Welcome!
When I had the chance to speak with one of OpenEvidence’s co-founders, I asked when more countries would have access to the platform.
This week, OpenEvidence and Anthropic announced that they are expanding access to around 100 low- and middle-income countries.
There are physicians working in remote areas, sometimes with very limited access to medical literature or specialist support, who are still expected to make difficult decisions every day.
I think giving them easier access to reliable medical information could make a real difference.
Of course, the information still needs to fit the local reality: guidelines, medications, resources, and the health system itself.
But expanding access like this is an important step.
Now it’s time to check this week’s newsletter.
🤖AIBytes
Two clinical studies that deserve a closer look.
Turning Routine Chest CT Into a Cancer Screening Opportunity
Researchers developed EAGLE to detect esophageal cancer and precancerous lesions on noncontrast chest CT. It was tested across several countries and real-world clinical settings.
🔬 Methods
EAGLE was trained on 6,813 patients from two centers.
External testing included 11,466 patients across eight centers in three countries.
17 radiologists reviewed 300 CT scans with and without AI support.
Real-world testing included 35,402 patients, followed by prospective validation in 17,446 patients.
📊 Results
In external testing, EAGLE detected 90.0% of cancers while correctly ruling out cancer in 98.5% of negative cases.
Detection was lower for earlier disease: 60.1% for stage I cancer and 52.5% for high-grade precancerous lesions.
With AI support, radiologists detected more cancers: sensitivity increased from 71.9% to 85.7%, while specificity improved from 79.6% to 91.7%.
In prospective testing, 38 of 90 AI-positive cases were confirmed, for a positive predictive value of 42.2%.
🔑 Key Takeaways
AI could add esophageal cancer detection to chest CT scans already being performed for other reasons.
AI support helped radiologists detect more cancers while also reducing false positives.
Early disease remains harder to detect, and larger international prospective studies are still needed. Follow-up was also under two years.
🔗 Zhou J, Guo G, Yao J, et al. Large-scale esophageal cancer screening through noncontrast computed tomography and artificial intelligence. Nat Med. 2026. doi:10.1038/s41591-026-04656-4
Can One Blood Test Detect Multiple Cancers Early?
PATHFINDER 2 tested a blood-based multi-cancer early detection test in adults 50 years or older without signs or symptoms of cancer.
🔬 Methods
Prospective, interventional study with 35,878 participants.
The test looks for cancer-related DNA methylation patterns in blood.
It also predicts where the cancer signal may be coming from to guide further testing.
Participants were followed for 12 months to assess performance and safety.
📊 Results
Among 32,007 participants with complete follow-up, 173 cancers were detected after a positive test.
The test had a 60.3% positive predictive value and 99.64% specificity.
Sensitivity was 39.3% across all cancers and 69.8% for a prespecified group of 12 cancers responsible for about two-thirds of U.S. cancer deaths.
The predicted cancer origin was correct in 91.3% of true-positive cases.
🔑 Key Takeaways
A single blood test detected cancers across multiple organs, including many without routine screening options.
False positives were uncommon, but the test still missed many cancers, especially when all cancer types were considered.
Randomized trials and longer follow-up are still needed.
🔗 Nabavizadeh N, McDonnell C, Kurbegov D, et al. Performance and safety of a multi-cancer early detection test: the PATHFINDER 2 study. Nat Med. 2026. doi:10.1038/s41591-026-04618-w
🧬AIMedily Snaps
Fast updates clinicians should not miss.
OpenAI released GPT-6 Sol and GPT-6 Luna, newer models designed to bring stronger performance at a much lower cost. (Link)
Abridge was selected for a nationwide ambient AI contract with the U.S. Department of Veterans Affairs, the country’s largest integrated health system. (Link)
The World Health Organization highlighted new guidance for reviewing the ethics of AI-related health research, including consent, bias, and oversight. (Link)
Sinai Hospital began using Sepsis ImmunoScore, an FDA-authorized AI tool that helps identify patients at risk for sepsis. (Link)
Stanford researchers built a virtual biotech with 37,000 AI agents that analyzed about 50,000 clinical trials and helped explore new drug candidates. (Link)
🧪Research Signals
New papers worth your time.
NEJM AI: A new framework explains how medical AI should be tested step by step, from early performance to real-world use and monitoring after deployment. (Paper)
Nature: AI analyzed brain cells from 584 donors to find patterns linked to Alzheimer’s disease and why some people stay cognitively stronger despite disease. (Paper)
JAMA: An AI chatbot improved HPV vaccine knowledge compared with standard information, vaccination decisions did not differ significantly. (Paper)
Nature: Researchers share what they learned while expanding clinical AI from one hospital to more than one million patients across India, Thailand, and Australia. (Paper)
npj: LLMs changed how urgently they prioritized patients when stigmatizing language was added to otherwise similar clinical cases. (Paper)
npj: A review of 41 chest X-ray AI models found that studies involving physicians were more likely to address clinical use and fairness. (Paper)
🦾TechTools
AI medical tools
HeartFlow (Link)
Uses coronary CT to create a 3D model of blood flow and help assess whether a coronary narrowing is functionally significant.
ArteraAI Prostate (Link)
Combines prostate biopsy images with clinical data to estimate prognosis and treatment benefit in localized prostate cancer.
📈 Productivity AI tool of the week:
Heptabase (Link)
A visual workspace for organizing papers, notes, PDFs, and ideas. Its AI can work across your own materials.
That’s it for today.
Thank you for taking the time to read.
You’re already staying ahead of what’s happening in medical AI. If you know someone who would enjoy AIMedily too, send it their way.
Itzel Fer, MD PM&R
Forwarded this email? Subscribe free.