Starting with 2027 payment, CMS no longer counts diagnoses from unlinked chart review records toward Medicare Advantage risk scores. Because 2027 risk scores draw on 2026 dates of service, the change already applies to the charts your teams are reviewing this year, and the plans feeling the change first are the ones with members whose care happens outside the claims stream entirely.
The policy is one more step in a direction the industry has been moving for several years, away from finding diagnoses after the fact and toward documenting them accurately at the point of care. When risk adjustment leaders closed The 27th Risk Adjustment Forum in Orlando with a roundtable on October 29, the conversation kept returning to the same shift. Here is what the change does, where the gap is widest, and what the shift upstream asks of providers, coders, AI tools and the teams around them.
What the 2027 change does
A chart review record is a diagnosis a plan submits from a medical record review rather than from a claim. When the record ties back to an encounter the plan received and processed, CMS treats the record as linked, and when no encounter exists to tie the record to, the record is unlinked. In the CY 2027 Rate Announcement, CMS finalized excluding diagnoses from unlinked chart reviews, along with diagnoses from audio-only encounters, from risk score calculation, with an exception for members who move from one MA organization to another.
Plans still have the option to submit unlinked records, but the diagnoses no longer affect payment. For most members the answer runs through the visit itself, since a diagnosis documented at an encounter your plan paid for stays linked. For some members, no such visit exists in your data.
Members treated at the VA show where the gap lands
Many veterans enrolled in Medicare Advantage receive some or all of their care through the Department of Veterans Affairs. The VA documents those visits in VA records but does not bill the Medicare Advantage plan, so the plan has no claim to link a chart review to, and VA care generally does not appear in the encounter data CMS uses to calculate risk scores.
Until now, plans have closed part of the gap by retrieving VA records and submitting supported diagnoses through chart review. Under the 2027 policy those diagnoses drop out, and roundtable panelists described the result for plans with large veteran populations: full financial responsibility for members whose documented chronic conditions no longer count toward their risk scores.
The options under discussion include data aggregation vendors experienced in reconciling VA records, and in-home assessments, which create a linked encounter. In-home assessments carry scrutiny of their own, from recent settlements involving diagnoses captured in the home to a provision in the proposed Save MEDICARE Act (H.R. 9544) to exclude health risk assessment diagnoses starting in 2028, so the approach holds up only when the assessment delivers real clinical care and the documentation supports every diagnosis on review.
Accuracy has to start at the point of care
Retrospective review is not going away, but regulators accept fewer retrospective findings each year, and the greater value now sits upstream, where a provider documents a condition fully during the visit. When documentation is right the first time, risk adjustment, quality measures, Star Ratings and care management all benefit from the same record.
Much of the work is closing the distance between clinical training and coding guidelines. As one panelist put the problem, many physicians learn in medical school a patient who has had cancer always has cancer, while coding guidelines and OIG audit methods treat a malignancy with no active disease and no current treatment as a history of cancer. Diabetes in remission, unconfirmed diagnoses carried forward on problem lists, and medication lists used as the only support for a diagnosis create the same kind of gap, and few providers learn the coding rules behind them in training.
How plans deliver the message matters as much as the message. Established physicians already carrying heavy administrative and EHR demands tend to respond better when documentation requirements come framed as regulatory changes rather than critiques of their care, while physicians new to practice need context on how coding guidelines apply to the patients in front of them. Many plans moved provider education to remote outreach during COVID-19, and the plans now putting educators back in clinics in person describe stronger relationships and more open conversations about documentation.
Coding teams face a version of the same problem internally. One calibration practice gives every coder the same de-identified encounter note, has each person code the note independently, and brings the team together to compare results, which surfaces differences in how coders read the guidelines before an auditor finds them.
AI belongs on the front end, with a human in the loop
Autonomous coding and natural language processing tools are moving fast, and a common risk is leadership treating them as a direct replacement for coding staff. Roundtable panelists expect OIG audit work examining codes generated by AI and NLP tools, and many want CMS guidance requiring a verified human-in-the-loop validation step, with every AI output accurate, defensible and traceable to the source record.
Ambient documentation shows why the human step matters. In one example a panelist shared, an AI scribe added two paragraphs of history the patient never said, and the error came out only because the physician reviewed the note with the patient’s caregiver before signing. Transcription tools also struggle across languages, regional dialects and accents, which makes clinician review part of the workflow rather than an extra step.
The strongest use cases sit before the claim goes out. A tool checking a documented chronic kidney disease stage against the patient’s eGFR results, for example, flags a mismatch while the provider still has the chance to add specificity, and leaves your QA team a clear trail to audit. Holding AI output to the same structured, continuous QA you apply to any vendor keeps the speed without giving up defensibility.
Silos cost plans money and cost members patience
Panelists called internal silos the single biggest source of wasted time and money they see. Quality teams, in-home assessment vendors, risk adjustment, case management and provider relations often work the same members and the same providers without coordinating, and the duplicated effort shows up in both budgets and member experience. Members receive disconnected calls, letters and portal messages from several teams, sometimes about issues already resolved with their physician, and many stop answering.
Roundtable panelists argued for a single point of contact for each member, one person coordinating screenings, assessments, care management and social needs. RISE looked at the leadership decisions behind bringing risk adjustment and quality improvement together earlier this month.
The runway into 2027
Diagnoses from 2026 dates of service set 2027 risk scores, so the unlinked chart review exclusion already applies to this year’s records, and the 2027 service year sets risk scores for 2028. Your team has the full 2027 service year to build the linked encounters, provider documentation habits and front-end validation the policy now rewards, and much of the work overlaps with the readiness RADV expansion already asks of you. Plans with large populations receiving care outside their networks, starting with members treated at the VA, have the most to gain from starting early.
Risk adjustment, compliance and audit readiness, and provider partnerships will be covered at RISE National 2027, March 16-18 at Caesars Forum in Las Vegas.