New AI Reads Doctor Notes to Spot Colon Cancer Risks Early

Colorectal cancer is becoming more common in people under 50, but early warning signs often hide inside messy, free-form doctor notes. A new research system called VERGE was created to automatically scan those notes and pull out six red-flag symptoms—things like abdominal pain, changes in bowel habits, or unusual bleeding—plus family history of the disease, which is a known risk factor.

Here's the clever part: VERGE doesn't just read once. First, it uses a method called retrieval-augmented generation (letting the AI look up relevant medical knowledge). Then it enters a check-and-correct loop. It proposes an answer, verifies that the answer is actually supported by the note, checks whether it makes medical sense, fixes problems, and rechecks itself. Only if the AI can't resolve something does it escalate the case to a human expert.

In a test on over 4,000 doctor-note segments, VERGE improved precision—meaning fewer false positives, where the AI flags a symptom that isn't really there—from about 76% to 85%. It still caught real symptoms about as well as before, and just 1.5% of its findings needed a human to double-check. For comparison, a simpler single-step AI was more error-prone.

Why does this matter to you? In medicine, false alarms cause unnecessary worry, extra tests, and delays. A system that reads clinical notes with fewer mistakes could help doctors identify which younger adults really should get a colonoscopy or further screening. It's not replacing your doctor—it's a supportive tool that could make risk assessment faster and more trustworthy, especially for a disease that's increasingly striking people in their 30s and 40s.