The Signal Before the Sample
Before CMS selects a single enrollee for RADV medical record review, the agency’s AI has already analyzed the plan’s entire submitted data set. Population-level pattern detection runs continuously against the data CMS receives from all MA contracts. The patterns it identifies influence which contracts receive larger sample sizes, which diagnosis categories are oversampled, and potentially which payment years are prioritized for review.
Understanding what these algorithms look for gives plans the ability to monitor their own data for the same signals. The goal isn’t to game the system. It’s to identify and fix the coding patterns that indicate underlying documentation problems before those patterns attract enhanced scrutiny.
Four Patterns That Attract Attention
The first is coding intensity divergence. CMS calculates a coding intensity adjustment at the national level, but the underlying analytics examine individual contract data. When a plan’s risk scores rise year over year without corresponding increases in hospitalizations, specialist referrals, or treatment costs, the divergence suggests codes are accumulating from coding activity rather than clinical complexity. This is the most watched population-level signal in MA risk adjustment.
The second is category concentration. Plans that submit a disproportionate share of their codes in high-value HCC categories relative to regional or national clinical prevalence display a distribution that suggests targeted coding rather than organic documentation. The concentration is especially visible in acute categories (stroke, MI) where clinical prevalence is relatively stable but coding rates vary widely across plans.
The third is add-to-delete asymmetry. Plans that submit thousands of supplemental additions and zero deletions across multiple review cycles produce a mathematical signature of one-directional coding. CMS can detect this without examining a single chart. The asymmetry itself tells a story about program design.
The fourth is unlinked diagnosis volume. CMS data shows that 85% of 88.8 million chart review diagnoses submitted in 2023 couldn’t be matched to encounters. Plans with high unlinked volumes are generating risk score credit from diagnoses that CMS has already decided to exclude starting in payment year 2027. The volume itself signals reliance on a coding pathway CMS is shutting down.
Running the Same Analytics Internally
Plans can compute each of these metrics from their own submitted data. Calculate your coding intensity trend over three to five years against your clinical utilization trends. Map your HCC category distribution against CMS’s published prevalence benchmarks. Compute your add-to-delete ratio from retrospective review data. Quantify your unlinked CRR volume as a percentage of total supplemental submissions.
Where the analytics reveal concerning patterns, the fix is specific. Coding intensity divergence indicates accumulated unsupported codes that need cleaning through two-way review. Category concentration calls for enhanced validation in the concentrated categories. Zero deletion rates require methodology restructuring. High unlinked volumes require encounter linkage verification before submission.
Seeing What CMS Sees First
CMS’s AI runs continuously. It doesn’t wait for audit cycles. Plans that run the same analytics against their own data see what CMS sees, with one critical advantage: they can act on it before CMS does. Fixing a pattern before radv audits target it is remediation. Fixing it after is defense. The cost difference between those two positions grows with every audit cycle.
