New: independent coding checks for insurers and providers. See how Lens works →

Back

Closing Care Gaps Without Crossing Into Upcoding

· 7 min read
Risk adjustment

OIG's 2026 Medicare Advantage guidance flags chart-review-only coding and HRAs as risks. See the seven controls that separate real gap closure from upcoding.

Risk adjustment compliance comes down to one question: does your program find diagnoses that are true and documented, or diagnoses that pay? Most programs say the first. Their processes, incentives and metrics often say the second.

The line isn't between adding diagnoses and not adding them. Real chronic conditions go undocumented every year, and capturing them is accurate coding. The line is whether every diagnosis comes from a clinician's actual evaluation, and whether the program corrects errors in both directions. In February 2026, the HHS Office of Inspector General put that line in writing in its compliance program guidance for Medicare Advantage. This article covers where the line sits, what recent enforcement shows, and the controls that keep a gap-closure program on the right side of it.

Where the line is

Closing a real care gap Crossing into upcoding
Flagging a chronic condition from last year so the clinician can evaluate it at this year's visit Prompting the clinician to add a specific code
Querying when the note is unclear, with the clinical indicators and an option to say "not present" A query that only offers the higher-paying answer
Reviewing records and both adding supported diagnoses and deleting unsupported ones Reviewing records only to find diagnoses to add
Documenting a condition at a face-to-face visit where it was assessed Adding an addendum months later so a diagnosis can be submitted
Paying vendors and coders for accuracy Paying vendors a share of the revenue their diagnoses generate
Following up an in-home assessment with care Using in-home assessments mainly to generate diagnoses

Every behavior on the right can be dressed up as the one on the left. That's why compliance has to live in processes and metrics, not in intent.

What OIG's 2026 guidance says

OIG released its Medicare Advantage industry segment-specific compliance program guidance on February 3, 2026. On risk adjustment, it names three practices as risk areas: relying solely on chart reviews to identify diagnoses that raise risk scores, conducting in-home health risk assessments (HRAs) to generate additional diagnoses, and failing to remove diagnosis codes that chart reviews have shown are unsupported.

It also sets expectations for controls:

  • Audit diagnosis data before and after submission, with extra scrutiny of high-risk codes
  • Pair every process that adds diagnoses with one that checks accuracy and ties the diagnosis to appropriate care
  • Educate employed and contracted providers on proper coding
  • Analyze outliers: provider coding intensity, risk scores and HCC prevalence benchmarked and tracked over time, potentially using algorithms or AI platforms
  • Monitor first-tier, downstream and related entities, including coding vendors

The guidance is voluntary. But it's what OIG and the Department of Justice will measure a program against when something goes wrong.

The data behind it is OIG's own. In 2019 it found that more than 99 percent of MA chart reviews added diagnoses rather than deleting them. In 2024 it estimated that diagnoses reported only on HRAs and HRA-linked chart reviews generated $7.5 billion in 2023 payments, with 20 companies driving 80 percent of it.

What recent enforcement shows

Three cases show where liability attaches and where it doesn't.

Case Outcome Alleged conduct
Kaiser Permanente affiliates $556 million settlement, January 2026 Mining records for new diagnoses; pushing clinicians to add addenda months or more than a year after visits; tying evaluations and incentives to risk adjustment targets; not removing diagnoses later found unsupported
Independent Health and a chart review vendor Up to $98 million, plus $2 million from the vendor's CEO, December 2024 A vendor paid up to 20 percent of the revenue its diagnoses generated; resolved or unsupported conditions, such as past heart attacks and transient depression, coded as current
UnitedHealth (Poehling) A special master recommended summary judgment for UnitedHealth in March 2025; DOJ objected Using chart reviews to add diagnoses without deleting unsupported ones. The special master found DOJ hadn't reviewed the underlying records to prove the codes were wrong

The settlements didn't include admissions of liability. The pattern across all three is still clear.

Knowledge plus inaction is the core risk. Kaiser's alleged failure to remove diagnoses later found unsupported, and the theory in Poehling, both turn on what an organization knew about unsupported codes and didn't correct.

Contingency pay is a red flag. A vendor paid a share of the revenue its diagnoses create has an incentive the program can't audit away.

Records decide cases. The Poehling recommendation turned on the government not reviewing medical records. The reverse is also true: an organization that can't show records supporting its diagnoses has no defense either. That's the same test a RADV audit applies, one HCC at a time.

Seven controls for compliant gap closure

1. Two-way review, always. Every review that can add a diagnosis must also be able to delete one, and must report both. If your retrospective program's delete rate is near zero, it's either reviewing perfect documentation or not looking.

2. The clinician decides. Prospective flags and queries present what to evaluate, never what to code. A clinician must be able to dismiss a flag as easily as accept it, and acceptance without an assessment in the note doesn't count. What compliant flags look like, and how they differ from leading prompts, is covered in retrospective vs prospective vs concurrent risk adjustment review.

3. No pay tied to diagnoses. Vendors are paid per chart reviewed or a fixed fee, not a share of RAF or revenue. Coder productivity measures include accuracy and deletes. Provider incentives reward completed visits and care quality, not HCC counts.

4. A time limit on addenda. Set a written policy for how long after an encounter a clinician may amend a note, and how. Late documentation needed for a diagnosis becomes a prospective flag for the next visit, not an addendum.

5. A delete and repayment process. When an audit finds an unsupported diagnosis that was already submitted, it's deleted, and any overpayment is reported and returned under the MA overpayment rules. The process is documented, has an owner, and runs on a schedule.

6. Outlier monitoring. Track coding intensity and HCC prevalence by provider, practice, vendor and HCC, against peers and over time. Investigate the outliers. This is the analysis OIG's guidance asks for, and the cheapest early warning available.

7. Audits that test what RADV tests. Sample submitted HCCs, not just charts, before and after submission. Use reviewers who didn't code the original records, and score against the actual documentation requirements, not just MEAT. The rubric is in MEAT documentation for chronic conditions.

What real care gap closure looks like

None of this means capturing fewer conditions. It means capturing them the right way.

Real gaps are conditions the patient has that haven't been evaluated and documented this year. The fix is a visit: an annual wellness or chronic care visit where the clinician sees a reliable list of what's due, evaluates each condition, documents the assessment and plan, and follows up with care. Specialist notes get coordinated so conditions managed elsewhere are visible. New diagnoses come from workup, not inference.

Programs built this way capture the conditions that are actually present, and every diagnosis can be traced to a face-to-face evaluation. That's the outcome the risk adjustment model was designed to pay for.

Where AI fits

AI can make a risk adjustment program more compliant or less, depending on what it's pointed at. OIG's guidance explicitly mentions algorithms and AI as tools for finding outliers. The same technology pointed at adding diagnoses scales exactly the conduct the guidance warns against.

Questions to ask of any AI coding tool, including your own:

  • Does it suggest deletes as readily as adds, and report both?
  • Is every suggestion linked to the specific note, date, provider and text that supports it?
  • Does a credentialed coder confirm every suggestion before submission, with the decision logged?
  • Does it surface documentation gaps as prospective flags or compliant queries, rather than codes?
  • How is the vendor paid? Per HCC found is the same red flag as contingency chart review.
  • Can you audit its accuracy in both directions, on your own data?

A ten-question self-assessment

Answer yes or no. Any "no" is a gap to close before an auditor finds it.

  1. Do we report delete rates alongside add rates, for every review program?
  2. Can clinicians dismiss prospective flags as easily as accept them?
  3. Are all queries non-leading, with "not clinically supported" as an option?
  4. Are vendors paid without reference to diagnoses or revenue generated?
  5. Are provider and coder incentives free of HCC or RAF targets?
  6. Do we have a written addendum policy with a time limit?
  7. Do we delete unsupported submitted diagnoses and return overpayments on a schedule?
  8. Do we monitor coding intensity outliers by provider, vendor and HCC?
  9. Do our audits sample submitted HCCs, before and after submission, with independent reviewers?
  10. Do we know what our AI tools suggest, how they're evaluated, and how they're paid for?

FAQ

What is upcoding in risk adjustment?

Upcoding in risk adjustment is submitting diagnoses that aren't supported by the medical record, or at a higher severity than documented, so a member's risk score and the plan's payment rise. It includes coding resolved conditions as current and adding diagnoses the clinician didn't evaluate.

Is closing care gaps the same as upcoding?

No. Closing a care gap means a clinician evaluates and documents a condition the patient actually has at a qualifying visit. Upcoding means submitting a diagnosis the record doesn't support. The difference shows in process: who decides, whether errors are deleted, and how people are paid.

What does OIG's 2026 Medicare Advantage guidance say about risk adjustment?

It names relying solely on chart reviews to raise risk scores, using in-home HRAs to generate diagnoses, and failing to remove unsupported codes as risk areas. It recommends audits before and after submission, outlier analysis, provider education and vendor oversight.

Are chart review vendors on contingency allowed?

They carry significant risk. A vendor paid a share of the revenue its diagnoses generate has a direct incentive to add codes, and that arrangement featured in the Independent Health settlement. Per-chart or fixed-fee arrangements with accuracy audits are the safer structure. Get counsel's view on any existing contract.

Do we have to delete unsupported diagnoses we find?

Yes. Diagnoses later found unsupported should be deleted, and identified overpayments reported and returned under the Medicare Advantage overpayment rules. Failing to remove known unsupported diagnoses is one of the three risk areas OIG's guidance names, and it featured in the Kaiser allegations.

This article is general education on risk adjustment compliance. It isn't legal advice; consult counsel about your organization's obligations.

More from the Journal

All articles

See it on your own charts.