What is WISeR?
The Wasteful and Inappropriate Service Reduction (WISeR) model was launched on January 1, 2026, by the Centers for Medicare & Medicaid Services (CMS) and is planned to run through the end of 2031. It aims to use advanced technology, artificial intelligence, and machine learning in the prior authorization process for specific Original Medicare services in the states of Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington.
Which services does the model target?
CMS states that the model targets specific services with a high risk of abuse, unnecessary utilization, or inappropriate application in particular. Examples include skin and tissue substitutes, certain electrical nerve stimulator implants, and arthroscopy for knee osteoarthritis. Emergency services, inpatient-only procedures, and services where delay would pose serious risk are excluded from the scope.
What did STAT’s documents reveal?
According to a report by STAT dated September 15, 2026, more than a thousand pages of documents and data obtained through an Electronic Frontier Foundation Freedom of Information Act lawsuit revealed that technical and operational issues occurred during the initial implementation of the pilot. The report noted that some care decisions were delayed and that a technology vendor warned that delivering a fully functioning product on the timeline requested by CMS was not realistic.
Why can delay be a clinical safety issue?
Although prior authorization may appear to be an administrative process, its outcome can affect the timing of a patient’s access to treatment. 2026 reports cited from hospitals in Washington State alleged that wait times for certain services increased two- to four-fold compared to the previous process. Such findings show that algorithmic efficiency metrics alone are not sufficient; the clinical impact of delay must also be monitored.
What is CMS’s official safety mechanism?
CMS states that under WISeR, technology companies may analyze requests; however, every decision to not approve a request must be reviewed by a licensed human clinician with relevant clinical expertise. In addition, resubmission, peer-to-peer review, and existing appeal rights are preserved. This structure mandates human oversight, at least at the design level.
What exactly does human-in-the-loop mean?
In the human-in-the-loop approach, artificial intelligence is not the ultimate authority, but a decision-support layer. The system can detect missing documentation, pre-assess alignment with criteria, or flag high-risk situations; however, clinical context, exceptions, and patient-specific circumstances are evaluated by a human professional. The critical point is that the human is not merely a symbolic figure who “presses the approval button,” but an authorized person who can truly question and modify the decision.
Risk of automation bias
Human oversight alone is not sufficient. If the clinician accepts the algorithm’s recommendation as correct by default, “automation bias” can occur. For this reason, interface design, displaying the model’s uncertainty, transparency of the data used, and presenting alternative explanations are important. Human oversight becomes a safety layer only when genuine cognitive independence is achieved.
Which metrics should be monitored?
In clinical AI systems, not only accuracy or processing speed, but also false rejection rate, false approval rate, performance differences across subgroups, delay durations, re-evaluation rates, clinician override frequency, and actual patient outcomes should be monitored. In addition, performance should be re-validated as model versions change.
Explainability and auditability
It must be clear which data a system uses, which clinical rule it applies, and how the decision can be reviewed. Particularly in high-impact areas such as reimbursement, insurance, or treatment access, the explanation “the algorithm said so” is not sufficient. The decision must be auditable retrospectively, and errors must be analyzable systematically.
Should artificial intelligence replace the doctor?
Current evidence and regulatory approaches point not toward a fully autonomous “AI replacing the doctor” model in high-risk clinical decisions, but rather toward systems that support the clinician’s decision quality and speed. Artificial intelligence can be very powerful in image analysis, risk stratification, documentation, and knowledge synthesis; however, patient values, clinical exceptions, multimorbidity, and uncertainty management require human judgment to remain central.
Our approach at our Bodrum clinic
We position artificial intelligence not as an authority that substitutes for clinical thinking, but as an auxiliary system that strengthens the physician’s decision. Organizing data faster and highlighting patterns that might otherwise be overlooked is valuable; the final clinical interpretation must be made within the physician’s responsibility, along with the patient’s history, physical examination, laboratory, and imaging findings.
References
- STAT, September 15, 2026: Implementation issues and delays in the Medicare WISeR pilot
- CMS: Official WISeR Model page
- CMS: WISeR FAQ — human clinician review, timeframes, and appeals processes
- CMS: Electronic prior authorization and medical professional oversight principles
Last medical review: September 18, 2026
