Healthcare AI Daily Briefing

Diagnostics & Pathology

Six‑predictor random forest model improves lymph‑node metastasis prediction beyond tumor stage.

Journal article

A multicenter retrospective study of 2,725 colorectal‑cancer patients built a six‑predictor random‑forest model (BMI, preoperative CEA, tumor site, cT, histology, differentiation) from routinely available data. In external validation the model’s AUROC was 0.776, higher than cT alone (0.707). The authors say it is for research‑stage adjunctive risk stratification and needs prospective validation before clinical use. [1]

Medical Imaging

Dual‑suggestion AI support boosts radiology residents’ accuracy, not non‑radiology.

Preprint

In a multicenter randomized trial of 123 residents with fewer than three years’ experience interpreting radiographs, dual‑suggestion AI support yielded higher accuracy than single‑suggestion support for radiology residents, while accuracy did not change for non‑radiology residents, suggesting the approach may lessen the impact of incorrect AI suggestions. [2]

Policy & Governance

Over 200 firms have CMS AI test approval; FDA cleared four for Medicare.

Federal pilots test AI tools that perform clinical functions traditionally done by physicians. Over 200 companies have CMS approval to test them; the FDA has allowed four for Medicare use and is accepting comment on evaluation rules through Oct 19 2026. The clinical question is what standards will determine whether AI is accurate and safe enough for less‑direct physician involvement. [6]

Industry & Products

Over 900 clinicians actively use ambient AI platform in oncology.

Podcast from Beyond the Chart: Exploring the AI Frontier for Oncology

The interview reports that over nine hundred clinicians are actively using the ambient AI platform, and describes the rollout as a new muscle that both the technology company and oncology practices had to learn together, characterizing it as a wild ride and a learning curve for both sides. [4]

Policy & Governance

About eighty percent of routine patient emails could be auto‑handled, leaving twenty percent for physicians.

Podcast from Claims Denied: A Hospitalogy Podcast

The interviewee estimates that roughly eighty percent of administrative patient‑email inquiries—such as prescription refills or appointment requests that occur outside office visits—could be managed by an automated system with a human‑in‑the‑loop for monitoring and feedback, while the remaining twenty percent would require physician attention. [5]

Operations & Workflow

AHA disputes insurer claim that AI coding raises inappropriate spending.

The American Hospital Association challenges insurers' assertion that AI-assisted hospital coding drives inappropriate spending, arguing that the insurer's analysis did not review medical records. [3]