Artificial intelligence can be trained to spot pancreatic cancer risk three years before diagnosis: study

According to a new study, artificial intelligence algorithms may be the key to identifying who is at the most risk of developing one of the most notoriously difficult cancers to diagnose early: pancreatic cancer.

In a study published Monday in the peer-reviewed journal Nature Medicine, researchers found that with the help of AI, they were able to identify those most at risk for developing pancreatic cancer around three years before diagnosis purely by using the patients’ medical records.

It could be a game-changer in battling this type of cancer, which is fast-growing and hard to detect, researchers say.

“One of the most important decisions clinicians face day to day is who is at high risk for a disease, and who would benefit from further testing, which can also mean more invasive and more expensive procedures that carry their own risks,” Chris Sander, faculty member in the Department of Systems Biology in the Blavatnik Institute at Harvard Medical School and study co-senior investigator, said in a press release.

“An AI tool that can zero in on those at highest risk for pancreatic cancer who stand to benefit most from further tests could go a long way toward improving clinical decision-making.”

A family history of pancreatic cancer and the presence of certain genetic mutations will flag individual patients for targeted screenings and early testing, but this still leaves many patients slipping through the cracks who had no way to know they were at a higher risk.

Identifying any cancer early on is important for recovery, but it’s especially urgent with pancreatic cancer, which is one of the most difficult cancers to detect in its early stages when it’s most curable.

It begins in the pancreas, an organ behind the stomach which creates enzymes and hormones to assist in digestion and blood sugar regulation respectively. But pancreatic cancer often doesn’t cause any symptoms until it has spread beyond the pancreas to other organs, at which point the chances of defeating it are much lower.

Clinicians have nicknamed the pancreas “the angry organ,” the release explained, because of how difficult it is to perform a biopsy on the pancreas.

According to the Canadian Cancer Society, only around 10 per cent of people diagnosed with pancreatic cancer in Canada will survive past the five year mark.

In this new study, researchers trained artificial intelligence models on clinical data from Denmark spanning 6.2 million patients over 41 years. Out of this sample, around 24,000 would be diagnosed with pancreatic cancer at some point.

By sorting through this enormous wealth of data, the AI algorithms were able to put together warning signs for pancreatic cancer within specific time frames. Once the AI had learned these “disease codes,” researchers found that the AI algorithms were substantially more accurate in predicting who would develop pancreatic cancer compared to population-wide estimates based on the incidence levels of the disease.

To test it further, they took the most successful algorithm and used it with a new cohort: three million patients from the U.S. Veterans Health Administration data set, spanning 21 years.

This data set contained 3,900 patients who were diagnosed with pancreatic cancer. Researchers found that their AI algorithm was slightly less predictive than with the Denmark cohort, which was a truly national sample, but that when retrained on U.S. data, its accuracy improved.

Pancreatic cancer is much more difficult to screen for compared to other cancers. While a mammogram, pap smear and blood test will easily allow doctors to search for breast cancer, cervical cancer and prostate cancer respectively, screening methods for pancreatic cancer are more expensive. Researchers pointed out that doctors are understandably less likely to order a CT scan or MRI for a patient to screen for pancreatic cancer without the family history usually used to assess risk.

Being able to identify those who truly need more testing and those who don’t will not only save time and resources, but could catch more cases of this aggressive cancer earlier.

If patients are diagnosed in the early stages, the five year survival rate surges to 44 per cent, the release said — but currently, only around 12 per cent of cases are diagnosed that early. If the tumour spreads beyond the pancreas, the five year survival rate can go as low as two per cent.

“That low survival rate is despite marked advances in surgical techniques, chemotherapy, and immunotherapy,” Sander said. “So, in addition to sophisticated treatments, there is a clear need for better screening, more targeted testing, and earlier diagnosis, and this where the AI-based approach comes in as the first critical step in this continuum.”

Researchers noted that the shift in accuracy for their AI algorithm when introduced to a new country’s data suggests that for this method to be successful, AI models need to be trained on large datasets or as locally as possible to capture specific demographic patterns for risk.

“Many types of cancer, especially those hard to identify and treat early, exert a disproportionate toll on patients, families and the healthcare system as a whole,” Søren Brunak, director of research at the Novo Nordisk Foundation Center for Protein Research at the University of Copenhagen and co-senior investigator, said in the release. “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.”