WASHINGTON – The Electronic Frontier Foundation reports thousands of refusals in the first three months of Medicare’s machine assisted approval programme. It reports one request that waited 83 days.
Those figures do not show whether the programme refuses payment more often than a person would. The two numbers that would settle that have not been published.
The records came from the Centers for Medicare and Medicaid Services through freedom of information litigation. The group reported on them on 8 September 2026.
The programme is called the Wasteful and Inappropriate Services Reduction model. It runs in New Jersey, Ohio, Oklahoma, Texas, Arizona and Washington, through 31 December 2031.
Contents
What the records show#
The figures below come from the Electronic Frontier Foundation’s reading of the records. They are that group’s account. The roughly 1,000 pages have not been read here.
Two companies together refused 5,944 requests in the first three months, the group reported. It also reported that Virtix Health, which works in Washington state, refused more requests than it approved in that period.
EFF reports that CMS expects a decision within 72 hours. No CMS document we hold states that number. EFF also reports a status report describing a request that went unanswered for 83 days.
A provider’s comment in EFF’s account reads: “I have had cases submitted and waiting over 1 1/2 months for a UTN to be generated… In the meantime patients are having to be cancelled for surgeries they need.”
The group also reported that a low quality score cuts a company’s payment by only 5 to 10 percent.
CMS names six participants, one in each state. They are Cohere Health in Texas, Genzeon in New Jersey, Humata Health in Oklahoma, Innovaccer in Ohio, Virtix Health in Washington, and Zyter in Arizona. No refusal counts have been published for the other four.
The payment arrangement is why the counts are worth arguing about. The Federal Register notice says participants are “compensated based on a share of averted expenditures.” The more spending a company averts, the more it earns.
Why a count is not a rate#
A count says how many refusals were issued. A rate says what share of requests were refused. A rate needs a denominator, and none has been published.
The group reports a count for two of the six companies over three months. It does not report the total requests those two decided.
Neither the Federal Register notice nor the CMS model page publishes a rate.
One line comes close. The group reported that one company refused more than it approved. That is a serious finding. It is still one company over one quarter, not a programme rate.
Two companies out of six is not a third of the programme, either. No request volume has been published for any participant, so how much of the programme these two handled is unknown. Counting companies is not counting requests.
The figures also cover the first three months only. The programme is a six year test, and that was its opening quarter.
That cuts both ways. Early numbers can overstate a problem that settles down. They can also understate one that grows.
The comparison that does not exist#
Even a full rate would not answer the question in the headline. There is nothing to compare it against.
The model does not simply impose advance approval. The notice says “submitting a prior authorization request is voluntary.” A provider who skips it has the claim reviewed before payment instead.
What is missing is any published refusal rate for these services, in these states, under whatever arrangement applied before 2026.
That number cannot come from the model. Running the programme forward produces figures about the programme. A before-and-after needs the before, and that would have to come from claims data CMS already holds.
What would settle it#
Three sets of figures would answer the question. None is public today.
First, a total request count by company and by quarter. That would turn 5,944 into a rate.
Second, an approval and refusal breakdown for all six companies. That would show whether Virtix is an outlier or the pattern.
Third, refusal rates for the same services in the same states before 2026. Those would show whether machine assisted review moved the number.
A fourth would help and is not on the list. Under the notice a refusal is not itself appealable. What can be appealed is the claim denial that follows, and we have found nothing published on how many of those were later overturned.
CMS has promised to watch some of this. The notice says the agency will hold participants “accountable for the accuracy and timeliness” of their decisions. It does not promise to publish the counts.
What happens next#
Two of the listed services are not being reviewed at all. CMS delayed them in April 2026, and says they wait for “a future date that will be announced in a subsequent Federal Register notice.”
A refused request can be resubmitted by the provider without limit. On a resubmission the notice says the provider “would have the opportunity to request a peer-to-peer review to inform the new determination.” It does not say who that peer is.
If a claim carrying a refusal is submitted, the notice says the contractor will deny it.
The provider “may then appeal the claim denial with the MAC under existing appeals procedures.”
Year two begins in January. The litigation that produced these records is public, and more may follow.
For a beneficiary whose request was refused, the rules have not changed while this is argued. A refusal must carry a licensed clinician’s decision. A provider may resubmit without limit. And if a claim is denied, beneficiaries “retain their administrative appeal rights.”
The same question keeps arriving wherever a rule or a tool sets an amount. Who is allowed to disagree with the result? It arrived when a scoring tool sets hours.
The companion story explains what the programme actually does. Background is in AI is deciding Medicare coverage.
Source: Federal Register notice 2025-12195, published 1 July 2025, full text read; Federal Register notice 2026-06616, published 6 April 2026, full text read; the CMS WISeR model page and provider fact sheet; and the Electronic Frontier Foundation’s account of records released in its litigation against CMS, read 8 September 2026.