Three Years of HCC Reform

The CMS-HCC risk adjustment model is the mechanism that converts clinical diagnoses into plan revenue. When CMS finalized the V28 model revision in the CY 2024 Rate Announcement, it restructured the classification system that determines what conditions are worth, how they are grouped, and what calibration data drives the payment weights. The three-year phase-in that followed, blending V28 with its predecessor V24 from 2024 through 2026, is now complete. CY 2026 is the first year plans experienced 100% V28 without a transition cushion.

This article is the technical reference for what changed, why it changed, and how V28 interacts with the chart review exclusion and the encounter-based risk adjustment future that the rest of this series examines. Understanding the CY 2027 rate environment requires understanding the model that generates the risk scores at its center. The 0.09% advance notice (MCR-02.01) layers the chart review exclusion (MCR-02.02) on top of a population that has already absorbed three years of model transition. The stacking effect explains why 2027 feels more disruptive than any single year of the phase-in.

What the CMS-HCC Model Is

#

Medicare Advantage plans are paid a per-member-per-month capitation rate adjusted for the health status of their enrolled population. The adjustment mechanism is the CMS-HCC model, a classification system that maps ICD-10 diagnosis codes submitted by plans to Hierarchical Condition Categories. Each HCC carries a coefficient, a numeric weight that represents the predicted incremental cost associated with that condition relative to demographic baseline factors like age, sex, Medicaid eligibility, and institutional status. The sum of a beneficiary’s demographic factors and applicable HCC coefficients produces an individual risk score. That risk score, multiplied by the county-level benchmark payment rate, determines the plan’s monthly capitation for that person.

The relationship between the model and plan economics is direct. Higher risk scores produce higher capitation payments. A beneficiary with diabetes, heart failure, and COPD generates a meaningfully higher risk score than a demographically identical beneficiary without chronic conditions, and the plan receives proportionally more revenue for that enrollment. Plan revenue is, in functional terms, a product of enrollment volume and average risk score. Anything that changes how diagnoses map to HCCs, what coefficients those HCCs carry, or how the resulting risk scores are normalized changes plan revenue, even if not a single patient’s clinical status has changed.

This is why model revisions are not technical footnotes. When CMS restructured the model in 2024, it changed the economics of every MA plan’s enrolled population without changing what those populations cost to serve. The gap between those two things, between what the model says a population is worth and what that population actually costs, is what every plan has been managing for the past three years.

What V28 Changed

#

The prior model version, V24 (formally the 2020 CMS-HCC model), was calibrated on 2014 diagnoses and 2015 expenditures and structured around ICD-9-based condition categories that had been crosswalked to ICD-10 codes when the coding system transitioned in 2015. V24 carried the legacy architecture of a classification system built on an older coding framework, mapped forward onto a newer one. CMS acknowledged that this crosswalk had limitations: the specificity and granularity available in ICD-10 could not be fully exploited through a system designed around ICD-9 groupings.

V28 rebuilt the classification from ICD-10 natively. It was calibrated on 2018 diagnoses and 2019 expenditures, using pre-COVID fee-for-service data to avoid the pandemic-era utilization distortions that would have contaminated a model trained on 2020 or 2021 data. The choice of pre-COVID calibration years was deliberate and defensible but carries a trade-off: V28 reflects pre-pandemic cost patterns that may not fully account for post-pandemic shifts in utilization intensity, chronic disease prevalence, and care delivery patterns.

The structural changes fall into four categories.

Classification restructuring. The number of payment HCCs increased from 86 in V24 to 115 in V28, reflecting more granular clinical groupings. At the same time, the total number of ICD-10 codes mapping to a payment HCC decreased from approximately 9,797 to 7,770. CMS added 268 new ICD-10 mappings while removing roughly 2,294 codes that previously generated payment. The net effect is a model that is more selective about which diagnoses count while creating more categories for those that do. Conditions eliminated from payment status included those CMS determined failed to predict costs accurately, had very small coefficients, were uncommon in the Medicare population, or lacked clear diagnostic coding criteria.

Coefficient constraining. V28 introduced a systematic approach called constraining, where related HCCs within a clinical hierarchy receive the same coefficient value rather than differentiated weights based on severity level. The most discussed example is diabetes. Under V24, diabetes with chronic complications carried a coefficient of 0.312, while uncomplicated diabetes carried a lower weight. Under V28, both conditions receive a constrained coefficient of 0.166. A patient’s clinical status has not changed, but the payment generated by that patient’s diabetes diagnosis has dropped significantly. CMS applied constraining where it concluded that the coding differential between MA and FFS was driven by discretionary coding variation rather than genuine clinical severity differences. MedPAC’s February 2026 comment letter on the CY 2027 advance notice documented that for 2024, the V28 model reduced coding intensity by an estimated 8.8 percentage points relative to V24, with the reduction falling most heavily on organizations with the highest coding intensity under the prior model.

Conditions gaining weight. Certain cancers received more granular staging classifications that increased coefficients for advanced-stage malignancies. Some vascular disease categories were reclassified to better reflect observed cost patterns. Severe chronic conditions with high cost variance saw updated weights. Notable additions to the payment model included codes for benign carcinoid tumors, anorexia and bulimia nervosa, post-polio syndrome, and severe persistent asthma.

The constraining methodology is CMS’s most direct response to the coding intensity problem. Under V24, the differentiated coefficients for severity levels within a clinical hierarchy created a financial incentive for plans to code at the highest defensible severity level, because the payment difference between, for instance, diabetes with and without complications was substantial. Constraining eliminates that differential for categories where CMS concluded the coding variation was driven more by documentation practice than by genuine clinical severity difference. The effect is to reduce the marginal return on aggressive coding within those hierarchies. For plans that had built their coding infrastructure around capturing the highest-severity variant of common chronic conditions, constraining reduced the revenue generated by that infrastructure even where the underlying coding was technically accurate.

Conditions losing weight. Beyond the diabetes constraining, depression and bipolar disorder categories saw over 50% of their mapped ICD-10 codes removed from the model. Conditions where coding intensity in MA was substantially higher than in FFS, suggesting discretionary coding rather than genuine clinical prevalence differences, were systematically targeted. Mild, unspecified, or in-remission mental health conditions were removed from payment status, with clinical expert panels agreeing that active relapse would be reflected through more severe active disorder codes. Some vascular disease subcategories similarly lost mapped codes, and categories related to renal dialysis dependence were reclassified.

The Three-Year Phase-In

#

CMS phased V28 in over three years to avoid a single-year revenue shock that would have destabilized plan operations and benefit design across the MA market.

The blend schedule operated on dates of service: for 2023 dates of service (Payment Year 2024), risk scores were calculated as 33% V28 and 67% V24. For 2024 dates of service (PY 2025), the blend shifted to 67% V28 and 33% V24. For 2025 dates of service (PY 2026), V28 reached 100%.

The practical effect was a graduated revenue impact. CMS projected that the V28 model would reduce MA risk scores by approximately 3.12% on a fully phased-in basis, translating to an estimated $11 billion in net savings to the Medicare Trust Fund in the first year. The blending meant plans absorbed roughly one-third of that impact in year one, two-thirds in year two, and the full weight in year three. Plans had to manage dual-model operations throughout the transition, calculating scores under both V24 and V28 simultaneously, maintaining coding teams that understood both classification systems, and adjusting bids and benefit designs to reflect the blended revenue trajectory.

The 2026 completion changed the rate calculus fundamentally. During the transition, the V24 component of the blended score provided a partial buffer. V24 coefficients were higher for many commonly coded conditions, and its broader ICD-10 mapping included diagnoses that V28 eliminated. Each year’s shift toward V28 reduced that buffer. When the transition ended and plans moved to 100% V28 for PY 2026, the buffer disappeared entirely. The CY 2026 rate announcement’s generous 5.06% increase partially masked the full V28 impact by providing a growth rate cushion that offset some of the model-driven risk score reduction. MedPAC’s January 2026 analysis noted that the V28 model had reduced coding intensity in recent years and that the reduction corresponded with reduced payments but stable supplemental benefits and high plan availability, suggesting the transition had proceeded without the market disruption some stakeholders had predicted.

The phase-in also produced differential effects across the plan landscape. Plans with high coding intensity under V24, particularly those that relied on capturing severity-differentiated diagnoses through aggressive documentation and chart review programs, absorbed larger risk score reductions than plans whose coding profiles were closer to FFS patterns. MedPAC’s analysis confirmed that V28 reduced coding intensity more for organizations with higher V24 coding intensity, meaning the model revision functioned as a targeted correction that fell most heavily on the plans whose coding practices diverged most from the FFS baseline.

The CY 2027 advance notice offers no such cushion. The 0.09% proposed increase layers the chart review exclusion and its $7.2 billion impact on top of a population already operating under full V28 with no transition blend remaining. Plans that managed the three-year phase-in by gradually adjusting bids and benefits now face an abrupt additional reduction without the graduated absorption the phase-in provided.

The Normalization Factor

#

Risk score normalization is the mechanism CMS uses to ensure that changes in coding patterns or model weights do not automatically increase aggregate program payments. If average risk scores rise, whether because of model changes, coding intensity trends, or demographic shifts, the normalization factor adjusts the denominator so that the average risk score remains close to 1.0. The effect is that risk score inflation does not translate directly into payment inflation.

The normalization methodology has been a persistent point of contention between CMS and the plan industry. For CY 2025, CMS developed and finalized a multiple linear regression approach to calculating normalization factors, replacing the prior simple linear regression method. The new methodology incorporates the most recent five years of average FFS risk scores and includes a COVID-19 indicator variable to account for the pandemic’s suppression of FFS risk scores in 2021 and 2022 (when reduced utilization produced fewer diagnosis submissions and artificially lower risk scores).

The industry argument, advanced most prominently by AHIP through analyses by Wakely Consulting Group, is that the normalization methodology overcorrects. By reducing the per-risk-score-point payment value, normalization compounds the effect of the V28 model’s lower coefficients and the coding pattern adjustment’s 5.9% reduction. Plans contend that the three mechanisms, model revision, normalization, and CPA, operate on overlapping aspects of the same coding intensity phenomenon and that their combined effect exceeds what any one mechanism would justify independently.

CMS applies separate normalization factors for MA-PD plans and standalone PDPs, a change introduced for CY 2025. The rationale is that the two plan types have different enrollment compositions, coding patterns, and cost structures, and a single normalization factor applied across both creates cross-subsidies that distort competitive dynamics. MedPAC has supported this approach, noting that the divergence in risk score trends between MA-PDs and PDPs meant the prior unified normalization tended to overpredict MA-PD plan costs and underpredict PDP costs, creating competitive asymmetry. For CY 2027, CMS proposes to continue the separate normalization methodology and extend it to the Part D RxHCC model as well.

PACE Transition

#

PACE organizations operate on a different transition timeline, reflecting the program’s distinct population characteristics, cost structure, and data submission infrastructure.

While standard MA plans completed the V28 phase-in at 100% for PY 2026, PACE organizations were at only 10% V28 and 90% V24 (actually the 2017 CMS-HCC model, an even older version) for PY 2026. For CY 2027, CMS proposes to accelerate the PACE transition to a 50/50 blend between the 2017 model and the proposed 2027 CMS-HCC model. The chart review exclusion will not apply to PACE organizations for CY 2027.

The slower timeline reflects PACE’s structural vulnerability to revenue disruption. PACE serves a frail, nursing-home-eligible population under a comprehensive capitated model that bundles acute, post-acute, and long-term care services. The populations are clinically complex, the per-beneficiary cost is high, and the margin for financial error is thin. An abrupt shift to V28 coefficients could destabilize PACE organizations’ ability to maintain the staffing and service intensity their enrollees require. CMS has been working with PACE organizations since 2024 on encounter data submission improvements, with a tentative timeline for full transition to encounter-data-only risk adjustment for PACE by CY 2029 (see MCR-09.06).

The separate PACE timeline also reflects practical data infrastructure limitations. PACE organizations historically submitted risk adjustment data through the legacy RAPS system rather than the encounter data system used by standard MA plans. The transition to EDS submission requires workflow changes, technology investment, and staff training that the smaller PACE enrollment base (approximately 75,000 beneficiaries nationally) makes proportionally more expensive per organization.

What V28 Means for the Encounter-Based RA Future

#

V28 is not the end of risk adjustment reform. It is the classification foundation on which the next phase will be built.

The V28 model updated the clinical classification and recalibrated coefficients, but it still operates on diagnosis submissions from multiple sources: encounter data, RAPS, and chart review records. The current system accepts diagnoses from all three pathways and uses them to calculate risk scores. The chart review exclusion proposed for CY 2027 (MCR-02.02) removes one of those pathways by excluding unlinked chart review diagnoses from risk score calculation. The encounter-based risk adjustment future (MCR-02.04) would formalize the principle that only diagnoses submitted through encounter data tied to a documented provider visit count for payment.

V28’s restructured HCC map and recalibrated coefficients will be the classification system applied when encounter-based RA arrives. The conditions that gained or lost weight under V28, the constraining methodology, the more selective ICD-10 mapping, all of these design choices will carry forward into a model where the data source narrows from three pathways to one. Plans that invested in encounter data quality during the V28 transition are better positioned for that narrowing. Plans that relied on chart reviews and RAPS submissions to supplement encounter data face a compounding problem: the model has already reduced what their diagnoses are worth, and the data source that supplemented their encounter submissions is being eliminated.

The convergence is three-step. V28 recalibrated what conditions are worth. The chart review exclusion eliminates one mechanism for capturing those conditions. Encounter-based RA, when it arrives, will make the provider encounter the sole authoritative source of payment-eligible diagnoses. Each step narrows the distance between what MA plans are paid and what the beneficiaries they enroll would cost in Traditional Medicare. V28’s role in that trajectory is foundational: it set the classification and the coefficient structure that everything else builds on.

The CY 2027 advance notice proposes to recalibrate V28 using more recent data, updating from 2018 diagnoses and 2019 expenditures to 2023 diagnoses and 2024 expenditures. This is not a new model version. It is V28 with refreshed calibration data. But the refresh matters: it will update the cost weights to reflect four additional years of clinical and spending patterns, including post-pandemic utilization shifts. Whether that recalibration produces higher or lower coefficients for specific conditions will depend on how spending patterns evolved between the pre-COVID and post-COVID periods. For plans, the recalibration is one more variable in a CY 2027 environment that has already stacked multiple revenue-reducing mechanisms into a single payment year.

Related Reading

#

MCR-10_02 Racial and Ethnic Health Equity in Medicare: HCC Coding Gaps, Benefit Disparities, and What the Data Shows MCR-04_07 Star Ratings in Transition: The Quality Bonus Payment Battlefield