Overview
- A study in Gut validated REDMOD on routine abdominal CT across multiple hospitals, finding 73% sensitivity for very early pancreatic cancer versus 39% for radiologists.
- The model detected disease a median 475 days before diagnosis and reached cases as far as about three years ahead of clinical detection.
- Specificity was about 88% for people without cancer, and repeat scans for the same patient produced consistent risk scores over time.
- Researchers plan a prospective trial called AI-PACED to test real-world use, false-positive rates, follow-up pathways, and outcomes in higher-risk patients.
- Authors note key limits including a less diverse cohort and call for targeted testing in high-risk groups before routine clinical adoption.