AI-assisted prior authorisation cut Medicare services by up to a fifth: what the JAMA WISeR study shows, and what it does not
A research letter published in *JAMA* on 26 October 2026 gives the first evidence on WISeR, the US Medicare pilot that uses technology companies, encouraged to apply artificial intelligence, to pre-approve selected services. In the six pilot states, use of the targeted services fell by about 15% to 20% in the first six months compared with the rest of the country. The study cannot say whether the care that disappeared was wasteful or needed, and that unanswered question is exactly the one European health systems and AI developers should focus on.
Prior authorisation requires a clinician to justify a service before the payer agrees to cover it. It is routine in US private insurance and in Medicare Advantage, but traditional Medicare has historically almost never used it. That changed on 1 January 2026, when the Centers for Medicare & Medicaid Services (CMS) launched the Wasteful and Inappropriate Service Reduction (WISeR) model.
What WISeR is
According to CMS, WISeR runs for six performance years, to 31 December 2031, in Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington. It applies only to traditional Medicare, not to Medicare Advantage. The study authors describe it as requiring prior authorisation for 41 services considered vulnerable to fraud, waste and abuse; CMS gives examples such as skin and tissue substitutes, electrical nerve stimulator implants and knee arthroscopy for osteoarthritis. Emergency and inpatient-only services are excluded.
The participants are six technology companies. They use AI and machine learning to review medical necessity, must employ licensed clinicians to make any recommendation for non-payment, and are paid a percentage of the spending on wasteful or inappropriate care they help avoid, adjusted for performance measures. That payment design is central to the controversy around the model.
What the study did
Hannah Neprash of the University of Minnesota and Michael Barnett of Brown University used Epic Cosmos, an electronic health record dataset covering more than 300 million patient records. They compared monthly encounters for WISeR-targeted services among traditional Medicare patients in the six pilot states with those in the other 44 states and the District of Columbia, across the two years before launch and the first six months after it.
As a check, they also examined services that had originally been slated for WISeR but were never implemented. If some unrelated trend had been driving down use, those services should have fallen too.
What it found
The results below are as reported by the authors and their institutions; the full research letter is behind the journal's paywall.
- Epidural steroid injections, the highest-volume service in the model, fell by about 15% relative to trends in non-pilot states.
- The other targeted services fell by nearly 20%.
- The effect weakened over time. For epidural steroid injections, the reduction was about 21% in the first quarter of 2026 and about 9% in the second. For other services it moved from about 21% to 18%.
- The check held. Services that were planned for WISeR but never implemented showed no significant decline, which supports the conclusion that prior authorisation itself drove the change.
The lead author put the central uncertainty plainly: "The key question we can't yet answer is whether the care not received was wasteful or necessary."
Limitations
Volume, not value. The study measures how many services were delivered. It does not measure pain, function, complications, delayed diagnoses or costs, so it cannot distinguish appropriate deterrence from harmful delay.
Observational design. States were not randomised. The comparison with non-pilot states and the never-implemented services make a causal reading plausible, but other state-level changes cannot be excluded entirely.
Six months of a six-year model. The fading effect for epidural injections may reflect early administrative friction being resolved or clinicians learning which requests succeed. Either way, the long-term effect is unknown.
The AI component is not isolated. The study evaluates the policy as a whole. It does not tell us how much of the effect came from algorithmic screening and how much from the simple existence of a pre-approval step.
Data coverage. Epic Cosmos is very large but includes only organisations that use Epic, so some care is not captured.
What it means for clinicians and patients
For clinicians in pilot states, the study confirms what many have reported: a new approval step changes practice quickly. For patients, the important question is whether the services they did not receive were ones they needed. That question remains open, and the authors call for research on appropriateness, outcomes, clinician burden and Medicare spending before CMS decides whether to continue or expand the model.
Why it matters in Europe
European health systems face the same pressures that produced WISeR: rising demand, constrained budgets and evidence that some procedures are overused. AI-assisted review of claims and referrals is an obvious temptation. The EU, however, has already set rules that would apply to a comparable scheme.
The AI Act. Annex III of Regulation (EU) 2024/1689 lists as high-risk AI systems used by or on behalf of public authorities to evaluate people's eligibility for essential public assistance benefits and services, including healthcare services, and to grant, reduce, revoke or reclaim them. High-risk status brings obligations on risk management, data quality, documentation, human oversight and accuracy. Following the Digital Omnibus on AI, Regulation (EU) 2026/1744, these Annex III obligations now apply from 2 December 2027.
The GDPR. Article 22 gives individuals the right not to be subject to a decision based solely on automated processing that significantly affects them, subject to limited exceptions and safeguards. WISeR's requirement that clinicians make the non-payment recommendation reflects a similar instinct, but European regulators will look at whether human review is meaningful or merely nominal.
Incentives. A payment model that rewards the reviewer for avoided spending would draw close scrutiny in Europe. The WISeR study shows why: an approval step can reduce volume quickly, and without outcome data no one can tell whether the reduction is a saving or a cost shifted onto patients.
What it means for industry
For manufacturers of devices and procedures that appear on prior-authorisation lists, the lesson is direct. Utilisation can fall by a fifth within months, regardless of a product's merits, when a payer adds a review step. The best defence is evidence that defines which patients benefit, with selection criteria a reviewer, human or algorithmic, can apply. For developers of AI review tools, the lesson is that regulators will ask not only how many requests a system screens but how often it is right, and for whom.
What this means for your organisation
WyteRock's AI co-pilot explainer sets out the principle that should govern any AI used in healthcare decisions: the human is the pilot and AI is the co-pilot, and every output is a consideration to check, not an order to follow. For manufacturers whose products face utilisation review, our clinical evidence planning service designs studies that establish appropriate use. If you need one scoped answer, such as how the AI Act's high-risk rules would apply to a specific payer-side or referral-triage tool, the fixed-price Task Request service is designed for it.
Alex Mansour is CEO of WyteRock Global, a France-based life-sciences advisory firm that supports manufacturers, health systems and investors on clinical AI, clinical evidence planning and EU market access.
Sources
- Neprash HT, Barnett ML. Changes in Service Use Following the Introduction of AI-Enabled Prior Authorization in Medicare. JAMA. Published online 26 October 2026. doi:10.1001/jama.2026.21230. https://jamanetwork.com/journals/jama/fullarticle/10.1001/jama.2026.21230 (PubMed: https://pubmed.ncbi.nlm.nih.gov/42932180/)
- University of Minnesota School of Public Health. "Medicare's prior-authorization pilot test led to an immediate drop in use of services, study finds", 26 October 2026. https://www.sph.umn.edu/news/medicares-prior-authorization-pilot-test-led-to-an-immediate-drop-in-use-of-services-study-finds/
- Medical Xpress. "Medicare's prior-authorization pilot test led to an immediate drop in use of services", 27 October 2026. https://medicalxpress.com/news/2026-10-medicare-prior-authorization.html
- DocWire News. "Medicare's AI prior auth pilot tied to 15%–20% drop in targeted services", October 2026. https://www.docwirenews.com/post/medicares-ai-prior-auth-pilot-tied-to-15-20-drop-in-targeted-services
- Centers for Medicare & Medicaid Services. Wasteful and Inappropriate Service Reduction (WISeR) Model. https://www.cms.gov/priorities/innovation/innovation-models/wiser
- Rathi VK, Dhruva SS, Ross JS. Expanding Prior Authorization in Traditional Medicare—The WISeR Model. JAMA. 2025. doi:10.1001/jama.2025.16326. https://doi.org/10.1001/jama.2025.16326
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 6(2) and Annex III, point 5(a). EUR-Lex: https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
- Van Eecke P, Regenhardt B. "Digital AI Omnibus Delays Key Deadlines, Introduces New Rules". Cooley, 3 August 2026 (on Regulation (EU) 2026/1744). https://cdp.cooley.com/digital-ai-omnibus-delays-key-deadlines-introduces-new-rules/
- Regulation (EU) 2016/679 (General Data Protection Regulation), Article 22. EUR-Lex: https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng
Get new articles by email. Sign up for the newsletter or join the CentiAgent Club; members receive an alert for every new article.