A retired shop teacher in Spokane did not expect a computer to have the last word on his care. He had already sat with his physician, reviewed the images, and agreed on a procedure his doctor believed was necessary. Then a letter arrived. The request had been turned down. Across the state, Medicare AI denials are no longer an abstract policy fight. Documents described by the Seattle Times show that a new prior authorization program rejected more claims than it approved, a result that has left patients, clinics, and lawmakers asking who is actually practicing medicine.
What the Washington documents reveal
The most striking fact is also the simplest. According to records reviewed by the Seattle Times, the tool said no more often than it said yes. The newspaper’s account describes a denial pattern steep enough to alarm people who thought prior authorization would be a narrow screen for waste, not a default barrier. A program that rejects more than it clears is not a quiet administrative tweak. It is a change in the odds that a recommended service will be paid.
Readers should hold that comparison carefully. Public documents do not, by themselves, prove that every rejected request was medically sound. They also do not prove that the approvals were wise. What they do show is a tilt. When a new system begins life by denying more often than it grants, the burden of explanation belongs to the people who built it.
How prior authorization reached traditional Medicare
Prior authorization is familiar to anyone who has fought an insurer over an MRI, a stay in a skilled nursing facility, or a course of physical therapy. A clinician asks permission. A reviewer, sometimes a nurse and sometimes a contractor, decides whether the service meets a set of rules. For years that machinery sat mostly in private Medicare Advantage plans. Traditional Medicare, the federal program older adults still think of as the straightforward version, used it far more sparingly.
That boundary is moving. Federal officials have argued that selected services are overused and that technology can sort appropriate care from spending that does not help patients. The promise is speed and consistency. The fear, now visible in Washington, is that consistency can mean consistent refusal. A rule applied at scale does not become fair merely because a machine applies it the same way every time.
What an algorithm can see and what it cannot
An automated review can read codes, dates, and checklist criteria. It can flag a request that lacks a required note. It cannot sit in the exam room when a patient describes pain that keeps her from sleeping, or notice that a standard alternative already failed. Medicine is full of cases that meet the spirit of a guideline and miss the letter, or the reverse.
That gap is why physicians bristle when a denial arrives with little clinical reasoning attached. A code string is not a conversation. If Medicare AI denials rest on incomplete charts, the fix may be better documentation. If they rest on rules that ignore ordinary medical judgment, documentation will not save the patient. The documents now public do not settle that distinction. They make it urgent.
Patients meet the letter, not the reviewer
For a person on a fixed income, a denial is not a data point. It is a postponed surgery, a longer stretch of pain, or a bill that may arrive anyway if the family proceeds without coverage. Many beneficiaries are in their seventies or eighties. Some have limited vision, unreliable internet, or adult children who live in another state. The appeal window does not pause for any of that.
I have watched relatives treat official mail as something to fear rather than something to answer. They assume the government has already decided. That instinct is understandable, and it is also how a high denial rate becomes a quiet one. People who never appeal do not appear in the success stories of a review process. They simply go without.
Clinics absorb a second job
Doctors and their staff now spend hours translating a human visit into language a screening tool will accept. The work is not glamorous. It is phone queues, portal messages, repeated faxed notes, and peer conversations that may never happen. Small practices feel it first. A rural clinic cannot hire a department devoted to authorization the way a large hospital system can.
There is a bitter irony here. A tool sold as efficiency can create more labor on the clinical side while removing labor from the review side. The savings, if they exist, show up in a federal ledger. The cost shows up as a medical assistant staying late and a patient waiting another week to learn whether treatment can begin.
Denials, appeals, and the clock
A denial is not always the end. Federal rules give beneficiaries paths to challenge a decision, and some challenges succeed once a human clinician looks again. That fact is used, fairly, by defenders of the model. It is also incomplete. An appeal that works after a month still delayed care. An appeal that works only for people with a persistent adult child is not an equal right. And an appeal that almost no one files cannot launder a rejection rate that exceeds the approval rate.
Speed matters as much as the final yes or no. Pain, infection risk, and loss of function do not wait for a queue. If the Washington program is fast at denying and slow at reconsidering, the design itself tilts against the patient even when the written rules look neutral.
The contractors behind the screen
Automation in public health programs rarely means a single government computer in a federal office. It usually means vendors, subcontractors, and proprietary criteria that clinics cannot fully inspect. When a denial cites a guideline, the patient deserves to know which guideline, who wrote it, and whether a clinician with relevant training reviewed the case.
Secrecy is not a technical necessity. It is a choice. If a company profits from processing reviews, the public should be able to see denial rates by service, by county, and by how often an appeal overturns the first decision. Washington’s documents are a start. They should not be the whole record.
Waste is real, and so is underuse
Honest reporting has to grant the other side of the argument. American health care does waste money. Some procedures are ordered out of habit. Some imaging adds little. Taxpayers and beneficiaries share the bill through premiums and general revenue. A program that never questions a request would not be a serious steward of Medicare.
The question is proportion. A screen aimed at clear waste should not deny more often than it approves unless the underlying requests are truly that poor. Nothing in the public description of these documents suggests officials have made that case in plain language. Until they do, the staggering pattern looks less like precision and more like a blunt instrument.
Washington as a warning for other states
Pilot programs are how Washington, D.C., tests ideas before they spread. That is the polite version. The less polite version is that people in the test states absorb the mistakes. If Medicare AI denials at this volume are treated as a successful proof of concept, other states will inherit the same letters, the same delays, and the same confused clinics.
Lawmakers in the other Washington should ask for the full tables, not a press summary. How many requests, by service type? How many denials were later reversed? How long did each step take? Were rural patients denied at different rates than urban ones? A model that cannot answer those questions is not ready to travel.
What beneficiaries can do this week
Policy arguments do not schedule a procedure. People dealing with a denial can still take practical steps. Read the letter for the exact reason and the deadline. Ask the doctor’s office whether a missing note, rather than a medical disagreement, triggered the refusal. Request the criteria used. File the appeal in writing and keep copies. If the delay threatens health, ask the clinic whether an expedited review is available.
None of that restores the promise that Medicare would be simpler than private insurance. It does keep a refusal from becoming final by default. Family members can help most by treating the paperwork as time sensitive, not as junk mail.
Trust is the scarce resource
Medicare has endured because older Americans believe the program will be there, and will be legible, when they are sick. A black box that denies more than it approves spends that trust quickly. Officials can rebuild it only with numbers they are willing to publish, clinicians in the review loop, and a standard that treats a wrongful delay as seriously as a wrongful payment.
The Seattle Times documents do not close the case. They open it. Medicare AI denials in Washington have already shown what happens when software meets real patients without enough public light. The next disclosure should not require another leak or another records request. It should be the ordinary habit of a public program that asks permission to touch people’s care.