• Not every patient with early-stage breast cancer needs chemotherapy. Genomic testing can help identify those who may safely avoid it, even when conventional clinical features suggest higher risk.
• Agendia’s MammaPrint 70-gene assay helps assess the risk of breast cancer recurrence and informs chemotherapy decisions, potentially sparing selected patients unnecessary treatment and toxicity.
• At ESMO 2026, Agendia presented an investigational AI model that estimates MammaPrint genomic risk from routine pathology slides, potentially expanding access to precision oncology.
• The AI model showed a strong correlation with MammaPrint results, but further validation is required before it can be used to guide clinical treatment decisions.
• Agendia’s FLEX registry, with approximately 25,000 enrolled patients, is investigating how tumor biology influences treatment response, resistance, and the selection of more individualized therapies.

For patients diagnosed with early-stage breast cancer, the decision to undergo chemotherapy is not always straightforward. Identifying who genuinely needs chemotherapy, and who may safely avoid its toxicities, has become an important focus of personalized breast cancer care.

Genomic testing is helping clinicians make that distinction. Agendia’s MammaPrint assay analyzes the expression of 70 genes to assess a tumor’s risk of recurrence, providing information that can influence treatment decisions. Now, researchers are investigating whether artificial intelligence can extract similar prognostic information from routine pathology slides, potentially extending access to molecular insights without requiring the same laboratory-based testing.

At the European Society for Medical Oncology (ESMO) Congress 2026 in Madrid, Agendia presented results from an investigational AI model trained on 70,000 digitized pathology images. In an evaluation involving 372 patients with early-stage breast cancer, the model’s risk scores showed a strong correlation with MammaPrint results (r = 0.954). Ten-year breast cancer-specific survival was 98.1% among patients classified as low risk by the AI model, compared with 86.7% among those classified as high risk. Although further validation is required before the technology can inform clinical treatment decisions, the findings suggest that routinely collected pathology images may contain valuable information about a tumor’s underlying biology.

In an interview with LA Times Studios at ESMO 2026, M. William Audeh, MD, Chief Medical Officer of Agendia and a breast medical oncologist, discussed how genomic testing is already helping oncologists avoid unnecessary chemotherapy, the potential and current limitations of AI-driven pathology, and the clinical questions being explored through Agendia’s FLEX registry, which has enrolled more than 25,000 patients.

Agendia had three studies at the congress. The AI result, he said, was set for an oral talk in an artificial intelligence session on the congress’s Saturday.

The model is not yet as precise as the tissue test, and Agendia will not offer it to clinicians until it matches that test. “At this point, it can certainly identify tumors that are low risk or high risk, but it is not yet achieving the same level of precision that the tissue based genomic assays can provide,” he said.

In a release after the talk, Agendia said the model was trained on 70,000 slide images and tested on 372 patients from the RASTER study, followed for a median of 10.3 years. The company reported that the model’s risk scores correlated with MammaPrint’s at r = 0.954, a measure where 1 is perfect agreement.

Patients the model rated low risk had 10-year breast cancer-specific survival of 98.1%, against 86.7% for those it rated high risk, according to the release. Agendia describes the AI assay as investigational, and the figures come from a company release and a congress talk, not a peer-reviewed paper.

What MammaPrint and BluePrint Measure

MammaPrint reads the activity of 70 genes in a sample of tumor tissue from early-stage breast cancer. They estimate the risk that the cancer will come back, which helps decide who can avoid chemotherapy.

BluePrint, Agendia’s second test, runs on the same tissue sample; it describes the factors that drive the tumor’s growth.

Audeh explained that MammaPrint has been available for many years and is used routinely for women and men with newly diagnosed breast cancer. It has been validated, he said, in studies at the highest level of evidence.

He said women whose tumors test as genomically low risk have 5-year and 8-year outcomes without chemotherapy “in the high 90% range.” That holds, he said, even for some women whose cancer has reached the lymph nodes or whose tumors measure 2 to 5 centimeters, features that would usually lead to chemotherapy.

“We’re curing them to live what we hope is going to be a long and healthy life, without a lot of leftover toxicity from therapy,” Audeh emphasized.

How a Slide Image Could Carry Genomic Information

The AI model works from an H&E slide, tissue stained with hematoxylin and eosin, the standard stain pathologists use to look at cells under a microscope. The slide is scanned, and the model reads the digital image.

The presentation, he said, would be the first to show genomic information interpreted from an H&E image. He said the work could bring genomic results to parts of the world where tissue-based genomic testing is harder to get than in the U.S. and Europe.

The next step, Audeh said, is to learn what else AI can find in the slide image that tissue genomics cannot. He named the cells around the tumor as one example, since they may affect how a cancer behaves and which treatments work best.

What the FLEX Registry Collects

Agendia’s FLEX study is an observational registry, a database that follows patients through routine care without assigning treatments. It is registered at ClinicalTrials.gov as NCT03053193.

FLEX has reached 25,000 enrolled patients, with a goal of at least 30,000, and every patient has a MammaPrint and BluePrint result, clinical details on how they were treated, and outcomes at 3 and 5 years, he said.

FLEX also records the transcriptome, the full set of genes expressed in the tumor, from the same sample. According to Audeh, that lets the company study questions that large randomized trials left open, because randomizing patients on them was not feasible or not ethical.

He said Agendia has published work showing that MammaPrint can identify which patients need anthracyclines, a class of chemotherapy drugs that includes doxorubicin and that he described as particularly toxic. The company is also studying gene pathways that may make tumors resistant to current therapies, he said.

Who Is in the FLEX Data

Most FLEX patients have hormone receptor-positive, HER2-negative breast cancer, the most common type, whose growth is fed by estrogen or progesterone and not by the HER2 protein. A significant share have HER2-positive disease, and a smaller group have triple-negative breast cancer, which lacks all three of those receptors.

Audeh also said FLEX includes nearly 3,000 women who self-report as Black and more than 2,000 who self-report as Latina. Self-reported race is an imperfect measure of genetic ancestry, he said, but the best one available.

Agendia’s studies comparing hormone receptor-positive tumors in Black and white women found significant differences that will inform treatment decisions, according to Audeh. Similar work is under way for Latina and Asian women, he said.

Knowing When Less Treatment Is Enough

After approximately 35 years in oncology, Audeh believes breast cancer patients have often received more treatment than they needed. Genomic testing is helping clinicians identify patients who may safely avoid chemotherapy, including some whose tumors appear high-risk by conventional measures.

With MammaPrint already informing clinical decisions, Agendia’s ongoing AI research and FLEX registry may offer additional ways to refine treatment selection. The question now is whether these advances can help more patients avoid unnecessary treatment without sacrificing its benefits.

See article via the Los Angeles Times