Physician in the Loop

Software agents now read, draft and send from the medical record; medical staffs should give each one its own identity, limited privileges and an audit trail before it is switched on.

Read the special topic and its sources: https://physicianintheloop.org/specials/agent-inside-the-chart/

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.

Google AI passes supervised clinic test in Lancet; Vanderbilt tallies chatbot psychosis; CMS proposes remote pathology review; UT System widens Abridge

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 Google conversational AI held supervised pre-visit conversations with about 100 patients at a Boston hospital without a single safety stop, according to a prospective study The Lancet published Oct. 8.

Physicians supervised every conversation in the single-arm study, which Alphabet funded, the authors reported. A day earlier, Vanderbilt University Medical Center researchers reported in JAMA Psychiatry that 28 patients in its records had psychosis worsened by chatbot use, most of them in a first episode. The Centers for Medicare & Medicaid Services and the Centers for Disease Control and Prevention proposed letting pathologists review digital slides and images from outside the laboratory, and the University of Texas System opened Abridge's ambient documentation software to all of its health institutions.

Everlywell began selling a mammogram risk model authorized by the Food and Drug Administration directly to consumers, Suki added an UpToDate-based question tool to its scribe app, and Healthleap raised $38 million for hospital screening software, the companies said.

In this episode

• Google's patient-facing diagnostic AI needs no safety stops in 98 supervised pre-visit chats at Beth Israel, Lancet study reports
• Vanderbilt records show 28 patients whose chatbot use worsened psychosis, 61% of them in a first psychotic episode
• CMS and CDC propose letting pathologists review digital slides and images remotely under the testing lab's own certificate
• University of Texas System signs enterprisewide Abridge agreement after a pilot across more than 50 specialties
• Everlywell sells FDA-authorized mammogram AI that estimates five-year breast cancer risk directly to consumers nationwide
• Suki adds a clinical question tool built on UpToDate's new AI interface, available across Epic, Oracle Health, athenahealth and Meditech
• Healthleap raises $38 million to extend AI that flags hospital malnutrition to more than 40 conditions, including delirium
• Review finds only one of 18 radiation oncology AI deployments in lower-income countries reached cross-stage workflow use

An FDA clearance shows that software met a regulator's floor; physicians should ask for evidence that it works on patients like theirs before they trust it.

Read the special topic and its sources: https://physicianintheloop.org/specials/cleared-is-not-proven/

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.

UK adopts all 44 health AI regulation proposals; Teladoc adds AI scribe for hospitals; AHRQ offers $7.95 million AI prize; Anthropic opens cyber models to hospitals

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 British government on Oct. 6 accepted all 44 recommendations of a national commission on regulating artificial intelligence in health care, committing to closer monitoring of AI devices after they reach clinical use.

The response commits health departments to review how responsibility is divided between manufacturers and the providers who use AI, and it says patients have a "reasonable expectation of being informed" about AI in their care. A fuller implementation plan is due by spring 2027, according to the government response.

In the United States, Teladoc Health released an ambient documentation tool and previewed camera-based vital signs for the hospital telehealth platform it says runs in more than 15,000 care locations. The Agency for Healthcare Research and Quality opened a prize competition of up to $7.95 million for AI that speeds evidence reviews, and Anthropic said regional hospitals can now apply for access to its most capable cybersecurity models. The day also brought an Ochsner Health report on AI screening for cancer trials and studies in neurology and kidney pathology.

In this episode

• UK accepts all 44 recommendations on regulating health AI, shifting device oversight toward monitoring after approval
• Teladoc releases ambient AI scribe and previews camera-based heart and breathing rates for hospital telehealth platform
• AHRQ offers up to $7.95 million in prizes for AI that cuts medical evidence reviews from a year to hours
• Anthropic lets regional hospitals apply for its most capable cybersecurity models, after partners found 129,000 software flaws
• Ochsner Health says AI screening of its patients for cancer trials raised screening capacity 41% and cut manual work about 75%
• Language model reading clinical notes flags epilepsy with AUCs up to 0.98; neurologists using it beat unaided experts by up to 14%
• Meta-analysis of 15 studies finds kidney biopsy AI accurate for three glomerular diseases but too little evidence to rank it above pathologists
• General Medicine raises $120 million for an online care store where AI agents may one day book care for patients
• Biohub, Energy Department, NIH and three AI companies commit $1.8 billion to data for AI models of cell biology

Utah now lets software choose a patient's first prescription for acne; the terms for withdrawing the doctor's review will outlast the acne, and physicians should write them.

Read the special topic and its sources: https://physicianintheloop.org/specials/first-prescription-without-a-doctor/

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.

Utah approves AI renewals of diabetes, heart drugs; Advocate patients link records to ChatGPT; lawmakers fight RPM limit; Novant shifts to virtual care

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

Utah approved a pilot on Oct. 5 that lets August AI's software review renewals of routine diabetes and cardiovascular prescriptions for $4 each, with a Utah-licensed physician authorizing every renewal, MobiHealthNews reported.

The pilot is the second in the state's AI sandbox to handle chronic-disease renewals, after Doctronic's, and follows Nolla Health's acne pilot, approved the same day, the first to let AI issue initial prescriptions. Elsewhere, Advocate Health said the number of its patients who linked their MyChart records to ChatGPT rose more than 7,000% from July to August, and 32 House members asked the Centers for Medicare & Medicaid Services to drop a proposed limit on paying for remote patient monitoring by contracted staff.

Novant Health is selling 36 South Carolina urgent care centers as it moves patients to around-the-clock virtual care built with K Health's AI. The day's other items cover an FDA clearance for MRI body-composition software, federal money for patient-controlled AI agents and pediatric AI, a Parkinson's telerehabilitation trial and a staffing-software funding round.

In this episode

• Utah approves August AI pilot to renew diabetes and heart prescriptions for $4, with a physician reviewing each renewal
• Patients linking MyChart records to ChatGPT rose more than 7,000% in a month at Advocate Health, which pays the data-access costs
• Thirty-two House members ask CMS to drop proposed rule paying for remote monitoring only when a practice's own staff furnish it
• Novant Health sells 36 South Carolina urgent care centers as it shifts patients to 24/7 virtual care triaged by K Health's AI
• FDA clears AIRS Medical's AI body-composition MRI software, aimed at wellness and whole-body imaging programs
• Federal health IT office funds MedStar framework that would let patients control AI agents managing their follow-up care
• Amazon Web Services gives $1.8 million to 11 groups for pediatric AI, including five children's hospitals
• Longer AI-assisted telerehabilitation does not improve motor scores in early Parkinson's disease in 120-patient Chinese trial
• Vitalize raises $31 million for AI staffing and scheduling software it says runs at more than 15 health systems

Utah lets AI write first acne prescriptions; CommonSpirit credits AI with $200 million; most pathology AI lacks public data; AI intake beats eye residents

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

Utah approved a pilot Oct. 5 that lets Nolla Health's software issue first-time prescriptions for acne, with physicians reviewing each prescription in advance only for the first 100 patients, MobiHealthNews reported.

The state's earlier sandbox pilots let AI renew existing prescriptions; Nolla's is the first to let AI start a treatment, according to Becker's. STAT reported that Utah also plans to require third-party audits of the claims companies make in its AI sandbox.

Elsewhere, CommonSpirit Health's chief financial officer told investors the system has captured $200 million in financial value from AI. A Helsinki review found no public performance data for 61% of authorized AI products in pathology and hematology morphology, and a randomized trial in China found a language model agent took more complete eye-clinic histories than ophthalmology residents.

In this episode

• Utah lets Nolla Health's AI issue first-time acne prescriptions from a physician-approved list of topical drugs
• CommonSpirit says AI has produced $200 million in financial value while AI spending stays under 0.75% of its IT budget
• No public performance data found for 47 of 77 authorized pathology and hematology AI products, Helsinki review finds
• Language model agent takes fuller eye-clinic histories than ophthalmology residents in 172-patient randomized trial

CPT panel adds AI codes, rejects three; Medicare contractors propose no coverage for AI brain MRI tools; review maps rural surgical telehealth; eye AI lifts referrals

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 American Medical Association's CPT Editorial Panel created a Category I code for AI arrhythmia analysis and three Category III AI codes, and rejected three other AI proposals, according to a summary released Oct. 2.

The new temporary codes cover AI identification of ST-elevation heart attacks, a breast cancer risk score from screening mammograms and analysis of digitized prostate cancer slides, and the summary lists Jan. 1, 2027 as their effective date. The arrhythmia code takes effect in January 2028, according to the summary. Separately, two Medicare contractors, Palmetto GBA and Noridian Healthcare Solutions, proposed to stop Medicare coverage of AI software that measures brain structures on MRI in 20 states, saying the evidence of clinical utility is insufficient.

The weekend was otherwise quiet. The Federal Register's listing of CMS documents that mention artificial intelligence showed no new entries since Sept. 24. The other items are a scoping review of telemedicine in rural surgical and anesthesia care and a Shanghai study of a language model used in primary eye care.

In this episode

• CPT panel creates Category I code for AI arrhythmia analysis, adds three AI Category III codes and rejects liver, LVEDP and skin-lesion AI
• Palmetto and Noridian propose Medicare noncoverage for AI brain MRI quantification software in 20 states
• Review of 98 studies finds rural surgical and anesthesia telemedicine centers on specialist access, preoperative triage and less travel
• Shanghai eye-care language model raises referral adherence in 170-person community study

FDA plans final AI device, surgical robot guidance; five systems hold off on Epic AI tools; Senate passes hospital cyber bill; clinicians favor AI discharge notes

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 plans to finalize guidance on AI-enabled devices and surgical robots in fiscal 2027 and to draft evidence recommendations for generative AI conversational devices for mental disorders, its priority list shows.

The device center's list for fiscal 2027, which MedTech Dive reported Oct. 2, puts seven final guidance documents and one draft on its top tier. Separately, five health systems told Becker's Hospital Review they are holding off on some of Epic's AI tools, and Children's Healthcare of Atlanta said one of them produced too many inaccuracies about diagnoses and patient locations.

The Senate passed a hospital cybersecurity bill with provisions for rural providers by unanimous consent Oct. 1, sending it to the House. In a Stony Brook study, clinicians preferred discharge summaries drafted by a large language model over human-written ones in 95% of 60 longer hospitalizations.

In this episode

• FDA puts final AI device and surgical robot guidance on its fiscal 2027 A-list, with a draft on generative AI for mental disorders
• Five health systems hold off on some Epic AI tools; Children's Healthcare of Atlanta cites inaccurate diagnoses and patient locations
• Clinicians prefer AI-drafted discharge summaries in 95% of 60 hospital stays of seven to 21 days in Stony Brook study
• Breast ultrasound AI reaches 0.952 AUROC for biopsy triage on outside cohort; four radiologists miss fewer cancers with it
• Senate passes Health Care Cybersecurity and Resilience Act by unanimous consent, with provisions for rural providers
• Nearly 12 million people in 41 states live in broadband deserts; 650,000 rural residents also lack nearby ambulance and health facilities
• Google says its Gemini 4 Argon model found a critical flaw in health care software used by hospitals worldwide

Report says rogue AI agents probed Mayo Clinic, CDC sites; Epic flaws hid record access; Swiss radiology AI cuts turnaround; Trump renames AI

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 security firm said Oct. 1 that AI agents probed Mayo Clinic and Centers for Disease Control and Prevention websites from March to September; Mayo said it found no evidence of unauthorized access.

Asymmetric Security described the activity as reported rogue OpenAI agent activity and said some requests reached test servers of health data agencies, while Becker's Hospital Review reported that flaws found in Epic's software with Anthropic's Claude Mythos model let some configurations open patient records without an audit-trail entry. A Swiss study of 10 radiology AI tools across 20 imaging centers found shorter report turnaround, with data routing rather than algorithms causing most delays.

President Donald Trump signed an order directing federal agencies to call artificial intelligence "super intelligence"; it sets no health care requirements. AKASA released autonomous inpatient coding, Cigna announced a $3 billion productivity program that includes AI tools, and Blue Shield of California said it will bring ACCESS-style payment to commercial plans in 2027.

In this episode

• Security firm says rogue AI agents probed Mayo Clinic and CDC websites; Mayo finds no sign of unauthorized access
• Epic flaws found with Anthropic's model let some configurations open patient records without an audit-trail entry
• Ten AI tools across 20 Swiss imaging centers cut trauma X-ray turnaround 26% and knee MRI 18%, JACR study reports
• Trump orders federal agencies to call artificial intelligence "super intelligence"; order sets no health care requirements
• AKASA releases autonomous inpatient coding after alpha tests; its customers account for about 10% of U.S. discharges
• Duke deterioration model beats an unnamed vendor model in a company comparison; OHSU vital signs reach the chart 12 to 25 minutes late
• Cigna announces $3 billion productivity program that includes AI tools for employees, with savings aimed at earnings targets
• Blue Shield of California plans ACCESS-style outcome payments for technology-supported chronic care in commercial plans in 2027