AI breakthroughs are meeting patients’ needs sooner
Pancreatic cancer remains one of the deadliest forms of the disease, with the American Cancer Society reporting a five-year survival rate of 13% in its 2024 statistics. It’s one of the lowest survival rates of any cancer, due in large part to the challenge of identifying pancreatic cancer early. But Mayo Clinic clinician-investigators have always believed it was possible to identify pancreatic cancer at a stage before symptoms become present — significantly boosting the odds of survival.
Time after time, early screening risk programs failed to make meaningful inroads. Individual review of patient imaging files is challenging — the human eye isn’t proficient at making the diagnosis that early, causing false-positive rates to soar.
But Mayo Clinic had a differentiator. It was a deep, longitudinal dataset dating back decades that included people who had been diagnosed with late-stage pancreatic cancer. The depth of the data meant that researchers could look back even earlier in patient records to review imaging taken 12 months or more prior to the discovery of cancer.
“The data allowed our teams to train an artificial intelligence (AI) model on that patient cohort that could pick up those early-stage pancreatic cancer cases with really high sensitivity and specificity,” says Matthew Callstrom, M.D., Ph.D., who serves as medical director of the Strategy Department and medical director of Mayo Clinic’s Generative Artificial Intelligence Program. “It could do this at a very high accuracy — 97%.”
Our aspirations for AI are to impact patient outcomes.
— MATTHEW CALLSTROM, M.D., PH.D.
Dr. Callstrom says the work of Ajit Goenka, M.D.; Panos Korfiatis, Ph.D.; and Eric Williamson, M.D., and their teams showed that Mayo Clinic could determine whether a person had pancreatic cancer through applied AI at an early point in disease progression. Currently in the U.S., lung cancer causes the most cancer deaths, followed by colorectal cancer and then pancreatic cancer.
“We think we can shift the needle on that,” Dr. Callstrom says.
While discovery in medicine is challenging, translating those discoveries into clinical practice can be even more complex, according to Dr. Callstrom. This year, Mayo Clinic research scientists including Aadel Chadhuri, M.D., Ph.D.; Suresh Chari, M.D. (emeritus); Dr. Goenka; and Mark Truty, M.D., M.S., are testing the pancreatic cancer algorithm in the AI-PACED clinical trial, an area where Mayo Clinic excels. By focusing on patients at high risk for pancreatic cancer, particularly those with a family history of the disease, Mayo Clinic can evaluate new AI-driven approaches with more impact.
Mayo Clinic embeds scientists directly in clinical practice where they work alongside clinicians who understand the problems patients face and have access to the data. Together, they develop and test AI solutions in real time.
“That’s across the entire organization, and there are incredible discoveries being made all over,” he says.
Like many clinicians at Mayo Clinic, Dr. Callstrom’s background is varied and unique. He was a chemical engineering major at the University of Minnesota who stayed to do a Ph.D. in chemistry, followed by a postdoctoral opportunity at Harvard University.
From there, he began teaching chemistry at The Ohio State University. When one of his friends who was also a colleague at Ohio State was diagnosed with colon cancer, the experience deeply affected him.
“I became very motivated to try to impact patients’ lives. I went into medical school and was very fortunate to get into Mayo Medical School (now called Mayo Clinic Alix School of Medicine),” Dr. Callstrom says. “I’ve done all my training here, and my clinical emphasis is on treating patients with cancer.
“So, I do interventional oncology treating patients and trying to help them through a very difficult period in their life. And through the other aspects of my work in AI, we are hopefully developing cures for them.”
In fact, Mayo Clinic has been looking at AI approaches for more than a decade, and its emergence in imaging over the most recent years dovetailed with Dr. Callstrom’s interests in figuring out how he could help the lives of as many patients as possible.
“He’s a pioneer in AI, working on machine learning back when most people weren’t thinking about that,” says Jim Rogers, CEO, Mayo Clinic Digital Pathology. “His singular focus has always been: ‘What can I do to improve care for folks?’
“There’s real courage and vision there. Because of him and others throughout the institution, Mayo now has more machine learning AI algorithms in actual practice — not just theory — than any other organization.”
Results from that early work are popping up all over Mayo Clinic, which has more than 60 AI models already deployed behind the scenes every day. This means that the solutions have been built and validated and are running automatically.
One area ripe for innovation is in cardiovascular medicine. Itzhak Zachi Attia, Ph.D.; Paul Friedman, M.D.; Francisco Lopez-Jimenez, M.D., M.S.; and Peter Noseworthy, M.D., M.B.A., among a team of many others, are using EKG data to train various models to look for issues. Already one algorithm has been found to be effective in accurately identifying atrial fibrillation (AFib) at an early stage. Using a similar approach to the clinical trial in pancreatic cancer, the team discovered that AFib could be detected more than six months before it becomes clinically important.
“The reason that’s impactful is that once atrial fibrillation starts, clots can form in the heart, and those clots can travel to the brain and cause a stroke,” Dr. Callstrom says. “So, if you can stop that or prevent it from happening, you avoid a debilitating outcome for a patient and get them on medication early.”
MAYO CLINIC EMBEDS SCIENTISTS DIRECTLY IN CLINICAL PRACTICE WHERE THEY WORK ALONGSIDE CLINICIANS WHO UNDERSTAND THE PROBLEMS PATIENTS FACE.
Another model can diagnose patients with low ejection fraction, a form of heart failure, before symptoms are present, allowing doctors to intervene before the issue becomes critical.
“That model runs on every Mayo patient who has an EKG now,” Dr. Callstrom says. “Our cardiovascular medicine team ran a clinical trial in our health system with 20,000 patients to find out that this did have an impact on patient outcomes. We were able to measure it, and it picked up many patients with unsuspected heart failure or AFib. It was pretty amazing.”
Jim Rogers, who also serves as the senior administrator for the Generative AI Program, emphasizes that Mayo Clinic’s AI approaches align with the organization’s Bold. Forward. strategy to enhance internal care delivery while pursuing broader healthcare transformation.
The work is purposefully iterative. While each project is initially focused on a specific application or disease, Mayo’s established structure allows new learnings to be expanded to other areas by encompassing disciplines like genomics, pathology, imaging, text analysis, voice recognition and more.
The goal? Practical integration of all these elements to benefit physicians’ abilities to solve the needs of patients.
“We’re learning from every activity and with each step forward,” Jim says. “We’re not doing this out of mere curiosity — we want practical impact as quickly as possible. When a patient walks into one of our rooms, they expect us to have all the information needed to effectively treat them.”
AI’s impact isn’t just limited to disease identification and treatment. It’s helping physicians connect on a deeper level with patients too.
Mayo Clinic is using AI-powered ambient listening technology to transform patient encounters. Instead of clinicians typing notes during conversations — which can detract from personal interaction — the AI system captures and summarizes the discussion automatically. Then, the clinician can rapidly review and approve the notes, ensuring accuracy in the patient’s record.
We’re not doing this out of mere curiosity — we want practical impact as quickly as possible.
— JIM ROGERS, CEO, MAYO CLINIC DIGITAL PATHOLOGY
Dr. Callstrom says this innovation eliminates a difficult choice many physicians face: either document during the visit or spend extra time recording notes later in the day, hours after the appointment. With ambient AI handling documentation in the background, providers can focus entirely on patient interaction, ultimately bringing more humanity into healthcare.
Even with all the advances, it’s what’s on the horizon that excites Dr. Callstrom the most.
Mayo Clinic is leveraging AI to analyze complex data in unprecedented ways — from genomics and digital pathology to cellular-level imaging. It’s all part of the organization’s mission in this new era to ensure the needs of the patient come first.
“Our aspirations for AI are to profoundly impact patient outcomes,” Dr. Callstrom says. “One of the things we always talk about is trying to identify disease at a state where we can intervene early. Previously, it’s been hypothetical — ‘If we had the right data, we could do this.’ It turns out with AI we can.”
Kari Turkowski, Ph.D., has never been one to shy away from a challenge.
AI-driven tools are accelerating new knowledge, solutions and technologies.
Bahman and Becky Hoveida believe in a future of healthcare where cancers are diagnosed earlier.