Start with the number UnitedHealth wants you to have: 97%.

That is the share of diagnoses from its HouseCalls in-home visit program that FTI Consulting found were supported by patients' records, in an audit the company released on July 7. The unsupported rate came out at 3.4%, which UnitedHealth noted is roughly three times lower than the error rate the Centers for Medicare and Medicaid Services found in its most recent audit of Medicare Advantage records. CEO Stephen Hemsley said the results validate the program's integrity, adding that the company is committed to doing better.

Now the number UnitedHealth did not offer: what those unsupported diagnoses were worth.

A spokesperson declined to say how much Medicare Advantage reimbursement the flagged diagnoses generated, or what a 3.4% error rate would represent if extrapolated across the company's book of business. That silence is the most informative part of the release, and it points at what the audit was actually built to do.

The audit answers a question nobody asked

The allegation against in-home health risk assessments is not that the diagnoses fail to appear in a chart. It is that the diagnoses appear in a chart, generate a payment, and then nothing happens.

That is the finding of the 2024 HHS Office of Inspector General report that put HouseCalls in the spotlight in the first place. OIG concluded that Medicare Advantage plans collected billions through diagnoses surfaced during health risk assessments, with the largest share tied to in-home visits, for conditions that saw no subsequent treatment, no follow-up encounter, and no sign of clinical action. Of the roughly $7.5 billion that flowed to MA plans through in-home assessments in 2023, UnitedHealth took in about $3.7 billion, per the report. The concern is not sloppy paperwork. It is diagnoses that exist to be billed.

The FTI review does not test that. It tests documentation. It asks whether a submitted diagnosis is backed by the medical record, which is the same question CMS asks in a risk adjustment data validation audit. That is a legitimate question. It is simply not the one under investigation.

And there is a circularity problem sitting underneath it. The HouseCalls clinician generates a medical record during the visit. If a diagnosis captured in that visit is then validated against the record produced by that visit, the exercise risks confirming that the company wrote down what it wrote down. Whether the patient had the condition, and whether anyone treated it, is a separate matter that a chart review is not designed to settle.

Watch what happened to the 17

The sample was 200 HouseCalls visits, covering 494 diagnoses. FTI initially found 17 diagnoses unsupported by the medical record.

UnitedHealth then reviewed those 17 and pushed back, arguing that most were supported by other data. FTI subsequently agreed that 12 of the 17 were backed by paid claims. That is how the headline number moved from 96.6% to 97%.

Two things about that. First, a paid claim is a billing artifact. It establishes that a code was submitted and money changed hands, which is a weaker form of evidence than a clinical note, and in a dispute about whether codes are being submitted improperly, it is close to arguing in a circle.

Second, and more simply: the audited party reviewed the adverse findings and the auditor revised most of them. That may be entirely proper. It is also worth stating plainly, because it did not make the headline.

The sample size deserves a mention too. Four hundred and ninety-four diagnoses is a thin base from which to certify a program that operates in all fifty states through more than 4,000 clinicians, inside a company Hemsley says makes more than 19 million home visits a year. A 3.4% rate drawn from that sample carries a confidence interval wide enough to drive a truck through, and the errors that matter are not distributed evenly. The OIG flagged high-value conditions specifically, things like vascular disease, heart failure and diabetes, because those are the codes that move a risk score. One unsupported diagnosis of a lucrative chronic condition is not the equivalent of one unsupported diagnosis of anything else. An error rate treats them as if it is.

Optum says the quiet part on its own website

For anyone wondering whether the coding upside of HouseCalls is a hostile inference by critics, Optum's own provider-facing page settles it.

The HouseCalls page on UHCprovider.com pitches the program to physicians with a list of benefits. Among them: "Comprehensive diagnosis documentation and coding." And, separately, that HouseCalls "can help improve performance on key Star measures."

There is nothing illicit about writing that. It is a plain description of what the program does for the business. But it does mean the company is marketing HouseCalls to doctors partly on its documentation and star-rating value, while telling regulators and the public that it is a preventive care program with unfortunate optics. Both things are being said, in public, at the same time.

To be fair to the clinical case: UnitedHealth also reported that HouseCalls visits coincided with up to 5% fewer inpatient stays and 4% fewer emergency room visits in the following year for older adults with conditions like diabetes and hypertension. Coincided is doing work in that sentence. The people who accept a free in-home visit are not a random sample of seniors, and the company has not published anything that would let an outsider assess the counterfactual.

Meanwhile, the other dial

Medicare Advantage payment runs on two levers. Risk scores, which is the HouseCalls fight. And star ratings, which determine quality bonus payments.

The second lever is currently being resolved in court, repeatedly.

In May, a federal judge in Georgia ruled for Clover Health, finding that CMS improperly included 20 measures in its 2026 star ratings calculation and ordering a recalculation. Clover had lost half a star on its largest plan and put the cost at roughly $120 million.

CMS recalculated Clover's rating. Then, in a June memo, it recalculated for everyone else too, but on different terms: it kept 10 of the 20 contested measures, and it only updated ratings for plans whose scores would go up.

Read that again. The remedy was structured so that no plan could lose.

It still was not enough. Elevance sued on July 1, saying the differential treatment cost it $115 million and arguing that if the 20 measures were legally defective for Clover, they were defective for everyone. This month SCAN and Alignment filed their own suits, claiming $125 million and $50 million respectively. CareFirst has a suit over roughly $32 million. This is the third consecutive year CMS has been forced to redo ratings after losing in court.

What the two stories have in common

Taken separately, these are a compliance story and a litigation story. Taken together they describe a payment system in which both dials are contested, and both, in practice, turn upward.

Star ratings litigation cannot lower anyone's rating, because CMS built the recalculation so that only increases apply. Risk adjustment auditing, when the company commissions it, tests documentation rather than clinical reality, and the audited party gets to appeal the findings. Quality bonuses have grown to around $16 billion a year. MedPAC estimates Medicare Advantage overpayments will run to roughly $76 billion this year. Researchers have questioned whether the star ratings program improves plan quality at all.

None of this means the FTI numbers are false, and none of it means Elevance and SCAN are wrong on the law. They may well win. That is rather the point. Each individual move is defensible, procedurally sound, and documented. The aggregate is a program where the metrics meant to discipline payment have become instruments for increasing it.

UnitedHealth still faces civil and criminal Department of Justice investigations into its Medicare billing. A 200-visit chart review commissioned by the company will not resolve them, and it was probably never meant to. It was meant to produce a number.

It produced 97%.

Further reading