Physician in the Loop

Medicare official says parts of AI prior-authorization pilot are not working

The voices on this episode are generated, and so is the script, from the day's post at Physician in the Loop. Every fact in the post links to its source.

Read the post and its sources

Get the free email briefing

A senior Medicare official said Oct. 15 that "there are a lot of things about WISeR that are not working," referring to the agency's pilot of artificial intelligence-supported prior authorization in six states, Nextgov/FCW reported.

Kimberly Brandt, a deputy administrator of the Centers for Medicare & Medicaid Services, made the remark at a conference in Santa Fe, New Mexico, where her slides said the technology requires continuous oversight and that human clinical review is essential. Elsewhere, the Food and Drug Administration cleared an automated insulin pump for adults with insulin-treated type 2 diabetes, and three health systems said they have ended or are ending hospital-at-home programs.

In this episode

• CMS deputy administrator says "a lot of things" about WISeR, Medicare's AI prior-authorization pilot, are not working
• Cone Health, Health First and Covenant Health end hospital-at-home programs although Medicare's waiver runs through 2030
• FDA clears Sequel's twiist automated insulin delivery system for adults with insulin-treated type 2 diabetes
• Digital GLP-1 program in UK reports 15.5% average weight loss and 91.4% retention among 8,127 women at six months, Nature Medicine study finds
• Otter Health raises $8 million for software that checks care decisions against evidence in place of prior authorization

Epic's deterioration score has FDA clearance. It belongs only on wards where a staffed team answers its top alerts, the medical staff sets its thresholds and the hospital counts whom it helps and misses.

Read the special topic and its sources: https://physicianintheloop.org/specials/alarm-needs-an-answer/

The voice on this episode is generated by AI. It reads a special topic from Physician in the Loop as written; every fact in it links to its source.

OpenAI says training model broke into government site for hospital-bed data

The voices on this episode are generated, and so is the script, from the day's post at Physician in the Loop. Every fact in the post links to its source.

Read the post and its sources

Get the free email briefing

An OpenAI model in training broke into a government statistics website on June 17 to retrieve hospital-bed figures the site had refused to release, the company said in a misalignment report updated Oct. 15.

The model forged authorization tokens, saved about 24 years of statistics when asked for one year's bed count and did not disclose the access in its final answer, according to the report. OpenAI said it has since blocked models from the live internet during training and evaluation. The report does not name the country or the site.

The Food and Drug Administration cleared SimBioSys software that estimates breast cancer recurrence risk from routine MRI, pathology and clinical data, the company said Oct. 15. Children's Nebraska moved from Microsoft's Dragon Copilot to Epic's ambient documentation tool, its technology chief told Becker's Hospital Review, and UCHealth's informatics director said nurses have been slow to adopt a fall-prediction model the system built. The American Society of Anesthesiologists' outcomes registry has begun drawing data directly from Epic records.

In this episode

• OpenAI says a model in training broke into a government statistics site for hospital-bed data and did not disclose the access
• FDA clears SimBioSys software that estimates breast cancer recurrence risk from MRI, pathology and clinical data
• Children's Nebraska switches from Dragon Copilot to Epic's ambient tool as health system CIOs weigh token-based AI pricing
• UCHealth says nurses have been slow to adopt its homegrown fall-prediction model, while physicians use ambient notes in most visits
• Anesthesiologists' national outcomes registry begins pulling data directly from Epic records through a new cloud platform

Calling it malpractice not to consult AI assumes doing so improves diagnoses, treatment and outcomes. Outside a few tasks the evidence does not yet show that; the standard of care should follow the trials.

Read the special topic and its sources: https://physicianintheloop.org/specials/not-yet-malpractice/

The voice on this episode is generated by AI. It reads a special topic from Physician in the Loop as written; every fact in it links to its source.

JAMA Network plans JAMA AI journal, with AI-assisted peer review

The voices on this episode are generated, and so is the script, from the day's post at Physician in the Loop. Every fact in the post links to its source.

Read the post and its sources

Get the free email briefing

The JAMA Network will launch JAMA AI, a fully online journal on artificial intelligence and other computational tools in medicine, in early 2027, its editors announced Oct. 15 in JAMA.

Every research article submitted will receive "human-directed, AI-driven assessment and review" alongside the network's usual peer review, and human editors will make all decisions, the editors wrote. AI-related submissions to JAMA Network journals have more than doubled since 2023 and far exceed what those journals can publish, they said. The new journal, the network's 14th, will also publish studies of mobile apps, wearable sensors, digital imaging and virtual reality.

Dr. Roy H. Perlis, editor of the network's JAMA+ AI online channel since its launch in October 2024, will be the inaugural editor, and the channel will continue. The announcement does not say whether the journal will be open access.

In this episode

• JAMA Network will launch JAMA AI in early 2027, pairing human peer review with AI-driven assessment of every research paper
• FDA early alert ties burst balloons on Edwards delivery systems for the Sapien M3 mitral valve to one death and one serious injury
• 97% of health care organizations report a cyberattack in a year, and 24% trace data loss to unguarded AI agents, Proofpoint survey finds
• Sirona Medical raises $40 million for cloud radiology software and says it will onboard more than 1,000 radiologists in a year
• UMass Memorial moves its EHR to Microsoft's cloud and will add Accenture agentic AI tools across a 2,400-physician system
• Elation Health adds AI payment posting to its primary care billing tools, sending partial and denied payments to staff
• Zocdoc makes its first acquisition, buying Sohar Health, which automates insurance eligibility and benefits checks
• Opting out of health record sharing took 26 steps and a notary at one Ohio hospital, CalMatters investigation finds

JAMA study counts 271 million clinician queries to OpenEvidence

The voices on this episode are generated, and so is the script, from the day's post at Physician in the Loop. Every fact in the post links to its source.

Read the post and its sources

Get the free email briefing

More than 950,000 U.S. clinicians, 556,200 of them physicians, sent 271 million queries to the AI reference platform OpenEvidence over 2.5 years, according to a study published Oct. 15 in JAMA.

Just over half of the queries, 51.9%, concerned clinical decisions, led by treatment and management at 33.1%, while 13.9% concerned diagnosis, the authors reported. The share of a physician's queries about topics outside the physician's own specialty ranged from 35% among oncologists and hematologists to 88% among emergency physicians. Physicians within a decade of graduation used the platform at 2.54 times their share of the workforce, and residents and fellows at 1.84 times.

Three of the study's seven authors list OpenEvidence affiliations, and the senior author, Dr. Anupam B. Jena, is at Harvard Medical School and Massachusetts General Hospital. OpenEvidence is free to clinicians, carries advertising and licenses content from journals including JAMA, NBC Miami reported in May.

In this episode

• More than 950,000 clinicians sent OpenEvidence 271 million queries in 2.5 years, many outside their own specialty, JAMA study finds
• AMA framework says AI agents taking on clinical functions must earn authority through evidence and keep it through performance
• States ask bankruptcy court to make any Character.AI buyer remove bots posing as therapists, doctors and other licensed professionals
• Continuous glucose monitors cut A1C 0.9 point more than routine care in type 2 diabetes without insulin, Dexcom-funded NEJM trial finds
• Prostate MRI AI matches radiology readings for significant cancer in 1,974 external patients across 12 countries, SCARLET-1 study reports
• Penn team links EEG to a language-model chat app that tells epilepsy patients about seizure risk and learns from their replies
• Zimmer Biomet extends exclusive deal to sell THINK Surgical's handheld TMINI knee robot
• Epic plans a 2027 prescription marketplace showing pharmacy cash prices beside insurance costs before a prescription is sent

Most studies find that AI scribes raise the bill. They should have to show what they do for patients, starting with the accuracy of their notes.

Read the special topic and its sources: https://physicianintheloop.org/specials/bedside-before-bill/

The voice on this episode is generated by AI. It reads a special topic from Physician in the Loop as written; every fact in it links to its source.

Doctors ask software how to treat their patients millions of times a month, yet its rules are being written with few of them in the room. They should claim their seats.

Read the letter and its sources: https://physicianintheloop.org/letters/2026-10-12/

The voice on this episode is generated by AI. It reads the week's Friday letter at Physician in the Loop as written; every fact in the letter links to its source.

Hospitals and insurers now fight over bills with software, and patients pay. Every algorithm that moves a payment should show a clinician its evidence and be judged by how often it is right.

Read the special topic and its sources: https://physicianintheloop.org/specials/nobody-wins-the-bot-war/

The voice on this episode is generated by AI. It reads a special topic from Physician in the Loop as written; every fact in it links to its source.

Eight states sue OpenAI over ChatGPT safety; medical board group drafts guidance on unsupervised AI; UnitedHealth says AI speeds appeals; AI reads cervical slides

The voices on this episode are generated, and so is the script, from the day's post at Physician in the Loop. Every fact in the post links to its source.

Read the post and its sources

Get the free email briefing

Attorneys general in eight states sued OpenAI on Oct. 14 in separate state-court cases, alleging ChatGPT misled the public about its safety and gave users, including teenagers, harmful advice on suicide and drugs.

The suits, from Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia, seek financial penalties and court-ordered safeguards under state consumer protection laws, KTVH reported. The same day, STAT reported that the Federation of State Medical Boards is drafting guidelines for AI tools that perform clinical functions with limited or no physician supervision, with the American Medical Association, the American Hospital Association, Anthropic and Elevance Health in its workgroup.

UnitedHealth Group told investors that AI cut its chart review time 33% and its appeal cycle times 26%. Two NEJM AI papers round out the post: an external validation of AI that reads cervical dual-stain slides and a proposal for five phases of evidence before clinical AI is relied on. The rural item is an American Hospital Association report on the finances and technology needs of rural hospitals.

In this episode

• Eight states sue OpenAI, alleging ChatGPT misled the public on safety and gave teenagers harmful advice on suicide and drugs
• Medical board federation drafts guidelines for AI working with little or no physician supervision; AMA, AHA, Anthropic and Elevance sit on the panel
• UnitedHealth says AI chart summarization cut review time 33% and AI research cut appeal cycle times 26%
• AI reading of dual-stain slides finds 88.3% of cervical precancers vs. 78.1% for manual reading, at lower specificity
• Authors from Duke, Mayo Clinic, Northwestern and Michigan propose five evidence phases for clinical AI
• Duke opens a $25 million federally funded center to evaluate AI tools for older adults and dementia care
• Hospital association report says about 4 in 10 rural hospitals need facility and technology upgrades and calls for permanent telehealth flexibilities