The Wire · 20 Nov 2026
The Wire: 20 November 2026
Ten developments in healthcare AI, summarised from published sources and read by an AI voice. News, not medical advice.
Listen
The ten stories
- Computer vision timestamps surgery within a minute of truth
JAMA Surgery · 18 Nov
Stanford's single-center validation of ambient OR documentation at scale: computer-vision timestamps for patient entry and exit landed within one minute of ground truth in 90-95% of cases; overnight manual entries averaged about 106 seconds of error versus 35 seconds for the vision system.
- GE puts feeding-tube-checking AI on the X-ray machine itself
GE HealthCare (press release) · 17 Nov
FDA 510(k) clearance for Critical Care Suite 3.0: AI embedded on the mobile X-ray device flags feeding tube tip and side-port position on the image, adult and now pediatric patients, no cloud connection needed. Mobile X-ray sees 75% of its use in EDs and 51% in ICUs.
- Residents preferred AI discharge summaries in 1,294 London cases
JAMA Network Open · 17 Nov
Five London hospitals, GPT-4o mini pipeline versus routine summaries: adjusted signoff with no or minor edits 92.2% vs 70.0%, noninferiority met, higher ratings for accuracy, completeness, clarity and tone. But omitted key information still surfaced in 23.5% of AI drafts.
- LLM turns trial full text into computable FHIR evidence
npj Digital Medicine · 17 Nov
FIRE-EVIDENCE converts full-text studies into structured, FHIR-aligned computable evidence: 94.4% F1 and 97.4% source faithfulness over 26 expert-annotated RCTs, replicated on 18 studies from five systematic reviews. Aimed first at women's health and menopause decision support.
- Emotional-support AI use rises with unmet mental-health need
JAMA Network Open · 17 Nov
Japanese two-wave internet panel, 6,852 adults without a usual source of care: regular emotional-support AI use climbed from 4.6% (lower need) to 16.3% (high need) and 20.7% (both). Those leaning on chatbots are those the system already fails to reach.
- Barts uses AI to answer patient complaints twice as fast
Barts Health NHS · 17 Nov
Six months after introducing AI-assisted drafting for patient complaints: response rates rose from about 60% to 92%, draft preparation time fell by more than half, and reopened complaints dropped to almost zero. Admin AI, quietly useful where backlogs hurt patients directly.
- Patients trust AI skin advice more when a doctor supervises
JMIR mHealth and uHealth · 17 Nov
Randomized trial, 879 patient-panel members rating skin-cancer app advice: trust scored 76.9 with teledermatologist supervision versus 65.7 for AI-only, and adherence intention odds were 1.7 times higher with the hybrid. Oversight is the product feature patients respond to.
- Open-source agentic AI does months of analysis in minutes
Nature Medicine · 16 Nov
OpenScientist, an open agentic co-scientist, completed five expert-reviewed biomedical case studies: quantified Alzheimer's blood biomarker performance, built a proteomic survival model that held in a hidden cohort, and matched an independent human analysis blinded. Open weights, air-gapped mode.
- Chest X-ray AI beats FRAX at fracture risk in 85,284 people
NEJM AI · 16 Nov
Deep-learning survival model reading routine chest radiographs, trained on 53,109 adults, externally validated on 85,284 at two institutions: C-index 0.843 versus 0.779 for FRAX with BMD; biggest edge in osteopenia (AUROC 0.779 vs 0.659). Screening from images already taken.
- Kettering's AI has now triaged 62,661 chest X-rays
Kettering General Hospital NHS · 15 Nov
AI now reviews chest X-rays for over-16s at Kettering and Northampton general hospitals, flagging studies that need priority attention - 62,661 radiographs since go-live. A routine NHS deployment milestone rather than a trial: this is what normal looks like now.