CMS-HCC V28: What the New Risk Adjustment Model Changes for Your RAF Scores

V28 is now the only model CMS runs MA risk through

CMS finalized the 2024 CMS-HCC risk adjustment model, what the industry calls V28, in the CY 2024 Rate Announcement. Plans had a three-year glide path: 33% weighting in 2024, 67% in 2025, and 100% in CY 2026. From 2026 onward, every MA plan’s risk scores run through V28 and V28 only (CMS, 2026 Advance Notice Fact Sheet, Jan 2025).

This is the biggest overhaul of the HCC model since CMS moved from the old CMS-HCC community model in 2014. The combined risk model revision and FFS normalization impact in CY 2026 is projected at -3.01%, the single largest downward pressure on MA payment in the 2026 rate cycle. MA payments still rise 4.33% year over year, about $21 billion; the risk model revision is the headwind pulling against that tailwind (CMS, 2026 Advance Notice Fact Sheet, Jan 2025).

Three structural changes, not a tweak

What changed under the hood is best read as three deliberate moves rather than a refresh.

• ICD-10 native. V24 was built on ICD-9 mappings, shoehorned into ICD-10 through the CMS crosswalk. V28 is native ICD-10. Condition categories now reflect how diseases are actually coded in modern clinical practice, not how they were mapped twenty years ago (CMS, 2026 Advance Notice Fact Sheet, Jan 2025).
• Updated data years. V24 used 2014 diagnoses and 2015 expenditures. V28 uses 2018 diagnoses and 2019 expenditures. The methodology is unchanged; the coefficients reflect more recent treatment patterns and cost relationships (CMS, 2026 Advance Notice Fact Sheet, Jan 2025).
• Fewer HCCs, a higher bar for coefficient significance. V28 consolidates condition categories where coding variation ran high and clinical specificity ran low. The model applies “longstanding principles to make clinical revisions focused on conditions that are subject to more coding variation” (CMS, 2026 Advance Notice Fact Sheet, Jan 2025), which is the agency’s careful way of saying that some plans had been working certain codes and the model is now tighter for it.

Fewer HCCs means the ones you keep carry more weight

V28 is a leaner model. The condition categories were restructured from the ground up on ICD-10 native logic, so some conditions that carried modest RAF weight in V24 got reclassified, consolidated, or dropped entirely. The per-member impact concentrates into fewer high-weight categories, and the practical consequence is that codes a plan used to chase for 0.1 to 0.2 RAF points may now be worth zero.

The flip side is structural. Accurately captured chronic conditions with strong clinical evidence now carry more relative weight, because the denominator of categories that count has shrunk. Capturing the right conditions matters more than capturing more conditions.

CMS also confirmed it is calibrating the model on MA encounter data, the diagnosis, cost, and utilization data plans submit, and “will have the option to start phasing in an encounter data-based risk adjustment model as early as CY 2027” (CMS, 2026 Advance Notice Fact Sheet, Jan 2025). The signal is large. The future model is not just leaner; it is sharper at detecting coding patterns that do not match actual utilization. A plan that has been padding RAF scores with high-prevalence, low-severity codes that never appear in encounter data as actual treatment is running a strategy with a written end date.

MEAT is the documentation that survives a RADV audit

Adding a code is easy; making it audit-proof is the hard part, and the part most vendors gloss over. The MEAT framework (Monitor, Evaluate, Assess, Treat) is the industry standard for documenting HCC diagnoses in a way that satisfies CMS under RADV scrutiny. It is not CMS regulation, but it is what every experienced coding auditor looks for, and the closest thing to a safe harbor in risk adjustment documentation.

What MEAT requires in the medical record:

The most common RADV failure mode is not that the patient did not have the condition. It is that the provider documented the condition exists and forgot to document what they did about it at that visit. The condition is there. The assessment might be there. But the Monitor, Evaluate, or Treat piece is thin or missing, and under CMS’s RADV methodology that diagnosis gets tossed (CMS, RADV Program).

A coding vendor that only looks for diagnoses to add, and never flags unsupported codes for deletion, is building a liability time bomb, not a compliance program. The same bidirectional principle drives the AI-assisted abstraction workflow, where a tool that only hunts for gaps and never flags unsupported codes answers only half of the question an auditor will eventually ask.

Prospective coding is no longer optional

The strategic trade-off between prospective and retrospective capture runs deeper than timing, and the prospective vs. retrospective risk adjustment post walks through it in full. The short version: prospective coding catches conditions before the assessment year closes. A plan reviews charts year-round, flags suspected conditions, and gets providers to document them at the point of care. It carries a higher hit rate on RAF and asks more of the provider.

Retrospective coding reviews last year’s charts after the year closes. The logistics are simpler; a reviewer sweeps through charts en masse. But the reviewer is limited to what was documented and has no ability to influence provider behavior for that assessment year. A program that runs only retrospectively learns what it missed too late to do anything about it.

V28’s leaner model and CMS’s encounter-data-driven future shift the math. Prospective coding is no longer a nice-to-have. It is the only way to capture the fewer, higher-weight conditions that drive RAF in V28 while building documentation that survives RADV. A plan cannot retroactively add a MEAT-compliant assessment to a chart from eight months ago. It can flag the gap prospectively and make sure the next encounter includes a proper evaluation.

Bidirectional review is the only defensible posture

CMS’s RADV program is the primary mechanism for recovering overpayments from MA plans, and its logic is direct: “CMS confirms that any diagnoses submitted by an MAO for risk adjustment are supported in the enrollees’ medical records. If diagnoses are unsupported by the medical records, CMS may collect overpayments” (CMS, RADV Program).

The federal government is projected to spend $9.2 trillion on MA payments over the next decade, $1.3 trillion of it on supplemental benefits and premium buy-downs alone (CMS, 2026 Advance Notice Fact Sheet, Jan 2025). CMS has every incentive to audit aggressively, and OIG has been publicly vocal about the scale of improper payments in MA.

This is where chart review has to be bidirectional. Most vendors operate on a model where they are paid a percentage of the RAF uplift they generate, and that structure creates an incentive to add codes and look the other way on unsupported ones. The plan gets the short-term RAF bump, but when a RADV audit arrives two years later, it finds diagnoses that should not have been there, and the overpayment gets clawed back, often with interest and extrapolation.

The defensible approach runs in both directions at once. Capture every code the patient actually has, and delete every code the record cannot support. Quality Health’s AI platform for quality & risk adjustment operates on this logic: it reads every chart for both capture and deletion, and keeps the audit trail an auditor eventually asks for. The net RAF score may run slightly lower than the add-everything approach, but it is real, it is defensible, and it does not evaporate under audit. That is a much better place to be when the RADV contractor shows up.

What to do before 2027

The transition is fully live, the encounter-data-based model is coming, and MEAT compliance is the bar. Five steps sort readiness from exposure.

• Re-score your population under V28 if you have not already. RAF assumptions from the V24 era are stale. Know which HCCs still carry weight and which do not.
• Audit your existing HCC codes for MEAT compliance. A code on the books without a supporting assessment in the chart is a RADV liability, not an asset.
• Shift coding resources toward prospective workflows. The conditions that drive RAF in V28 require contemporaneous clinical documentation, and a plan cannot backfill MEAT compliance retroactively.
• Run a bidirectional chart review. A vendor that only adds codes and never recommends deletions leaves half the risk exposure unmeasured. Surface the unsupported diagnoses now, before CMS does.
• Track your encounter data submissions. The 2027-and-later model will be calibrated on what a plan submits to CMS. If the encounter data is incomplete or inconsistent with risk adjustment submissions, that gap will eventually cost the plan.

Sources

• CMS, “2026 Medicare Advantage and Part D Advance Notice Fact Sheet,” January 10, 2025. cms.gov
• CMS, “Medicare Advantage Risk Adjustment Data Validation Program.” cms.gov