Quick Answer: Retrospective risk adjustment is the review of a Medicare Advantage member’s past medical record to confirm which diagnoses were truly documented for a payment year, then correct the codes submitted to CMS. The model has changed. A defensible program today does two-way coding: it adds supported diagnoses and removes unsupported ones. Add-only chart review, which finds new codes but ignores codes that should be deleted, is now a documented compliance risk under 2026 OIG guidance and recent DOJ enforcement. RAAPID’s Clinical AI Platform cuts review time to 8 to 12 minutes and reaches 92% out-of-box accuracy, with two-way coding built in.*
Key Takeaways
- Retrospective risk adjustment now means two-way coding: adding supported diagnoses and removing unsupported ones, not add-only chart mining.
- An OIG review found chart reviews drove an estimated $6.7 billion in Medicare Advantage payments for 2017, and over 99% of those reviews only added diagnoses instead of deleting them. [1]
- 2026 OIG guidance and a March 2026 DOJ settlement treat add-only review that ignores deletable codes as a compliance failure. [2][3]
- An OIG audit found 247 of 271 sampled enrollee-years carried unsupported high-risk diagnoses, a 91% error rate, often history-of conditions coded as active. [4]
- RAAPID’s two-way retrospective program reaches 92% out-of-box accuracy and over 98% after human quality review, with chart review time of 8 to 12 minutes and a 60 to 80% productivity gain for coding teams.*
The Old Retrospective Model Is Closing
Retrospective risk adjustment used to be a search for more codes. Find diagnoses in old charts, submit them, raise the score. That approach is ending. In 2026, the safest risk adjustment programs prove the codes they keep and delete the ones they cannot support.
The data shows how lopsided the old model was. OIG found that chart reviews accounted for an estimated $6.7 billion in Medicare Advantage payments for 2017, and more than 99% of those reviews added diagnoses rather than removing them. [1] That one-way pattern is now a regulatory red flag, not a best practice.
Two-way coding is the fix. Add-only review submits new diagnoses but never removes the unsupported ones a review surfaces. Regulators read that pattern as intent to inflate payments. A March 2026 Department of Justice settlement with Aetna resolved exactly this: a program that added codes but failed to delete unsupported ones the same reviews had already identified. [2]
This guide explains how retrospective risk adjustment works in practice, why add-only review is now a liability for health plans and providers, and how RAAPID’s two-way retrospective coding adds and removes HCC codes with a clear evidence trail behind every decision.
What Is Retrospective Risk Adjustment?
Retrospective risk adjustment is the systematic review of the medical record after patient encounters to confirm, correct, and document the diagnosis codes that reflect a member’s true health conditions for a given payment year. It looks backward at encounters that already happened.
Coding teams examine clinical documentation, lab results, specialist consultations, and medication lists to find chronic conditions that were missed or miscoded during initial claims processing. Every validated HCC code must be supported by MEAT documentation: evidence that the condition was Monitored, Evaluated, Assessed, or Treated during a qualified face-to-face encounter. [4]
Why it matters: the diagnoses health plans submit drive each member’s risk score, and those scores must survive a Risk Adjustment Data Validation (RADV) audit. A code without supporting clinical evidence is a code a plan may have to repay.
The work splits into two jobs. The first is finding real conditions that were documented but never coded. The second, the one most legacy programs skip, is finding coded conditions that the chart does not actually support. A defensible program does both, working to close gaps in documentation and capture missed conditions so the record supports appropriate reimbursement for the care delivered. For how diagnoses map to HCCs in the first place, see our guide to risk adjustment coding.
Why Retrospective Risk Adjustment Matters for Health Plans in 2026
Three forces are converging to make retrospective programs a regulatory priority rather than a revenue function for health plans and providers. The financial impact of getting risk adjustment wrong has never been higher.
Regulatory enforcement is accelerating. CMS confirmed in its January 2026 RADV memo that Payment Year 2020 audits began by February 2026, with variable sample sizes of 35 to 200 enrollees per contract and a restored five-month medical record submission window. [5] CMS is also adding an AI-powered coder support tool to streamline reviews, though final determinations stay with human certified medical coders. [5]
OIG guidance raised the bar. The February 2026 OIG Medicare Advantage Industry-Specific Compliance Program Guidance, the first major update of its kind since 1999, flags specific practices that draw federal scrutiny. [3] These include using chart reviews only to add diagnoses without removing unsupported ones, running health risk assessments that generate diagnoses unused in patient care, and prompting clinicians through EHR platforms to add risk-adjusting diagnoses that members do not have. [3]
Financial performance now depends on risk adjustment accuracy, not volume. OIG found that chart reviews drove an estimated $6.7 billion in Medicare Advantage payments for 2017, with over 99% adding diagnoses rather than deleting them. [1] Health plans and providers that want financial sustainability in value-based care have to prove their risk adjustment programs produce accurate documentation, not inflated risk scores. Patient outcomes suffer when the focus shifts from care quality to revenue capture.
Add-Only Chart Review Is Now a Compliance Risk
Add-only chart review submits new diagnosis codes to CMS but does not remove the unsupported diagnoses the same review identifies. It is a one-directional process, and 2026 guidance treats it as a red flag.
OIG’s February 2026 guidance flags chart reviews that add codes without removing unsupported ones as a suspect practice. [3] Failing to delete a code you know is unsupported counts as a regulatory failure, not an oversight. The guidance also tells health plans to review any software used in risk adjustment, including vendor tools, to confirm it is not designed mainly to increase risk scores without supporting clinical validity. [3]
The enforcement record backs this up. The March 2026 DOJ settlement with Aetna resolved allegations that a PY2015 add-only program submitted additional diagnoses but failed to withdraw unsupported diagnoses its own reviewers had flagged. [2] The lesson is blunt: if your review finds a code that should come out and you leave it in, that decision is now evidence.
Audit data shows how common unsupported diagnoses are. An OIG audit of one Medicare Advantage organization (report A-07-22-01207, March 2026) found 247 of 271 sampled enrollee-years carried unsupported high-risk diagnoses, a 91% error rate. [4] The most frequent error was a history-of condition coded as active, such as a past stroke coded as an acute stroke. Those are exactly the unsupported diagnoses a two-way review removes.
Autonomous Retrospective Risk Adjustment Solution
One platform. Every HCC validated. Revenue secured.
Two-Way Retrospective Coding: Add and Delete
Two-way retrospective coding both adds supported diagnoses and removes unsupported HCC codes in the same review, with MEAT evidence behind every decision. MEAT (Monitor, Evaluate, Assess, Treat) is the documentation convention that shows a diagnosis was actively managed during an encounter.
This is RAAPID’s position: deletion carries the same weight as addition. A program that only adds is a revenue engine. A program that adds and deletes is a clean record. Here is how the two compare.
| Dimension | Add-only chart review | Two-way retrospective coding |
| Supported codes added | Yes | Yes |
| Unsupported codes removed | No | Yes |
| Evidence trail per code | Often missing | MEAT-linked for every code |
| 2026 OIG view | Flagged as risky | Aligned with guidance |
| RADV audit posture | Exposed | Defensible |
Two-way coding is the practical form of defensible coding. Every diagnosis a plan keeps is encounter-linked and evidenced; every diagnosis it removes is documented with a reason. That is what makes the difference at audit time. For how this feeds an audit-ready risk score, see our work on building defensible RAF scores.
How the Retrospective Risk Adjustment Process Works
A defensible process follows structured steps that build an evidence trail for every HCC code. RAAPID’s risk adjustment solution runs this as decision support, not automation: the Neuro-Symbolic AI recommends, and certified coders decide.
- Strategic chart selection. Rather than reviewing every record, the platform uses predictive analytics to identify high-value charts: members with multiple chronic conditions, members on medications that suggest undocumented conditions, and members whose risk scores dropped unexpectedly. This focuses coding teams on the records most likely to hold missed diagnoses or codes that lack support.
- Clinical documentation review and HCC validation. Coders and the AI analyze the medical record, lab results, imaging reports, and unstructured clinical data to identify relevant diagnoses. Each suspected condition maps to a valid HCC at the highest ICD-10-CM specificity. Under the CMS-HCC V28 model in effect for payment year 2026, the set of hierarchical condition categories changed, so coders revalidate HCC codes against current mappings. Accurate HCC capture depends on coders who understand the clinical context behind each code.
- MEAT evidence capture and two-way review. Every diagnosis links to MEAT evidence from a qualified encounter, or it is flagged. Equally important, the review identifies unsupported HCC codes for deletion, not just correct codes to add. Accurate documentation requires both adds and deletes.
- Quality assurance and submission. Multi-level quality checks confirm every code has defensible evidence before data reaches CMS. This verifies face-to-face encounter compliance, confirms proper documentation of patient health status for each chronic condition, reconciles documentation across patient encounters when care spans a provider network, and gives each kept and deleted code an audit-ready trail.
The result is a single source of truth for member risk, built on encounter-linked documentation rather than guesswork.
Industry First RADV Audit Solution
AI-powered solution enables health plans to efficiently manage and streamline RADV audits
Prospective, Retrospective, and Concurrent Risk Adjustment
Health plans use three approaches across a complete risk adjustment strategy in value-based care. Understanding the difference helps healthcare organizations match the right method to the right point in the workflow.
- Prospective risk adjustment captures diagnoses before or during the visit, at the point of care, so real conditions are documented in the moment.
- Concurrent risk adjustment reviews charts during or shortly after a care episode, giving care teams a real-time safety net to correct issues fast.
- Retrospective risk adjustment reviews completed charts after encounters as the final quality and audit layer on every HCC code submitted to CMS.
The most effective risk adjustment programs integrate all three. Prospective coding sets the foundation during patient care; concurrent review provides a real-time check; retrospective coding confirms every code reflects patient health accurately and is properly documented. Retrospective protects and cleans the record, while prospective risk adjustment grows the future safely. Most healthcare organizations need both, and RAAPID’s business mix runs roughly 80% retrospective and 20% prospective today, which tracks where risk adjustment compliance pressure sits across healthcare organizations.
How AI-Powered Technology Transforms Retrospective Reviews
The limits of manual review are well documented. Complex charts take 30 to 45 minutes each, accuracy drops after long sessions, and coding teams burn out on repetitive work across unstructured clinical data. Improving accuracy through manual effort alone is no longer sustainable.
AI-powered clinical tools change the equation for plans and providers. Neuro-Symbolic AI, which combines large language models with structured clinical reasoning, addresses three failures that limited older technology.
First, it understands clinical context. Neuro-Symbolic AI recognizes that a metformin prescription paired with an elevated A1C points to active diabetes management. It separates “family history of diabetes” from “diabetes with peripheral neuropathy,” a distinction that directly changes HCC codes and risk scores.
Second, it provides explainable results. Every suggested diagnosis links to specific evidence in the patient’s chart, creating a transparent audit trail. With seamless EHR integration, these AI-powered tools connect to existing workflows and cut administrative burden on coding teams and healthcare providers.
Third, it supports two-way coding. RAAPID’s risk adjustment solution flags unsupported HCC codes for removal with the same rigor it uses to surface missed diagnoses. That reflects the outcome regulators expect: accurate data, not maximized risk scores.
Productivity and Accuracy You Can Defend
Defensible does not mean slow. RAAPID’s retrospective program reaches 92% out-of-box AI accuracy, independently validated, and over 98% coding accuracy after human quality review.* The 92% figure is raw AI output before review; the 98%+ figure is final quality accuracy with a coder in the loop. We keep those two numbers separate on purpose.
Review time runs 8 to 12 minutes per chart, and coding teams report a 60 to 80% productivity gain over manual work.* That speed comes from cutting search time, not scrutiny. Medical coders spend their hours judging evidence instead of hunting for it.
The compliance-first framing matters more than raw throughput. A faster add-only program just produces exposure faster. A faster two-way program produces a cleaner, more defensible record at scale for plans and providers. That is the outcome a CFO worried about RADV liability and a chief compliance officer worried about evidence trails both need.
Preparing for RADV Audits
CMS’s accelerated RADV strategy means health plans and providers should treat every review as potential audit evidence. The January 2026 memo confirmed PY2020 audits on a roughly quarterly cadence, with sample sizes of 35 to 200 enrollees and a five-month medical record window. [5] For Medicare Advantage plans, the window to get this right is open now.
To build risk adjustment programs that hold up, draw on these practices from OIG’s guidance: [3]
- Pair record review with data accuracy controls and filtering logic that flags anomalies in diagnosis data.
- Benchmark HCC prevalence rates across years to spot unusual coding patterns.
- Analyze provider coding intensity across the provider network and educate providers where patterns suggest overcoding.
- Make sure retrospective reviews generate both adds and deletes.
- Report unsupported codes to CMS and address overpayments under Medicare overpayment law (42 U.S.C. 1320a-7k). [3]
For complex chronic conditions that span multiple clinicians, the review has to reconcile documentation across providers so each member’s risk adjustment data is accurate and supported across the patient population.
What Risk Adjustment Leaders Are Asking Right Now
- What is two-way retrospective coding, and why does it matter in 2026?
- Is add-only chart review still compliant after the 2026 OIG guidance?
- How long does an AI-assisted chart review take per chart?
- What does accelerated CMS scrutiny mean for retrospective programs?
- How is retrospective different from prospective and concurrent risk adjustment?
- Does AI replace human coders in retrospective review?
Ready to Evaluate a Collaborative Path?
Over a Partnership Evaluation Call.
Frequently Asked Questions
Retrospective risk adjustment reviews a member’s past medical record to confirm which diagnoses were truly documented for a payment year. A defensible program adds supported codes and removes unsupported ones before submission to CMS, creating an evidence trail for every diagnosis it keeps or deletes. It supports fair reimbursement that reflects real patient complexity.
Concurrent risk adjustment happens during or shortly after a care episode, allowing faster correction of coding issues. Retrospective coding happens after the encounter is complete, serving as the final quality and audit check on every HCC code submitted to CMS. Many plans also use prospective coding at the point of care to cut downstream corrections.
See Defensible Retrospective Review in Action
For healthcare organizations, add-only chart mining is now a liability you can measure. Two-way retrospective coding is how you fix it: add what is supported, remove what is not, and keep an evidence trail for both. Book a RAAPID demo to see RAAPID’s risk adjustment solution apply two-way coding on your own charts.
Source
[5] Centers for Medicare & Medicaid Services, RADV Health Plan Management System memo, January 27, 2026.
*Internal RAAPID benchmarks from client engagements. Accuracy and productivity figures vary by client, data quality, and chart mix.
Wynda Clayton, MS, RHIT, CRC
Director of Risk Adjustment Coding & Compliance, RAAPID
Wynda Clayton, MS, RHIT, CRC, is Director of Risk Adjustment Coding & Compliance at RAAPID and a former CMS RADV auditor. She advises Medicare Advantage health plans on defensible coding, two-way retrospective coding, and audit readiness.
