AI-Driven Medicare Pilot Yields Alarming Denial Rates and Delays, Lawsuit Reveals

A lawsuit filed by the Electronic Frontier Foundation (EFF) has uncovered significant problems with a Medicare pilot program that uses artificial intelligence to approve medical claims, revealing alarmingly high denial rates and substantial delays in patient care.

The program, officially known as the Wasteful and Inappropriate Service Reduction (WISeR) model, was launched in January across six states—Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington. Its stated objectives are to reduce healthcare costs and streamline administrative processes for the 6.4 million traditional Medicare beneficiaries affected. However, documents obtained through a Freedom of Information Act (FOIA) request show that the implementation has been plagued by errors, with some vendors denying more claims than they approved.

“There’s alarmingly high denial rates, and that is known by CMS,” said Lena Cohen, a staff technologist at the 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 highlighted one instance where a prior authorization request went unanswered for 83 days.

Corrective Action for High Denial Rates

The released records indicate that Virtix Health, the vendor responsible for prior authorization decisions in Washington state, was placed on a “corrective action plan” by the Centers for Medicare & Medicaid Services (CMS) after initially denying a higher volume of requests than it approved. CMS stated that it closely monitors model implementation and that Virtix’s turnaround times have since improved and are tracking toward the mandated three-day requirement.

Another vendor, Innovaccer, raised internal concerns prior to the launch, citing changing CMS requirements, unclear governance, and insufficient time for end-to-end testing with the medical community. Gretchen Jacobson, vice president and Medicare expert at the Commonwealth Fund, criticized the approach, noting that “testing shouldn’t occur on the go” and that mature systems should be in place before rollout.

Financial Incentives and Scope of Care

The WISeR model requires additional approvals for more than a dozen common procedures, including epidural steroid injections for pain management, treatments for spinal stenosis and osteoarthritis, and nerve-stimulation therapies for sleep apnea and incontinence. CMS defends the program by asserting that patient safety is a core tenet, aiming to prevent clinically unsupported or fraudulent procedures that could cause harm.

Critics, however, point out that the vendors are paid based on the costs they reduce, creating a financial incentive to deny coverage. Miranda Yaver, an assistant professor of health policy and management at the University of Pittsburgh, described the situation as a “dramatic manifestation of profit-seeking.” She noted that while some denials are appropriate, many are not, and called the program’s early performance “unimpressive.”

The EFF found that low quality scores currently reduce vendor payments by only 5% to 10%, a penalty critics argue is too weak to counter the incentives for denial.

Potential for Expansion

Beyond the current six states, internal planning documents suggest the program could expand to cover air ambulance transport, cancer treatments, MRI scans, and medications. EFF’s Cohen expressed concern over this potential commitment to incorporating AI into medical decision-making.

In response to inquiries about expansion, CMS stated it does not anticipate extending the WISeR model beyond the current six states. The EFF continues to seek additional details regarding how often denials are reversed on appeal, how the AI systems were trained, and what measures were taken to protect against bias. Jacobson emphasized that data should be released frequently and publicly, stating, “It shouldn’t take a FOIA.”