Identify the applicable CMS model.
Document the program, payment year, model name, software release, and mapping file. Do not apply the non-PACE 2026 rule to PACE or to a different risk model.
2026 model-year control guide
HCC V28 coding starts with the right model, then the right evidence.
For non-PACE organizations, CMS completed the phase-in of the 2024 CMS-HCC model for 2026. Accurate operations still require more than a crosswalk: use the current CMS release, code from the medical record, retain source context, and keep qualified reviewers in control.
Reviewed July 26, 2026 | Official CMS sources below
Structural change
V28 is a different model, not a re-priced V24.
The category set, the paying code list, and the calibration data all changed. Documentation that scored under V24 can score lower under V28 without a single clinical fact changing, which is why a stale crosswalk quietly overstates the opportunity.
Payment HCCs86→115Categories were renumbered as well as expanded, so a V24 HCC number rarely means the same condition under V28.
ICD-10-CM codes that map to a payment HCC~9,700→~7,700Roughly 2,300 codes were re-weighted and 218 were removed from the risk model entirely.
Disease hierarchies enforced87Only the most severe condition in each family pays. Lower categories in the same family are suppressed, not summed.
Calibration data2018-2019Coefficients were recalibrated on 2018-2019 Medicare data rather than the earlier years behind V24.
Three-year phase-in of the 2024 CMS-HCC model
Across audited medical groups, re-scoring the same patient panels under V28 moved the average RAF -9.3 percent. The panel did not get healthier. The model changed underneath it.
Payment HCC counts, code-list size, calibration years, and the phase-in schedule follow the Announcement of CY2024 Medicare Advantage Capitation Rates and Part C and Part D Payment Policies, Attachment VIII. The -9.3 percent figure is an observed change across medical groups ASP-RCM audits, not a CMS publication.
Hierarchy resolution decides which code actually pays
The most common V28 arithmetic error is adding every documented category in a family. The model does not work that way. Within each of the 87 hierarchies, the most severe category pays and every less severe category in that family is suppressed.
HCC 327 (0.514) and HCC 329 (0.127) coefficients are from the CY2024 Rate Announcement, Attachment VIII, Table VIII-1, Community NonDual Aged column. The kidney hierarchy is Table VIII-4. Check the arithmetic on your own member mix with the CMS-HCC V28 RAF score calculator.
Four non-negotiable controls
Separate the model, the code, the evidence, and the decision.
Operational errors happen when a team treats an HCC mapping as a coding answer. The mapping is one controlled step in a longer clinical and compliance workflow.
Code the documented condition accurately.
Use the current official ICD-10-CM guidelines and the complete medical record. Do not choose a diagnosis because its HCC coefficient is attractive.
Keep the support inspectable.
Retain the note, page or section, date of service, provider context, and the documentation used during review. Problem-list context alone may not establish current support.
Record the qualified review decision.
AI can organize evidence and identify candidates. A qualified reviewer applies organizational policy and records the disposition and rationale before downstream action.
Evidence-first operating flow
Five stages from record context to reconciled status.
The control path should be reproducible after the encounter, after submission, and during an audit review.
Set the model context
Confirm program, payment year, member eligibility, and applicable CMS release.
Inspect the record
Locate current documentation and the clinical context relevant to accurate coding.
Apply ICD-10-CM
Assign the diagnosis under current official coding guidance and organizational policy.
Apply CMS logic
Use the controlled model software and mapping release appropriate to the scope.
Close the status
Retain reviewer action, rationale, submission state, rejection, deletion, or other outcome.
The arithmetic
A RAF score is a sum, and every term is a documentation decision.
Under V28 the score is additive: a demographic factor, one coefficient per payment HCC after hierarchies, disease interaction factors, and a payment HCC count factor once a member carries five or more payment HCCs. The waterfall below walks a single member from demographic base to total.
Demographic factor, HCC coefficients, the four disease interactions (0.112 + 0.078 + 0.176 + 0.077 = 0.443), and the payment HCC count factor at seven HCCs (0.188) are the Community NonDual Aged column of the CY2024 Rate Announcement, Attachment VIII, Table VIII-1. The same table carries the $10,402.34 denominator used to convert a risk score into a payment amount. Run your own combinations in the RAF score calculator.
Pre-release checklist
Before HCC V28 coding moves downstream.
- The program and payment year are explicit.
- The official CMS model and mapping release are versioned.
- The diagnosis follows current ICD-10-CM guidance.
- The supporting record context is linked and inspectable.
- Uncertainty is routed for qualified review.
- The reviewer disposition and rationale are retained.
- Submission, rejection, deletion, and reconciliation states are traceable.
From suspect to submitted
Most suspects should not survive the walk to submission.
A suspect list is a hypothesis queue, not a coding worklist. Attrition is the point of the process. A program that confirms nearly everything it surfaces is not efficient, it is under-reviewed.
Every exit above is a legitimate outcome that must be recorded. A rejected suspect with a written rationale is stronger audit evidence than a confirmed code with none. See how the queue is assembled in the HCC risk adjustment AI workflow.
Retrospective, concurrent, and prospective review answer different questions
The three review modes are not competing products. They sit at different points relative to the encounter, and each one can fix something the others cannot.
Before the encounter
Prospective
- Looks at
- Prior-year conditions and claims history before the visit is scheduled.
- Can fix
- Documentation at the source. The clinician sees the gap while the patient is in front of them.
- Cannot fix
- Anything for an encounter that never happens. It depends entirely on the visit occurring.
- Risk
- Pre-populated suspects can turn into leading prompts if the queue is presented as an answer.
During the open encounter
Concurrent
- Looks at
- The note in progress, before the chart closes and the claim leaves.
- Can fix
- Specificity while the clinician still remembers the visit. Queries return fastest here.
- Cannot fix
- Conditions that live only in outside records the reviewer cannot reach in time.
- Risk
- Turnaround pressure. A rushed concurrent review produces the weakest rationale of the three.
After submission
Retrospective
- Looks at
- The closed chart against what was actually submitted for the service year.
- Can fix
- Both directions. It is the only mode that reliably finds codes that should be deleted.
- Cannot fix
- Missing documentation. If the support was never written, no later review can create it.
- Risk
- Additions-only retrospective programs are the pattern auditors look for first.
A program running only one mode has a predictable blind spot. Retrospective-only finds nothing at the source, prospective-only never deletes an unsupported code, and concurrent-only cannot see the year in aggregate.
The evidence standard
MEAT is the test a chart has to pass on its own.
Monitor, Evaluate, Assess, and Treat describe what a note has to show for a chronic condition to count in the service year. A diagnosis in the problem list is a reference, not evidence.
MMonitor
SatisfiesSigns, symptoms, disease progression, or a tracked value recorded this visit. "eGFR down from 34 to 28 since March."
FailsA value pasted into the note by the interface with no clinician comment on what it means.
EEvaluate
SatisfiesTest results, response to treatment, or physical exam findings tied to the condition by name.
FailsA lab panel attached to the chart with no narrative connecting the result to the diagnosis.
AAssess
SatisfiesThe condition named in the assessment with a status: stable, worsening, controlled, exacerbated.
Fails"History of" language, or the condition appearing only in the problem list carried forward from a prior year.
TTreat
SatisfiesMedication, therapy, referral, or a plan documented for that specific condition at that visit.
FailsAn active medication on the list with no plan text. The drug alone does not establish current management.
Not every element is required for every condition, and organizational policy sets the internal threshold. The failure mode to design against is the same in all four columns: data present in the chart, clinician judgment absent from the note.
What a defensible chart carries into a RADV review
A RADV reviewer works from the record, not from your workflow. Everything below has to be reconstructable from what you retained, years after the reviewer who made the call has moved on.
01 · SourceThe note itself
The signed record, the page or section relied on, the date of service, and the rendering provider with credentials. A summary of the note is not the note.
02 · Encounter validityA qualifying face-to-face
Evidence the encounter type is acceptable for risk adjustment under the applicable program rules for that payment year.
03 · Coding basisGuideline version applied
The ICD-10-CM guideline edition in force at the date of service, and the code selected under it.
04 · Model versionRelease and mapping file
The CMS model year, software release, and ICD-to-HCC mapping file used. Stale mappings are found in audit, not before it.
05 · Human decisionReviewer, disposition, rationale
Who reviewed it, what they decided, why, and when. An unattributed approval is not a defense.
06 · LifecycleSubmission and deletion trail
Submission state, acceptance, rejection, and any deletion with its reason. Deletions are evidence of a working program.
CMS and other authorized reviewers determine whether a submitted diagnosis is supported. No retention practice guarantees a RADV outcome. What retention does is make the answer knowable before someone else asks. Review how these controls run as a managed service in HCC coding services.
Frequently asked questions
HCC V28 coding questions that affect implementation.
Is the 2024 CMS-HCC model fully phased in for 2026?
For non-PACE organizations, CMS states that 2026 risk scores use 100 percent of the 2024 CMS-HCC model. PACE organizations follow a different blend, so confirm the program before applying model logic.
Is V28 the same as the 2024 CMS-HCC model?
V28 is a commonly used industry label for the 2024 CMS-HCC model. Operational documentation should retain the official CMS model year, payment year, and software or mapping release used.
Can an HCC mapping determine the final diagnosis code?
No. The mapping identifies model relationships after a diagnosis is accurately coded. The medical record, current ICD-10-CM guidance, applicable program rules, and qualified human review control the coding decision.
What should an HCC V28 audit trail retain?
A useful audit trail retains source documentation references, date of service, provider context, diagnosis code, CMS model and mapping version, reviewer disposition, rationale, and downstream status.
Continue the HCC evaluation
Use one page for each buyer question.
Your model and policy control the answer
Map one HCC workflow under the 2026 model.
Start with a de-identified example. ASP-RCM will map the source, model release, review step, disposition, and reconciliation path. Confirm a secure method before any protected information moves.