Molecular Data Is Now Picking Pancreatic Cancer's First Drug
Half of pancreatic cancer patients may be getting harsher chemotherapy than they need. AI is identifying which half.
When a pancreatic cancer diagnosis lands, the first decision an oncologist faces is which of two chemotherapy regimens to prescribe. One, FOLFIRINOX, is an aggressive four-drug combination with severe side effects including nerve damage. The other, gemcitabine/nab-paclitaxel, is generally more tolerable. Historically, that choice has come down to one question: does the patient look strong enough to handle FOLFIRINOX?
The tumor's biology has rarely entered that conversation. New data from Irving, Texas-based Caris Life Sciences suggests it should. According to the company's validation dataset, roughly 55% of patients with pancreatic ductal adenocarcinoma (PDAC) may not need the more intensive regimen, a figure Caris shared with AIM Media House ahead of publication.
The tool making that determination is already in wide use. Caris says it profiled approximately 8,500 PDAC samples through its MI Cancer Seek platform in 2025, ordered by more than 9,800 physicians across more than 3,900 sites. The validation data is arriving after broad deployment.
What the Tool Shows Physicians
The Caris signature, embedded inside MI Cancer Seek, reads each tumor using two sequencing methods. Whole Exome Sequencing maps the mutations in a tumor's DNA. Whole Transcriptome Sequencing reads which genes are actively switched on or off at the time of testing. That combined data feeds into CodeAI, Caris's proprietary AI platform, which the company says draws on molecular profiles from more than 550,000 patients to generate a treatment prediction.
"Clinicians have historically had to make difficult treatment decisions with limited biological guidance," David Spetzler, President of Caris Life Sciences, says to AIM Media House.
It goes further than a recommendation. Spetzler says patients are categorized as high risk or standard risk, with a suggested treatment between the two regimens. Physicians also receive Kaplan-Meier curves showing outcomes for patients treated with each regimen across four possible risk-and-treatment combinations, letting them locate their specific patient within that grid.
Caris says the model was trained using real-world clinical claims data acquired from a third party, with time to next treatment used as a surrogate for progression-free survival, and overall survival evaluated as a secondary endpoint.
A Field Moving in the Same Direction
Independent trial data published in September 2025 offers context for what Caris is claiming. The PASS-01 trial, a randomized phase II study published in the Journal of Clinical Oncology, compared modified FOLFIRINOX and gemcitabine/nab-paclitaxel head-to-head in 160 patients with previously untreated metastatic PDAC, the first such comparison in a North American population. Its investigators conclude that first-line biomarker-selected strategies are urgently needed.
Across the full trial population, overall survival and safety trends favored gemcitabine/nab-paclitaxel. Treatment-related serious adverse events were 13% for modified FOLFIRINOX versus 3% for gemcitabine/nab-paclitaxel. When the data was broken down by tumor subtype, classical PDAC patients had a median overall survival of 13.9 months on gemcitabine/nab-paclitaxel versus 9.7 months on modified FOLFIRINOX.
The published evidence already trends toward the gentler option in an unselected population. What molecular risk stratification adds is something the pending validation study will be expected to show.
Other companies are building in the same direction. Chicago-based health technology company Tempus AI published a validation of its PurIST algorithm in JCO Precision Oncology in 2025, using RNA sequencing to classify PDAC tumors as classical or basal subtypes for treatment selection. Perthera, a McLean, Virginia-based precision oncology company, launched PDACai in February 2025, an AI signature designed to guide the same treatment decision. Perthera presented validation data at ASCO GI 2025. Neither algorithm carries FDA clearance.
Caris differentiates on that basis. The company says MI Cancer Seek is the only platform combining full DNA and gene activity sequencing to hold FDA approval for molecular profiling of solid tumors, with the PDAC guidance signature delivered through that existing approved assay and no additional tissue sampling required.
Why the Default Has Been 'More'
The clinical tendency to treat aggressively is well documented. A January 2025 study published in ESMO Open found that between 10% and 40% of cancer patients receive chemotherapy in the final weeks of life with no meaningful benefit to survival. The study identified optimism bias and pressure from patients and families as contributing factors.
In PDAC specifically, that default has had little biological pushback. Treatment decisions have tracked patient fitness, not tumor behavior.
What molecular tools like the Caris signature offer is a different kind of justification for recalibrating that default. The read comes from the tumor's molecular behavior, not a physician's clinical impression of patient fitness. The pending validation study will determine how far that shift goes.