In May 2026, the HHS Office of Inspector General published a nationwide Medicare Advantage audit of high-risk acute stroke diagnosis codes. OIG found that all 97 sampled Acute Stroke HCCs were not validated: 93 medical records did not support the diagnosis, and 4 records could not be located. OIG estimated $461.96 million in potential net overpayments for payment year 2021. The high-risk pattern was straightforward: selected acute stroke diagnosis codes appeared on physician data records, but no corresponding acute stroke diagnosis appeared on an inpatient or outpatient hospital data record in the same service year.
What is the acute stroke coding error in Medicare Advantage?
The acute stroke coding error in Medicare Advantage occurs when acute stroke diagnosis codes (ICD-10 I63.x) are submitted on physician data records for patients whose stroke occurred in a prior year and whose current documentation supports either a personal history of stroke code (Z86.73) or, when residual deficits are documented, an appropriate sequelae code (I69.x). Acute stroke generally requires emergency department or inpatient hospital evaluation and should not appear as a stand-alone physician-office diagnosis without corresponding hospital evidence. OIG sampled 97 enrollees nationally and found that none of the sampled Acute Stroke HCCs were validated by the associated medical records.
The error has a specific mechanical cause that explains why it persists across so many plans and so many years: the EMR problem list. When a patient has an acute stroke, that diagnosis (coded as I63.x) gets entered into the medical record. It is correct at the time. But the acute event resolves. The patient is discharged, recovers, and begins years of follow-up visits. The stroke code on the problem list does not automatically update to reflect that change. At each subsequent office visit, the provider documents “CVA” or “cerebrovascular accident” on the encounter note. The coder, or the chart review vendor’s algorithm, sees the problem list, sees the HCC-eligible code, and submits it.
The result is an acute stroke code on a physician data record with no corresponding acute stroke diagnosis on an inpatient or outpatient hospital data record in the same service year. That is the pattern OIG identified. It is not a clinical gray area. According to the American Academy of Professional Coders, “odds are you will never code an acute stroke in a physician’s office.” Acute stroke generally requires hospital-level evaluation, diagnostic workup, and contemporaneous documentation. A physician-office record that carries an acute stroke code without corresponding hospital evidence is a clear audit red flag.
Three distinct factors drive the error at scale. First, copy-forward: EHR systems carry active problem list diagnoses from visit to visit, and providers may not update the code when the patient’s status changes from acute to history. Second, problem list auto-pull: EHR platforms often populate the assessment field with the full problem list, presenting outdated codes as current diagnoses without prompting validation. Third, retrospective chart review: vendors surfacing HCC suspects algorithmically from historical data will find the acute stroke code in the record and flag it without verifying whether the current clinical picture supports it. As AAPC has noted, risk adjustment coding decisions should not be driven by what is on the problem list. They need to reflect what the provider actually assessed, monitored, evaluated, or treated during the encounter.
There are three coding paths after a stroke event, and each has different risk adjustment consequences. Getting this distinction right is both a compliance requirement and, in some cases, an opportunity to capture HCC value that is actually more defensible than the incorrect code.
Acute stroke (I63.x) applies during the acute event: the emergency department evaluation, hospitalization, diagnostic imaging, and contemporaneous acute-stroke workup. Selected acute stroke codes map to the Ischemic or Unspecified Stroke HCC and generate risk adjustment payment. They are almost never the correct codes on a later routine physician-office record without corresponding hospital evidence.
Stroke sequelae (I69.x) applies when the record documents ongoing residual effects after the acute event, such as hemiplegia, hemiparesis, monoplegia, aphasia, dysphagia, or cognitive impairment. Some documented stroke sequelae may support HCC capture under the applicable CMS model, depending on the specific code and model year. A patient with documented residual hemiparesis, for example, may support a more accurate and more defensible risk-adjusting code than an unsupported I63.x acute stroke code. That makes the acute stroke error a double failure: it creates compliance exposure while potentially obscuring the patient’s actual clinical complexity.
Personal history of stroke (Z86.73) applies to patients who have recovered from a stroke with no current neurological deficits. It carries no HCC value. It is the correct code when the current record reflects prior stroke history without documented residual deficits. In the OIG audit, the submitted acute stroke codes were not validated by the associated medical records.
The practical implication: when auditing acute stroke submissions, plans shouldn’t just ask whether the I63.x code is wrong. They should ask what the current documentation actually supports. In some cases, the correct code is Z86.73, which generates no HCC payment and may require a deletion of the unsupported acute stroke diagnosis. In others, the record may support an appropriate I69.x sequelae code, which can generate defensible risk adjustment value when the residual deficit is documented, assessed, and supported..
The $462 million figure made headlines, but the underlying pattern has been documented for years. OIG’s 2020 report A-07-17-01176 found just over $14.4 million in inaccurate payments tied to a specific transfer scenario: beneficiaries who were covered under traditional Medicare in one year and enrolled in Medicare Advantage the following year. OIG focused on selected acute stroke diagnosis codes reported on one physician’s claim without being reported on a corresponding inpatient claim. That report recommended physician education and policies and procedures to evaluate whether acute stroke diagnosis codes complied with Federal requirements.
What followed was a steady escalation. OIG conducted 34 individual plan-level audits of the same pattern. Across those audits, 1,146 of 1,185 sampled acute stroke diagnoses were not supported by medical records, a 96.7% error rate. A 2025 audit roundup covering three MA plans found a combined $8.4 million in potential net overpayments for high-risk diagnosis codes including acute stroke.
In December 2025, OIG published the results of its audit of a Humana Medicare Advantage contract in Louisiana, finding 218 of 240 sampled diagnoses unsupported, with the extrapolated overpayment for that single contract exceeding $10 million. The pattern was consistent: high-risk diagnosis codes on physician records, including acute stroke, without encounter-linked documentation.
In December 2023, OIG published a self-audit toolkit with SQL code that MA organizations could adapt to identify high-risk diagnosis-code patterns in their own data. Acute stroke was one of the high-risk categories included in the toolkit. In the 2026 nationwide acute stroke audit, OIG reported that 12 of the 63 MA organizations associated with the sampled enrollees had begun using the toolkit to focus on acute stroke diagnoses.
Then came the May 2026 nationwide report: 240,401 enrollees in the high-risk sampling frame, 97 sampled, no sampled Acute Stroke HCCs validated, and $462 million in estimated potential net overpayments. This was not a discovery. It was the escalation of a documented pattern that plans had ample warning to address.
CMS’s detection methodology is structurally simple, and plans can replicate it exactly. The signal is the combination of two data points CMS has always seen simultaneously: a physician data record with an acute stroke code, and the absence of an acute stroke diagnosis on any inpatient or outpatient hospital facility record in the same service year. When those two conditions are present, the pattern is algorithmically visible without opening a single chart.
CMS is now expanding RADV at a scale it previously had not applied. As part of that expansion, CMS announced that newly initiated audits will include all eligible MA contracts for each payment year, supported by enhanced technology and a medical-coder workforce increasing from roughly 40 reviewers to approximately 2,000 by September 1, 2025. The acute stroke pattern — physician data record only, with no corresponding acute stroke diagnosis on an inpatient or outpatient hospital data record in the same service year — is purpose-built to surface in that kind of analysis. It requires no clinical judgment to flag. A query returns it.
For RADV readiness, acute stroke should be treated as a priority risk category. Across OIG’s 34 prior MA audits of this pattern, 1,146 of 1,185 sampled acute stroke diagnoses were not supported by medical records. Plans with acute stroke codes on physician data records and no corresponding hospital evidence in the same service year should assume those codes are highly exposed if reviewed.
RADV Audits Are Changing the Rules for Every Medicare Advantage Plan covers the full scope of the audit expansion: what the new scale means, how HCC selection works, and what the review process looks like for plans in the audit cycle.
The OIG published the methodology. It published the SQL. Plans that have not run this analysis on their own data have not done so by choice, not by ignorance of the method. Here is how to do it.
Step 1: Pull all physician data records with selected ICD-10 I63.x acute stroke codes. Pull all physician or professional service records with an acute stroke diagnosis for service years 2020 forward. Keep inpatient and outpatient hospital data available as the comparison set; those records are necessary to determine whether there is corresponding hospital evidence in the same service year. What remains after that cross-reference is the universe that needs review.
Step 2: Cross-reference against inpatient and outpatient hospital data records in the same service year. For each enrollee in the pull, query whether they also have an acute stroke diagnosis on an inpatient or outpatient hospital data record in the same service year. Enrollees with corresponding hospital evidence may be defensible and should still be reviewed for documentation support. Enrollees with a physician-data-record-only acute stroke diagnosis fall into OIG’s high-risk category.
Step 3: Flag and pull medical records for physician-only submissions. For flagged enrollees, pull the actual medical records and conduct a coding review against a specific question: does the record support an acute event at the time of the physician’s service, confirmed by diagnostic studies? Or does the note reflect a patient presenting for routine follow-up whose clinical status is post-stroke?
Step 4: Apply the correct code based on current clinical status. The answer to Step 3 drives three possible outcomes: I63.x if the record genuinely supports an acute stroke event with contemporaneous hospital-level evidence; an appropriate I69.x sequelae code if the record documents ongoing residual neurological deficits, such as hemiplegia, hemiparesis, monoplegia, aphasia, or dysphagia; or Z86.73 if the patient has recovered without current documented deficits and the visit reflects prior stroke history. The code follows the clinical reality in the note, not what is on the problem list.
Step 5: Submit deletions or corrections for codes that cannot be supported. CMS guidance states that when a plan sponsor determines through internal review that submitted diagnosis codes do not meet risk adjustment submission requirements, the sponsor is responsible for deleting those codes as soon as possible. Identifying the error and correcting it proactively is the compliance-appropriate response, and it places the plan in a materially better position under RADV than being found first.
Step 6: Use the OIG Toolkit as an ongoing detection framework. The December 2023 OIG Toolkit contains the SQL code OIG used to run this analysis against CMS’s data. MA organizations can adapt that code to query their own internal data systems and run the same analysis on a recurring basis, not as a one-time remediation but as a standing compliance control. Preparing for RADV Audits: Building a Defensible Documentation Strategy covers how internal audit frameworks like this fit into the broader RADV readiness posture.
The acute stroke error frequently originates upstream of the plan itself. Retrospective chart review vendors surfacing HCC suspects algorithmically will find old acute stroke codes in the medical record and flag them as valid suspects, because the code is there and it maps to an HCC. Without a validation step that asks whether the current physician data record is the only record in the service year, and whether the clinical note actually supports an acute event, the incorrect code moves through the pipeline.
This is not a hypothetical risk, but it should not be treated as a vendor-only problem. Plans relying on vendor-sourced suspects need vendor contracts and QA standards that explicitly address high-risk code categories, with acute stroke first among them, and require encounter-level clinical validation before submission, not after.
The CY2027 unlinked chart review exclusion adds a second layer of exposure for plans in this situation. Starting with CY2027 risk score calculation, CMS is excluding diagnoses from unlinked chart review records, with an exception for beneficiaries who switch from one MA organization to another. Plans with acute stroke codes from unlinked retrospective sources may be carrying two separate vulnerabilities simultaneously: a documentation accuracy problem and a policy exclusion. Those are not the same fix. $462 Million in Unsupported Payments and PY2021 RADV on the Calendar covers the broader RADV enforcement context this pattern sits inside.
Building an Audit-Ready Documentation Culture: From Reactive to Proactive addresses what it means to build a compliance posture that finds these gaps systematically rather than discovering them in an audit finding.
HCC Validator is built to flag exactly this kind of pattern before submission. It identifies OIG high-risk code submissions, including acute stroke codes on physician-only records, validates that diagnoses are encounter-supported, and builds the structured evidence trail that RADV auditors require. The goal is for the Validator to find what an auditor would find, before the auditor does. For plans running the self-audit described above, HCC Validator is what converts a one-time remediation into a standing operational control.
HCC Assistant addresses the problem at its source: the encounter. When a provider is seeing a patient with a prior stroke history, HCC Assistant surfaces the coding distinction in real time. Is this an acute event? What does the documentation support? Is there a documented neurological deficit that warrants an I69 sequelae code? Catching the specificity error at the point of care, before the code enters the claim, is where the fix is most durable. It is also where the clinical reality and the coding decision are closest together.
AI Chart Assistant reduces the chance the provider walks into the encounter without understanding the patient’s current status. It synthesizes chart history into a pre-visit clinical overview that surfaces the stroke documentation trajectory: when the acute event occurred, what deficits were documented, how the patient’s status has changed. That preparation is what enables the provider to make an accurate coding decision instead of relying on what an outdated problem list says.
Quality Assistant closes care gaps at the encounter level, aligning risk adjustment and quality through the same clinical interaction. For patients with a prior stroke, it supports the documentation of current functional status, including depression screening, medication reconciliation, and functional assessment, producing the clinical picture the correct code requires. Accurate coding flows from complete documentation. Quality Assistant supports the documentation.
Plans that run the acute stroke self-audit using the OIG methodology should expect to find a subset of physician-record-only codes that don’t hold up. Some will require deletions. The code is simply wrong, the patient recovered, and Z86.73 was the correct path. Others will reveal something different: patients with documented residual deficits who should have been coded with I69.x sequelae codes all along, codes that are both more accurate and, in many cases, more risk-adjusting than the I63.x that was submitted.
Plans that implement ongoing controls, including pre-submission validation against OIG high-risk categories, second-level coding review, and point-of-care specificity support, should expect fewer acute stroke errors entering the submission pipeline, reduced RADV audit risk in the specific category OIG has been targeting for a decade, and a defensible posture if audit selection occurs. Building a Defensible Risk Adjustment Program in a Post-V28 Environment covers what that long-term program posture looks like beyond this specific code category.
What is the acute stroke coding error in Medicare Advantage? The acute stroke coding error occurs when acute stroke diagnosis codes (ICD-10 I63.x) are submitted on physician data records for patients whose stroke occurred in a prior year, instead of the correct personal-history code (Z86.73) or, when residual deficits are documented, an appropriate sequelae code (I69.x). Acute stroke generally requires emergency department or inpatient hospital evaluation and is almost never appropriate as a stand-alone outpatient physician-office diagnosis without corresponding hospital evidence. OIG sampled 97 MA enrollees nationally and found that none of the sampled Acute Stroke HCCs were validated, estimating $462 million in potential net potential net overpayments for payment year 2021.
Why do providers and coders submit acute stroke codes in physician office settings? The most common driver is EMR copy-forward: an old acute stroke code on the problem list gets carried into each subsequent visit without being updated to reflect the patient’s current status. Chart review vendors also surface outdated HCC suspects without validating whether the code is appropriate in the current year. In both cases, the acute code moves through to submission even though the clinical reality changed years earlier.
What is the difference between acute stroke (I63.x), sequelae (I69.x), and history of stroke (Z86.73) in HCC coding? Acute stroke (I63.x) applies during the acute event and generally requires emergency department or inpatient hospital evaluation, diagnostic workup, and contemporaneous documentation. Sequelae (I69.x) applies when the record documents ongoing residual neurological deficits, such as hemiplegia, hemiparesis, aphasia, or dysphagia; some sequelae codes may support HCC capture under the applicable CMS model. History of stroke (Z86.73) applies to patients who have recovered with no current documented deficits and carries no HCC value. The error occurs when I63.x is used in situations where Z86.73 or an appropriate I69.x code is clinically correct.
What should MA plans do if they find unsupported acute stroke codes in their submission history? Plans should submit deletions or corrections for diagnosis codes that do not meet CMS risk adjustment submission requirements. CMS guidance states that when a plan sponsor determines through internal review that submitted diagnosis codes do not meet those requirements, the sponsor is responsible for deleting them as soon as possible. Plans should audit physician-record acute stroke codes back to service year 2020, conduct an internal coding review for flagged enrollees, submit deletions or corrections for unsupported codes, and implement pre-submission validation and second-level coding review to prevent recurrence. Finding and correcting the error proactively places the plan in a materially better position under RADV than being found first.
What should MA plans do if they find unsupported acute stroke codes in their submission history? Plans are required under federal regulations to submit deletions for diagnosis codes not supported by medical records, regardless of when the error is identified. Plans should audit physician-record acute stroke codes back to service year 2020, conduct an internal coding review for flagged enrollees, submit deletions for unsupported codes, and implement pre-submission validation and second-level coding review to prevent recurrence. Finding and correcting the error proactively places the plan in a materially better position under RADV than being found first.
The acute stroke coding error is not ambiguous. Acute stroke generally requires emergency department or inpatient hospital evaluation, and the clinical standard, coding guidance, and documentation requirement are well-established. OIG has been publishing audit results on this pattern since 2020. It released a self-audit toolkit with SQL code in 2023. It has now estimated the nationwide exposure at $462 million in potential net potential net overpayments, while CMS is scaling RADV to audit all eligible MA contracts for each payment year in newly initiated audits.
The plans that found unsupported acute stroke codes in OIG’s individual audits did not have uniquely bad coding practices. They had the same problem that exists at scale across the industry: a clinical status change that happened at the encounter level but never made it into the submission. That disconnect is fixable. The data query is published. The methodology is documented. The window to fix it before RADV gets there is narrowing, and for some plans it may already be closed.
Contact Inferscience to audit your high-risk HCC exposure and build the pre-submission validation posture this environment demands. Request a walkthrough to see how HCC Validator identifies OIG-flagged code patterns before they become RADV findings.
