Mayo Clinic AI Model Predicts Pancreatic Cancer Risk Years Before Diagnosis

**Researchers at Mayo Clinic have designed an artificial intelligence model that can potentially predict an individual’s risk of developing pancreatic cancer years before diagnosis. **

Research will be presented at the American College of Surgeons (ACS) Clinical Congress 2026, held from September 26-29 in Washington. Thousands of surgeons convene at the annual event to advance surgical quality, patient safety, and access to care.

According to the official press release, shared with the Investing News Network (INN), pancreatic cancer is rare but highly deadly, accounting for about 3 percent of all new cancers but 8 percent of all cancer deaths.

Data from the American Cancer Society, notes that there have been about 67,000 new diagnoses and 52,000 deaths so far in 2026.

“Pancreatic cancer can be curable, but only when we catch it early, and fewer than one in five patients is diagnosed in time,” said Mayo Clinic surgical oncologist and study co-author Cornelius Thiels, DO, MBA, FACS. “As a result, survival for many patients is still measured in months, not years.”

---

Dr. Thiels said his team set out to develop an AI model that can identify patients at greatest risk of developing cancer of the pancreas because universal screening for pancreatic cancer “isn’t feasible”.

“We know that pancreatic cancer forms over five to seven years, but the things that a clinician or patient sees don’t happen until it’s too late.”

The researchers built the AI model using Mayo Clinic electronic health records and routine lab test results to analyze 6,066 pancreatic cancer patients and 33,396 control subjects, each with up to 19 years of medical history, to detect early risk indicators.

To evaluate its ability to predict pancreatic cancer three years before diagnosis, researchers measured the model's accuracy. It achieved an AUROC of 0.853 (where 1.0 is perfect accuracy) and an AUPRC of 0.712, demonstrating strong predictive performance with few false positives.

The model showed strong calibration, with a calibration slope of 1.08, meaning its predicted risk closely matched what actually happened to patients.

“Our model showed that a greater than 50 percent risk of pancreas cancer predicted by our model indicated an 88 percent likelihood of being diagnosed with pancreatic cancer in one year,” Dr. Varghese, a surgical data scientist at Mayo Clinic in Rochester, explained.

“We built this to be as generalizable, scalable, and easy to put into practice as possible,” Dr. Varghese added. The data inputs the model relies on are captured almost universally in hospital systems worldwide, Dr. Varghese said. “If it’s shown to work, it could be used in almost any setting,” he added.

According to Dr. Thiels, the model is currently being deployed on a research basis. “We’re proving that we can move this from a retrospective research tool into our clinical environment and run it prospectively for validation,” he stated.

Dr. Thiels noted that efforts are underway to validate the model further, both prospectively within Mayo and at an external healthcare system this year. “We are also working on developing more advanced machine learning architectures, which appear to improve the performance even more,” he added.

Earlier this year, a study appearing in the journal Gut described a Mayo-built AI model called REDMOD that read ordinary CT scans from people who were later diagnosed to look for early signs of pancreatic cancer.

The AI caught most of those hidden cancers, often more than a year before diagnosis, about twice as many as specialists caught looking at the same scans. The gap was even bigger for scans taken more than two years before diagnosis.

A follow-up trial called AI-PACED will test the tool in real care for high-risk patients. It will also track false alarms and whether finding the cancer earlier improves outcomes.

“The greatest barrier to saving lives from pancreatic cancer has been our inability to see the disease when it is still curable,” said the study’s senior author Dr. Ajit Goenka.

​What investors are watching

Lu Zhang, founder and managing partner of Fusion Fund, has been watching AI-powered diagnostics closely. At Web Summit Vancouver last year, she pointed to advances in digital diagnostics for conditions like cancer, heart disease and mental health.

She said healthcare is entering its “prime time for innovation.” In her view, the core goal is to “improve the quality of life, how to really enable the future of healthcare to be personalized…and also be able to do super early diagnostics and reduce the healthcare burden in the long term.”

Zhang also noted that less than 5 percent of healthcare data is currently being used. Mayo Clinic’s model is built on electronic health records and routine lab test results.

In a recent conversation with the INN earlier this month, Zhang said large AI labs are paying high prices for high-quality healthcare data. They are also hiring PhDs and domain experts to label it.

For Zhang, healthcare is one of the clearest examples of where AI’s promise and its constraints collide. She repeatedly comes back to the sector as a case where high-quality, tightly controlled data makes a real difference — and where governance and deployment choices are non‑negotiable.

On the infrastructure side, she stresses that healthcare is part of the huge chunk of the economy that can’t just ship everything to the public cloud.

That, in her view, is why architecture design and small, efficient models matter so much: enterprise buyers in healthcare often want on‑prem or private‑network deployment, not generic cloud AI.

Zhang also highlights healthcare as a leading example of vertical, data‑driven AI moving fast precisely because the data is specialized and curated.

“They are able to directly use high-quality data, not a huge amount of data, but highly specialized healthcare data to fine-tune their model.”

She points to Google's (NASDAQ:GOOGL) AlphaFold as one reference point, but says the dynamic is broader. Large AI labs are actively competing to secure top‑tier medical datasets and expert feedback.

That mix of private, regulated environments; expensive but highly informative data; and expert human feedback makes healthcare a kind of proving ground for the approach Zhang favors: small, vertical models tuned on curated industry data and deployed inside tightly governed infrastructures.

Don’t forget to follow us @INN_Lifescience for real-time news updates!

Securities Disclosure: I, Meagen Seatter, hold no direct investment interest in any company mentioned in this article.