Medicare’s AI Prior-Authorization Test Produced Less Care. Was It Better Care?

In health care, a vanished procedure can mean two opposite things.

Presented By Our Partners

It can mean that a patient was spared an unnecessary intervention: one less injection with little chance of helping, one less operation whose risks outweighed its benefits, one less bill paid for care that should never have happened.

Or it can mean that friction won. A request sat in a queue. A clinician decided the paperwork was not worth another round. A patient in pain lost the stamina to keep calling. The care did not become unnecessary; it simply became unreachable.

A new study in JAMA shows that Medicare’s experiment with AI-enabled prior authorization made some care disappear. It cannot yet tell us which of those two stories occurred.

That distinction is the most important test for the next generation of health-care automation. An algorithm can reduce utilization. A policy can reduce spending. A dashboard can show faster decisions. None of those results, alone, proves that patients received better care.

The first measurable effect of WISeR

Medicare’s Wasteful and Inappropriate Service Reduction model, mercifully shortened to WISeR, began in January 2026. It brought prior authorization to selected services in traditional Medicare across Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington.

The six-year test focuses on services CMS considers vulnerable to fraud, waste, or inappropriate use, including epidural steroid injections and several nerve-stimulation and surgical procedures. Participating technology companies may use artificial intelligence and machine learning to review requests. CMS says those tools operate alongside human clinical review and that licensed clinicians make recommendations against payment.

Researchers from the University of Minnesota and Brown University used Epic Cosmos, an electronic-health-record dataset covering more than 300 million patient records, to examine what happened next. They compared monthly use of targeted services among traditional Medicare patients in the six WISeR states with use elsewhere, drawing on two years before the program and its first six months.

The change was immediate. Epidural steroid injections, the highest-volume service studied, fell 15.0% relative to nonparticipating states. Other targeted services fell 19.5%.

The effect diminished but did not disappear. For epidural injections, the estimated reduction narrowed from 20.7% in the first quarter to 9.3% in the second. For the other services, it moved from 21.4% to 17.6%.

The researchers also examined services that had initially been planned for WISeR but were not implemented. Those did not show the same decline. That strengthens the case that the authorization policy, rather than an unrelated national trend, changed care use.

Featured Partner

Invest in the Infrastructure Behind Modern Medicine

As healthcare expands beyond hospital walls, the buildings and campuses supporting that shift are generating compelling returns for investors who move early. The Healthcare Real Estate Fund offers qualified investors direct access to a curated portfolio of medical office, outpatient, and specialty care facilities.

Learn More →

It is meaningful evidence. It is not evidence that patients benefited.

Less is a result, not a verdict

The study could not see why an individual service did not happen. It could not separate requests that were never submitted from those that were delayed, rejected, abandoned, or reconsidered after a useful clinical conversation. It did not determine whether the foregone care met medical-necessity criteria. It did not measure pain, mobility, wound healing, complications, or quality of life.

Nor did it isolate the role of AI in each decision. WISeR is an AI-enabled administrative and payment model, not a clean trial of an algorithm against human reviewers. Saying that “AI denied 20% of care” would outrun the evidence.

The honest conclusion is narrower and more consequential: the policy changed behavior before the public could tell whether the resulting behavior was clinically better.

This is a familiar weakness in health-care technology. Systems are often judged by what they can count most easily: alerts generated, charts reviewed, authorizations completed, minutes saved, denials issued, or dollars averted. Patients live at the far end of a longer chain. They need the right service to be requested, approved, scheduled, completed, and followed by a better outcome.

The chain can break at every link.

The incentive hiding inside the workflow

CMS describes WISeR as a way to reduce unsupported care while protecting taxpayers and patients. That aim matters. Wasteful treatment can expose people to complications without improving their health, and public insurance should not pay unquestioningly for every bill submitted to it.

But the model’s incentives deserve equal attention. CMS says participating companies receive a share of expenditures associated with care their reviews avert, with adjustments for measures such as provider experience. The structure may reward finding waste. It may also create pressure to classify ambiguous care as avoidable.

That makes oversight more than a ceremonial “human in the loop.” The relevant questions are whether clinicians receive the full record, have enough time and independence to disagree with the system, reverse questionable recommendations, and leave an audit trail that patients and regulators can examine.

Records obtained through litigation by the Electronic Frontier Foundation add urgency to those questions. EFF’s October release describes vendor-readiness problems and CMS inquiries concerning denials that appeared inaccurate or inconsistent with coverage criteria. The records are an incomplete and partly redacted view, and EFF is an advocacy organization. They should not be treated as a comprehensive audit. But they make it harder to accept generic assurances about human review as proof that oversight works.

What patients in the six states can do

If you have Original Medicare in a WISeR state and a clinician recommends a targeted service, ask three questions early:

1. Is this service included in WISeR? The model does not cover every procedure, every setting, or Medicare Advantage.
2. Was prior authorization submitted, and on what date? Keep the request number, decision, and any explanation in one place.
3. What is the plan if the request is not affirmed? Ask who will provide the clinical documentation, whether a human reviewer considered the request, and what appeal or alternative-care path is available.

Do not assume silence means a clinical decision was made. If symptoms worsen while a request is pending, contact the treating team. Emergency services are outside the model’s prior-authorization process.

These steps do not fix the system. They make its failures easier to see and challenge.

The scorecard that should come next

The next WISeR report should not celebrate or condemn a 15% to 20% decline by itself. It should connect the administrative record to the patient’s actual course of care.

For every targeted service, the public needs to know:

• how many requests were submitted, affirmed, not affirmed, and withdrawn;
• how long each step took;
• how often human reviewers reversed an automated recommendation;
• how many patients appealed and how often they succeeded;
• whether patients ultimately completed the requested service, received an appropriate alternative, or received no care;
• whether pain, function, complications, hospitalization, and other outcomes changed;
• whether delays and noncompletion differed by race, disability, income, language, rurality, or practice size.

That is the difference between measuring a gate and measuring a care pathway.

WISeR may ultimately show that carefully governed technology can remove low-value treatment without harming patients. It may show that prior authorization shifts work onto clinicians and deters necessary care. It may show both, depending on the service, vendor, patient, and practice.

The new study tells us the gate is powerful. It does not yet tell us whether the people on the other side are healthier.

Sources

• Neprash HT, Barnett ML. “Changes in Service Use Following the Introduction of AI-Enabled Prior Authorization in Traditional Medicare.” JAMA. Published online October 26, 2026. doi:10.1001/jama.2026.21230.
• Centers for Medicare & Medicaid Services. “WISeR (Wasteful and Inappropriate Service Reduction) Model.”
• Cohen L, Guariglia M. “New Records Obtained by EFF Expose More Problems with Medicare’s AI Experiment.” Electronic Frontier Foundation. October 27, 2026.
• Hughes J. “Prior authorization in the real world: Takeaways from Epic ahead of looming API deadline.” Healthcare Dive. October 26, 2026.

This article is for general information and does not provide medical or legal advice.

Free Daily Briefing

The Latest Longevity Science.
Delivered Every Morning.

Join researchers, physicians, and health professionals getting daily breakthroughs in AI-driven medicine, epigenetics, and longevity research.

Support the research that powers this editorial

Subscribe Free

No spam. Unsubscribe anytime. We respect your inbox.

Keep Reading on Healthcare Discovery

• The Science of Breath: How Vagus Nerve Research Is Rewriting Stress, Recovery, and HRV in 2026
• The Universal Code of Aging: How a Nature Study Mapping 11,000 Gene Profiles Is Rewriting Longevity Science
• Ultra-Processed Food Is Rewiring Your Brain: What 385,000 People Reveal About Diet, Depression, and the Gut-Brain Axis