For a study published earlier this month, researchers at Harvard Medical School (HMS) and the University of Copenhagen gave The AI ​​tool has access to 9 million patient records3 in the Danish medical system and Virginia hospitals in the USA. They learned to work with tools to read diagnostic codes and identify patterns between a cancer diagnosis and other previous diagnoses. He then fed the tool a new set of medical records and asked it to determine each patient’s risk of pancreatic cancer at 3 months, 6 months, 1 year, 2 years, and 3 years.

When assessing short-term risk, the tool flagged more obvious diagnostic codes, such as unspecified jaundice, biliary tract disease, abdominal and pelvic pain, weight loss, and digestive tract neoplasms, which researchers believe may actually be symptoms of pre-existing cancer. But when asked to assess long-term risk, the tool picked up diagnoses that aren’t directly related, such as type 2 diabetes and non-insulin-dependent diabetes.

The researchers believe the tool is more accurate than current estimates for the entire population, and at least as accurate as genetic testing, which is currently limited to those already identified as high-risk.

In one part of the experiment, the researchers gave the tool an example of a real population of 1 million patients and asked it to identify the 1,000 patients with the highest risk of pancreatic cancer. Out of 1,000 selected, 320 of them got pancreatic cancer. And while some of the selected patients would have been classified as high-risk by doctors, the researchers believe that at least 70 of them would have been newly identified as high-risk by the AI ​​tool.