The Cancer That Hides Until It Is Too Late
Pancreatic cancer is one of the most feared diagnoses in medicine. Not because it is the most common — it is not — but because it is almost always found late.
By the time most people have symptoms, the cancer has usually spread beyond the pancreas (the organ tucked behind the stomach that produces digestive enzymes and insulin). At that stage, treatment options are limited and survival rates are low.
Who Gets Screened Today — and Who Doesn't
Right now, formal pancreatic cancer surveillance (regular testing to catch cancer early) is only offered to a narrow group of people: those with a known genetic mutation like BRCA2, or a strong family history of the disease.
But here is the uncomfortable truth: most people who develop pancreatic cancer do not have those known risk factors. They are not being watched. They are not being screened.
For years, researchers have tried to find a way to identify these "hidden high-risk" people before their cancer grows.
From Paper Records to Risk Scores
PRISM is an AI model built to read a person's electronic health record (EHR) — the digital file that holds diagnoses, lab results, medications, and doctor visits — and calculate a score estimating the likelihood of developing pancreatic cancer within the next few years.
Think of it like a financial credit score, but for health risk. The model looks at subtle patterns across dozens of data points — none of which alone signals cancer — and combines them into a single risk number.
PRISM was first shown to work in a retrospective study (one that looked backward at existing records). The new study tested it going forward, in the real world, across 44 U.S. health care organizations starting in April 2023.
A Study of More Than Six Million People
More than 6.2 million adults aged 40 and older received a PRISM risk score. Researchers then tracked which patients were later diagnosed with pancreatic cancer over 30 months.
Among all those patients, 3,609 developed pancreatic cancer during the follow-up period.
The highest-risk tier identified by PRISM had 30 times more pancreatic cancer cases than the average patient population — a dramatic enrichment that shows the model is picking up on something real.
The model assigned patients to risk tiers. At the most sensitive threshold, roughly 1 in 284 high-risk patients developed pancreatic cancer within 30 months. At the highest-risk threshold, that number dropped to about 1 in 46 patients.
One of the most promising findings was timing. The median time from a patient's first high-risk flag to their eventual diagnosis was 3.5 years. That window — three and a half years — is exactly the kind of lead time that could make a difference between catching cancer early and catching it too late.
Why Earlier Detection Matters So Much Here
Pancreatic cancer caught at an early, localized stage has a dramatically better prognosis than cancer found after it has spread. Surgery is often possible in early-stage cases. It is rarely possible in advanced ones.
If PRISM can reliably identify high-risk patients years before diagnosis, and if those patients then receive surveillance (such as MRI or endoscopic ultrasound), some cancers could be caught while they are still operable.
PRISM is already deployed in some U.S. health systems, but it is not yet a standard part of care. If you are over 40 and have never been told you are at risk for pancreatic cancer, this research does not mean you should panic. It means that tools to better identify risk are getting sharper — and that may eventually lead to more personalized screening recommendations.
If you have concerns about pancreatic cancer risk — especially if you have had recent unexplained weight loss, new-onset diabetes, or a family history of the disease — it is worth raising those concerns with your doctor.
The Study's Limitations
This study was observational (watching what happened, not randomly assigning treatment). While PRISM identified high-risk groups accurately, the study did not test whether screening those patients actually improves survival. That is the critical next step. There is also the risk of overdiagnosis — flagging people as high-risk when they will never develop cancer — which can cause anxiety and unnecessary procedures.
The researchers describe this as proof that EHR-based risk stratification is scalable and real-world ready. The next phase will involve prospective clinical trials where high-risk patients identified by PRISM are offered surveillance imaging, and outcomes are tracked. If those trials show that earlier detection translates to longer lives, it could reshape who gets screened for pancreatic cancer in the United States and beyond.
