Pancreatic cancer is a particularly aggressive and usually lethal malignancy. Pancreatic ductal adenocarcinoma (PDAC), the most common type, is the third largest cause of cancer death in the United States, even though it is uncommon. Early disease has very subtle or no symptoms, and pancreatic cancer is therefore often diagnosed at advanced stages. Meanwhile, clinicians lack a standard diagnostic lab tool or established method for early detection. Because the pancreas is located deep in the abdomen, hidden behind other organs, imaging is also difficult. Clinicians long for minimally invasive screening methods and accurate early diagnostic methods, especially blood tests.
Early detection efforts now focus on high-risk patients with genetic mutations known to cause pancreatic cancer, older people, and those with family history of the disease. However, about 75% of pancreatic cancer occurs in patients who are not considered high-risk. The United States Preventive Services Task Force in 2019 recommended against pancreatic cancer screening of asymptomatic adults, citing lack of data.
Two recent studies highlighted the ongoing search for early detection methods. One assessed pancreatic cancer risk using using CA19-9 and bilirubin concentrations in the blood to distinguish early-stage pancreatic cancer from benign neoplasms. (JAMA Netw Open 2023; doi:10.1001/jamanetworkopen. 2023.31197.) The other described an artificial intelligence (AI) model that may point to a population screening method to prompt monitoring and expedite diagnosis and treatment (Nat Med 2023; https: doi: 10.1038/s41591-023-02332-5).
These are just two examples of research advancing early diagnosis, said Peter Allen, MD, professor of surgery and chief of the division of surgical oncology at Duke University School of Medicine. The outlook for early diagnosis is improving given “new understanding of the ability to image the pancreas, pancreatic cancer biology, and our ability to block a predominant KRAS mutation, which is thought to be undruggable,” he said.
NEW RESEARCH ON DIAGNOSTICS
The JAMA paper describes a study in nearly 500 adult patients almost evenly split between development and validation cohorts at four academic hospitals in Italy, the Netherlands, and the United Kingdom. Both cohorts involved patients in their late sixties. In external validation, the prediction model showed an area under the (AUC) curve of 0.89 (95% CI, 0.84–0.93) for early-stage pancreatic cancer versus benign periampullary diseases, and outperformed CA19–9 (difference in AUC [ΔAUC], 0.10; 95% CI, 0.06–0.14; P < .001) and bilirubin (∆AUC, 0.07; 95% CI, 0.02–0.12; P = .004). In the subset of patients without elevated tumor markers, the model showed an AUC of 0.84 (95% CI, 0.77–0.92).
At a risk threshold of 30%, decision curve analysis showed that performing biopsies based on the prediction model was equivalent to reducing the biopsy procedure rate by 6% (95% CI, 1%–11%), without missing early-stage pancreatic cancer in patients, the researchers noted. They said the model could be used to assess the added diagnostic and clinical value of novel biomarkers and prevent potentially unnecessary invasive diagnostic procedures for patients at low risk.
The method is practical and cost-effective because it relies on readily available routine biomarkers, said corresponding author Elisa Giovannetti, MD, PhD, associate professor at Vrije Universiteit University Medical Center in Amsterdam.
The Nature Medicine paper described how researchers trained of an AI algorithm on 41 years’ worth of Danish National Patient Registry records of 6.2 million patients, 23,985 of whom developed pancreatic cancer. The algorithm associated future pancreatic cancer risk based on disease trajectories and was able to detect the cancer up to 3 years early using only these records.
For example, gallstones, anemia, type 2 diabetes, and other gastrointestinal problems were associated with greater risk for pancreatic cancer within 3 years. Then the researchers tested their algorithm on 21 years of U.S. Veterans Health Administration data. This data encompassed almost 3 million records spanning 21 years, including 3,864 individuals diagnosed with pancreatic cancer.
Training AI models on high-quality data, large representative datasets of clinical records aggregated nationally and internationally, and on local health data in the absence of globally valid models is crucial, the researchers noted.
COULD AI BEAT TRADITIONAL BIOMARKERS?
Michael Goggins, MBBCh., MD, professor of pathology at Johns Hopkins School of Medicine, said that the blood biomarker test might be applied to high-risk patients with pancreatic imaging abnormalities. The test provides some useful information but would not drive changes in patient care in its current form, he predicted.
Giovannetti said that a potential process for expanding the biomarker score would involve examining other biomarkers, including mutations commonly associated with pancreatic cancer, inflammatory or metabolic markers like LDH or GLUT1, or maybe specific microRNA profiles.
Meanwhile, Goggins own research has found that genetic factors influence the levels of CA19-9 circulating in blood, a finding worth considering when using the biomarker, he said. (Clin Cancer Research 2023; doi: 10.1158/1078-0432.CCR-23-0655.)
“Even if this test catches an early cancer, it is likely lethal,” Allen said. “A better approach would be removing high-risk lesions before they become truly invasive. An ounce of prevention is worth more than a pound of cure here. That applies to pancreatic cancer more than any other disease we currently study.”
For this reason, he and Goggins were intrigued by the AI method. The Nature paper is “proof of principle that machine learning can be applied to medical records and potentially prevent some cancer,” Goggins said.
The AI method identified high-risk patients in need of monitoring, especially those with pre-neoplastic lesions that could advance to cancer, noted Brian Haab, PhD, professor of cell biology at Van Handel Institute Graduate School. A viable AI test could spur lab analysis of pancreatic fluid to determine development of high-risk lesions, he added.
IMAGING AND BIOMARKERS WILL WORK TOGETHER
Currently, choices for detecting pancreatic seem to keep changing. Immunovia, which in 2021 received Food and Drug Administration (FDA) approval for its IMMray PanCan-d test focused on early detection of pancreatic cancer, has discontinued it. The company plans to focus on developing a next-generation pancreatic cancer detection method intended to work equally well across multiple patient risk groups, including those who do not produce CA19-9. The company says the forthcoming test will be performed on a widely used commercial platform.
Meanwhile, ClearNote Health’s Avantect Pancreatic Cancer Test has FDA Breakthrough Device designation for a method based on 5-hydroxymethylcytosine (5hmC) profiling of cell-free DNA (cfDNA). And Grail’s Galleri test uses next-generation sequencing and machine-learning algorithms to analyze methylation patterns of cfDNA to screen for multiple cancers, including pancreatic cancers.
Randall Brand, MD, professor of medicine at University of Pittsburg, noted that a 22-gene panel developed at University of Pittsburgh, PancreaSeq, classifies pancreatic cysts as potentially cancerous or benign. Based on mutations in KRAS and GNAS, PancreaSeq diagnosed mucinous cysts accurately in 90% of cases in a recent study (Gastroenterology 2022; doi: 10.1053/j.gastro.2022.09.028).
Brand also pointed to radiomics, which involves an advanced image analysis technique to study a cyst or surrounding pancreatic tissue beyond what is visible to the human eye. His own study integrates radiomics and genomics to characterize the biology of pancreatic cysts and improve clinical management. The goal is to determine whether a combination of radiomic and genomic biomarkers is superior to each alone for detecting mucinous cysts and advanced neoplasia.
Allen noted that Johns Hopkins University researchers are developing novel imaging techniques that are “more sensitive than what we currently have.” They rely on a suite of algorithms, called FELIX, that recognize pancreatic lesions from CT images without human input.
Lab and imaging tests combined will be key to improving pancreatic cancer treatment, Brand emphasized. “We need better early detection to improve our chance for a cure. It’s not just treatment."
Disclosures: Allen serves on the scientific advisory board for the Lustgarten Foundation. Brand has submitted a research proposal to Biologic Dynamics.
Deborah Levenson is a freelance writer in College Park, Maryland. +Email: [email protected]
A PERSONALIZED Vaccine for Pancreatic Cancer?
A small trial recently found that half of pancreatic ductal adenocarcinoma (PDAC) patients who received a personalized mRNA cancer vaccine after surgery had no tumor recurrence 18 months later (Nature 2023; doi: 10.1038s41586-023-06063-y).
The vaccine is designed to help immune cells recognize specific neoantigens on patients’ pancreatic cells. Previous research has shown that pancreatic cancer survivors had a stronger response to neoantigens from T cells than those who do not survive.
Researchers at Memorial Sloan Kettering Cancer Center (MSKCC) used mRNA technology to target 19 pancreatic cancer surgery patients’ own tumor neoantigens. Five had stage 1 disease, eight had stage 2, and six had stage 3 cancer. After removal of tumors, the researchers shipped samples to BioNTech in Germany, where the company analyzed the genetic makeup of neoantigens.
BioNTech produced personalized vaccines designed to train each patient’s immune system to attack the tumors using mRNA. All patients received the immune checkpoint inhibitor atezolizumab before getting the vaccine in nine doses over several months. After the eighth dose, patients also received standard chemotherapy drugs, followed by a ninth dose. Sixteen of 19 patients remained well enough to get at least some of the vaccine doses. In half these patients, the vaccines activated T cells that could recognize the pancreatic cancer specific to the patient.
Using a novel computational strategy, the researchers showed that T cells that recognized the neoantigens were not found in patients’ blood before vaccination. Among eight patients with strong immune responses, half had T cells that targeted more than one vaccine neoantigen.
After 18 months, patients who had strong T cell responses to the vaccine were cancer-free. Among patients whose immune systems didn’t respond to the vaccine, the cancer recurred within an average of just over a year. In just one patient with a strong response, T cells produced by the vaccine seem to have eliminated a small tumor that had spread to the liver.
These results suggest that the T cells activated by the vaccines kept the pancreatic cancers in check, researchers noted.
In the paper, the authors say their results must be replicated in larger studies. In October, MSKCC announced a new trial to test the vaccine in 260 patients at nearly 80 sites around the world.