Glossary
September 25, 2026

Mayo Clinic Doctors Build an AI Program That Spots Signs of Pancreatic Cancer Up to Three Years Before Diagnosis

Researchers at the Mayo Clinic built an AI program that reads ordinary medical records and routine blood-test results and picks out people at high risk of pancreatic cancer, one of the deadliest cancers, up to three years before they are diagnosed. It is early work, not yet checked by outside experts, but it points at a way to catch the disease while it can still be cured.

Doctors at the Mayo Clinic announced today that they have built an AI program that can pick out people at high risk of pancreatic cancer years before the cancer is found. The work will be presented at a large meeting of the American College of Surgeons in Washington that starts this weekend.

Pancreatic cancer is fairly rare, but it is one of the most deadly cancers because it usually isn't found until it has already spread. "Pancreatic cancer can be curable, but only when we catch it early — and fewer than one in five patients is diagnosed in time," says Dr. Cornelius Thiels, a cancer surgeon at the Mayo Clinic who worked on the study. The researchers note that the cancer takes about five to seven years to form, which leaves a window when it could, in principle, be caught.

The program doesn't need any special scan or new test. It reads the kind of information almost every hospital already keeps: a patient's medical history and the results of routine lab work, gathered over a decade or more. The team trained and tested it on the records of nearly 40,000 people. About 6,000 of them went on to develop pancreatic cancer and about 33,000 did not, and the program learned to find the small, scattered clues that set the first group apart.

In the team's testing, the program was good at telling high-risk people from low-risk people up to three years before their diagnosis. When it gave someone a better-than-even chance of having the disease, 88 out of 100 of those people were in fact diagnosed with pancreatic cancer within a year.

There are real limits. This is a summary presented at a medical meeting, not yet a full study reviewed by outside experts. It was tested by looking back at past patients' records from one health system, and a program that works well on old records can do worse on new patients or at other hospitals. Nobody is using it to make decisions about patients yet. The team says its next step is to run it on real patients going forward, on a research basis, to see whether it holds up.

If it does, the hope is a kind of quiet early-warning system: a program running in the background of ordinary checkups that tells a doctor which few patients deserve a closer look, at a stage when the cancer may still be curable.

Sources

  1. American College of Surgeons (via Newswise) — Artificial intelligence model predicts pancreatic cancer risk 3 years before diagnosisnewswise.com · primary
  2. Medical News Today — AI model may identify pancreatic cancer risk years earlymedicalnewstoday.com
Follow the timeline
Get an email when new entries are added.
© 2026 Sugarpine