As healthcare leaders celebrate the FDA’s approval of daraxonrasib — a pancreatic cancer drug that nearly doubled survival in clinical trials — two harsh realities remain: most pancreatic cancer patients are still diagnosed at a late stage, and those treated with daraxonrasib will eventually develop resistance to it.
Pancreatic cancer is the third-leading cause of cancer death in the U.S., representing 3.2% of new cancer cases but 8.4% of cancer deaths. Only 14.6% of pancreatic cancers are diagnosed before they spread to other parts of the body, according to SEER data.
To catch pancreatic cancer before it spreads, researchers are turning to AI.
Here are four notes on the state of AI and pancreatic cancer detection:
- An AI model developed by researchers at Rochester, Minn.-based Mayo Clinic identified 73% of prediagnostic pancreatic cancers on about 2,000 abdominal CT scans, a median of about 16 months before clinical diagnosis, according to a study published April 28 in Gut.
The model, called the Radiomics-based Early Detection Model, was tested on scans from multiple institutions, imaging systems and protocols. It is designed to flag elevated pancreatic cancer risk in patients who undergo CT scans for other reasons. Mayo Clinic researchers are testing the model in a prospective clinical study called AI-PACED, according to a news release from the health system. - Most health systems are not yet prepared to act on AI-generated risk signals, Ajit Goenka, MD, senior author of the Mayo Clinic study, told Becker’s.
“Most health systems are built to respond to cancer after it declares itself, not to act on a signal from an AI model before a lesion is even visible,” he said.
Dr. Goenka said systems still need to determine who receives an AI flag, how cases are escalated when a scan appears normal, how risk is communicated to patients without symptoms and how payers cover a workup prompted by a model rather than a symptom.
“These are solvable problems, but they require deliberate investments in our infrastructure,” he said. “The science of detection is advancing. The science of deployment needs to keep pace.”
- A blood test developed by researchers at Duarte, Calif.-based City of Hope detected 87% of stage 1 and 2 pancreatic cancers in nearly 1,800 patients, according to a study published Sept. 16 in Nature Medicine.
The test, called PANXEON, combines circulating microRNAs, exosomal microRNAs and the protein CA19-9 into a single AI-generated risk score. Its false-positive rates were 3% and 16% among low- and high-risk patients, respectively.
The test also detected precancerous high-grade dysplasia in more than 64% of cases. Researchers said the test is designed to complement existing diagnostic strategies by flagging high-risk patients for further evaluation. - AI analysis of EHR data could help health systems identify patients at elevated cancer risk before diagnosis, Anirban Maitra, MD, director of New York City-based NYU Langone’s Perlmutter Cancer Center, told Becker’s. Dr. Maitra, also a pancreatic cancer researcher, said the approach has the greatest potential to change how health systems deliver cancer care over the next decade.
Unlike liquid biopsy tests for colorectal cancer, early detection tests for pancreatic cancer are not covered by insurance, Dr. Maitra said. The Nancy Gardner Sewell Medicare Multi-Cancer Early Detection Screening Coverage Act, which Congress passed as part of an appropriations bill, clears a path for those tests to seek FDA approval and CMS reimbursement, he said.
An FDA advisory panel voted Sept. 23 to back Grail’s Galleri test, which would be the first multi-cancer early detection blood test to receive FDA approval.
The post AI and pancreatic cancer detection: 4 notes appeared first on Becker's Hospital Review | Healthcare News & Analysis.