Caris AI Identifies Pancreatic Cancer Patients Who Need Less Chemo

Only 8% of patients diagnosed with pancreatic ductal adenocarcinoma, the most common form of pancreatic cancer, are alive five years after diagnosis.

When that diagnosis lands, an oncologist typically faces the same narrow choice: put the patient on one of two standard chemotherapy regimens, both of them grueling, and hope it works.

That choice has come down largely to a doctor's read of how sick the patient looks. Other AI tools designed to guide that decision have entered the market, but none are delivered through an FDA-approved sequencing assay.

This week, Irving, Texas-based Caris Life Sciences launched a guidance signature embedded inside MI Cancer Seek, the only platform combining full DNA and gene activity sequencing to hold FDA approval for molecular profiling of solid tumors.

The company says a forthcoming study will go further, showing that for a meaningful group of patients, the data suggests neither option may be necessary at its most aggressive level. David Spetzler, President of Caris Life Sciences, described the gap the tool is designed to fill: "Clinicians have historically had to make difficult treatment decisions with limited biological guidance."

A Field That Defaults to More

Across cancer care broadly, the tendency to treat and keep treating 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.

Dr. Nathan Cherny, a senior oncologist at Shaare Zedek Medical Center in Jerusalem, Israel, and a lead author of the study, described what that looks like on the ground in an interview with Healio: "It's clear that the treatments are more likely to cause harm than good. I see nurses who are in a situation of moral distress, having to administer treatments that they feel are not in the patient's best interest."

The study identified optimism bias and pressure from patients and families as contributing factors.

There is also a financial dimension: chemotherapy administration accounts for approximately 65% of revenue in a typical oncology practice.

In pancreatic cancer specifically, the two standard options are FOLFIRINOX, an aggressive four-drug combination known for severe side effects including nerve damage, and gemcitabine/nab-paclitaxel, a regimen generally associated with a more manageable side effect profile.

Doctors have historically chosen between them based on whether the patient looks strong enough to handle FOLFIRINOX. What the tumor itself might respond to has rarely entered that conversation.

Caris is entering a market where at least one competitor is already operating. Perthera, a McLean, Virginia-based precision oncology company, launched PDACai in February 2025, an AI signature designed to guide the same FOLFIRINOX versus gemcitabine/nab-paclitaxel decision, at over 600 cancer centers across the U.S.

PDACai operates as a clinical decision support tool and carries no FDA approval. Caris's signature is delivered through MI Cancer Seek, which differentiates itself by being a simultaneous DNA and gene activity sequencing platform to hold that regulatory clearance for solid tumor profiling.

What the Tool Actually Does

Earlier AI tools in oncology promised to solve similar problems but fell short. An older 2018 investigation by STAT News found that IBM Watson for Oncology had been trained on hypothetical patient scenarios rather than real patient outcomes, producing recommendations that did not hold up in clinical practice. Caris's approach starts from a different foundation.

To generate a recommendation, the Caris tool reads the tumor at a molecular level using two methods. The first, Whole Exome Sequencing, maps the mutations in a tumor's DNA, specifically the genetic errors that make it cancerous. The second, Whole Transcriptome Sequencing, reads which genes inside the tumor are actively switched on or off at the time of testing, revealing how the cancer is behaving right now.

A 2022 paper in JCO Precision Oncology found that combining both methods gives doctors a more complete picture of a tumor than either approach alone.

That combined data feeds into CodeAI, Caris's proprietary AI platform, which the company claims looks for patterns across molecular profiles from more than 550,000 patients.

The AI identifies which of the two regimens patients with similar profiles have historically responded to better, and generates a personalized prediction for that specific cancer.

The pending publication will identify a group of pancreatic cancer patients who, based on how their cancer behaves at a molecular level, may not need FOLFIRINOX's full intensity. Caris has not yet released that data for peer review.

What is available now is the selection tool itself, built into an existing clinical report that oncologists can already order with no additional tissue sampling required.

Caris, which went public on Nasdaq in 2025, is projecting revenues of $1.0 to $1.02 billion for 2026.