AI shown to predict risk of pancreatic cancer well before symptoms appear

Scientists have found that artificial intelligence could be an effective tool in predicting pancreatic cancer before a single symptom appears, according to a study published in the journal Nature Medicine on May 8.

A team of researchers led by Copenhagen University Hospital in Denmark and Harvard Medical School in Boston completed a sweeping study to determine whether AI could flag a person’s risk of developing the disease.

The results exceeded their expectations, with the model successfully predicting risk up to three years before diagnosis.


In 2023, about 64,050 people in the U.S. will be diagnosed with pancreatic cancer and about 50,550 will die from the aggressive disease, the American Cancer Society (ACS) says.

The five-year survival rate across all stages is just 12% in the U.S.

While early screening and detection can improve outcomes, a vast majority of cases are diagnosed at advanced stages.

In the study, the researchers used AI and machine learning methods to analyze medical data from six million patients in Denmark and three million patients in the U.S.

“AI is very good at learning from large databases, even if they’re somewhat noisy, but you need lots of data in order for it to be effective,” study co-author Dr. Chris Sander, PhD, professor of cell biology at Harvard Medical School, told Fox News Digital in an interview.

Only a very small portion of those patients ended up developing pancreatic cancer.

The researchers’ goal was to use AI to find the differences between the two paths — those who were ultimately diagnosed and those who remained disease-free.

“AI-based screening is an opportunity to alter the trajectory of pancreatic cancer.”

The technology scanned the data for up to 2,000 disease codes across each patient’s medical history that could predict the likelihood of developing cancer within a certain time frame.

The timing of the diseases — many of which weren’t even related to the pancreas — was an important factor in predicting risk.

“The study aimed to see whether the patient was on a path to pancreatic cancer,” said study author Dr. S?ren Brunak, a Danish biological and physical scientist at the University of Copenhagen, in an interview with Fox News Digital.

Every patient has a complete disease trajectory, Brunak explained, comparing it to an “entire movie of all diagnoses and procedures.”

He added, “We were not just asking which diseases the patient had before, but also in what order they appeared, so we could identify any predictive signals.”

“We looked for risk factors from their past that might have an impact on whether they would get this rare form of cancer.”

Said Brunak, “We looked for risk factors from their past that might have an impact on whether they would get this rare form of cancer.”

Ultimately, the goal was to learn how pancreatic cancer actually develops, the biology behind it, which genes can predict risk and what other factors can make someone predisposed to the disease, Brunak said.

Potential to ‘improve patient outcomes’

When the researchers applied their AI model to predict the 1,000 patients who were at the highest risk, they found that about 320 of them eventually developed pancreatic cancer.


Different versions of the AI models predicted risk within different time frames — six months, one year, two years and three years before diagnosis.

The accuracy increased for the shorter time frames, Sander explained.

“Similar to the weather, the prediction was more accurate one year or one month out,” he said. “The prediction for shorter time scales was quite good.”

Dr. Harvey Castro, a Dallas, Texas-based board-certified emergency medicine physician and national speaker on artificial intelligence in health care, was not involved in the study but was impressed by its findings.

“The study’s results have the potential to inform the design of surveillance programs for patients at elevated risk, which could improve patient outcomes and quality of life,” he told Fox News Digital.

“The study can significantly impact treatment options and patient outcomes by focusing on the early detection of pancreatic cancer,” he added.

Current screening methods could miss cases

Early detection and treatment are key to improving pancreatic cancer survival rates, experts agree — but the current screening methods have some key limitations, they also say.

Most doctors rely on imaging tests, endoscopic ultrasounds, tissue biopsies and blood tests, according to the Mayo Clinic.

These types of targeted tests are usually not conducted until a physician already suspects that a patient might have the disease.

Additionally, with the high cost of such screenings as MRIs, CT scans and ultrasounds, these sophisticated tests may not be available to people who don’t have symptoms or proven risk factors, noted Sander, the study’s co-author.

“If we can move even a fraction of cancer care to earlier detection and treatment, it will have a huge benefit.”

Another problem with the current screenings is that they are notorious for generating false positives, Brunak pointed out.

“This overloads the health care system and patients get concerned without reason,” he said.

The new study suggests that by applying AI-based screening to a broader population, those who were unknowingly at a higher risk of the deadly disease could get earlier diagnoses and faster treatment before the cancer progresses to more advanced stages.

‘Consider rolling it out to broader community’

The current study was retrospective, looking back at existing data sets over a period of time in the past.

Next, Sander said they will apply what they learned in a prospective, forward-facing way.

“We will move forward with clinicians and try it out in the health system, start out small and see how well it works,” he told Fox News Digital.


“Then, based on how it performs, we would consider rolling it out to a broader community.”

AI screening in a clinical setting won’t happen overnight — it could take a few years, Sander said — but he doesn’t believe it would take as long as producing new cancer drugs.

“If we can move even a fraction of cancer care to earlier detection and treatment, it will have a huge benefit — not only for the patient, but also economically, given how expensive late-stage cancer is,” he said.

“AI-based screening is an opportunity to alter the trajectory of pancreatic cancer, an aggressive disease that is notoriously hard to diagnose early and treat promptly when the chances for success are highest,” Brunak said, per a press release published by Harvard Medical School.

In the meantime, Sander stressed the importance of understanding family history, requesting genetic testing and watching for early signs, such as unexpected weight loss or late-onset diabetes.


“Although not as powerful as the AI method, these are still important,” he said.

Certain lifestyle modifications, such as refraining from smoking, exercising regularly and adhering to a nutritious diet, can also help reduce risk.

Dr. Castro noted that while the study has several key strengths, it also presents some limitations and concerns, including the challenging treatment landscape for pancreatic cancer.

“The complexity of the disease and the need for further advancements in treatment options should be acknowledged alongside the potential benefits of early diagnosis,” he said.


“Further research and exploration of alternative approaches may help improve the effectiveness and generalizability of these models, ultimately contributing to better treatment options and outcomes for patients with pancreatic cancer,” he said.

The study was funded in part by grants from Stand Up To Cancer, the Lustgarten Foundation, the Novo Nordisk Foundation and the National Institutes of Health.