Harnessing artificial intelligence, clinical data and genomics, Ludwig Johns Hopkins researchers develop a test that could help reduce unnecessary surgeries
JULY 17, 2019, New York- An international scientific team led in part by Ludwig Johns Hopkins Co-director Bert Vogelstein has shown that a laboratory test using artificial intelligence tools has the potential to more accurately sort out which people with pancreatic cysts will go on to develop pancreatic cancers. The test, dubbed CompCyst (for comprehensive cyst analysis), incorporates measures of molecular and clinical markers in cyst fluids, and appears to be on track to significantly improve on conventional clinical and imaging tests, the research team says. A description of the work is published in the July 17 issue of Science Translational Medicine.
Pancreatic cysts are found in 4% of people in their 60s and 8% of people over age 70. Though only a small fraction of cysts progress to cancer, it is difficult to know which ones without surgical intervention. Surgeons rarely miss the more dangerous types of cysts when they operate but many patients also undergo surgeries that are later found to have been unnecessary.
CompCyst, which would complement existing clinical and imaging criteria for cyst evaluation, was much better than current methods in identifying which patients needed and were more likely to benefit from surgery. It was also better at capturing which were unlikely to benefit from surgery or only required further monitoring. The researchers found that using the test would have spared from surgery more than half of patients who underwent cyst removal later deemed unnecessary because the cysts were unlikely to have caused cancer.
Developed by Ludwig Johns Hopkins and other Sidney Kimmel Comprehensive Cancer Center investigators, CompCyst was created with patient data-including clinical impressions and symptoms-images from CT scans and molecular features of cyst fluid. The researchers evaluated the molecular profiles, like DNA mutations and chromosome changes, of 862 pancreatic cysts. They then fed the molecular information, along with clinical and radiologic data, into a computer-based program that used artificial intelligence to classify patients.
Based on histopathological analysis of the cysts, the researchers found that surgery was not needed for 45% of the patients who underwent surgeries because clinicians could not determine if the cysts were dangerous. If CompCyst had been used on these patients, the researchers estimate that 60% to 74% of the patients (depending on the cyst type) could have been spared unnecessary surgeries.
The study had several limitations, the researchers note, including that pancreatic cyst fluid was obtained at the time of surgery, and that the cysts evaluated are more atypical than those seen in routine clinical practice.
More detail about these findings is available in the Johns Hopkins release from which this summary is derived.