Skip to main content
Premium Trial:

Request an Annual Quote

PreciseDx Begins Limited Rollout of 'Quantitative' Test for Grading Invasive Breast Cancer

Premium

NEW YORK – PreciseDx is making its image-based test for grading invasive breast cancer available to patients on a limited basis as the firm works to establish its clinical utility and expand its applications.

Breast cancer grade is a measure of tumor aggressiveness, where grade 1 tumors are less likely to spread and grade 3 tumors indicate aggressive disease that is more likely to metastasize. Tumor grade is also important for making disease management decisions, such as choosing between neoadjuvant therapy and surgery, and it can help contextualize genetic test results.

Pathologists currently grade breast cancer after evaluating the tumor's histologic features, but discordance rates between pathologists in grading the same tumor can be as high as 25 to 30 percent, particularly when it comes to moderately differentiated grade 2 cancers. 

New York-based PreciseDx sought to improve on existing methods of grading breast cancers using an image-based deep learning approach. That effort led to the development of PreciseBreast, a digital laboratory-developed test that predicts the probability of breast cancer recurrence within six years for newly diagnosed patients with early-stage invasive breast cancer.

"The challenge in breast cancer today is accurate characterization of the disease process," said Michael Donovan, PreciseDx cofounder and chief medical officer. All major breast cancer treatment guidelines from the College of American Pathologists, National Comprehensive Cancer Network, and the American Society of Clinical Oncology discuss tumor grading as a critical step in patient care. The grade captures not only the size of the tumor but also significant features of the tumor cells indicating properties such as states of differentiation or proliferation. While a pathologist's evaluation and scoring of those features can be subjective, Donovan said PreciseDx's model provides a quantitative measure.

PreciseDx is the brainchild of Donovan and several colleagues from his time at Mount Sinai's Icahn School of Medicine and other biotechs, mainly the pathologists Carlos Cordon-Cardo and Gerardo Fernandez, and the microscopy technology expert Jack Zeineh. They all shared an interest in systems pathology, a nascent field that integrates mathematics with biology and first yielded integrative risk-based modeling systems for cancer in 2004.

"PreciseDx is really one of those derivations of a systems pathology effort, which utilizes various [image-based] tools," Donovan said.

According to PreciseDx CEO Wayne Brinster, the company, which first focused on prostate cancer and now has moved into breast cancer, wasn't founded so much as "hatched" in Cordon-Cardo's lab at Mount Sinai and funded by a grant set up by Hamilton James, former president of the global asset management firm Blackstone. In 2021, Cordon-Cardo's lab spun out PreciseDx as an independent company, and since then, the firm has raised $17 million, including $10.8 million through Series A financing. PreciseDx is in the process of initiating a Series B round.

The company plans to use these funds to further develop and commercialize its AI-powered grading test in breast cancer. Donovan and his collaborators began developing the deep learning model underlying the test while at Mount Sinai using a set of hematoxylin- and eosin-stained formalin-fixed paraffin-embedded tissue resection samples collected from patients between 2004 and 2016. Patients were at least 23 years old with infiltrating ductal or mixed ductal and lobular breast cancer, and researchers had a median of six years' follow-up data on them. Patients who received neoadjuvant therapy or had a prior history of breast cancer were excluded.

The deep learning algorithm analyzed the slides and found cell and tissue features related to cancer grade and risk of recurrence. The researchers then developed the breast cancer grading model based on 40 tumor features and a set of clinical variables including age, tumor size, stage, and lymph node status. The algorithm did not choose pathologists' tumor grades, estrogen and progesterone receptor status, and HER2 amplification, although that data was available.

"The interaction between the [tumor] cell and the epithelial cells, the interactions between epithelial cells and the stroma, all of that is a demonstration of the phenotype of invasive breast cancer," Donovan said. "That was part of what we were trying to achieve, to extract as much information as possible that's currently not done in a quantified, standardized way."

That development process yielded a model that could give a patient's breast cancer a grade between zero to 100 based on cell, tissue, and clinical features, which corresponded to her risk of disease recurrence — the higher the score, the higher the risk of recurrence. Researchers used a score of 58 in the training cohort to stratify patients into groups with low or high risk of breast cancer recurrence within six years.

With that cutoff, the breast cancer grading test had a negative predictive value of 95 percent and positive predictive value of 32 percent. In a separate validation cohort, the negative predictive value of the test was 94 percent and the positive predictive value was 24 percent, which the study authors attributed to the lower prevalence of events in the low-risk cohort.

Donovan said that a key differentiating factor in the development of the model was the use of outcome-based modeling. In the outcome-based model, each feature included in the final model was curated by evaluating its individual association with a recurrence event, such as local metastasis. Then, in an additional filtering step, researchers ascertained the significance of each feature for predicting the likelihood of recurrence.

After assigning weights to individual tumor features based on their significance in predicting recurrent disease, the researchers used them to construct a multivariate model called a morphology feature array. "Most modeling approaches focus on refining features based on histology alone and using pathologists to confirm, while our approach is multilayered and interrogates each individual feature not only for histologic accuracy, but also its potential for the spread of disease," said Donovan.

Ideally, he said PreciseDx would like to see its test being used as a tool that supports pathologists' and oncologists' assessments when diagnosing new breast cancer patients. If there were sufficient tissue available, Donovan said, the patient could also receive genomic testing, and the morphology feature array could assist with understanding the implications of those results.

While patients may lack access to genomic testing due to inadequate tissue, physician uncertainty, or lack of insurance, Donovan underscored that no additional tissue is required to run the grading algorithm. PreciseBreast has been reimbursed at $700, which he noted is much lower than the price tag for some genomic tests. "The initial goal is not to replace genomic strategies, but to enrich the data provided by all genomic tests and thereby improve and personalize risk stratification to ultimately guide informed treatment decision-making," he noted.

According to Donovan, the PreciseDx breast cancer test may also help resolve uncertainty for patients with ambiguous results on other tests, such as Exact Sciences' Oncotype DX breast cancer recurrence test.

The TAILORx clinical trial, in which Oncotype DX was used to decide whether to give adjuvant chemotherapy with hormone therapy to breast cancer patients, showed that postmenopausal women with risk recurrence scores between 11 and 25 on a 100-point scale could forgo chemotherapy. However, the picture was more complex for premenopausal patients in the intermediate risk category, where some patients with scores above 15 appeared to benefit from the addition of chemotherapy.

"After TAILORx, it became clear that there is a population where the use of chemotherapy is uncertain," Donovan said. Doctors can use the RSCLin tool, which combines the Oncotype DX score with clinical data such as age, tumor size, and grade, to clarify treatment decisions for patients in the intermediate group.

"PreciseBreast utilizes the same clinical features that are promoted in the RSCLin studies, but with a very strategic difference in that the AI grade by Precise is objective, automated, quantitative, and standardized versus the histologic grade in the clinical risk model, which is subjective and often discordant between and within pathologists," Donovan said. When Donovan's group compared PreciseBreast with a method based on the clinical features included in RSCLin, PreciseBreast improved risk and event discrimination.

PreciseDx is exploring clinical use case scenarios and building evidence to support an independent predictive claim for the test. "We monitor a lot of things that we don't use for scoring, currently," said Brinster. "We're not sure exactly where that's going to take us, but our plan is to run the test on a wider net of patients with really good information, looking at outcomes just like we did with recurrence."

The company is also assessing the test on different sample types, including biopsy samples, and in other cancer types, including prostate cancer. For now, the PreciseBreast is available on a limited release basis, but the company is continuing to study its clinical utility, working with payors to ink coverage contracts, and is in early discussions with development partners.