OIG Extrapolates $180M From $1.4M of Sampled MA Overpayments as the Rule That Lets CMS Do the Same Sits Vacated

The HHS Office of Inspector General wants $130.9 million back from Humana and $46.9 million from UnitedHealthcare after finding that 178 of 220 sampled HumanaChoice enrollee-years and 183 of 250 UnitedHealthcare of Wisconsin enrollee-years carried high-risk diagnosis codes their medical records did not support for payment years 2020 and 2021 (HHS OIG, A-05-24-00010 and A-07-24-01214, September 10, 2026). The sampled overpayments behind those two figures total $1.39 million.

The gap between the sample and the recommendation is the whole dispute. OIG estimates at the lower limit of a two-sided 90 percent confidence interval, both plans say the design that produced the estimate is invalid, and the 2023 rule that gave CMS the same extrapolation power was vacated by a Texas district court in September 2025 and is on appeal (RISE, September 26, 2025; Veradigm, November 2025).

Key Takeaways

• $130,922,653 is the precise Humana figure: $130,868,965 from the 200 statistically sampled enrollee-years plus $53,688 from 20 nonstatistical cases in the mis-keyed code group, against $669,237 actually found in the sample (OIG A-05-24-00010, September 2026).
• 79 percent of the $165.7 million CMS paid HumanaChoice for the 68,701 enrollee-years in the sampling frame is what the recommendation implies was overpaid, and 73 percent of United's $64.5 million frame, ratios that follow directly from the 81 and 73 percent sample error rates.
• 0.35 percent of the roughly $37.8 billion CMS paid HumanaChoice across the two payment years is the size of the recommended refund, which is why a contract-level audit reads as material to the frame and immaterial to the contract.
• 20 of 20 acute stroke enrollee-years and 16 of 20 embolism cases failed at HumanaChoice, mostly because a history of a prior event was coded as an acute one at a physician visit (OIG A-05-24-00010, September 2026).
• $4.7 billion over ten years is what CMS projected the vacated RADV rule would recover through extrapolation from payment year 2018; OIG says the vacatur does not bind it, and United says the 2012 method with a fee-for-service adjuster now governs (American Action Forum, 2023; OIG A-07-24-01214).

Eleven Diagnosis Groups and Two Sampling Frames

OIG built both audits the same way. From payment years 2020 and 2021 it pulled every enrollee whose submitted codes mapped to one of 11 high-risk groups: acute stroke, acute myocardial infarction, embolism, lung, breast, colon, prostate and ovarian cancer, sepsis, pressure ulcer, and potentially mis-keyed codes. Each is a diagnosis that raises an HCC score but rarely survives a chart read when it appears only on a physician or outpatient claim without a matching inpatient stay, treatment, or medication (OIG A-05-24-00010, September 2026).

The HumanaChoice frame held 68,701 enrollee-years and $165,672,058 of payments tied to those codes. OIG drew 200 enrollee-years by stratified random sample from the first ten groups, 20 per group, and 20 more nonstatistically from the 232 mis-keyed cases. The United frame was 28,410 enrollee-years and $64,458,294, sampled as 200 plus 50. An independent medical review contractor read the records; where the unsupported code should have been a less severe one in the same disease family, OIG credited the replacement HCC before counting the overpayment.

The pattern inside the groups is the same one OIG has reported since its first Humana audits on 2015 data. All 20 HumanaChoice acute stroke cases failed, typically because a patient with a stroke years earlier was coded as acute at a routine visit. Sixteen of 20 embolism cases failed on the same logic, a documented history of deep vein thrombosis carried forward as a current vascular-disease HCC. The audits credit the history code where one exists, which is why the sampled dollar amounts are lower than the raw HCC values.

How the Lower Limit Multiplies a Sample by 195

Stratified sampling with 20 units per stratum produces a wide interval, and OIG uses the bottom of it. Its stated policy is to recommend recovery at the lower limit of a two-sided 90 percent confidence interval, which it describes as "designed to be less than the actual overpayment total 95 percent of the time" (OIG A-05-24-00010, September 2026). Humana argued the agency should have used the lower bound of a 95 or 99 percent interval because CMS has never fixed a confidence level for RADV; OIG answered that the legal standard is a statistically valid method, "not the most precise methodology."

The arithmetic behind the ratio is unremarkable once the frame is understood. An 81 percent error rate applied to $165.7 million of code-linked payments points toward roughly $134 million; the lower limit lands at $130.9 million. The multiplier from $669,237 to $130.9 million, about 195 times, is simply the frame divided by the sample. United's objection that the standardised normal formula "is generally not recommended for use with very small sample sizes, such as those at the HCC-level in this audit" attacks the interval width, and OIG's reply cites Departmental Appeals Board and federal court decisions upholding samples smaller than 100 (OIG A-07-24-01214, September 2026).

The fee-for-service adjuster is the argument with a court behind it. United cites the actuarial equivalence requirement in 42 U.S.C. 1395w-23(a)(1)(C), notes that the 2023 RADV final rule eliminated the adjuster for payment years 2018 onward, and points to the September 2025 ruling that vacated that rule for inadequate notice of its reasoning (RISE, September 26, 2025). CMS filed its appeal on November 21, 2025 (Veradigm). OIG acknowledges the vacatur and says it "does not impact our findings," since "CMS has not issued any requirements that compel us to reduce our overpayment calculations." That leaves two federal bodies applying opposite extrapolation regimes to the same payment years.

Against CMS's own coding-intensity actions the numbers are small. The CY 2027 exclusion of unlinked chart review diagnoses removes $7.12 billion, 1.53 percent of aggregate MA payments, in a single year, as the site set out in May. Two contract audits totalling $177.8 million cover two years, and the Humana figure is 0.35 percent of the $37.8 billion CMS paid that contract over the period. What the audits add is a per-code error rate on the diagnoses the risk model rewards most.

The Refund United Says It Already Paid

United's last objection is the one with reserving consequences. It told OIG that the payment year 2020 overpayments "fail[ed] to account for the MLR rebate mechanism," that it had remitted to CMS an amount "more than ten times greater than the overpayment alleged by the OIG," and that CMS "is not entitled to a second recovery of the same funds" (OIG A-07-24-01214, September 2026). OIG placed the point outside its scope and referred it to CMS's audit resolution process.

The mechanism is real. A Medicare Advantage contract below the 85 percent minimum medical loss ratio remits the shortfall. If $46.9 million of revenue is later clawed back, the denominator of the 2020 MLR shrinks, the ratio rises, and the rebate already paid was too large by some fraction of the recovery. Neither audit describes a true-up, so the contingent liability a plan actuary books is the OIG figure less an MLR offset with no defined calculation behind it, on an appeal timetable nobody controls.

Humana's version of the same problem runs through underpayments. It argued OIG's method is "systematically skewed towards identifying overpayments rather than underpayments" because the review only tested codes actually submitted, and OIG agreed that finding unsubmitted HCCs was "beyond the scope of our audit." Both plans disagreed with every recommendation, and OIG's findings are not final determinations; CMS decides whether an overpayment exists. The audits therefore hand the plans a number that is contract-immaterial, frame-material and legally unsettled, and a rebate calculation from 2020 that may already have paid part of it.

Further Reading

• CMS Excludes Unlinked Chart Review Diagnoses From CY 2027 MA Risk Scores – The $7.12 billion prospective cut that sits beside these retrospective audits, and the OIG work plan item aimed at the same records.
• CMS Defers the V28 Recalibration for CY 2027 – The payment-side arithmetic of the risk model these diagnosis codes feed.
• ASOP 45's Proposed Revision Adds Coding Intensity to Risk Adjustment Practice – The standard that would govern how a plan actuary treats a documented error rate like these.
• UnitedHealth Sells WellMed Stake as Optum Targets 6% Margin – The delegated-risk structure under which many of the audited diagnoses were captured.

Sources

1. HHS OIG, "Medicare Advantage Compliance Audit of Specific Diagnosis Codes That HumanaChoice (Contract H5216) Submitted to CMS," A-05-24-00010, September 10, 2026
2. HHS OIG, A-05-24-00010 complete report (PDF), including Humana, Inc. comments and OIG responses
3. HHS OIG, "Medicare Advantage Compliance Audit of Specific Diagnosis Codes That UnitedHealthcare of Wisconsin, Inc. (Contract H5253) Submitted to CMS," A-07-24-01214, September 10, 2026
4. HHS OIG, A-07-24-01214 complete report (PDF), including United comments and OIG responses
5. Healthcare Dive, "Federal watchdog accuses Humana, UnitedHealthcare Medicare Advantage plans of upcoding," September 17, 2026
6. RISE, "Federal judge vacates 2023 RADV final rule: What this means for MA plans," September 26, 2025
7. Veradigm, "CMS Appeals Court Ruling on the RADV Final Rule," November 2025
8. American Action Forum, "CMS' New RADV Rule," February 2023