You probably know that pancreatic cancer is one of the hardest cancers to catch early. Most people have no idea anything is wrong until the disease has already spread. By then, treatment options are narrow and survival odds are poor.
Researchers at Mayo Clinic have built an artificial intelligence model that may flag high pancreatic cancer risk up to three years before diagnosis. It reads routine blood work and medical records already sitting in a patient's file.
The research was presented at the American College of Surgeons Clinical Congress 2026 in Washington in late September 2026. The findings have not yet been peer-reviewed.
Pancreatic cancer is the deadliest major cancer in the United States. Only about 13 percent of patients are alive five years after diagnosis, largely because the disease is almost never caught before it has spread, according to the American Cancer Society.
Also read: WHO reveal official guidelines for Ebola and Marburg diseases
When the cancer is still confined to the pancreas, five-year survival rises to 44 percent. Once it has reached distant parts of the body, it drops to 3 percent.
How the AI reads your old health record
The model was developed by surgical data scientist Chris Varghese and surgical oncologist Cornelius Thiels at Mayo Clinic Rochester. They analyzed data from almost 40,000 patients.
The group included 6,066 people with pancreatic cancer and 33,396 without, each with roughly 7.5 to 19 years of health history on file.
The AI looked for subtle patterns in things most hospitals already measure. That includes complete blood counts and other routine lab values, past diagnoses of diabetes or pancreatitis, and the way those numbers drifted over the years.
Also read: Marburg kills about half the people it infects. Two vaccines are now being tested
The tool scored 0.853 on a standard accuracy scale where 1.0 would be perfect. When it put someone's near-term risk above 50 percent, Varghese and his colleagues reported that 88 percent of those people were diagnosed with pancreatic cancer within a year.
A separate preprint from the same team, posted on PubMed Central in May 2026, showed the model held up on external testing, with somewhat lower accuracy the further out it was asked to predict.
What this could mean for future screening
Universal pancreatic cancer screening is not practical. The disease is relatively rare, and the imaging tests used to look for it are expensive or invasive.
The researchers hope their AI could first narrow down a high-risk group. Downstream tests like CT scans, MRI, or blood biomarkers could then focus on the people most likely to benefit, raising the odds of catching tumors while they are still operable.
Also read: Inside the bat cave where scientists watched Marburg virus prepare to jump to humans
Pancreatic cancer can be curable, but only when it is caught early, and fewer than one in five patients is diagnosed in time, Thiels noted in the Mayo Clinic announcement.
The team is now moving to prospective testing inside a healthcare system. That step has to be cleared before the tool could be used in routine care.
The American Cancer Society estimates that about 67,530 Americans will be diagnosed with pancreatic cancer in 2026. Roughly 52,740 are expected to die from it.
Anyone with persistent unexplained weight loss, new-onset diabetes later in life, or ongoing upper abdominal or back pain should bring it up with their doctor.
Also read: RFK Jr. launched a health movement: here is what MAHA actually stands for
Also read: Labs handling the world's deadliest viruses have doubled in a decade, and most are now in cities
This article is made and published by Aksel Kryhlmand, who may have used AI in the preparation.