V28 Risk Model Goes Full-Weight in PY 2026, Compressing Medicare Advantage Risk Scores by $11 Billion

Payment year 2026 is the first in which CMS calculates Medicare Advantage risk scores using 100% of the 2024 CMS-HCC model. The three-year blend is over, V24 is retired, and the partial cushion that softened the prior two years is gone.

CMS projected a 3.12% average risk score reduction from V28, roughly $11 billion of net savings to the Medicare Trust Fund. The average is not the operative number; the dispersion around it is.

Key Takeaways

• 100% V28 in PY 2026 removes the last third of V24 weight, and the 67% to 100% step is the largest of the three because there is no remaining offset.
• 9,797 to 7,770 valid diagnostic codes, a net removal of 2,027 and a 20.7% reduction in the codes that generate risk-adjusted payment.
• 0.590 to 0.166 is what happens to a diabetes-with-peripheral-vascular-disease patient's risk adjustment factor between the two models.
• $39.6 million a year is the revenue effect of a 3.12% risk score reduction on a 100,000-member plan at a $1,000 county benchmark with a 5% quality bonus.
• $7.12 billion more would come out under the proposed unlinked chart review exclusion, a further 1.53% reduction, on a capture channel 58% of MA contracts used in 2022.

The Step That Removes the Cushion

The transition was staged so plans could adjust, and the staging is why the final year hits hardest.

V24 assigned higher coefficients to many conditions than V28 does, so during PY 2024 the 67% V24 weight offset most of the compression and plan encounter runs showed only part of it. At 33% V24 weight in PY 2025 the slope steepened. The move to 100% removes the remaining third at once, which is the largest single increment of the three.

Two adjustments define the effective rate on top of it. CMS applies a normalization factor of 1.067 to the V28 model for PY 2026, which offsets part of the compression, and the statutory MA coding pattern adjustment of 5.90% remains, reducing payment for the diagnostic coding intensity difference against fee-for-service. Neither is discretionary and both sit inside the 2027 bid.

Where the Coefficients Moved

The model did two things at once, and they push in opposite directions for different populations.

It cut the valid code set from 9,797 to 7,770, removing about 2,294 codes and adding 268 for a net loss of 2,027, a 20.7% reduction. And it expanded hierarchical condition categories from 86 to 115, creating finer clinical groupings that redistribute coefficient weight rather than simply removing it.

Diabetes is the clearest case of the first effect. Under V24 a patient documented with diabetes and peripheral vascular disease carried a combined risk adjustment factor of roughly 0.590, being 0.302 for the diabetes HCC plus 0.288 for the vascular HCC. Under V28 the same patient generates about 0.166, because the model constrains diabetic disorders so complicated and uncomplicated presentations contribute similarly. Uncomplicated type 2 diabetes actually rose, from 0.105 to 0.166, a 58% gain. The direction of a plan's exposure therefore depends entirely on documented complication prevalence in its own book.

The expansion runs the other way for some populations. Three new pressure ulcer categories carry high coefficients: HCC 379 at 1.965, HCC 381 at 1.075, HCC 382 at 0.838. Plans with skilled nursing or complex wound care populations capture value there. Plans with oncology concentrations lose it, because V28 removed the interaction between immune disorders and cancer HCCs.

The revenue translation is direct. On a $1,000 county benchmark with a 5% quality bonus, a 3.12% risk score reduction is roughly $33 PMPM, which annualizes to about $39.6 million on a 100,000-member plan before normalization or coding pattern effects. Against that, the CY 2027 bid deadline in June 2026 falls after only one quarter of full-weight V28 encounter data. A plan seeing a 4% risk score decline in that quarter has to separate the mechanical mapping effect, which persists, from provider coding adjustment, which should improve as documentation adapts, on a single quarter of experience.

Three Channels Closing at Once

V28 is one compression among three, and each targets a different part of the same capture chain.

The model narrows what can be coded. The proposed unlinked chart review exclusion narrows where diagnoses can come from: CMS estimates $7.12 billion, a further 1.53% reduction relative to 2026, on a practice 58% of MA contracts used in 2022. And excluding audio-only encounters, identified by modifiers 93 and FQ, narrows the modality, hitting plans that built telephonic assessment programs after the pandemic.

Those channels overlap in exactly the wrong way. The codes most often recovered through retrospective review were the complication-specific diabetes and cardiovascular codes, which are the codes V28 eliminated or constrained. Supplemental data sources accounted for roughly $7.5 billion of MA payments in 2023, and MedPAC put total 2023 overpayments at $27 billion with $24 billion attributed to chart review and coding intensity. A plan relying on all three channels does not lose them additively; it loses the same dollars counted through several routes, and the offsets it might have used are gone together.

CMS softened rather than removed the effect. The CY 2027 Advance Notice estimated the combined risk model revision and normalization impact at negative $15.22 billion, and the final Rate Announcement came in at 2.48% growth against 0.09% proposed, a 239 basis point swing. The chart review exclusion and audio-only restriction stayed. With expected risk score growth around 2.45%, the effective growth rate reaches roughly 4.98%, which still trails medical cost trend in most markets, and the benefit side has already absorbed the difference: Milliman recorded total value added down over 7% year over year, the largest single-year decline in the program's history.