BIG THINK
An AI that can identify the patients most at risk of pancreatic cancer could lead to earlier detection of the deadly disease, which currently kills 88% of patients within five years of diagnosis.
The challenge: People with pancreatic cancer typically don’t experience any symptoms until a tumor is large or the cancer has spread to other organs. As a result, most aren’t diagnosed until their cancer is advanced and much harder to treat.
Screening people without symptoms for pancreatic cancer could lead to earlier detection, but the only tests for it are expensive or invasive. As a result, they’re reserved for the few people doctors believe are at high risk of pancreatic cancer, due to a family history of the disease, for example.
The AI looks for patterns of health problems that often precede pancreatic cancer.
What’s new? Now, researchers at Harvard Medical School and the University of Copenhagen have developed an AI that can identify which of a healthcare system’s patients are most at risk of pancreatic cancer based solely on medical records, flagging them for screening.
“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,” said Chris Sander, the study’s co-senior investigator.
How it works: The researchers trained and tested their AI using the health records of 6.2 million people in Denmark. The records spanned 41 years and included about 24,000 people with a pancreatic cancer diagnosis.
The AI can identify the patients within a larger population at highest relative risk of the disease.
From the training data, the AI learned to identify patterns of health problems — including gallstones, anemia, and type 2 diabetes — that people with pancreatic cancer often experience in certain sequences prior to their cancer diagnosis.
“Cancer gradually develops in the human body, often over many years and fairly slowly, until the disease takes hold,” Sander told the Register. “The AI system attempts to learn from signs in the human body that may relate to such gradual changes.”
These issues are all quite common, of course, and pancreatic cancer is very rare, so by themselves they don’t indicate much about an individual’s risk. But the AI seems to be able to pick up on patterns in health events that aren’t easily noticeable by skimming a patient history.
The results: While the AI can’t just look at one person’s record and say they have an absolute X% chance of developing pancreatic cancer in the next year, it can identify the patients within a larger population at highest relative risk of the disease, according to the team’s study…
