The researchers used data from patient medical records in both the United States and Denmark from 1977 to 2020. They surveyed a group of 6.2 million Danish patients, of whom 23,985 were ultimately diagnosed with pancreatic cancer, and through the Department of Veterans Affairs, where 3 million veterans are undergoing treatment, 3,864 of whom I was finally diagnosed.
The researchers used machine learning models to analyze the data and learn to predict cancer risk based on symptoms and various diagnostic codes in patients’ medical records.
Some of the symptoms associated with predicting higher risk have not been previously associated with pancreatic cancer. Gallstones, type 2 diabetes, anemia, and gastrointestinal symptoms such as vomiting and abdominal pain were all associated with an increased risk score three years before diagnosis.
The researchers wrote that in a real-world scenario, about 320 out of every 1,000 people the AI model identified as high risk would develop pancreatic cancer. The tool could make screening more affordable by targeting high-risk patients for surveillance, the researchers write.
Currently, the U.S. Preventive Services Task Force does not recommend pancreatic cancer screening for asymptomatic people. However, screening high-risk patients has a higher chance of long-term survival.
“AI tools that can focus on those at highest risk of pancreatic cancer and who would most benefit from further testing could greatly help improve clinical decision-making,” study co-authors said. said biologist Chris Sander. A Harvard Medical School lab is dedicated to using machine learning and other technologies to solve biological problems, according to a news release.
Such tools, if applied at scale, could extend life and improve outcomes, Sander said.
