By Deborah Borfitz
August 12, 2019 | Parallel efforts at two universities are getting closer to developing a tool for identifying which people with pancreatic cysts may go on to develop pancreatic cancer.
Washington University is the lead institution on a study where an antibody biomarker, called mAb Das-1, recently outperformed current clinical guidelines by more than 20 percentage points in identifying pancreatic cystic lesions (PCL) at high risk for harboring cancer among patients undergoing surgical resection. Among the patent holders for its possible use in the detection of pancreatic cancer is Koushik K. Das, M.D., assistant professor of medicine in the university’s division of gastroenterology. The monoclonal antibody was discovered several decades ago by his father, gastroenterologist Kiron M. Das, M.D., in a totally different context.
Meanwhile, in another just-completed trial spearheaded by researchers at Johns Hopkins, an artificial intelligence-driven laboratory test known as CompCyst has demonstrated its ability to help physicians identify patients with benign cysts who can be safely discharged and potentially decrease the number of unnecessary operations.
The Johns Hopkins study had collaborators from 15 other medical centers across the US as well as in Japan, Korea and Europe. The aim was to develop a test to correctly identify three groups—those with benign cysts that will never develop cancer, those with cysts that harbor cancer or high-grade dysplasia (HGD) who require surgery, and those with mucin-producing cysts that need surveillance, says Anne Marie Lennon, M.D., Ph.D., professor of medicine and director of the Johns Hopkins multidisciplinary pancreatic cyst clinic.
"Current tools are not accurate enough to differentiate the three categories of cysts," says Lennon. "Therefore, some patients with benign cysts get multiple scans, follow-up or even surgery that is not needed, and some patients with mucin-producing cysts undergo surgical resection which, in hindsight, was not needed at that time [i.e., on pathology there was no HGD or cancer]."
Physicians are in a quandary when cysts are incidentally identified, as they often are, via CT and MRI scans—for example, when patients are being tested for kidney stones or low back pain, says the younger Das.
Mucin-containing cysts, as opposed to simple liquid cysts, are precancerous lesions, but not all of them progress to cancer, Das says. Thus, doctors must decide between complex surgery, ongoing surveillance or discharging patients based on limited information.
It's hard to determine the percentage of all pancreatic cysts that are cancerous, since the denominator is unknown. "What we're seeing is probably just the tip of the iceberg," says Das. Researchers have estimated that 2%-4% of patients ages 50 to 70 have pancreatic cysts, increasing to 8%-9% after age 80.
Multiple Measures
The first step for clinicians is determining the type of cyst a patient has since that dictates the treatment path, Lennon says. Unfortunately, the diagnostic tools currently used—largely a combination of imaging and fluid analysis—are imperfect, so many people undergo surveillance and occasionally surgery in the mistaken belief that they have a precancerous or cancerous lesion. It has been estimated that one-quarter of cyst patients needlessly undergo surgical resection, she notes.
A more reliable set of tests would not only prevent unnecessary high-risk surgeries but also save people who don't need follow-up surveillance the cost and inconvenience of ongoing CT or MRI imaging, Lennon says. Currently, many experts recommend lifelong surveillance of mucin-producing cysts with MRI or CT because it is hard to distinguish those with and without malignant potential.
To that end, Johns Hopkins Kimmel Cancer Center-led investigators developed CompCyst and put it to the test in a global trial involving 862 patients, she says. As described in the July 17 issue of Science Translational Medicine, the test protocol was trained on half of the patients (436) and CompCyst was then validated on the rest of participants (426).
The test involves the successive application of three composite markers—comprising mutations in 11 genes, loss of heterozygosity, aneuploidy, two protein markers (carcinoembryonic antigen and vascular endothelial growth factor A) and a solid component observed upon imaging.
According to the American Cancer Society, pancreatic cancer is expected to become the second leading cause of cancer-related death in the US by next year, surpassing colorectal cancer. For all stages combined, 91% of pancreatic cancer patients die within five years of diagnosis.
The composite test was developed using clinical data, including symptoms, images from CT, MRI and endoscopic ultrasound scans and molecular features such as DNA alterations in the cyst fluid, Lennon says. The information was fed into a computer-based program using a previously published Multivariate Organization of Combinatorial Alterations (MOCA) machine learning algorithm to classify patients into a no-risk, low-risk (surveillance recommended) or high-risk (surgery recommended) groups.
MOCA sought out the best combination of features for the highest predictive value for each of the three groups, Lennon says. "Most of the clinical features we currently use to determine who to send to surgery [including cyst size, dilation of the pancreatic duct, and pancreatitis] are not specific enough and therefore not included."
The composite marker panel outperformed current clinical management by correctly identifying 60% (versus 19%) of patients as suitable for discharge and 49% (versus 34%) of patients in need of monitoring. It was on par with the current standard of care in correctly identifying when surgery is indicated, she adds.
More than 193 patients (45%) in the validation cohort who underwent surgical resection did not require the surgery, says Lennon. The use of CompCyst in that group could potentially have decreased the number of patients who went under the knife by 60% since their cyst would have been identified as low- or no-risk.
Overall, CompCyst had a significantly higher accuracy (69%) for classifying patients into one of the three groups—surgery, surveillance or discharge, she says.
Validating a Biomarker
There is a dearth of highly accurate biomarkers available to risk-stratify pancreatic cysts, so a series of clinical guidelines has emerged to help clinicians make a preoperative risk assessment, says Das. Among their recommendations are assessing clinical symptoms such as jaundice, pancreatitis or enlarged lymph nodes; imaging characteristics based on the size of the cyst and whether the pancreatic duct is involved; and if previous cytology results were concerning or positive for malignancy.
The problem with the guidelines is that one set may be incredibly sensitive but non-specific—meaning they will pick up almost every single patient who has cancer but also result in a lot of needless operations—and another set are not nearly as sensitive but much more specific, so some cancers get missed but everyone who gets surgery truly needs it. With Das-1, "we're trying for something that is both highly sensitive and highly specific," he says.
The mAb Das-1 biomarker had previously been found to be highly specific for adult normal colon epithelium. Several years ago, Das began working with his colleague at Massachusetts General Hospital (MGH), Mari Mino-Kenudson, M.D., whose earlier work had shown the strong association of an "intestinal-type" of pancreatic cyst being associated with cancer in PCL.
"We realized we had a biomarker reactive to intestinal cells that may be useful in the context of PCL as well," says Das. In subsequent research published in 2014, the biomarker strongly correlated with pancreatic cysts at high risk of becoming cancerous.
Das is first author of the most recent validation study involving mAb Das-1, results of which were published online June 5 in the journal Gastroenterology. In this study, researchers collected fluid from the PCL of 169 patients who opted for surgical removal. Das worked with collaborators from MGH, Johns Hopkins School of Medicine (including Lennon), Memorial Sloan Kettering Cancer Center and the Rutgers-Robert Wood Johnson Medical School.
When patients' cyst fluid was analyzed using a test to detect mAb Das-1, the biomarker was 95% accurate at predicting the cysts that were likely to become cancerous, he says. The best physicians can do now, following guidelines of various professional societies, is achieve a 74% positive predictive rate.
Study results may have been affected by the fact that participants were all healthy enough and willing to undergo surgery, he adds, and the goal is to have a test that is useful for all patients with a pancreatic cyst. The next step is a new prospective study to see if the biomarker can identify cysts likely to become cancerous before a patient undergoes surgery.
Das says he has been collecting cyst fluid samples from patients and will continue doing so until he has amassed a quantity large enough to validate the biomarker's performance on "all comers" on a prospective basis. As Das notes, "Even in a large referral center like ours where we do over 100 pancreas resections a year, it takes time to see the outcomes of patients who undergo surgeries versus those who do not."
Clinical practice is also in flux, Das says. Over the past decade, the pendulum has swung toward more judicious use of the surgical option.
Eventually, Das hopes to license the patent to the mAb Das-1 test so that it can become available more broadly to the market.
Next Steps
"As someone who looks after patients with pancreatic cysts, trying to identify pancreatic cancer earlier, I would say both of these studies are a huge step forward," says Lennon, adding that she is particularly excited about the level of collaboration they represent across institutions and geographies. Prospective validation studies will reveal exactly how well each tool performs in a scenario closer to real-world clinical practice.
The CompCyst test will be available to patients at Hopkins in another nine to 12 months as the next step toward regulatory approval, Lennon says. The study is likely to enroll over 1,000 patients, with solid results possible within two years. She suspects the assay will do even better in routine cases than the atypical ones in the proof-of-concept study where all patients had gone to surgery.