More than 99% of pancreatic cysts detected by doctors do not progress into pancreatic tumors. But distinguishing between low- and high-risk cysts is very difficult. A research team led by Johns Hopkins Co-director Bert Vogelstein has now harnessed machine learning to create a test for this purpose that coordinately assesses selected clinical features, imaging, and genetic and biochemical markers in cyst fluid to produce an answer. CompCyst was trained on data from 436 patients to identify people who needed surgery, those who should be monitored and those who needed no further surveillance. It was then evaluated in an international, multicenter study of another 426 patients. The researchers compared CompCyst’s predictions against the gold standard diagnostic: post-surgical histopathological evaluation of the cysts. CompCyst correctly predicted 60% of patients who should have been sent home, as opposed to the 19% detected using standard preoperative criteria. If CompCyst had been used to decide care for these patients, 60% to 74% of them might have avoided unnecessary surgery. CompCyst detected 49% of people who should simply have been monitored, compared to 34% using current methods, and 91% who needed surgery, compared to 89% using existing methods. The findings were published in July in Science Translational Medicine.
This article appeared in the November 2019 issue of Ludwig Link. Click here to download a PDF (1 MB).