Written and maintained by CASRAI Editorial Board
Last updated
Search for “CMS MEAT guidelines” and you will find a great deal of confident writing about a document that does not exist. MEAT is not a CMS standard. It is an industry mnemonic — Monitor, Evaluate, Assess/Address, Treat — that coders and CDI teams adopted because it is a serviceable shorthand for a requirement CMS states in quite different words. Understanding the difference is not pedantry. In a records review, you will be defending your documentation against the actual requirement, not against the mnemonic.
This guide covers what CMS actually requires for a diagnosis to support risk adjustment, which model version is in force and for which payment year, why conditions have to be re-established every calendar year, and how to use MEAT correctly as a drafting tool rather than as a legal standard.
What CMS actually requires
The requirement is stated plainly in the CY 2026 Rate Announcement, where CMS reiterates guidance including its April 15, 2022 HPMS memorandum, “Reminder of Existing Obligation to Submit Accurate Risk Adjustment Data”:
All diagnoses submitted for risk adjustment must meet CMS requirements that diagnoses are documented in the medical record as a result of a face-to-face visit and coded in accordance with the International Classification of Diseases, Clinical Modification (ICD-CM) guidelines for coding and reporting, which apply equally to diagnoses resulting from health risk assessments and chart reviews.
Three obligations sit inside that sentence, and they are the whole standard:
- A face-to-face visit. The diagnosis must arise from an encounter, not from a problem list, a claims history or a risk-score gap report.
- Documentation in the medical record. The encounter note must actually contain the clinical substance supporting the diagnosis.
- Coding in accordance with the ICD-10-CM Official Guidelines for Coding and Reporting. This is the operative coding standard — the one with actual regulatory force behind it.
Notice the final clause. These requirements “apply equally to diagnoses resulting from health risk assessments and chart reviews.” Diagnoses harvested from an HRA or a retrospective chart review get no relaxed standard. They must trace back to a documented face-to-face encounter, coded per the Official Guidelines, exactly as any other diagnosis must.
Which model, which payment year
Any statement about which conditions map to an HCC is meaningless without a model version and a payment year attached, because the mapping changed substantially and recently.
CMS names its models by year; the industry names them by version number. The model the industry calls v28 is what CMS calls the 2024 CMS-HCC model. The predecessor, v24, is the 2020 model. A third, v22, is the 2017 model, still relevant to PACE organizations.
The 2024 model was not a tweak. As CMS explained when finalizing it, the underlying condition categories were rebuilt around ICD-10: “a reclassification had to occur to develop HCCs aligned with the ICD-10 classification system,” and “as a result of this reclassification, the number of HCCs in the payment model increased from 86 to 115.”
CMS also warned against the comparison most organizations reach for first. Because the model was reclassified and recalibrated simultaneously, “generally comparing HCC coefficients across both models would not be an appropriate measurement.” A coefficient that moved between models did not necessarily change in clinical significance; the denominator moved too.
The phase-in, and where it landed
CMS phased the 2024 model in over three years rather than switching at once. From the CY 2024 Rate Announcement: for CY 2024, risk scores were “a blend of 67% of the risk scores calculated with the current model (the 2020 model) and 33% of the risk scores calculated with the updated model (the 2024 model)”; for CY 2025, 33% of the 2020 model and 67% of the 2024 model; and for CY 2026, “100% of the risk scores to be calculated with the 2024 model.”
The CY 2026 Rate Announcement confirmed that landing. For non-PACE organizations, “CMS is completing the phase-in of the 2024 CMS-HCC model … by using 100 percent of the risk score calculated using the 2024 CMS-HCC risk adjustment model.”
| Payment year | 2020 model (v24) | 2024 model (v28) |
|---|---|---|
| CY 2024 | 67% | 33% |
| CY 2025 | 33% | 67% |
| CY 2026 | — | 100% |
PACE organizations are on a different schedule entirely. For CY 2026, CMS began phasing out the 2017 model for PACE, calculating risk scores as “a blend of 10 percent of the risk score calculated using the 2024 CMS-HCC risk adjustment model and 90 percent of the risk score calculated using the 2017 CMS-HCC risk adjustment model.” A PACE organization in CY 2026 is overwhelmingly still on the 2017 model. ESRD models run on yet another track: for CY 2026, non-PACE organizations continue using the 2023 CMS-HCC ESRD models for beneficiaries in dialysis, transplant and post-graft status.
Everything below describes the 2024 CMS-HCC model (v28) as applied at 100% for payment year 2026 to non-PACE Medicare Advantage organizations. If you are working PACE or ESRD populations, check your model before applying any of it.
Two adjustments that sit on top of your coding
Two CY 2026 parameters are worth knowing, because they explain why accurate coding does not translate one-for-one into revenue:
- The MA coding pattern difference adjustment is 5.9 percent for CY 2026 — the statutory minimum, which the Act has set as a floor for 2019 and each year after. CMS is emphatic about what this is not: “the coding pattern adjustment is not an adjustment for inaccurate or fraudulent coding, but rather is a program-wide adjustment designed to account for the impact on MA risk scores of the differential coding patterns between MA and FFS.”
- Normalization factors for CY 2026 were calculated using a five-year multiple linear regression and average historical FFS risk scores from 2020 through 2024: 1.067 for the 2024 CMS-HCC Part C model and 1.187 for the 2017 model.
Neither is something a coder or clinician influences. Both are reasons why a raw RAF figure is not a revenue figure.
Why recapture is annual
The single most misunderstood feature of risk adjustment is that a condition documented last year does not carry forward. The CMS-HCC model is prospective: diagnoses collected during a data collection year determine the risk score CMS pays against in the following payment year. Each payment year is calculated fresh. A beneficiary’s amputation, transplant status or chronic condition does not persist in the model simply because it persists in the patient.
This is why “recapture” is the operative word. Every chronic condition that should influence the risk score has to be re-documented at a face-to-face encounter and re-submitted within each collection year — not because CMS doubts the condition resolved, but because the payment mechanism is built on the current year’s submitted data.
The practical consequences follow directly:
- A patient who does not present during the year generates no diagnoses, and every one of their conditions drops out of the following year’s score.
- A condition addressed at one visit and omitted from every subsequent note is captured — but a condition merely carried in the problem list and never addressed is not.
- Status conditions that are permanent in reality still require annual documentation.
The corollary matters just as much: recapture is not a licence to code from the prior year’s list. The condition must be addressed at this year’s encounter and documented there. Copying last year’s assessment forward produces exactly the record that fails review.
Using MEAT correctly
With the actual standard established, MEAT becomes useful for what it is: a drafting checklist that helps a clinician write a note demonstrating the condition was genuinely addressed. It is a way of satisfying the ICD-10-CM Official Guidelines, not a substitute for them.
| Element | What it asks | Documentation that carries it |
|---|---|---|
| Monitor | Are signs, symptoms or disease progression being tracked? | “Diabetic neuropathy — numbness stable in both feet since last visit, no new ulceration on monofilament exam.” |
| Evaluate | Were findings, results or medication response reviewed? | “CKD stage 4 — eGFR 24, down from 27 in March; reviewed today’s BMP with patient.” |
| Assess / Address | Was the condition’s status judged and recorded? | “COPD — worsening; two exacerbations since last visit, now GOLD group D.” |
| Treat | Was a therapy, referral or plan ordered? | “Heart failure with reduced EF — increasing torsemide to 40 mg daily; cardiology referral placed.” |
Any one element, properly documented, demonstrates the condition was addressed. The failure mode is not “only three of four” — it is documentation containing none of them: a diagnosis appearing in a problem list, a past medical history recitation, or an isolated code with no clinical narrative attached anywhere in the note.
What consistently fails
- The unlinked problem list. Fifteen conditions in a sidebar, none mentioned in the assessment. The visit addressed two of them; the record supports two.
- History versus current. “History of CVA” and “residual hemiplegia following CVA” are different clinical statements with different risk implications. Write the one that is true.
- Copied-forward assessments. Identical text across four visits signals to a reviewer that nothing was assessed at any of them.
- Unspecified codes where specificity exists. The 2024 model’s ICD-10 realignment made specificity more consequential, not less.
- Diagnoses that appear only in an HRA. CMS’s requirements apply equally to health risk assessments; an HRA finding still needs a documented face-to-face encounter behind it.
What “hierarchical” actually does
The H in HCC is a ranking rule. Within a disease hierarchy, the most severe manifestation that a patient’s coded diagnoses support supersedes the less severe ones, and only that category contributes to the score. Documenting both a milder and a more severe form of the same disease does not add two categories. This is why coding more diagnoses is not the objective — coding the accurate one is. Conditions in different hierarchies do accumulate, and some combinations carry additional interaction effects in the model.
Certification, correction and the sixty-day clock
Risk adjustment documentation is not a best-practice aspiration; it is attached to affirmative legal obligations that CMS set out in the same passage of the CY 2026 Rate Announcement.
Organizations submitting risk adjustment data “are required to certify as a condition of receiving monthly payment that the data submitted under 42 CFR 422.310 are accurate, complete, and truthful based on best knowledge, information, and belief (see 42 CFR 422.504(l)).” And “if an organization receives information about inaccurate diagnoses, the organization must correct its data.”
The Part C/D Overpayment Rule at 42 CFR 422.326 supplies the consequence. As CMS summarizes it, quoting the D.C. Circuit: the rule “establishes that, if a Medicare Advantage insurer has received a payment increment for a beneficiary’s diagnosis and discovers that there is no basis for that payment in the underlying medical records, that is an overpayment that the insurer must correct by reporting it to CMS within sixty days for refund.” That formulation comes from UnitedHealthcare Ins. Co. v. Becerra, 16 F.4th 867, 869–870 (D.C. Cir. 2021), which upheld the relevant portions of the rule.
Read together: discovering that a submitted diagnosis lacks support in the record is not a neutral finding to be corrected at the next reconciliation. It starts a sixty-day clock. That is the reason internal audit findings need a defined remediation path rather than a backlog.
Where this does not apply
The CMS-HCC model described here governs Medicare Advantage. Commercial ACA marketplace plans run on the separate HHS-HCC model, which uses different condition categories, is concurrent rather than prospective, and is calibrated on a commercial population. Medicaid managed care programs use state-selected models that vary by state. Do not carry a v28 mapping into any of them.
Within Medicare Advantage itself, individual plans may impose documentation expectations beyond the CMS floor as a contract term. Those are contractual, not regulatory — but they are still enforceable against you.
Frequently asked questions
Are MEAT criteria a CMS requirement?
No. MEAT is an industry mnemonic, not a CMS standard, and there is no CMS document titled “MEAT guidelines.” The requirement CMS states is that diagnoses be documented in the medical record as a result of a face-to-face visit and coded in accordance with the ICD-10-CM Official Guidelines for Coding and Reporting. MEAT is a useful way of meeting that requirement, and nothing more.
Which CMS-HCC model version is in effect?
For payment year 2026, non-PACE Medicare Advantage organizations are on 100% of the 2024 CMS-HCC model — the version commonly called v28 — following a three-year phase-in that ran 33% in CY 2024 and 67% in CY 2025. PACE organizations in CY 2026 are on a blend of 90% of the 2017 model and 10% of the 2024 model.
How many HCCs are in v28?
115. CMS stated that the ICD-10 realignment increased “the number of HCCs in the payment model … from 86 to 115.”
Do chronic conditions need to be documented every year?
Yes. The model is prospective and each payment year’s risk score is built from that collection year’s submitted diagnoses. Conditions do not carry forward, which is why annual recapture at a documented face-to-face encounter is necessary — including for conditions that are permanent.
Can a diagnosis be captured from a health risk assessment or chart review?
Only if it satisfies the same requirements as any other diagnosis. CMS states that the face-to-face and ICD-10-CM coding requirements “apply equally to diagnoses resulting from health risk assessments and chart reviews.”
Does documenting a condition in the problem list support the code?
No. A problem list entry is not evidence that the condition was addressed at the encounter. The assessment and plan must show the condition being monitored, evaluated, assessed or treated.
What happens if we discover a submitted diagnosis was unsupported?
Under 42 CFR 422.326, a payment increment with no basis in the underlying medical records is an overpayment that must be reported to CMS within sixty days for refund. Submitting organizations also certify under 42 CFR 422.504(l) that data submitted under 42 CFR 422.310 are accurate, complete and truthful, and must correct data on learning of inaccurate diagnoses.
Why did our RAF scores fall when nothing changed clinically?
Several CY 2026 parameters move scores independently of documentation: the completion of the 2024 model phase-in at 100%, the normalization factor of 1.067 for that model, and the 5.9 percent MA coding pattern difference adjustment. The 2024 model also reclassified condition categories, so some previously mapping diagnoses no longer do.
Related guides
- Number and Complexity of Problems Addressed: The First MDM Column — the parallel question of what “addressed” means in E/M documentation.
- Chronic Care Management vs the Office Visit: Which Minutes Go Where — chronic condition documentation in the primary care encounter.
- Risk of Complications and Morbidity: The MDM Column With No Numbers.
- G2211 Billing Guidelines and Documentation: What CMS Actually Requires — longitudinal care documentation on the fee-for-service side.
- RAC Audit: Triggers, Review Types, ADR Timelines, and the Five-Level Appeal.
Sources
- Centers for Medicare & Medicaid Services, Announcement of Calendar Year (CY) 2026 Medicare Advantage (MA) Capitation Rates and Part C and Part D Payment Policies (CY 2026 Rate Announcement), released 7 April 2025.
- CMS, Announcement of Calendar Year (CY) 2024 Medicare Advantage Capitation Rates and Part C and Part D Payment Policies (CY 2024 Rate Announcement), released 31 March 2023.
- CMS, Risk Adjustment — 2026 Model Software / ICD-10 Mappings (CMS-HCC software releases for the 2024 and 2017 models).
- CMS HPMS memorandum, “Reminder of Existing Obligation to Submit Accurate Risk Adjustment Data,” 15 April 2022.
- 42 CFR 422.310; 42 CFR 422.326; 42 CFR 422.504(l).
- UnitedHealthcare Ins. Co. v. Becerra, 16 F.4th 867 (D.C. Cir. 2021).
