What Is a RADV Audit? Everything Medicare Advantage Plans Should Know

What Is RADV

A RADV audit is CMS's Risk Adjustment Data Validation program, through which the agency reviews Medicare Advantage plan medical records to confirm that diagnosis codes submitted for risk adjustment payments are supported by clinical documentation, with findings that can trigger financial recoveries extrapolated across the full contract population under rules finalized in the 2023 RADV final rule.

Risk Adjustment Data Validation (RADV) is CMS's audit program designed to verify that diagnosis codes submitted by Medicare Advantage plans for risk adjustment payment purposes are supported by medical record documentation. RADV ensures that CMS pays plans only for the clinical complexity their members' records substantiate.

CMS has the legal authority under the Medicare Modernization Act and subsequent regulations to conduct RADV audits on any MA contract. The program serves as the primary check against inaccurate or unsupported risk adjustment submissions that would result in improper payments.

• Statutory Authority: Section 1853(a)(1)(C)(ii) of the Social Security Act authorizes CMS to adjust MA payments based on the health status of enrollees. RADV is the enforcement mechanism ensuring those adjustments are accurate.
• Scope: RADV audits evaluate whether submitted HCC diagnoses are supported by medical records from face-to-face encounters with qualified providers during the applicable payment year. Every HCC contributing to a member's RAF score is subject to validation.
• Financial Impact: CMS estimates that approximately 9.5% of MA payments are improper due to unsupported diagnoses. With total MA payments exceeding $460 billion annually, RADV's financial stakes are measured in billions of dollars across the program.

How RADV Audits Work

RADV audits follow a structured methodology that CMS has refined over multiple audit cycles. Understanding the mechanics helps plans prepare effectively.

• Contract Selection: CMS selects MA contracts for RADV using a combination of random sampling and targeted selection. Targeted selection criteria include outlier risk score levels, rapid year-over-year RAF growth, high coding intensity relative to benchmarks, and patterns identified through data analytics. CMS does not publicly disclose its exact selection algorithm.
• Beneficiary Sampling: Once a contract is selected, CMS draws a statistically valid random sample of enrolled beneficiaries for audit. The sample size is designed to support extrapolation — typically 200 members per contract, though sample sizes can vary.
• HCC Identification: For each sampled beneficiary, CMS identifies all HCC diagnoses submitted for the applicable payment year. Every submitted HCC is subject to review — not just the highest-value ones.
• Medical Record Request: CMS requests the medical records supporting each submitted HCC from the plan. The plan must produce records from contracted providers within a specified timeline, typically 45-90 days. Records not produced within the deadline are automatically coded as unsupported.
• Independent Coding Review: CMS-contracted certified coders review the medical records and independently determine whether each submitted HCC is supported by the documentation. Coders apply strict criteria: face-to-face encounter, qualified provider, definitive diagnosis, clinical evidence, and current-year date of service.
• Error Rate Calculation: CMS calculates the error rate — the percentage of submitted HCCs that are not supported by documentation. This rate is calculated for the sampled beneficiaries and, under extrapolation, applied to the full contract population.

The RADV Audit Process

The RADV audit process spans multiple phases over several months, with defined touchpoints for plan engagement.

• Phase 1 — Notification: CMS notifies the selected MA organization that its contract has been chosen for RADV audit. The notification specifies the payment year under review, the beneficiary sample, and the timeline for medical record submission.
• Phase 2 — Record Collection: The plan initiates medical record retrieval from its provider network for every sampled beneficiary's HCC-supporting encounters. This phase tests the plan's chart retrieval infrastructure — plans without established retrieval agreements face significant operational challenges.
• Phase 3 — Record Submission: The plan submits collected medical records to CMS by the specified deadline. Incomplete submissions or records submitted after the deadline result in those HCCs being classified as unsupported.
• Phase 4 — Coding Review: CMS's contracted coding reviewers evaluate each submitted record against RADV documentation standards. Each HCC is classified as "supported" or "not supported." This review is independent — the plan's internal coding determinations are not considered.
• Phase 5 — Preliminary Results: CMS issues preliminary audit results to the plan, identifying which HCCs were found unsupported and the calculated error rate. The plan has an opportunity to respond, provide additional records, or dispute specific findings.
• Phase 6 — Final Determination: After considering the plan's response, CMS issues final audit results with the confirmed error rate. If extrapolation applies, CMS calculates the projected payment recovery amount based on the extrapolated error rate across the full contract population.
• Phase 7 — Payment Recovery: CMS recovers the determined overpayment amount. Plans may appeal through the CMS administrative appeals process, but the payment adjustment takes effect while appeals are pending unless a stay is granted.

RADV Extrapolation Explained

Extrapolation is the mechanism that transforms RADV from a sample-level audit into a contract-level financial event. The 2023 RADV final rule made extrapolation the defining factor in RADV's financial impact.

• What Extrapolation Does: CMS takes the error rate found in the audited sample and projects it across the plan's entire contract population. If 200 sampled members show a 12% unsupported HCC rate, CMS assumes that same 12% rate applies to all 50,000+ enrollees.
• The 2023 Final Rule: After years of industry debate, CMS finalized its RADV extrapolation methodology in January 2023. The rule confirmed that CMS will apply extrapolation to RADV findings for payment year 2018 and all subsequent payment years. Prior payment years use a non-extrapolated approach.
• FFS Adjuster Eliminated: CMS's final methodology does not include a Fee-for-Service (FFS) adjuster — a provision the industry had advocated for that would have offset RADV error rates against similar documentation error rates in Traditional Medicare. Without this adjustment, RADV error rates apply at face value.
• Financial Magnitude: Consider a plan with 60,000 members, an average RAF score of 1.10, and a county base rate of $950. If RADV finds a 10% unsupported HCC rate, extrapolation could translate to a payment recovery of $10-15 million or more, depending on the RAF impact of the unsupported conditions.
• Statistical Methodology: CMS uses a lower-bound confidence interval approach, meaning the extrapolated recovery amount uses a statistically conservative estimate. Even so, extrapolated amounts typically dwarf the direct financial impact of the sample findings alone by a factor of 100x or more.
• Strategic Implication: Extrapolation means every single HCC submission matters. A documentation failure for one member in the audit sample affects the payment calculation for the entire contract. This fundamentally changes the economics of compliance — investing in pre-submission validation for every HCC is now cost-justified by the extrapolation math alone.

Common RADV Findings

Published RADV results and OIG reports reveal consistent patterns in the types of HCCs that fail audit validation. These patterns are predictable and preventable.

• No Medical Record Available: The plan cannot produce a medical record for the sampled member's HCC. This occurs when chart retrieval processes fail, provider records are inaccessible, or the encounter documentation was never created. This is the most straightforward failure — and the most operationally preventable.
• Diagnosis Not Supported by Clinical Evidence: A medical record exists and contains the diagnosis, but the record lacks objective clinical evidence (lab values, exam findings, imaging results) supporting the condition. The diagnosis appears in the assessment but has no clinical foundation in the encounter note. This is the most common RADV finding, accounting for 35-40% of unsupported HCCs.
• Condition Not Addressed in Current Year: A chronic condition appears in the patient's medical history or problem list, but the encounter note for the payment year does not evaluate, manage, or address the condition. Problem list carry-forward without active encounter documentation does not meet RADV standards.
• Non-Qualifying Encounter Type: The HCC was submitted based on a non-qualifying encounter — lab results, radiology reports, or ancillary services without an associated face-to-face provider visit. Only encounters with qualified providers (MD, DO, PA, NP) meeting CMS-defined encounter criteria qualify for risk adjustment.
• Coding Specificity Mismatch: The ICD-10 code submitted requires a level of specificity that the documentation does not provide. The record supports a general diagnosis but not the specific manifestation, laterality, or severity that the submitted code implies.
• Signature and Authentication Issues: The encounter note is clinically complete but lacks a valid provider signature, contains an unsigned addendum, or has authentication that does not meet state or CMS requirements for electronic signatures.

How to Prepare for RADV

RADV preparation is not a project — it is an ongoing operational capability that must be embedded into risk adjustment workflows year-round. The comprehensive RADV audit checklist provides a detailed 10-item framework.

• Pre-Submission Validation: The highest-impact preparation step. Screen every HCC submission against RADV documentation standards before data reaches CMS. Automated scrubbing technology can flag unsupported diagnoses, specificity gaps, and non-qualifying encounters at the point of submission — eliminating the highest-risk HCCs before they enter the RADV audit universe.
• Chart Retrieval Readiness: Establish and maintain chart retrieval agreements with all contracted provider groups. Test retrieval timelines quarterly. Plans that cannot produce records within 45 days of an audit notification will fail those HCCs by default. Build retrieval infrastructure before you need it, not after.
• Internal Mock Audits: Conduct at least one full-scale mock RADV audit annually using CMS methodology: random member sample, independent coding review, error rate calculation, and extrapolation modeling. Mock audits quantify your actual exposure and identify specific remediation priorities.
• Provider Documentation Training: Train providers on the specific documentation elements RADV reviewers evaluate: face-to-face encounter evidence, definitive diagnosis language, clinical support indicators, treatment or monitoring plans, and proper signature authentication. Providers who understand RADV criteria document more defensibly.
• Technology Solutions: Deploy RADV scrubbing technology that automates compliance screening at scale. Manual chart-by-chart review is neither scalable nor sufficient for the volume of HCCs a typical MA plan submits annually. Technology addresses the gap between submission volume and review capacity.
• Continuous Compliance Culture: The organizations with the lowest RADV exposure treat audit readiness as an operational standard rather than a periodic initiative. Every submission is audit-ready by default. A risk adjustment analytics platform tracks compliance metrics in real time alongside financial and clinical metrics. Audit preparation is not a cost center — it is a revenue protection function.

Understanding how to reduce RADV audit exposure through systematic compliance measures is the difference between plans that view RADV as a crisis and plans that view it as a routine quality validation.