The Rules Changed. Most Programs Didn’t.
Risk adjustment coding in Medicare Advantage is supposed to reflect real patient complexity. CMS uses submitted diagnosis codes to calculate risk scores that determine how much a plan gets paid for each member. Sicker members generate higher payments. The system depends on accurate coding to function correctly.
For years, the industry treated this system as a revenue optimization exercise. Find more codes. Submit more diagnoses. Increase risk scores. Collect larger payments. Vendors sold chart review programs based on how many HCCs they could add. Plans measured success by RAF score uplift. The compliance implications were an afterthought.
That approach hit a wall. The DOJ collected over $670 million from Kaiser and Aetna in settlements tied to inflated risk adjustment submissions. CMS-HCC V28, fully implemented as of January 1, 2026, eliminated 2,000+ ICD-10 codes and restructured the model to reduce the impact of coding intensity. OIG audits routinely find error rates above 80% in sampled records. The enforcement environment has fundamentally changed, but many coding programs haven’t caught up.
What Accurate Coding Actually Requires Now
Under the current regulatory framework, every submitted diagnosis needs three things: encounter linkage (the diagnosis was documented during a clinical visit), clinical evidence (the chart note shows active management through MEAT criteria: Monitoring, Evaluation, Assessment, or Treatment), and explainability (there’s a documented trail showing why the code was assigned and what evidence supports it).
That’s a higher bar than most programs were built to clear. Traditional chart review programs focused on finding diagnoses in historical records and submitting them. Whether the documentation met evidentiary standards was checked inconsistently, if at all. Whether the code could be explained and defended in an audit was someone else’s problem.
The OIG’s March 2026 audit of BCBS Alabama illustrates what happens when programs fall short. Out of 271 sampled enrollee-years, 247 had unsupported diagnosis codes. Acute stroke and myocardial infarction categories had 100% error rates. The most common failure: history-of conditions coded as active diagnoses without any evidence of current clinical management. These weren’t coding mistakes. They were documentation failures that the coding program didn’t catch.
The Technology Shift
AI-assisted coding tools have changed what’s possible. Explainable AI can scan clinical notes and map documentation to MEAT criteria, flagging where evidence is present, absent, or ambiguous for each suspected HCC. Coders see the specific lines in the note that support or fail to support a diagnosis. Audit simulation tools score defensibility before submission and predict which codes are at risk.
The critical requirement is explainability. Systems that output “submit this code” without showing the underlying clinical evidence are creating new risk rather than reducing it. When CMS audits a submitted diagnosis, they ask for proof. The AI needs to have assembled that proof before the code was ever submitted, and the coder needs to have validated it.
Two-way review capability is equally important. Programs that only add codes replicate the exact pattern that generated the Kaiser and Aetna settlements. AI that identifies both missed diagnoses and unsupported ones gives coding teams the ability to clean their submissions before regulators do it for them.
What Success Looks Like in 2026
The metric that matters has shifted from codes added to codes defended. Plans that measure their risk adjustment programs on accuracy, documentation quality, and audit readiness are aligned with where CMS is heading. Plans still measuring on volume and RAF uplift are running programs designed for a regulatory environment that no longer exists.
Effective Risk Adjustment Coding in 2026 means every submitted diagnosis is encounter-linked, clinically evidenced through MEAT criteria, and supported by an explainable audit trail. It means running two-way retrospective reviews that add and remove. It means deploying prospective tools that generate defensible codes at the point of care. The plans that build this way are the ones that will survive the next five years of regulatory pressure without writing settlement checks.
ALSO READ: GlobeInsightBlog: Global Trends, Technology, Business & Future Insights Explained
