In partnership with
🤖 AIBytes
Patients are already turning to AI first—for navigating care, symptoms and decisions-making —especially when access to clinicians is limited.
🔬 Methods
Observational study of real-world AI use.
617,827 health-related conversations (January 2026).
Global users (22% US; 45% English).
Conversations classified into 12 health intent categories.
Classification validated vs clinicians (84% agreement).
Analysis included:
Type of questions (intent)
Time of day
Device (mobile vs desktop)
Topic clustering (subset n=10,000)
📊 Results
Most queries are general health information
40.8% of all conversations.
Personal health use is common
1 in 5 conversations involve symptoms or conditions.
Caregiver use is significant
1 in 7 symptom or condition queries are about someone else.
Higher-risk queries increase at night
Symptom questions: 10.6% → 13.4% (morning → night).
Emotional well-being: 3.3% → 5.2%.
Clear difference by device
Mobile: more personal use (symptoms 15.9%).
Desktop: more professional use (research 16.9%, paperwork 15.7%).
Healthcare navigation is a major reason for use
Many queries involve finding care, insurance, and access.
🔑 Key Takeaways
AI is already a first point of contact for health questions.
Use rises when healthcare access is lowest (evenings and nights).
A meaningful portion of use involves caregivers, not just patients.
Patients rely on AI for system navigation, not only clinical information.
The study shows how AI is used, not whether it improves outcomes.
🔗 Costa-Gomes B, Tolmachev P, Taysom E, et al. Public use of a generalist LLM chatbot for health queries. Nature Health. 2026. doi.org/10.1038/s44360-026-00117-x
Researchers developed a model that generates chest X-rays from text and annotations to support training and evaluation of medical AI systems.
🔬 Methods
Model development and evaluation study.
Dataset: 960,000 chest X-ray–report pairs (OpenChest).
Model: vision–language model.
Inputs:
radiology text (what disease is present)
bounding boxes (rough location of disease)
masks (exact shape of disease)
The model generates images based on these inputs.
Evaluated with image metrics and radiologist review.
📊 Results
Generated images were closer to real X-rays than prior models.
Radiologists became more accurate when using them.
Adding synthetic images improved AI performance.
Models trained with little real data still improved.
Synthetic data reproduced real-world bias patterns.
🔑 Key Takeaways
The model generates X-rays from clinical descriptions.
Synthetic data improves training with limited real data.
It can help test and reduce bias in AI models.
Clinical use still needs validation.
🔗 Ji Y, Lin D, Wang X, et al. A Generative Foundation Model for Chest Radiography. NEJM AI. 2026. https://doi.org/10.1056/AIoa2500799
🦾TechTools
Explores medical literature as a citation network, making it easier to see how key papers connect.
Helpful for finding seminal studies, reviews, and influential authors quickly.
Best when you want to expand on one strong paper.
Searches clinical trials, systematic reviews, and guidelines.
Finds relevant evidence fast without complex PubMed filters.
Useful for quick clinical lookups, while still leaving interpretation to the clinician.
📈Productivity Tool of the Week: Codex
OpenAI agent built to carry out work across files, tools, and workflows.
Can help with tasks like building, fixing, and completing projects;
ChatGPT helps you think, Codex helps you do.
🧬AIMedily Snaps
Nvidia: State of AI in Healthcare and Life Sciences: 2026 Trends (Link).
FDA: Announces Major Steps to Implement Real-Time Clinical Trials (Link).
OpenAI HealthBench Professional: Evaluating Large Language Models on Real Clinician Chats (Link).
AMA: AI Prior Authorization System Causing Delays in Medicare (Link).
Utah Medical Licensing Board calls for suspension of AI prescription renewals (Link).
AMA urges Congress to strengthen safeguards for AI chatbots (Link).
🧪Research Signals
NEJM: Trust, Scrutiny, or Collaboration? A Performance-Based Framework for Human–AI Interaction in Medicine (Link).
Nature: Show us the evidence for the value of medical AI (Link).
The Lancet: AI-based pathological model for pan-cancer lymph node metastasis detection: a multicenter diagnostic study with retrospective and prospective validation (Link).
Nature: DeepFAN, a transformer-based model for human–AIcollaborative assessment of incidental pulmonary nodules in CT scans: a multireader, multicase trial (Link).
The Lancet: Identification of drug repurposing candidates for amyotrophic lateral sclerosis using electronic health records: a retrospective cohort study (Link).
Nature: Seven deadly sins in AI for digital medicine (Link).
🧩TriviaRX
Which disease was the first to be eradicated worldwide through vaccination?
A) Polio
B) Smallpox
C) Measles
D) Diphtheria
Now, the answer from last week’s TriviaRX ✅ A) Excess oxygen therapy
In the 1940s and 1950s, high concentrations of supplemental oxygen in premature infants were later identified as the main cause of what was then called retrolental fibroplasia, now known as retinopathy of prematurity.
Want to keep up with AI in medicine in less than 5 minutes a week?
Subscribe free to AIMedily.
Itzel Fer, MD PM&R
You think 4x faster than you type. Your IDE should keep up.
Wispr Flow lets you dictate prompts, acceptance criteria, and bug reproductions inside Cursor or Warp — with automatic file name and variable recognition. Say user_id, get user_id. Say useEffect, get useEffect.
Paste directly into GitHub, Jira, or Linear. Give coding agents the full context they need without typing a novel.
89% of messages sent with zero edits. Millions of developers use Flow daily, including teams at OpenAI, Vercel, and Clay. Free on Mac, Windows, and iPhone.