Mayo Clinic and Stanford Develop Blood Test to Predict Cancer Treatment Response
Researchers at Mayo Clinic and Stanford Medicine have developed an artificial intelligence framework that can detect tumor microenvironment signals in blood samples, potentially allowing oncologists to predict which patients will respond to immunotherapy before imaging shows results.
The discovery, published in Nature this month, addresses a longstanding clinical gap. Until now, doctors had no way to monitor how a tumor's surrounding environment-the normal cells that tumors recruit to help them survive-responds to treatment over time.
The Problem: Cancer Research Has Ignored the Microenvironment
Current FDA-approved blood tests focus only on circulating tumor DNA from cancer cells themselves. These tests work for roughly 5% of cancer patients, said Dr. Aadel Chaudhuri, a radiation oncologist at Mayo Clinic.
"We're not learning anything about the microenvironment," Chaudhuri said. "That motivated us to develop this new technology to visualize the tumor microenvironment from blood plasma."
The tumor microenvironment consists of normal cells-immune cells, fibroblasts, and others-that tumors manipulate to survive. These cells lack cancer mutations, making them invisible to standard genetic tracking methods.
How the AI Framework Works
Chaudhuri and Aaron Newman, an associate professor of biomedical data science at Stanford, developed a two-part machine learning approach over eight years.
The first model, called Spatial Ecotyper, analyzed tissue samples from 17 cancer types and identified nine distinct "spatial ecotypes"-recurring patterns of cell organization the researchers call "cancer neighborhoods." These patterns appeared consistently across different cancer types, from lung and breast cancers to melanomas.
The second model, Liquid Ecotyper, detects these same patterns in blood samples by reading chemical tags called methylation markers on cell-free DNA. When cells in the tumor microenvironment die, they shed DNA fragments into the bloodstream. The AI learns to recognize which methylation patterns correspond to specific cell types and their activities.
Transparency Built Into the Model
Unlike many deep learning systems that function as "black boxes," this framework shows its work. When the model makes a prediction, it identifies the exact methylation sites it used and mathematically proves they correspond to known human biology.
"You can't just use off-the-shelf AI models," said Newman. "We devised an approach that gives us an exact readout of which methylation sites the model deems important to make a prediction."
This transparency matters clinically. Doctors need to understand why an AI system recommends a treatment decision, not just accept its output.
Early Clinical Evidence
In one patient case, the blood test detected a shift in tumor microenvironment composition during combination therapy that imaging alone missed. The model showed that resistance-associated cells dropped sharply while benefit-associated cells rose dramatically.
Standard CT imaging can create false impressions during immunotherapy treatment. Inflammation can make a tumor appear worse when it's actually responding-a phenomenon called pseudo-progression. A blood test measuring microenvironment changes could prevent oncologists from abandoning effective treatments prematurely.
"The oncologist didn't know what was happening in the tumor microenvironment based on imaging," Newman said. "They had no idea the combination therapy was priming the patient to respond."
What Remains Unknown
The initial research focused on treatment-naive tumors. Researchers still need to understand whether active therapy creates new spatial ecotypes or changes existing ones.
Chaudhuri said preliminary data suggests the test can detect microenvironment changes months to a year before imaging shows shifts. That finding is the focus of current studies.
The team is also exploring how spatial ecotypes respond to non-immunotherapy treatments like antibody-drug conjugates and examining blood-based detection in blood cancers like multiple myeloma.
Clinical Timeline
Chaudhuri and Newman are conducting clinical trials to determine whether serial blood monitoring of the tumor microenvironment actually improves treatment decisions and patient outcomes.
The goal is to make Liquid Ecotyper analyses widely available to patients by 2030, Chaudhuri said.
For healthcare professionals evaluating new diagnostic tools, this work represents a shift from tumor-focused analysis to understanding the ecosystem tumors depend on. Learn more about AI for Healthcare and AI Data Analysis applications in clinical settings.
Your membership also unlocks:
