Mayo Clinic researchers are exploring whether patterns in routine health records can identify people who would benefit from earlier investigation.
An artificial intelligence model developed by Mayo Clinic researchers could help identify people at elevated risk of pancreatic cancer years before diagnosis, potentially creating an opportunity for earlier, targeted evaluation.
The research analysed electronic health records and routine laboratory results from 6,066 people with pancreatic cancer and 33,396 controls. Participants had between 7.5 and 19 years of medical history available, allowing researchers to examine how health patterns changed over time. The findings were presented at the American College of Surgeons’ Clinical Congress in September 2026.
The approach addresses a difficult problem: pancreatic cancer is often diagnosed at an advanced stage, while screening everyone is not currently feasible. Researchers hope routinely collected information could help narrow the group of patients who might benefit from additional assessment.
The prediction timeline requires careful interpretation. The conference announcement described risk identification up to three years before diagnosis. In a subsequent interview, lead researcher Chris Varghese discussed updated analyses extending to five years. These are research findings measured relative to a recorded diagnosis; they do not establish that the model reliably predicts cancer a fixed number of years before symptoms begin.
Separate Mayo Clinic research is examining medical imaging. An AI system called REDMOD analysed nearly 2,000 abdominal CT scans, including earlier scans from patients subsequently diagnosed with pancreatic cancer. It identified 73% of prediagnostic cancers at a median of approximately 16 months before diagnosis, according to an April 2026 announcement. That imaging study was published in Gut.
The two approaches use different information: one examines health histories and laboratory results, while the other searches for subtle tissue changes on CT images. Neither finding establishes a standalone screening test ready for widespread use.
The health-record research was presented as a conference abstract and had not yet undergone journal peer review, according to the conference announcement. Researchers are undertaking prospective validation, including evaluation beyond Mayo’s own healthcare system.
For the imaging approach, further clinical testing will assess false positives, early detection and patient outcomes. The practical question is whether these tools can guide useful investigations and improve care—not simply identify patterns in historical data.
Sources:
