Medicare Advantage has a problem that its architects recognized from the beginning. If the federal government pays a private insurer the same flat monthly amount for every Medicare enrollee — healthy or sick, 65 or 85, routine checkups or advanced cancer — the insurer has a powerful financial incentive to enroll only the healthiest beneficiaries and avoid the sickest. The healthy enrollee generates margin. The sick one generates losses. Without some mechanism to compensate insurers for taking on sicker beneficiaries, the program would collapse into a race to enroll the healthy and exclude everyone who actually needs care.
Risk adjustment was designed to solve that problem. The federal government adjusts the monthly capitated payment upward for enrollees with more documented health conditions — paying more for sicker beneficiaries and less for healthier ones — to neutralize the financial incentive to cherry-pick. The mechanism is sound in principle. It addresses a real problem in a logical way.
In practice, risk adjustment has been systematically exploited to generate tens of billions of dollars in annual overpayments. The reason is a design vulnerability that was either not anticipated or not adequately corrected: the payment system rewards documented diagnoses regardless of whether those diagnoses resulted in any treatment. An insurer that adds more diagnosis codes to an enrollee’s record receives a higher payment. An insurer that treats more of its enrollees’ conditions does not receive any additional payment for doing so. The incentive is to diagnose — specifically, to document diagnoses — not to treat.
That distinction between rewarding documentation and rewarding care is the origin of the $84 billion annual overpayment the Medicare Payment Advisory Commission reported to Congress in March 2025. Article 05 documents the mechanisms by which that overpayment is generated. This article documents the payment architecture that makes it possible.
How the HCC Model Works
The risk adjustment system CMS uses for Medicare Advantage is built on a coding framework called the Hierarchical Condition Category model — HCC for short. The model translates patient diagnoses into a numerical risk score, and that risk score into a payment multiplier applied to the base capitated rate.
The process works in layers. First, diagnoses are expressed as ICD-10-CM codes — the standardized classification system used throughout American medicine to record conditions. Not all diagnosis codes carry risk adjustment value; only those associated with conditions that statistically predict higher future healthcare costs are mapped into the HCC framework. Under the current model — Version 28, fully operative as of January 1, 2026 — 7,770 ICD-10-CM codes map to 115 HCC categories. Conditions within the same clinical family are grouped hierarchically, so that only the most severe condition in a given hierarchy counts toward the risk score. A patient with both diabetes and severe diabetic complications counts the complications, not both conditions separately.
Each HCC category is assigned a relative weight — a coefficient that represents its expected contribution to healthcare costs. Those weights are added together, combined with a demographic score based on age, sex, disability status, and living situation, to produce a Risk Adjustment Factor score, known as the RAF score. An RAF score of 1.00 means the enrollee is expected to use average Medicare resources. A score of 1.50 means the enrollee is expected to cost 50 percent more than average. CMS multiplies the RAF score against the base capitated rate to calculate the monthly payment the insurer receives for that enrollee.
The math creates a direct and powerful financial incentive: every additional HCC code that can be legitimately documented for an enrollee increases the RAF score, which increases the monthly payment, for every month of the year. A single additional HCC code worth a relative weight of 0.30 applied to an enrollee with a base capitated rate of $1,000 per month generates $300 per month in additional government payment — $3,600 per year — whether or not the documented condition was ever treated.
The Data Submission Structure
The RAF score that determines what the government pays for an enrollee is based on diagnoses submitted to CMS by the Medicare Advantage plan. The submission covers diagnoses from the prior calendar year — a plan submits 2025 diagnoses in early 2026, and those diagnoses determine payment rates for 2026.
The diagnoses can come from two sources. The first is the treating provider — the physician, specialist, or hospital clinician who actually examined and treated the beneficiary and documented conditions in the medical record as part of delivering care. The second is the MA plan itself, through insurer-initiated processes that review medical records or conduct assessments specifically to identify diagnoses for risk adjustment submission.
OIG documented the consequences of this two-source structure in a 2020 report that examined the highest-value risk adjustment diagnoses — the conditions that generate the largest payment increases when documented. The finding: 99 percent of the highest-value risk adjustment diagnoses submitted by Medicare Advantage plans were submitted by the plan rather than by the treating provider. The conditions that produce the largest payments were documented not by the physician who examined and treated the patient, but by processes the insurer initiated specifically to find and document those conditions.
That finding is the bridge to Article 05, which documents the specific mechanisms — retrospective chart reviews, in-home health assessments, chart review addenda — through which MA plans generate insurer-initiated diagnoses at scale. What the finding establishes here, at the level of the payment architecture, is that the risk adjustment system as operated does not primarily reward insurers for enrolling sicker beneficiaries and managing their care. It rewards insurers for finding and documenting conditions in medical records, whether or not those conditions were treated.
The Coding Intensity Gap
CMS has long recognized that MA plans document diagnoses at higher rates than traditional Medicare providers document for the same conditions in the same population. This phenomenon — MA plans generating higher average risk scores than the health status of their enrollees would justify based on traditional Medicare data — is called coding intensity.
The coding intensity gap has been documented consistently for over a decade. MA enrollees have higher average RAF scores than comparable traditional Medicare beneficiaries not because they are sicker, but because MA plans generate more diagnosis codes per beneficiary through the insurer-initiated documentation processes that treating providers do not replicate.
CMS applies a mandatory minimum coding intensity adjustment — a downward adjustment to MA risk scores intended to account for the documented gap between MA and traditional Medicare coding patterns. For 2026, CMS applies a 5.9 percent minimum adjustment, which its own analysis estimates partially but not fully offsets the coding intensity gap. MedPAC’s March 2025 analysis calculated that coding intensity alone accounts for $40 billion of the $84 billion MA overpayment — meaning the gap between what MA plans are paid and what covering the same beneficiaries under traditional Medicare would cost is roughly half attributable to the diagnosis inflation the adjustment is meant to correct, and still substantially uncorrected after the adjustment is applied.
The Recalibration Problem
CMS has attempted repeatedly to recalibrate the risk adjustment model to more accurately reflect actual beneficiary health status and reduce the overpayment. The industry has resisted each attempt through litigation and lobbying, and the political economy of recalibration has systematically favored the industry’s position.
The most recent significant recalibration is the transition from HCC Version 24 — which governed MA payments for roughly a decade — to Version 28, fully implemented in 2026. V28 reduces the number of valid ICD-10-CM diagnosis codes mapped to HCC categories from 9,797 to 7,770, eliminates some conditions from the risk adjustment model that the industry had used to inflate scores, and is projected to reduce average MA risk scores by approximately 3.12 percent. The industry challenged the V28 transition through lobbying and regulatory comment, arguing that the reduction in valid diagnosis codes would reduce payments to plans serving complex patients. CMS phased the transition over three years — one-third in 2024, two-thirds in 2025, fully in 2026 — rather than implementing it immediately, in part to manage the political resistance to rapid payment reduction.
The Risk Adjustment Data Validation audit — RADV — is CMS’s primary tool for verifying that the diagnoses MA plans submit for payment are actually supported by medical records. RADV audits compare submitted diagnosis codes against the underlying medical record documentation to identify diagnoses that were billed without clinical support. Early RADV audit results found that average overpayment rates for audited plans were well over 10 percent of risk-adjusted payments — meaning more than one in ten dollars paid based on diagnosis coding was not supported by the medical record.
The RADV audit program has been limited in scope and slow in implementation. The methodology for extrapolating audit findings to recover overpayments was contested in litigation for years. The insurance industry challenged CMS’s authority to recover overpayments beyond the specific contracts audited, and the litigation delayed large-scale recovery. The reform proposals relevant to RADV are addressed in Article 12; what matters here is that the primary audit mechanism for correcting the overpayment has operated at a fraction of the scale needed to recover documented losses.
What Accurate Risk Adjustment Would Produce
The purpose of risk adjustment is to pay MA plans based on the actual health status of their enrollees — compensating for genuinely sicker beneficiaries while not rewarding diagnosis inflation. If the risk adjustment system functioned as designed, MA payments would reflect actual beneficiary health status rather than the insurer-documented diagnoses that currently drive them.
MedPAC has consistently found that MA plans are paid more per enrollee than traditional Medicare would cost for the same beneficiaries. The overpayment has two components: the benchmark rate structure established in 2003 that sets MA payment rates above projected traditional Medicare costs even before diagnosis coding is applied (documented in Article 02), and the coding intensity gap that inflates those already-above-market rates further through insurer-initiated diagnosis documentation.
Accurate risk adjustment — risk scores that reflect actual beneficiary health status rather than MA plan documentation practices — would substantially reduce the second component of the overpayment. It would not eliminate the first component, which is a function of the benchmark rate structure rather than the coding model. Correcting both components simultaneously is the subject of the payment reform proposals in Article 12.
What accurate risk adjustment would not do is change the fundamental structure of the payment system. The HCC model would still reward documented diagnoses. The insurer-initiated documentation processes that generate the highest-value codes would still be financially rational. The incentive to diagnose rather than treat is built into the architecture of a payment system that measures health status through diagnosis codes rather than through health outcomes or care delivered. Recalibrating the model reduces the magnitude of the distortion; it does not change the direction of the incentive.
The Structural Conclusion
Risk adjustment was designed to solve one problem — preventing insurers from avoiding sick beneficiaries — and has created another: a payment structure that rewards the documentation of diagnoses over the delivery of care. The design flaw is not subtle. The financial incentive it creates is not subtle. And the evidence that the incentive has been acted on at scale — documented by MedPAC, GAO, OIG, and the Department of Justice — is not ambiguous.
The $84 billion annual overpayment that this payment architecture produces is examined in full in the next article. What this article establishes is the architecture itself: a system in which the government pays private insurers based on what conditions they document in their enrollees’ medical records, in which the highest-value conditions are documented almost entirely by insurer-initiated processes rather than by treating providers, and in which three decades of recalibration attempts have reduced but not eliminated the gap between what MA costs and what covering the same beneficiaries under traditional Medicare would cost.
The insurer’s financial interest is in maximizing the RAF score. The beneficiary’s financial interest is in receiving care. The payment system rewards one of those interests directly and the other not at all.
The complete Medicare Advantage series
01 — What Medicare Advantage Actually Is — and How It Replaced Traditional Medicare
02 — The Political History: How Private Insurers Got Into Medicare
03 — The Marketing Machine: How Enrollment Works and Who It Targets
04 — Risk Adjustment: The Payment System That Rewards Diagnosis, Not Treatment
05 — The $84 Billion Overpayment: How Upcoding Works at Scale
06 — The Profit Extraction Model: What Insurers Take Before Care Is Delivered
07 — Prior Authorization in Medicare Advantage: What OIG Found
08 — The Denial and Appeals Record: What Happens When Enrollees Push Back
09 — Network Adequacy and the Coverage Gap
10 — The Extra Benefits Myth: Dental, Vision, and What the Fine Print Says
11 — When Medicare Advantage Fails: Disenrollment at the End of Life
12 — The Reform Proposals: From Audit Reform to Elimination
13 — What the Evidence Resolves — and What It Doesn’t
This article was researched and drafted with AI assistance under human review. See our full AI and editorial practices.