When a doctor notices a saclike pocket of fluid (known as a cyst) on a patient’s pancreas during a checkup, they are obliged to recommend surgery to remove it because it could be cancer. There is still no good way of telling which cysts harbor a deadly form of cancer and which are benign.
The unfortunate thing is, these surgeries come with a 50% chance of complication and a 5% chance of death and around 78% of them end up having been unnecessary because the cyst was not cancerous after all. Around 800,000 patients in the United States alone have to go through this situation every year.
Luckily for these people, there is a new machine-learning algorithm that could help. Surgeons and computer scientists at Johns Hopkins University have built a test called CompCyst (for comprehensive cyst analysis) that is significantly better than today’s standard of care (medical imaging and human observations) at predicting whether patients should be sent home, monitored, or undergo surgery. They have published the study of the new AI in the journal Science Translational Medicine.
Senior author Anne Marie Lennon, director of the pancreatic cyst program at the Johns Hopkins Kimmel Cancer Center, says she expects to offer the test to Hopkins patients within 6 to 12 months and hopes to make it commercially available following a larger, prospective clinical trial. “We are extremely excited about the results of this,” she said at a press conference related to the study.
Study author Christopher Wolfgang, director of surgical oncology at the Kimmel Cancer Center said that the vast majority of pancreatic cysts are benign, but right now doctors track them all. “We need to follow all patients, on the order of hundreds of thousands of patients, with expensive and, in some cases, invasive tests to find those few patients who will progress to cancer,” he said.
Since follow-up testing can involve radiation exposure and complications, as well as provoke anxiety, the team of researchers set out to build a tool to sift through patient information in the hopes of identifying patterns to distinguish low-risk from high-risk cysts. They began by gathering data from hundreds of patients at Hopkins and 15 medical centers around the world who were diagnosed with a cyst and then underwent surgery to have it removed. Every cyst removed was then examined and classified as having either no risk, a small risk, or a high risk of progressing to cancer.
CompCyst is focused around a machine learning algorithm called MOCA, for Multivariate Organization of Combinatorial Alterations. It’s a test that combines molecular data—including DNA mutations and chromosomal changes—with protein information from extracted cyst fluid and imaging tests. Co-author Marco Dal Molin, a postdoctoral research fellow at Hopkins, said that the algorithm tests millions of combinations of data points to predict the right treatment pathway with high sensitivity and specificity.
The algorithm the CompCyst test uses was trained with data from 436 patients. It was then tested on a second, separate set of data from 426 patients. Overall there were three sets of patient groups and CompCyst outperformed the standard-of-care that doctors use today in all of them.
It correctly predicted:
- 60% of patients who should have been sent home versus 19% using standard-of-care.
- 49% of patients who should have been monitored versus 34% using standard-of-care.
- 91% of patients in need of surgery versus 89% using standard-of-care.
Overall, an estimated 60% to 74% of the patients would have avoided unnecessary surgery if CompCyst had been used.
Co-author Bert Vogelstein, a professor of oncology and co-director of the Ludwig Center at Hopkins, said:
Combining clinical and genetic features using machine learning is the wave of the future to inform clinical judgment not only about pancreatic cysts but about many other diseases.
