CMS delays the audio-only and unlinked CRR exclusions to the CY2027 Midyear model run

Your first-half-2026 audio-only and unlinked chart review captures just got a six-month reprieve, and a retroactive clawback date to go with it. In an HPMS memo dated July 9, 2026, CMS moved both CY2027 diagnosis exclusions from the CY2027 Initial model run to the CY2027 Midyear model run. The exclusions still bind payment year 2027 and 2026 dates of service. Only the run that applies them moved. CMS says the delay is to "streamline implementation and allow organizations sufficient time to acclimate to the unlinked CRR and audio-only diagnosis exclusions".

The January 1, 2027 start date we reported is superseded

Our CY2027 Rate Announcement piece said both exclusions start January 1, 2027. That framing was ours, not a CMS effective date, and it no longer holds. Diagnoses from unlinked CRRs and from audio-only encounters with modifier "93" or "FQ" will not be excluded from risk score calculations during the CY2027 Initial model run. Nothing about the policy slipped a year, and the exclusions themselves are untouched.

Only half of 2026 actually gets the reprieve

This is the part worth slowing down on, because two different clocks are running.

The exclusions apply to 2026 dates of service. The Initial run, though, only reads diagnoses from July 2025 through June 2026. Intersect those and only your January through June 2026 captures are in the Initial run at all. Those are the ones that ride into the January 2027 risk score and then fall out at Midyear.

July through December 2026 dates of service are never scored by the Initial run. They first appear at the Midyear run, which reads January through December 2026, and by then the exclusions are already on. There is no window where those count. If you are sizing the first-half-2027 cushion, that distinction is the whole number.

Initial run versus Midyear run: what each one excludes

The audio-only filtering logic is all-or-nothing on the line items

CMS states the test at the record level, not the diagnosis level. Diagnoses from professional and outpatient encounter data, chart review records, and FFS claims with 2026 dates of service are excluded "if all risk adjustment eligible CPT/HCPC line items associated with the record/claim have a modifier of '93' or 'FQ'".

The flip side matters more at the desk. If one or more risk adjustment eligible service lines did not carry modifier "93" or "FQ," all header diagnoses stay in for risk score consideration. So a mixed record is not partially excluded. It is fully included.

CMS also restates the underlying rule: diagnoses must result from face-to-face encounters to be considered for risk adjustment, and the exclusion applies across all provider types. The change is for CY2027 (2026 dates of service) and does not change filtering logic or submission guidance for prior years. Our explainer on telehealth diagnoses as a risk adjustment source covers the underlying modality rules.

Unlinked CRRs: the one exception, and how CMS actually checks it

For non-PACE organizations, diagnoses from unlinked CRRs for 2026 dates of service are excluded from risk score calculation, with an exception for beneficiaries who switch from one Medicare Advantage contract to another in a different parent organization.

The test is narrower than it sounds. CMS verifies that the parent organization effective on the date of service is different from the parent organization effective at the time of submission, so a contract-to-contract move inside one parent organization does not qualify. Diagnoses that clear that check show on the MAO-004 as allowable. For background on how the ban took shape, see our coverage of the proposal to ban unlinked chart reviews.

On linking itself, CMS did not issue new instructions. Plans should keep following existing guidance for CRR linking in the Encounter Data Submission and Processing Guide and the MAO-004 User Guide.

Two sets of MAO-004 reports and the new Allow/Disallow Reason Code "U"

CMS says it "is working on updates to the MAO-004 report" to address the CY2027 exclusions, so treat the report mechanics below as CMS's stated plan rather than a shipped layout.

For the months of January through September 2026, CMS will issue two sets of MAO-004 reports. The first set will not reflect the exclusions, matching the Initial run. The second set, including reissued reports for January through September submissions, will be released during November 2026 and will reflect the exclusions, matching the Midyear run. Beginning with the November report for all submissions during October 2026, MAO-004 reports will reflect the exclusions.

Two reason codes carry it. The updated report adds new Allow/Disallow Reason Code "U" for diagnoses disallowed because of the unlinked CRR exclusion where the parent organization switcher exception is not met, and that code applies to non-PACE organizations only. The audio-only exclusion is folded under the existing Allow/Disallow reason code "H" for both PACE and non-PACE submissions.

That November release is the first MAO-004 output that reflects the exclusions, and the reissued January through September set is where a plan can see which of those submissions fall out at Midyear. CMS has not yet published the updated layout.

Payments move in January, then get adjusted in July 2027

Payments for January through June 2027 are based on the CY2027 Initial model run and do not reflect either exclusion. Payments for July through December 2027 are based on the Midyear run and do reflect both.

Any retroactive adjustments to January through June payments caused by risk score changes between the Initial and Midyear runs will be made during July of 2027. Where a Midyear score differs from the Initial score, the Midyear score is used from July through the end of the year. Our CY2026 midyear risk score explainer walks through how that reconciliation reads on the payment side.

Models, MOR record types, and the MMR hold; here are the CY2027 normalization factors

Nothing in this memo changes which codes map. CY2027 still runs the 2024 CMS-HCC model, listed at model version V28, so a capture that mapped last week maps the same way today.

The rest is plan-side. CMS continues the 2023 ESRD model at V24 for beneficiaries in dialysis, transplant, and post-graft status, and the 2027 RxHCC model (2023/2024 calibration) at V08, with separate continuing enrollee segments for MA-PD plans and stand-alone PDPs. The memo describes these as previously described in the CY2027 Rate Announcement.

CMS calculated the CY2027 CMS-HCC normalization factors using a four-year simple linear regression and average FFS risk scores from 2022 through 2025. The published values are 1.079 for the 2024 CMS-HCC model and 1.202 for the 2017 CMS-HCC model, which for CY2027 appears only in the 50/50 PACE blend; 1.072 and 1.145 for the 2023 and 2019 ESRD Dialysis models; and 1.119 and 1.209 for the 2023 and 2019 ESRD Functioning Graft models. For RxHCC, the 2023/2024 calibration is 1.109 for MA-PD plans and 1.005 for PDPs, with 1.237 for the 2018/2019 calibration used only by PACE organizations. Those are the factors any CY2027 risk score projection has to be built on.

On the reporting side, there are no new MOR record types for CY2027, and there will be no updates to the MMR for CY2027. RxHCC MOR record types 6 and 7 carry over from CY2026 to the CY2027 RxHCC models.

Risk adjustment model software moves from SAS to Python

CMS says it is continuing work toward transitioning the risk adjustment model software away from SAS by CY2028. Python software is available for the CY2027 Initial software, with CY2027 Midyear/Final software to follow later in 2026.

The hard edge is CY2028. From the CY2028 Initial software forward, CMS intends to only create and release risk adjustment model software in Python. If anyone on your analytics side still runs the SAS package, CY2027 Midyear/Final is the last run that has one.