One vendor was put on a 'corrective action plan' for denying more claims than it approved

Artificial intelligence is being used to approve care for 6.4 million Medicare beneficiaries in six states.

A pilot program using artificial intelligence to approve care for Medicare beneficiaries has resulted in high levels of denials and delayed decisions and prompted warnings from a vendor about a lack of readiness, according to a lawsuit seeking information about the program.

The program, called the Wasteful and Inappropriate Service Reduction model, or WISeR, was created with the aim of cutting costs and easing administrative burdens and is slated to run through the end of 2031 in six states.

The Freedom of Information Act lawsuit, which was filed in March by the digital-rights organization Electronic Frontier Foundation against the Centers for Medicare & Medicaid Services, recently produced about 1,000 pages of documents detailing medical providers' complaints about the program, contracts with technology vendors, and records of denials and delays of care.

Some of the information released under the lawsuit had been previously flagged by medical providers and lawmakers, but the extent of problems, the excessive denials by one vendor and the concerns raised before the launch by another vendor were not previously known.

The Freedom of Information Act is a U.S. law that gives anyone the right to request access to records from any federal agency. A FOIA lawsuit is a legal action filed in federal court when a government agency fails to properly fulfill or respond to a request for public records.

"There's alarmingly high denial rates, and that is known by CMS," said Lena Cohen, a staff technologist at EFF. "A significant number are not getting responses within three days [as required]. Many are waiting over a month to receive a reply, and that's really concerning." Cohen noted that one prior authorization request went unanswered for 83 days.

More than 11,000 people in the U.S. turn 65 every day, and more than 70.5 million are enrolled in Medicare, according to the latest administrative data from CMS.

A 65-year-old retiring in 2026 can expect to spend an average of $185,500 on healthcare and medical expenses throughout retirement, according to Fidelity Investments. That figure is up 7.5% from a year ago amid rising prices for care, greater use of healthcare services as the population ages, and growing costs tied to managing chronic conditions.

Meanwhile, there is a rising level of concern about the use of artificial intelligence in everyday life.

"These records really confirm what advocates, providers and lawmakers have been warning about since WISeR was launched. It's troubling that CMS knew about much of this and didn't communicate it to the public," Cohen said.

Under the program, which launched in January, prior authorizations for about 6.4 million Americans enrolled in traditional Medicare in New Jersey, Ohio, Oklahoma, Texas, Arizona and Washington state are being carried out by AI.

The WISeR model requires additional approvals for more than a dozen common medical procedures, such as epidural steroid injections for pain management, minimally invasive treatments for spinal stenosis in the back and osteoarthritis in the knees, and nerve-stimulation therapies for conditions such as sleep apnea and incontinence.

"A core tenet of the WISeR model is ensuring patient safety," CMS said. "Items and services were selected that are both non-emergent and highly susceptible to fraud, waste, and abuse to help patients avoid clinically unsupported, unnecessary, and inappropriate procedures that may cause potential harm, like pain, bleeding, infection, anxiety, or other adverse outcomes. Protecting beneficiary access to Medicare-covered, medically necessary care is a top priority."

A 'rocky' launch

"Some of the decision delays were quite long for patients," said Gretchen Jacobson, vice president and Medicare expert at the Commonwealth Fund, adding: "You should think that new technology would make the process almost instantaneous. What everyone wants is for Medicare beneficiaries to access the care they need in a timely way. Technology should make prior authorizations easier, quicker and more seamless - and that does not seem to be the case."

The WISeR program is part of the Trump administration's efforts to cut what it calls waste, fraud and abuse in government. The companies contracting to do the review process receive payments when they reduce costs, which critics have said incentivizes them to deny service. CMS has said a qualified human clinician must review all denials of care.

"We learned that early implementation was somewhat rocky," said Matthew Fiedler, a senior fellow at the Center on Health Policy at the Brookings Institution. "It was among the first glimpses of how this model is operating in practice. But we don't have a great picture on whether it was resolved or whether it will continue to be an issue."

Fiedler added: "What does seem clear here is that the model was not ready to operate as planned on Day 1. It's unclear whether there was a rush on the CMS side to get the model off the ground or whether this was normal growing pains. This did not operate smoothly out of the gate."

WISeR vendors have been denying claims at unusually high rates, the documents show. One company, Virtix Health, which handled prior authorization decisions in Washington state, was required by CMS to submit a "corrective action plan" because it initially denied more requests than it approved.

"CMS closely monitors the implementation of all models, including system performance and denial patterns, and is actively addressing issues as they arise," CMS said in a statement to MarketWatch. "Since implementing those corrective actions, Virtix's turnaround times have improved and are on track to meet the model's three-day requirement."

Virtix did not immediately respond to a request for comment.

CMS has said it protects against inappropriate denials by linking vendors' payments to quality scores that take into account the timeliness and accuracy of decisions. However, EFF said that low scores reduce payments by only 5% to 10%.

"The financial incentive to deny coverage is a dramatic manifestation of profit-seeking. Some denials are appropriate. Many are not," said Miranda Yaver, an assistant professor in the department of health policy and management at the University of Pittsburgh and the author of a book titled "Coverage Denied: How Health Insurers Drive Inequality in the United States."

"This is a really big change to traditional Medicare," Yaver said. "The report card is pretty unimpressive. We all could have foreseen that this wasn't going to go great."

Meanwhile, another vendor, Innovaccer, raised concerns about the pilot program's launch, citing CMS's changing requirements and expectations, unclear governance processes and the lack of time provided for end-to-end testing with the medical community.

"The report highlights that some vendors were not quite ready to have the technology rolled out. These should be mature systems. Testing shouldn't occur on the go," Jacobson said.

Innovaccer did not immediately respond to a request for comment.

Analysts raised concerns that the WISeR program could be eventually expanded to review more medical conditions, further changing the nature of Medicare, which requires fewer prior authorizations than Medicare Advantage, the alternative to traditional Medicare that is run by private insurance companies.

"The internal planning documents show there's discussion of expanding this model. That shows a concerning commitment to incorporating AI into medical systems and decision-making," EFF's Cohen said.

An expansion could include air ambulance transport, cancer treatment, MRI scans and medications, EFF found.

CMS said it does not anticipate any expansion of the WISeR model beyond the current six states.

EFF is seeking more details from CMS, such as how often denials were reversed on appeal, how the AI systems were trained and how they were tested to protect from bias, Cohen said.

Overall, EFF's "findings are certainly concerning. AI systems should be agreeing to cover the care that should be covered," Jacobson said. "Ideally, the data should be released in an ongoing, frequent and predictable fashion, publicly, so we can understand what's happening more quickly. It shouldn't take a FOIA."

-Jessica Hall