Hello!
Stanford just released its third report from the 100 Year Study on AI, a project that examines how AI is evolving and affecting society, with a new report every five years.
A lot has changed since the last report in 2021. Generative AI has become part of our daily lives; it is helping advance scientific research, being integrated into our workflows, and its applications in medicine continue to grow.
The report also identifies important challenges, including inaccurate information, bias, and potential risks.
I think we need to understand both the opportunities and the risks. AI has great potential to improve patient care, but we also need strong evidence to understand where it can make a real difference. (Link)
Let's get into this week's updates!
🧬AIMedily Snaps
Fast updates clinicians should not miss.
Google Cloud introduced Gemini for Healthcare, a new AI platform designed to connect clinical records, guidelines, and administrative systems. (Link)
Oura, the company behind the smart ring, acquired Counsel Health to bring together wearable health data, medical AI, and access to physicians. (Link)
Stanford experts discussed how AI could change medical education and why clinical judgment and critical thinking remain important. (Link)
Anthropic released Claude Fable 5.2 and Mythos 5.2, reporting improvements in scientific research, with potential applications in medicine. (Link)
OpenAI introduced BioBasicsBench, a set of 100 tasks to test how well AI agents analyze lab results and identify experimental errors. (Link)
Mayo Clinic launched an AI clinical decision support tool covering more than 35 specialties, with clinical guidance available within EHR workflows. (Link)
🤖AIBytes
Two clinical studies that deserve a closer look.
Can AI Help Diagnose Autoimmune Diseases Earlier?
A positive antinuclear antibody (ANA) test does not always mean a patient has an autoimmune disease. Deciding who needs a rheumatology referral can be challenging.
Researchers tested whether an AI-based prediction model could help identify high-risk patients and reduce diagnostic delays.
Methods
A two-year pragmatic randomized study included 2,385 adults with positive ANA tests. After exclusions, 1,960 were analyzed.
Intervention: An EHR-based model identified high-risk patients who were offered expedited rheumatology evaluation.
Control: Standard care, without acting on the model's risk scores.
The model used routinely collected clinical information, including ANA titers, autoantibodies, and symptoms.
Results
Median time to diagnosis was 43 days with AI-guided referrals vs. 138 days with standard care among patients diagnosed with autoimmune disease.
Patients in the intervention group were diagnosed significantly sooner.
Among patients seen by rheumatology within six months, 23% in the intervention group received a diagnosis vs. 16% in the control group, but the difference was not statistically significant.
Key Takeaways
AI-based risk assessment combined with faster rheumatology referrals reduced diagnostic delays, particularly in high-risk patients.
The model performed well in clinical practice using a simple, interpretable approach.
The study could not separate the effects of AI from faster referrals or assess long-term health outcomes.
Barnado A, Moore RP, Domenico HJ, et al. Randomized Evaluation of an Autoimmune Disease Prediction Model — AURA (AUtoimmune Risk Assessment). NEJM AI. 2026. doi:10.1056/AIoa2600112.
AI-Powered Smartwatch Tested for Hypertension Screening
Many people with hypertension are unaware they have it. Could a smartwatch help identify those who need further evaluation?
A study in Nature (preview) tested an AI system that uses smartwatch sensors to detect signs of hypertension without requiring calibration with a blood pressure cuff.
Methods
An AI model was developed using smartwatch data and tested in a prospective study across five U.S. sites.
Of 296 adults enrolled, 196 had sufficient data for analysis.
Participants wore a smartwatch for approximately seven days.
The AI analyzed optical pulse signals and wrist movement to identify signs of hypertension.
Researchers compared its performance with initial office blood pressure screening, using 24-hour ambulatory blood pressure monitoring as the reference standard.
Results
Sensitivity: 65.8% for AI vs. 55.3% for office screening.
Specificity: 90% for both methods.
Positive predictive value: 80.6% for AI vs. 77.8% for office screening.
The difference in sensitivity was not statistically significant.
Key Takeaways
AI identified signs of hypertension using smartwatch data without requiring calibration with a blood pressure cuff.
Its performance was comparable to initial office screening, although it missed about one-third of hypertension cases.
Larger studies are needed to determine if this approach improves hypertension detection and outcomes.
The system could help identify people who need further evaluation, but it cannot replace cuff-based blood pressure testing.
Di Achille P, Cai L, Wu J, et al. Opportunistically Detecting Signs of Hypertension on a Consumer Smartwatch. Nature. 2026. doi:10.1038/s41586-026-11336-3.
🦾TechTools
AI medical tools
FDA-cleared AI software that analyzes coronary CT angiography to assess the functional significance of coronary stenosis.
FDA-authorized AI device that uses spectroscopy to assess suspicious skin lesions in adults, helping non-dermatologist physicians decide if referral is needed.
📈Productivity AI tool of the week:
An AI research assistant that lets you ask questions about scientific papers and review their content.
🧪Research Signals
New papers worth your time.
JAMA: AI predicted delirium risk across 33,095 hospital admissions, maintaining strong performance at a second hospital. (Link)
JAMA: AI scribes were associated with improved same-day note completion across 209,009 visits, with differences by patient language. (Link)
JAMA: A randomized trial involving 400 participants found that an AI maternal-health chatbot did not significantly improve anxiety or depression after 12 months. (Link)
npj: An AI system identified all 324 brain CT scans with clinically important findings and marked 18% of the 2,001 scans as lower priority for radiologists to review. (Link)
The Lancet: Clinical AI evaluations should consider whether a system is ready for the context in which it will be used, not just how well it performs on benchmarks. (Link)
NEJM AI: A perspective analysis of Medicare claims from nearly 590,000 physicians across 56 specialties examines how the need for physical patient care may limit AI's ability to replace physicians. (Link)
That’s it for today.
Thank you for taking the time to read.
You’re already ahead of the curve in medical AI—don’t keep it to yourself. Forward AIMedily to someone who’d find it useful.
Itzel Fer, MD
Forwarded this email? Subscribe free.