit hierarchy specified in the DSA so adjudicators are not making ad hoc calls patient-by-patient.
Lost-to-follow-up patients (8–15% typical) create an actuarial problem: if missing patients are disproportionately non-responders (plausible — non-responders may discontinue therapy and disengage from care), naively excluding them inflates the measured response rate and understates the rebate owed.
Contract-specified handling approaches, roughly in order of how conservative they are toward the payer: • Complete-case analysis — missing patients excluded from both numerator and denominator (simplest, but biased toward the manufacturer if dropout correlates with non-response) • Non-responder imputation (NRI) — any patient without a valid follow-up value is counted as a non-responder by default; the industry-standard convention in dermatology biologic trials, increasingly carried into VBC adjudication • Last-observation-carried-forward from an interim assessment, where a mid-window checkpoint exists • Discontinuation-triggers-rebate clauses — if a patient discontinues therapy before the follow-up window for a reason coded as lack of efficacy (rather than administrative or unrelated-AE reasons), the discontinuation itself is treated as a non-response event, closing the most common loophole in outcome-based deals
The adjudication methodology (complete-case vs. NRI vs. hybrid) is one of the most heavily negotiated clauses in the entire contract, because a shift from complete-case to NRI can move the measured response rate by 5–10 percentage points on a typical specialty cohort — directly moving the rebate total.
Adjudication converts a continuous or categorical clinical measurement into a contractual fact: responder or non-responder. Because real money follows this classification, most mature VBCs route it through an independent third-party adjudicator rather than letting either counterparty self-report, and back it with a statistical audit sample.
A typical adjudication pipeline runs in four deterministic steps per patient, executed by a neutral vendor against the pre-specified statistical analysis plan (SAP) written into the contract at signing:
1. Eligibility check — confirm the patient has a valid baseline and a follow-up value inside the contracted window; apply the missingness rule (complete-case vs. NRI) agreed in Stage 3 2. Metric calculation — compute percent change from baseline (for continuous measures like LDL-C or HbA1c) or confirm the categorical response tier (PASI75, ACR50, MRD-negative) directly from the captured value 3. Threshold comparison — flag responder if the calculated metric meets or exceeds the contracted threshold (e.g. ≥75% severity reduction); intention-to-treat framing is standard, meaning the flag is based on all enrolled eligible patients regardless of adherence, unless the contract explicitly carves out an adherence gate 4. Confidence/audit flag — patients whose value sits within a pre-defined margin of the threshold (e.g. within 3 percentage points) are automatically routed to manual chart-review adjudication rather than algorithmic classification, since these borderline cases carry the highest risk of dispute
The adjudicator then issues a batch reconciliation file to both counterparties simultaneously, with per-patient responder/non-responder flags but without exposing individual clinical detail beyond what the DSA permits — typically just the binary outcome flag and the eligibility status.
Cigna's 2017 outcomes-based agreement with Novartis for the heart-failure drug Entresto based the rebate not on any single patient's result but on the aggregate reduction in heart-failure hospitalization across Cigna's entire treated population relative to a pre-specified historical benchmark — an early example of population-level rather than patient-level adjudication, chosen specifically because individual hospitalization events are too noisy to adjudicate reliably one patient at a time.
Because both counterparties have a direct financial stake in the classification, adjudication is treated less like a clinical judgment and more like a financial audit:
• Sampling — a random 10–20% sample of adjudicated patients (plus 100% of borderline cases from step 4 above) is re-verified against original source documents (EHR chart, lab report) by an auditor independent of the routine adjudication vendor • Agreement statistic — inter-rater agreement between the adjudicator's automated flag and the audit re-verification is tracked as Cohen's kappa, with κ ≥ 0.8 ("almost perfect agreement" on the Landis-Koch scale) as the typical contractual quality bar; falling below triggers a full re-adjudication of that batch • Dispute escalation — unresolved disagreements escalate to a joint clinical review committee with named representatives from both counterparties, and, in a minority of contracts, a pre-agreed third-party arbitrator for cases that remain unresolved • Coding and DRG-manipulation risk — for claims-proxy endpoints, adjudicators specifically screen for coding pattern shifts around the measurement window (e.g. an unusual spike in outpatient-observation coding rather than inpatient admission for the same clinical event) that could artificially suppress a hospitalization-based non-response signal
The adjudication cycle for a typical quarterly batch runs 4–8 weeks from window-close to final reconciliation file — the biggest single driver of how quickly a manufacturer can recognize the rebate liability on its own financial statements.
The final stage closes the financial loop: rebates for non-responders are calculated and paid on a fixed reconciliation cadence, both parties true up any disputed patients from the audit process, and the aggregate real-world performance becomes the evidence base for renegotiating — or terminating — the contract at renewal.
Settlement translates the adjudicator's responder/non-responder flags into an actual cash transfer:
1. Rebate calculation — for each non-responder, the contracted rebate amount (flat percentage of WAC, tiered amount, or full refund depending on contract type) is calculated against that patient's actual billed amount for the measurement period 2. Netting — rebates are typically netted against the manufacturer's existing supplemental rebate or 340B/Medicaid rebate payments on the same claims stream where contractually permitted, rather than processed as a fully separate payment rail 3. Invoice and dispute window — the payer issues a reconciliation statement; the manufacturer has a contractually defined window (commonly 30–45 days) to dispute individual line items before the rebate is deemed final 4. Payment — cleared rebates are paid via the standard rebate-processing infrastructure already used for volume-based supplemental rebates, meaning most of the operational cost of running a VBC is the adjudication layer, not the payment mechanics themselves
For durability-linked gene therapies, settlement is structured as staged installment payments rather than post-hoc rebates: the manufacturer is paid a fraction of the total contracted price up front and additional tranches only as the patient remains a durable responder at pre-defined checkpoints (commonly years 1, 2, and 5) — spreading the payer's financial risk across the same multi-year window the clinical durability risk actually plays out over.
The aggregate response rate and rebate flow observed across a full contract term feed directly into the next negotiation cycle:
• Real-world ICER re-estimation — the observed real-world response rate (which is almost always lower than the pivotal trial's efficacy rate, an "efficacy-to-effectiveness gap" widely documented across specialty drug classes) is used to recompute the incremental cost-effectiveness ratio actually being delivered, which ICER and payer HEOR teams then compare against the $100,000–$150,000/QALY benchmark used at initial listing • Renewal decision — roughly two-thirds of active VBCs are renewed at term, typically with the threshold, rebate percentage, or measurement window adjusted based on the observed data rather than the deal being torn up entirely • Portfolio scaling — successful bilateral VBCs increasingly graduate into standardized, multi-payer or multi-state templates; the CMS Innovation Center's Cell and Gene Therapy Access Model, launched in January 2023, negotiates outcomes-based agreements on behalf of participating state Medicaid programs for sickle cell gene therapies (Casgevy from Vertex/CRISPR Therapeutics and Lyfgenia from bluebird bio), removing the need for each of 50 state Medicaid programs to separately negotiate the same clinical endpoint • Termination triggers — contracts specify a minimum enrolled-cohort size below which statistical noise makes adjudication meaningless; falling persistently below it, or hitting an aggregate response rate far outside the range assumed at pricing, are the two most common reasons a VBC is allowed to lapse rather than renew
Following its 2015 launch, Amgen's outcomes-based agreement for the PCSK9 inhibitor Repatha with Harvard Pilgrim Health Care refunded the health plan if a fully adherent patient's LDL-C did not fall as expected, and separately guaranteed a refund if the patient nonetheless suffered a heart attack or stroke — one of the earliest US payer-manufacturer deals to combine a biomarker-based rebate with a hard clinical-event money-back guarantee in the same contract.