A Mayo Clinic artificial intelligence model identified people at high risk of pancreatic cancer up to three years before they were diagnosed, using only information already stored in their electronic health records. The findings were presented at the American College of Surgeons Clinical Congress in Washington, D.C., and have not yet been peer-reviewed, according to the college's announcement of the research.
The model is a research tool. It is not available to patients or doctors, and it has not been tested outside Mayo Clinic.
Early detection matters greatly for this cancer. Pancreatic cancer usually causes few symptoms until it has spread, and most people are diagnosed too late for surgery that could cure it. A tool that flags risk from routine records could, if proven, send more patients to closer monitoring while treatment can still work.
Reading Years of Records for Hidden Patterns
Researchers led by Dr. Chris Varghese, a surgical data scientist, and Dr. Cornelius Thiels, a surgical oncologist, trained the model on Mayo Clinic records from 6,066 people who developed pancreatic cancer and 33,396 who did not. Patients had between 7.5 and 19 years of clinical history.
The model reads a patient's history over time, including diagnosis codes and routine lab test results, to spot combinations of changes that may come before cancer. HealthDay described it as a transformer-based neural network, the same broad type of AI behind many language tools.
"We know that pancreatic cancer forms over five to seven years," Thiels said in the announcement. That long window is what makes early prediction possible, since subtle changes in health records may appear years before a tumor is found.
The team designed the model around routine data. "The data inputs the model relies on are captured almost universally in hospital systems worldwide," Varghese said. "We built this to be as generalizable, scalable, and easy to put into practice as possible." That design could make it cheaper to deploy than imaging-based approaches, if it holds up.
Accuracy Figures and What They Mean
For predicting a diagnosis within three years, the model scored 0.853 on a standard accuracy measure called the area under the curve, where 0.5 equals a coin flip, and 1.0 is perfect. Its precision-recall score, which matters when a disease is uncommon, was 0.712.
"Our model showed that a greater than 50% risk of pancreas cancer predicted by our model indicated an 88% likelihood of being diagnosed with pancreatic cancer in one year," Varghese said. A HealthDay summary published by Drugs.com reported that at a lower 16% risk threshold, 63% of people flagged went on to be diagnosed, while 94% of those not flagged were cancer-free.
Those numbers need context. The study group included far more cancer cases than a typical patient population would, which tends to make prediction scores look stronger than they would in a real clinic. HealthDay also reported some details differently than the college's release, including the number of controls, the length of patient histories, and the prediction window, discrepancies the full paper should resolve.
MedicalDaily Evidence Check: This was a retrospective study of about 39,500 Mayo Clinic patients, presented as a conference abstract without peer review. It showed the model can separate future cases from non-cases in records. It did not show that using the model leads to earlier diagnosis, longer survival, or fewer unnecessary scans. The announcement did not describe funding sources.
Why Earlier Detection Could Change Outcomes
The American Cancer Society estimates about 67,530 Americans will be diagnosed with pancreatic cancer in 2026 and about 52,740 will die from it. It accounts for about 3% of new cancers but is the third leading cause of cancer death.
Only about 17% of patients are diagnosed while the cancer is still localized, according to the society's figures as summarized by the Lustgarten Foundation. Five-year survival for those early cases is about 44%, compared with 13% overall.
"Pancreatic cancer can be curable, but only when we catch it early," Thiels said, adding that "fewer than one in five patients is diagnosed in time." The U.S. Preventive Services Task Force recommends against screening adults without symptoms, which is why a records-based risk flag has drawn interest.
Validation Ahead Before Any Clinical Use
The researchers said the model is being tested prospectively at Mayo Clinic and that they plan to validate it at health systems outside Mayo this year. That step is essential, since patient populations, coding habits, and lab practices differ between hospitals, and a model that works in one system can falter in another.
Even a validated model would raise practical questions, including who would review the alerts and how health systems would decide which flagged patients to contact first. Flagged patients would need follow-up imaging, which carries costs, insurance hurdles, and the risk of finding harmless abnormalities that lead to more tests.
Readers should not seek this tool, which is not available. People with known risk factors for pancreatic cancer, including a strong family history, certain inherited gene changes, or chronic pancreatitis, can ask a doctor whether they qualify for existing high-risk surveillance programs. Researchers are also studying new-onset diabetes after age 50 as a possible early warning sign. Symptoms such as yellowing skin or eyes, dark urine, or persistent upper abdominal pain need prompt medical evaluation.
The model spotted risk patterns years ahead in records, and outside testing will decide whether it can help patients.
Key Questions Answered
What did the Mayo Clinic model do? It used routine health record data to identify people at high risk of pancreatic cancer up to three years before diagnosis.
How accurate was it? It scored 0.853 on a standard accuracy scale, and a predicted risk above 50% corresponded to an 88% chance of diagnosis within a year.
Is the tool available to patients? No. It is a research model that has not been peer-reviewed or tested outside Mayo Clinic.
Why is early detection so important? Five-year survival is about 44% when pancreatic cancer is caught early, compared with 13% overall.
Who should ask about screening now? People with a strong family history, certain inherited gene changes, or chronic pancreatitis can ask about high-risk surveillance programs.
What symptoms need prompt attention? Yellowing of the skin or eyes, dark urine, and persistent upper abdominal pain.
Published by Medicaldaily.com
