Risk adjustment in 2026: What V28 and RADV audits mean for health plans
Sponsored content. This article was supplied by a third party and Business Money has been paid to publish it.
Medicare Advantage plans spent three years preparing for the V28 model and a separate stretch preparing for expanded audits. Both arrived at full force in 2026, within weeks of each other.
That timing is the story. A stricter payment model and a broader audit program reward the same thing, which is documentation that holds up when somebody else reads the chart.
Key takeaways
- CMS-HCC V28 reached 100% implementation on 1 January 2026, and the V24 blend no longer applies.
- V28 uses 115 payment categories against V24’s 86, while cutting valid diagnosis codes from 9,797 to 7,770.
- CMS moved from auditing roughly 60 contracts a year toward every eligible contract, around 550 in total.
- A federal court vacated the 2023 RADV extrapolation rule in September 2025, and CMS has appealed.
- Audits continue on a quarterly cadence regardless of how the extrapolation question resolves.
- MEAT-supported documentation is the common requirement across both changes.
V28 finished phasing in on 1 January
CMS finalized V28 in the 2024 Rate Announcement and phased it across three payment years. Risk scores ran on a 33% V28 blend in 2024, shifted to 67% in 2025, and reached 100% this year.
The blend was the cushion, and it is gone. A plan still operating on V24-era coding habits absorbed only a third of the impact in 2024 and now absorbs all of it.
What actually changed in the model
The structural changes are larger than a version number suggests. V28 expands to 115 payment HCC categories from V24’s 86, renumbering and splitting many of them, so a condition that mapped to one category now often maps somewhere else entirely.
The bigger shift is what no longer counts. Valid ICD-10-CM diagnosis codes fell from 9,797 to 7,770, a net reduction of 2,027 after 2,236 removals and 209 additions.
V28 also introduces constraining, which forces some clinically related categories to share a coefficient.
That removes the incremental credit plans previously earned for documenting greater severity within the same disease family.
CMS projected an average RAF decline of roughly 3.12% from these changes. The effect is uneven, falling hardest on panels weighted toward the condition families that lost the most ground.
RADV audits scaled up at the same time
In May 2025, CMS announced it would audit every eligible Medicare Advantage contract annually rather than a limited subset. That moves the program from roughly 60 contracts a year toward approximately 550.
The resourcing followed. CMS expanded its certified coder capacity from around 40 reviewers to approximately 2,000, and signaled it would use AI-assisted tooling with final determinations still made by human coders.
CMS also set out to clear a backlog running from Payment Year 2018 through 2024. That timeline has slipped, but the direction has not changed. A January 2026 memo confirmed the program is proceeding.
Payment Year 2020 audits began in February 2026 on a roughly quarterly cadence, with sample sizes between 35 and 200 enrollees and a restored five-month record submission window.
The extrapolation question is still open
This is where a lot of published guidance is out of date. On 25 September 2025, the US District Court for the Northern District of Texas vacated the 2023 RADV final rule in Humana Inc. v. Becerra.
The ruling was procedural rather than substantive. The court found CMS had failed to satisfy notice-and-comment requirements under the Administrative Procedure Act, and did not rule on whether extrapolation or removal of the fee-for-service adjuster were lawful in themselves.
CMS filed notice of appeal on 21 November 2025, so the position is unresolved. Extrapolated recovery is paused for now, but the underlying question has not been decided.
The practical takeaway is unchanged either way. Audits are proceeding on schedule, and a plan that cannot evidence its diagnoses still faces recovery on the sampled charts, whatever happens on appeal.
Documentation is now the whole exercise
Both changes converge on the same requirement. Under V28, fewer codes carry weight, and under expanded RADV, every submitted diagnosis may need to be defended against the record.
MEAT is the standard framework here, covering whether a condition was monitored, evaluated, assessed, and treated within the documentation. A diagnosis on a claim without MEAT support in the chart is exactly what an auditor removes.
This is why a modern risk adjustment solution is judged on evidence linkage rather than code volume.
Platforms such as RAAPID work in both directions, adding diagnoses the record supports and removing those it does not, with MEAT evidence attached to each HCC.
Two-way review matters more under current conditions than it used to. Submitting an unsupported code now creates audit exposure that can outweigh the RAF value it adds.
What to look for in a platform
Ask how each suggested code traces back to the chart. A system that returns a code list without linked evidence leaves your coders doing the defensibility work manually.
Check whether the tool removes codes as well as adding them. Many were built during a period when the incentive was capture volume, and that design no longer fits the audit environment.
Confirm the model is current. Any platform still reasoning on V24 mappings is working from a code set that retired at the start of this year.
Throughput matters more than it did. With audits on a quarterly cadence and record windows measured in months, a review process that takes weeks per project leaves little room to respond.
Finally, look at the security posture. RAAPID publishes HIPAA compliance and SOC 2 attestation, and any vendor handling protected health information should be able to evidence both.
Conclusion
The two changes are usually discussed separately, but they function as one problem. V28 narrowed what counts, expanded RADV increased how often somebody checks, and neither trend has an obvious reversal ahead of it.
Plans that treated risk adjustment as a coding volume exercise are the ones feeling this hardest. The organizations coping well moved earlier to evidence-linked review, where each code arrives with the chart support already attached. Whichever way the extrapolation appeal resolves, that shift is the durable part.
Frequently asked questions
Is the V24 model still used at all in 2026? No. Payment Year 2026 is the first year scored entirely on V28, so the blended calculation no longer applies to risk scores.
Does the court ruling mean RADV audits have stopped? No. The vacated rule concerned extrapolation and the fee-for-service adjuster, and CMS confirmed in January 2026 that audits continue, with Payment Year 2020 reviews beginning in February.
Will extrapolation come back? That depends on the appeal, which CMS filed in November 2025 and which remains pending. Plans are generally advised to prepare as though it may return rather than assume it will not.
Why did our RAF scores drop without any change in our population? Most likely the model. V28 removed a net 2,027 diagnosis codes from risk adjustment and constrains some related categories to shared coefficients, which lowers scores for identical documentation.
What does MEAT actually require? Evidence in the record that the condition was monitored, evaluated, assessed, and treated. A diagnosis listed without that supporting detail is vulnerable in an audit.