Healthcare AI Daily Briefing

Policy & Governance

FDA and CMS scramble to resolve regulatory overlap as states test AI in clinical settings.

Federal officials are moving behind the scenes to address conflict between state and federal jurisdiction over AI that performs clinical tasks such as writing prescriptions, according to STAT reporting. States regulate medical practice and licensing, while AI acting without human supervision likely falls under FDA device authority, creating overlapping oversight that could stall adoption. [1]

Operations & Workflow

Healthcare IT Today podcast previews Hello Patient AI voice agents for outpatient clinics.

The Life Sciences Today podcast episode announcement introduces Alex Cohen, co-founder and CEO of Hello Patient, and describes the company's AI voice agents for scheduling, follow-ups, billing, and referrals at multi-location clinic groups. The post frames the discussion around the work required to make the agents useful beyond a scripted demo, but does not provide interview details, specific operational constraints, or independent validation of the claims. [2]

Diagnostics & Pathology

Nanosensor fingerprint enriched rare proteins tied to brain-tumor signal.

Podcast from Health and Explainable AI Podcast

The interview describes washing proteins off the most informative nanosensors and finding rare proteins rather than abundant ones such as albumin; in brain-tumor cases the enriched proteins included some known cancer markers, as reported in the interview. [3]

Industry & Products

Redesign Health acquires Sheba's ARC Global to commercialize clinician ideas.

Redesign Health acquired ARC Global, Sheba Medical Center's innovation arm, to help health systems turn clinician ideas into companies and deploy technologies such as AI across hospitals worldwide, HIStalk reports. [4]

Operations & Workflow

Study launched on AI-enabled prior authorization in Medicare using Epic Cosmos data.

Journal article (JAMA)

Researchers are using Epic Cosmos data to assess changes in medical service use by Medicare recipients after the Center for Medicare & Medicaid Innovation implemented AI-enabled prior authorization requirements under its WISeR model. [5]

Medical Imaging

Recruiting trial tests AI reports for indeterminate lung nodule management.

Trial registry record

A recruiting trial on ClinicalTrials.gov compares management of indeterminate pulmonary nodules with and without an AI report that may influence surveillance versus invasive investigation decisions. [6]

Medical Imaging

Domain adaptation beats fine-tuning only when labeled data are scarce.

Journal article (Journal of medical systems)

In a retrospective study of 3,123,836 CT and MR reports, domain-adapted Llama-3-8B models fine-tuned on 1,280 impression pairs scored higher on expert ratings than similarly fine-tuned OpenBio-8B (24.3 vs 22.8, P = 0.0004), but the advantage disappeared with 499,814 pairs (24.9 vs 25.0, P = 0.89). Large-scale fine-tuned open-source models outperformed Gemini-1.5-Flash-002 on expert-rated quality and harmful output rates. Radiologist agreement was low (ICC 0.22–0.49) and automated NLP metrics correlated weakly with human evaluation (ρ < 0.32), indicating standard metrics poorly reflect clinical relevance and that the data-efficiency threshold helps define cost-effective deployment. [7]

Medical Imaging

Registry records retrospective multicenter study of high-resolution CT plus AI for pulmonary nodule diagnosis.

Trial registry record

A ClinicalTrials.gov registry entry describes a retrospective, multicenter observational study evaluating the clinical value of combining high-resolution (1024×1024 matrix) targeted CT scans with a deep learning AI method for diagnosing pulmonary nodules, with an 'Active, not recruiting' status indicating data collection or analysis is underway but no new participants are being enrolled. [8]