Medicare AI Pilot Program’s Lack of Details Worries Critics

Lawmakers and Medicare rights groups say the Trump administration is withholding details of a pilot program that will use artificial intelligence to decide whether Medicare will pre-approve coverage for a limited group of procedures, a move they warn could lead to an increased risk of coverage denials.

The Centers for Medicare & Medicaid Services on Jan. 1 quietly launched its Wasteful and Inappropriate Service Reduction (WISeR) model, a six-year trial that will see AI algorithms determine the medical necessity of procedures ranging from skin substitutes to diagnostic knee surgeries that the agency believes pose a high risk for fraud, waste, and abuse.

The prior authorization pilot, which will run through 2031, is rolling out across Texas, New Jersey, Oklahoma, Ohio, Washington, and Arizona, and could affect nearly 6.5 million patients on traditional Medicare. Under the new trial, services that the Centers for Medicare & Medicaid Services have determined to lack “little to no clinical benefit” would see coverage reviewed in advance by six artificial intelligence contractors.

The trial comes amid CMS efforts under Administrator Mehmet Oz to crack down hard on wasteful spending in Medicare. According to an agency fact sheet, the Medicare program spent up to $5.8 billion in 2022 on services that were either unnecessary or inappropriate.

By focusing on services vulnerable to fraud and waste, the CMS says the model will protect beneficiaries by decreasing clinically inappropriate care.

Medicare patient groups, however, say they’ve been denied important details on how the program will operate, leaving many concerned the AI could go beyond flagging fraud and deny services that were routinely covered.

“This isn’t about easing access to care at all. This is about trying to save programmatic funds,” said David Lipschutz, co-director of law and policy for the Center for Medicare Advocacy, a nonprofit law organization representing the interests of Medicare beneficiaries. “The question is whether those savings are going to be for inappropriate care as the model characterizes, or whether it’s actually going to restrict medically necessary care.”

Concerns and Controversy

Although the pilot’s rollout will be limited to six states, Lipschutz fears the data gleaned from the trial could be used to justify expanding similar programs across more states and a wider range of services.

“The branch of CMS that’s testing this has a lot of discretion to do different things in the Medicare program, including waiving almost any provision of the Medicare statute. And in certain conditions, they can ramp up demonstrations and make them geographically wider,” he said, referring to the CMS Innovation Center.

Medicare’s use of prior authorization isn’t without controversy. A 2022 report from the US Department of Health and Human Services Office of Inspector General found insurance companies contracting with the Medicare Advantage program used the strategy to routinely deny care that met Medicare’s coverage criteria.

These concerns came to a head in 2023 when insurance giant UnitedHealth faced a class action over its use of Nh Predict, a proprietary AI algorithm that class members alleged overrode physician recommendations and denied elderly Medicare Advantage patients’ claims for stays in extended care facilities.

Jeremy Friese, the founder and CEO of one of contractors, Humata Health, said in an interview that the company’s AI model will be informed by Medicare’s existing coverage criteria and won’t be able to deny claims directly.

“Our computers can say ‘yes’, then we hand that off to a physician or nurse to make that final decision,” Friese said in an interview.

“Under no circumstance ever can our computers say ‘no’ when we’ve gotten one of these submissions,” he added.

Both Friese and the CMS declined to disclose specific details behind the logic or programming powering Humata’s algorithm or the financial incentive structure the company negotiated. The agency said in a statement, however, that any denials will be finalized by licensed clinicians and that Medicare beneficiaries and their providers will be able to appeal denied claims.

Friese said each of the companies selected will run its own proprietary algorithm, a move that raised concern from a prominent Medicare lobbyist.

The lobbyist, who asked not to be identified in order to speak freely on the subject, said having multiple algorithms could affect beneficiaries’ access to care because it could invite the possibility of inconsistent results across states.

Push for Transparency

Lawmakers say the introduction of prior authorization in traditional Medicare could have a chilling effect where providers self-police how they render care to avoid getting their claims denied.

“We assume, based on what we know, that the actual doctors will be on the hook when they provide care and some AI tech company denied the claim,” Rep. Greg Landsman (D-Ohio) said in an interview. “That will lead to one of two things: doctors and health-care providers going bankrupt, or these doctors and healthcare providers just not providing the care.”

In December, Landsman and Rep. Bonnie Watson Coleman (D-N.J.) introduced the Ban AI Denials in Medicare Act (H.R. 6361). The bill would prohibit the CMS from testing WISeR or other models that would introduce prior authorization to traditional Medicare.

When it was introduced, Diane Archer, founder and CEO of Just Care USA, lauded the bill and said in a statement that it “helps guarantee that Traditional Medicare’s enrollees get the care their treating physicians deem necessary.”

So far, the CMS has not disclosed any details to lawmakers of the contract terms negotiated with the AI companies or their financial incentives, Landsman said.

He called on the agency to publicly disclose the algorithm used to make coverage decisions, the financial agreements negotiated with AI companies, and specific reasons why the companies were selected.

“Fully disclose everything. These states did not volunteer, the providers did not volunteer, the patients did not volunteer,” Landsman said. “Tell us now.”