HCC risk adjustment coding works differently from fee for service billing, and practices that treat the two the same way lose money without ever seeing a denial. There is no rejected claim to work and no remittance line to appeal. The revenue simply never arrives, because the conditions a patient carries were treated during the year and never made it into the coded record in a form that counts.
Medicare Advantage plans are paid based on how sick their enrolled population is. Hierarchical Condition Categories are how that gets measured, and the measurement runs entirely on documented and coded diagnoses. A patient with several serious chronic conditions generates a higher payment than a healthy patient, but only if the record proves those conditions were addressed.
How the Payment Actually Works
Each patient carries a risk adjustment factor score built from demographic details and the conditions mapped to HCC categories during the year. That score drives the payment the plan receives, and in arrangements where a practice shares risk, it flows through to the practice.
The part people miss is that the score depends on coded data rather than clinical reality. A patient with advanced kidney disease who has been managed carefully for years still scores as though that condition does not exist if nobody codes it in the current year with support in the note.
Version 28 Is Now the Entire Calculation
The model changed, and 2026 is the first year running on it alone.
CMS phased the CMS-HCC Version 28 model in over three years, running it alongside the older Version 24 model at a shifting ratio. Payment year 2026 uses Version 28 at one hundred percent. The cushion that softened earlier years is gone, so any gap between older coding habits and what the current model requires now shows at full size.
What Changed in the Model
Two shifts matter for daily coding work.
The count of ICD-10-CM codes that map to a payment category dropped from roughly 9,800 to about 7,800. Codes that carried risk weight under the old model may carry none now, which means a practice can code diligently and still see scores fall.
At the same time, the number of payment categories grew from 86 to 115. The model covers more conditions at higher specificity while dropping many of the vaguer codes that used to contribute. The practical effect is that unspecified codes carry less value than they did, and specific ones carry more.
The response that works is better records of the conditions clinicians are already treating, rather than a search for additional diagnoses to add.
Diagnoses Reset Every January
This catches many practices every year. Risk adjustment operates on a calendar year basis, and conditions do not carry forward. A diagnosis documented and coded last year contributes nothing to this year’s score.
Every chronic condition a patient carries has to be addressed and coded again in the current year, supported by a qualifying encounter. Diabetes managed for a decade still needs to appear in this year’s record with current support. Practices that assume the problem list is doing this work find out at year end that most of it never got coded.
This is why annual wellness visits and scheduled chronic care follow-ups matter beyond the visit revenue. They create encounters where conditions get addressed and documented.
The Documentation Standard
Listing a condition in the chart does not support it. Auditors look for evidence the condition was actively addressed, and the shorthand most coding teams use is MEAT.
The condition should be monitored, with the note showing signs, symptoms, or disease progression. It should be evaluated, with test results, response to treatment, or examination findings recorded. It should be assessed or addressed, with the clinician’s thinking about status visible. It should be treated, through medication, therapy, referral, or another plan.
What Does Not Support a Code
Any one of these supports the code when it appears in the note in the clinician’s own words. What does not support it is a diagnosis appearing only in a problem list, a medication list, or a copied section carried forward from a prior visit without current commentary.
The Encounter Itself Has to Qualify
Documentation quality is not enough on its own. The encounter has to meet structural conditions, and audits fail records on these points regularly.
The visit has to be face to face, performed by an acceptable provider type, with the signature and credentials of the person who rendered the service. The record has to match the correct patient and the correct date of service. Records get invalidated for reasons that have nothing to do with clinical care: a missing signature, missing credentials, a mismatched date, or a document that does not qualify as a medical record at all.
Diagnoses captured only through health risk assessments without corresponding clinical support have drawn particular scrutiny.
Audits Now Carry Extrapolation
Risk Adjustment Data Validation audits test submitted diagnoses against the medical record, and the terms changed in a way that raises the stakes considerably.
CMS can now extrapolate audit findings across a contract rather than limiting recovery to the sampled records. One unsupported code in a sample can drive a payment recovery far larger than the value of that single code. Audit activity has expanded, and a federal compliance audit published in March 2026 found unsupported codes in the large majority of sampled high-risk diagnoses it reviewed.
The takeaway for a practice is that a lower score fully supported by the record is a stronger position than a higher score that collapses under review.
Coding Conditions That Have Resolved
Review has to run in both directions, and most programs only run one.
Review Has to Run in Both Directions
Teams look for conditions that were treated and never coded. Fewer look for conditions still being coded after they are resolved. A condition carried forward year after year on a problem list, coded as active while the record shows it cleared, is exactly the pattern an auditor pulls. Acute conditions from prior years are the usual offenders, along with conditions that were ruled out but never removed.
Removing unsupported codes lowers the score. It also removes the part of the score most likely to be taken back with extrapolation attached.
Building a Program That Holds Up
Start with the schedule. Patients with chronic conditions need at least one qualifying encounter during the year, and the ones who have not been seen should be identified early rather than in December.
Give clinicians documentation prompts that fit the workflow rather than adding to it. Most gaps come from conditions that were genuinely managed but written up in a way that does not show the management.
Audit prospectively instead of retrospectively where possible, since correcting a note before submission is far simpler than defending it afterward. Review both directions on every audit, adding what is supported and removing what is not.
Track the specific things that predict audit outcomes: the share of chronic conditions recaptured in the current year, the rate of unspecified codes where a specific option exists, signature and credential completeness, and the count of conditions coded without qualifying encounter support.
If you want a clearer picture of where your risk adjustment documentation stands under the current model, our team can review a sample of encounters and show what the record supports and what it does not.
