AI Medicare Review Pilot Puts Patient Access Under Pressure

The US government’s attempt to automate parts of Medicare’s prior-authorization process is becoming a test of how much risk public healthcare can absorb in the name of efficiency.

The WISeR program—short for Wasteful and Inappropriate Service Reduction—launched in January across New Jersey, Ohio, Oklahoma, Texas, Arizona, and Washington. It applies prior authorization to selected services including nerve stimulation, epidural steroid injections, cervical fusion, wound-care skin substitutes, and treatments for incontinence and impotence. Many of these services previously did not require doctors to obtain such approval for Medicare patients. The pilot is scheduled to run through 2031, with expansion planned.

What the released documents show:

• The rollout began before all of the technology was ready. Innovaccer reportedly asked for a delay; when that did not happen, it temporarily configured its system to automatically approve requests rather than create an unprocessed backlog.
• Zyter experienced data discrepancies for months after apparently failing to distinguish between Medicare Part A and Part B claims. The company later said it was fully operational across both categories.
• WISeR is supposed to issue decisions within 72 hours, yet some cases took weeks or months. At least one request was still pending after 83 days.
• In one weekly report, Virtix reviewed 6,096 authorization requests, approved 2,863, and denied 3,233—a 53 percent denial rate. Virtix later said it had reduced average authorization turnaround to 1.18 days after a corrective-action process, which ended on August 14.

The consequences are not merely administrative. Delayed authorization can postpone surgery or pain treatment for older and medically vulnerable patients. Feedback collected from providers and released by the Electronic Frontier Foundation described patients crying while waiting for treatment for painful spinal compression fractures. Another Ohio provider said that even a three- or four-day delay could be difficult for vulnerable patients, while a lack of communication lasting weeks was unacceptable.

The program also exposes a difficult accountability gap. Virtix said its decisions rely on National Coverage Determinations and Local Coverage Determinations established by the Centers for Medicare & Medicaid Services, rather than rules created by the vendor. That distinction does not solve the practical problem facing a doctor or patient who receives a confusing denial and cannot reach a human representative. When automated systems fail, it must be clear who is responsible for correcting the decision and compensating for the delay.

The business model is another concern. Vendors deploying AI in this setting may face an incentive to deny as many claims as possible, particularly if rejection is easier or cheaper than careful review. That does not prove that every denial is improper, but it makes transparent performance data and independent auditing essential.

The Government Accountability Office concluded in May that Trump administration officials had not followed proper procedures when establishing WISeR, casting doubt on its legality. Lawmakers have sought more documents and tried to halt the program, but one effort was blocked by Republican members of a committee. The pilot nevertheless appears to be moving forward.

WISeR illustrates why healthcare automation cannot be judged only by processing speed or projected savings. A system that delivers a fast, explainable, reviewable decision may support clinicians. A system that delays care, produces opaque denials, and leaves no accessible human escalation path can turn automation into a barrier to treatment. Before expansion, regulators should publish denial and appeal data, vendor performance by service type, error rates, processing times, and the results of human review. Medicare patients should not bear the cost of an unfinished experiment.

Source: Ars Technica AI