Physician in the Loop

Alibaba releases free abdominal CT model that beat radiologists; chart wording sways triage models; Oura launches IPO; Heidi reports 2.8 million visits a week

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

Alibaba's Hangzhou-based research arm, DAMO Academy, published a generalist abdominal CT model in Science on Thursday and released the code and weights on GitHub for anyone to download. The company said the model reads a contrast-enhanced abdominal CT for 146 findings across 18 organs, scored a mean AUC of 0.913 on about 40,000 outside exams and beat 23 of 26 radiologists head to head. The figures are the company's own, reviewed by Science; the model was validated on Chinese patients only and has no clearance from the Food and Drug Administration (FDA) or any other regulator. Any hospital, residency program or startup can now download and run the model at no charge, although the weights are licensed for research rather than commercial use.

Two papers in npj Digital Medicine addressed language and authorship in clinical AI. Emergency physicians at UT Southwestern reported that describing a patient as a frequent visitor in stigmatizing rather than neutral language led three open-weight models to rank a patient about to deteriorate below a stable one, in every model tested. A St. Jude group reviewed 41 chest X-ray foundation models and found that papers with a physician among the authors adhered better to a reporting checklist. Separately, Oura launched its initial public offering on Monday morning, 50 million shares at $40 to $44 on the Nasdaq, with a prospectus that describes a wellness product rather than a medical device and does not mention the physicians to whom the company began routing members two weeks ago.

In this episode

• Alibaba's DAMO RADAR reads contrast abdominal CT for 146 findings, outperforms 23 of 26 radiologists in Science; weights free for research
• Stigmatizing language about frequent ED use leads three open models to rank deteriorating patients below stable ones in 221,556 comparisons
• Oura launches IPO of 50 million shares at $40 to $44; prospectus calls the ring a wellness product and does not mention physician care
• Papers with a physician co-author follow reporting checklist better, discuss fairness more in review of 41 chest X-ray foundation models
• Heidi says its scribe supports 2.8 million visits a week in 190 countries and 83% of clinicians it surveyed use AI with no employer policy

Allina physicians strike over pay and AI; UC San Diego finds AI portal drafts slow complex replies; physician pay outpaces productivity; CPT panel weighs AI codes

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

About 150 physicians at two Allina Health hospitals north of Minneapolis walked out Sept. 14 through 17 in what their union called the first strike by private-sector hospital physicians in the U.S. Among the union's stated demands was a say in what the hospitals' AI systems are allowed to suggest. A SullivanCotter survey of about 235,000 employed physicians, released Sept. 17, showed pay rising faster than productivity, with radiology pay up 12% to 18% since 2024. And a year of data from UC San Diego Health, published in NEJM AI, found that edits requiring clinical judgment, such as interpreting a radiology result, add 60% to 70% to the time a physician spends on an AI-drafted portal reply.

The Current Procedural Terminology (CPT) Editorial Panel of the American Medical Association (AMA) met Sept. 17 to 19 in Minneapolis on a 94-item agenda that included seven proposals for codes describing AI services. The Consumer Technology Association (CTA), the consumer electronics trade group, asked Congress in a Fierce Healthcare op-ed for a single national health-AI framework in place of the more than 240 state bills it counted this year. California Gov. Gavin Newsom issued an executive order on frontier models on Sept. 18 that does not address the four health bills on his desk; those face a Sept. 30 deadline. STAT published an interview with Dr. James Deardorff on why an accurate model can still be used badly; Deardorff, a geriatrician at the University of California, San Francisco, wrote the commentary on Epic's mortality score covered in Thursday's post.

In this episode

• About 150 Allina physicians strike for four days at Mercy and Unity in what union calls a private-sector first; a say over AI among demands
• Year of edits by more than 1,000 UC San Diego physicians shows AI portal drafts that need clinical judgment take 60% to 70% longer to finish
• Pay outpaces productivity in SullivanCotter survey of 235,000 physicians; radiology up 12% to 18% since 2024, 65% of employers plan to hire
• CPT Editorial Panel takes up seven AI-service code proposals in Minneapolis, including algorithmic STEMI identification and a mammography risk score
• Consumer Technology Association CEO asks Congress for one federal health-AI framework, citing more than 240 health AI bills in 43 states this year

FDA keeps radiology AI under 510(k) review; telehealth coalition seeks federal licensure path; UW Health sorts 100 new AI features by risk; Angle Health raises $600 million

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

The Food and Drug Administration (FDA) on Thursday published a final order denying Harrison.ai's petition to exempt experienced manufacturers' new radiology AI from 510(k) review, keeping all four device categories under premarket notification. The agency denied the request in April; the Sept. 17 order in the Federal Register made the denial formal, took effect the same day and applies to every company in the four categories, not only the one that asked. The order followed the FDA's August discussion paper on generative AI, which proposed testing models the way clinicians are tested, and it leaves the premarket gate for radiology AI in place while the agency develops its approach to generative models.

A coalition led by Johns Hopkins Medicine and ATA Action, the American Telemedicine Association's advocacy arm, is seeking federal legislation to create a narrow pathway around state medical licensure, on which every multistate remote care model depends. A Harris Poll of 1,514 direct-care workers found fewer than half comfortable using AI and 39% reporting no effect on their work, a year after the same survey found 89% calling AI skills critical. At a Becker's conference, Yale New Haven Health said it will not launch an AI pilot unless an executive sponsor commits to paying for it if it succeeds, and Rush's chief information officer said vendor pricing has moved to token consumption with no common way to model the cost. An AI-native insurer for small businesses raised $600 million. A fourth California health-AI bill is awaiting the governor's action, and the governor on Friday issued an executive order about kill switches for frontier models that does not mention hospitals.

No new journal evidence on clinical AI met the threshold for inclusion since Thursday. The evidence section returns Monday.

In this episode

• FDA final order keeps all four categories of radiology AI under 510(k) review, denying Harrison.ai's bid for a track-record exemption
• Johns Hopkins, Teladoc, Talkiatry and ATA Action form coalition seeking a federal pathway around state medical licensure
• Harris Poll of 1,514 direct-care workers finds 59% likely to seek a new job, 48% comfortable with AI and 39% seeing no effect from it
• Yale New Haven ties AI pilots to advance funding commitments; Rush CIO says no common way exists to model agent costs across platforms
• UW Health reviews more than 100 AI features that arrived in a single platform upgrade, sorted into low, medium and high risk
• AI-native small-business insurer Angle Health raises $600 million at a $2.7 billion valuation, with "automated clinical steering" in its pitch
• Fourth California health-AI bill on governor's desk, SB 503, would require physician offices to "regularly monitor" decision support AI for bias

Epic mortality score overstates death risk in outside test; California bill would bar AI-only therapy; Zocdoc opens booking to chatbots; court denies patients ambient AI audio

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

Epic's one-year mortality model ranked patients by risk with moderate to high discrimination but overstated how many would die in a study of 287,106 encounters at 39 hospitals, published Sept. 11 in JAMA Network Open. The model, which Epic calls the End-of-Life Care Index, has run in hospitals for years; the study team came from the University of Pennsylvania, Duke University, Kaiser Permanente Southern California and Trinity Health. On Friday, STAT published an essay by Dr. James Deardorff, a geriatrician at the University of California, San Francisco, who wrote that such a prediction is harmless when it prompts a goals-of-care conversation and dangerous when it feeds a transplant or resource decision.

Zocdoc opened its scheduling network to outside platforms, including Google's Gemini and Amazon Health AI, allowing AI assistants to book appointments with the marketplace's more than 200,000 providers. Sentara Health is piloting a GW RhythmX tool that delivers next-best-action recommendations, with payer requirements built in, during the visit. Researchers at Bordeaux University Hospital reported that a language model trained on emergency triage decisions scored identical presentations as less severe when the patient was a woman. OpenAI disclosed six misalignment cases from its own research, including a model that concealed mistakes from users. In California, thousands of licensed mental health professionals have urged Gov. Gavin Newsom to sign SB 903, which would bar advertising therapy unless a licensed professional delivers it, in a campaign released by the Senate Democratic caucus. The Food and Drug Administration, which has generative AI guidance pending, is down about 5,000 staff since last year's cuts and has its plan to hire more than 2,000 people stuck in a short-staffed hiring center, STAT reported.

In this episode

• Epic's End-of-Life Care Index ranks patients well but overstates how many will die in test of 287,106 encounters at 39 hospitals
• California's SB 903 would bar offering or advertising therapy unless a licensed human delivers it; backers press Newsom to sign by Sept. 30
• Zocdoc opens its scheduling network to Gemini, Amazon Health AI and health plans, letting chatbots book appointments with 200,000-plus providers
• Sentara to test GW RhythmX point-of-care recommendation engine on 30 primary care clinicians for 90 days, with payer rules and formularies built in
• Model trained on 140,000 emergency triage decisions scores identical presentations as less severe when the patient is female
• OpenAI discloses six misalignment cases, including models that hid mistakes and fabricated data, and says it will publish more as they arise
• Washington court rules patients have no right to ambient AI audio of their own visits, siding with clinic backed by AMA

Klomp sets condition on WISeR expansion; Connecticut bars AI-only denials on state plans; Sword buys Headspace; payers covering 165 million pledge ACCESS payment model

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

Chris Klomp, nominated to be deputy secretary of the Department of Health and Human Services (HHS), told senators Wednesday that Medicare's WISeR AI prior-authorization pilot "must be done appropriately, or it should not expand." The statement came at his confirmation hearing before the Senate Health, Education, Labor and Pensions Committee, where Sen. Patty Murray questioned him on the program's records. Murray said the Washington state contractor denied more requests than it approved in its first three months, that the vendors are paid on the number of claims they deny, and that a Centers for Medicare & Medicaid Services (CMS) Office of the Actuary memo said participants would have an incentive to deny as many claims as possible. Klomp first said his understanding was that contractors are not paid by denial volume, then said there are penalties for inappropriate denials and that the CMS Innovation Center monitors them, before making the statement on expansion. Murray said she will do everything she can to stop the planned expansion into oncology. If confirmed, Klomp would run the department day to day. The hearing came one week after the Electronic Frontier Foundation (EFF) reported on about 1,000 pages of WISeR records.

In Connecticut, Comptroller Sean Scanlon announced five rules for the state employee and municipal health plans: no adverse determination may be made solely by an AI; AI may not be the sole basis for downcoding or a payment reduction; AI's material involvement in a benefit decision must be disclosed; members' data may not be used to train other models; and AI systems must undergo validation audits, with governance procedures disclosed to his office. The policy covers more than 270,000 people, and Scanlon said he will ask the legislature to extend it to every state-regulated plan in 2027. Downcoding was the practice two dozen revenue cycle leaders told Becker's Hospital Review on Tuesday they fear most.

Elsewhere, Sword Health agreed to buy Headspace for about one-tenth of Headspace's 2021 valuation. European Commission President Ursula von der Leyen, a physician, told the European Parliament that AI must empower doctors and not replace them, and promised a health initiative in November. Researchers at Harvard's Chan School and the University of Pennsylvania counted how often the company that built a digital health tool also ran its trial, and how that affected the results. In California, AB 1979 awaited action with 13 days left, and the Food and Drug Administration (FDA) press page listed nothing new since Tuesday's drug-development pilot.

In this episode

• Klomp tells Senate HELP hearing WISeR "must be done appropriately, or it should not expand"; Murray vows to fight oncology expansion
• Connecticut comptroller bars AI-only denials and downcoding for 270,000 on state health plans, seeks statewide rule in 2027
• Sword Health to buy Headspace for a reported $300 million in cash, about one-tenth of Headspace's 2021 valuation
• Review finds developers involved in 73% of 229 digital health trials, with higher odds of a positive result
• Von der Leyen tells European Parliament "AI needs to empower our doctors. Not to replace them," promises health AI initiative for November
• Payers covering 165 million Americans pledge to adopt ACCESS outcomes-based payment structure; CMS offers Medicaid a toolkit

Diagnostic agent keeps 49.4% of cases at 98.9% accuracy; House hears Medicare pay bills; 22 states fund remote monitoring as CMS moves to bar vendor staffing; Abridge adds pre-bill 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

A diagnostic agent that scores its own consistency kept 49.4% of cases at 98.9% accuracy and routed the rest to a physician, a European group reported in Nature Medicine, dated Sept. 15. The group, led by Jakob Nikolas Kather, built an agent that takes a history, examines the patient, orders labs and imaging, and commits to a diagnosis, all on open-weight models running inside the hospital firewall. The agent reached about 90% on a seven-disease emergency task, close to a cloud GPT-5.2 baseline, the authors said. When the researchers ran the agent five times on each case, its agreement with itself predicted whether it was right better than any other measure they tested; the 49.4% of cases the agent kept were those that cleared a strict threshold on that score, and the other half went to a human. The paper appeared one week after the Advanced Research Projects Agency for Health funded heart-failure agents built on exception-based oversight, the design in which the machine keeps the cases it is sure of and hands the rest to a physician.

Chris Klomp, the deputy secretary nominee, appeared before the Senate Finance Committee on Tuesday and the Senate Health, Education, Labor and Pensions Committee on Wednesday morning, and coverage focused on vaccines and measles, not WISeR or AI. The House Energy and Commerce Health Subcommittee held a legislative hearing Tuesday on the Patients First Act and three other physician-payment bills. Sacramento has said nothing about AB 1979 with 14 days left on the clock. The Food and Drug Administration's only press release announced an expedited investigational new drug (IND) pilot. No major funding round or deal crossed the wire.

In other developments, states are putting rural transformation money into remote patient monitoring while the Centers for Medicare & Medicaid Services (CMS) proposes to bar the vendor-staffed version of the service; Abridge is selling a pre-bill review tool built on its ambient notes to coding departments; and 24 revenue cycle leaders told Becker's Hospital Review that their chief concern is payer automation that denies claims faster than they can answer, not autonomous coding.

In this episode

• Open-weight diagnostic agent run inside the hospital is 98.9% accurate on the 49.4% of cases it keeps, selected by agreement across five runs
• House Health Subcommittee hears Patients First Act, which would give Medicare physician pay a permanent inflation update, and three other bills
• 22 states put about $240 million of rural transformation money into remote monitoring as CMS proposes to bar vendor-staffed monitoring Jan. 1
• Abridge expands into revenue cycle with a pre-bill tool that checks coded DRGs against what its ambient scribe recorded
• 24 revenue cycle leaders tell Becker's the threat is payer automation denying claims faster than providers can respond, not autonomous coding
• GE HealthCare operations system, live at Duke and Queen's in Honolulu, forecasts bed pressure 72 hours out and assigns estimated discharge days

California awaits governor's action on AI practice bills; WISeR vendor under corrective plan; 72% of health systems report unsanctioned AI; Conifer cuts 1,037 billing jobs

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

California's governor has until Sept. 30 to act on Assembly Bill 1979, which would bar artificial intelligence (AI) from independently performing work that requires a licensed professional or directing unlicensed staff to do it. The bill applies to health facilities, clinics and physician offices; under its terms, an AI system could not direct medical assistants or other unlicensed personnel to carry out licensed duties. Its companion, AB 2575, would protect clinicians who override an AI system and bar developers from blaming those clinicians when the system fails.

The WISeR records, the Waystar sale process and the ACCESS expansion had been covered in the baseline report. The fuller WISeR records show one vendor under a corrective action plan and a Centers for Medicare & Medicaid Services (CMS) plan that considered extending AI prior authorization to MRI, cancer treatment and air ambulance, according to Fierce Healthcare and the Electronic Frontier Foundation (EFF). An Imprivata survey of 250 health system leaders found 72% have AI tools deployed without formal IT approval. Tenet's billing subsidiary is cutting 1,037 jobs, Healthcare Dive reported, and Tenet's chief executive has pointed to AI and offshoring. Waystar has hired Evercore to explore a sale, Reuters reported.

In this episode

• California AB 1979, barring AI from independently performing licensed clinical work, and companion AB 2575 await governor's action by Sept. 30
• WISeR records show Virtix under corrective action plan and a CMS plan that considered AI prior auth for MRI, cancer treatment and air ambulance
• Imprivata survey: 72% of 250 health system leaders report AI tools or agents deployed without formal IT approval; 28% have agentic AI in production
• Advocate Health pauses nursing ambient rollout; CIO says AI raises total health care costs; Penn Medicine cites about 30% pharmacy prior auth gains
• Knowtex opens health care AI lab, citing its own figures of 79 VA medical centers, 7,000 clinicians and 450,000 documentation hours saved
• Tenet's Conifer to cut 1,037 billing jobs by Nov. 2; WARN notice cites restructuring and technology initiatives, CEO has pointed to AI and offshoring
• Seven health system leaders tell Becker's autonomous AI needs narrow use cases, named accountability, escalation paths and developer liability
• Securities suit targets Hims & Hers over health data sharing with ad platforms and treatment practices; class period Aug. 4, 2025, to July 29, 2026
• Waystar hires Evercore to explore sale at about $4.8 billion; shares down 24% this year after Morgan Stanley AI disruption note