PreciseDx's digital pathology AI on par with genomic testing in predicting breast cancer recurrence risks: study

A study from PreciseDx showed that its artificial-intelligence-guided pathology test could help predict a patient’s chances of their breast cancer returning on par with a leading genomic approach.

Working with real-world cancer data company COTA Healthcare, the former Fierce 15 winner put up its PreciseBreast test against Exact Sciences’ Oncotype DX, which delivers a personalized recurrence score on a scale of 0 to 100 based on the expression of different genes, including 16 linked to early-stage breast tumors.

PreciseDx, meanwhile, analyzes the physical features of cancer cells by visually inspecting tissue sample slides and incorporating other clinical data. The company—born out of digital pathology work at Mount Sinai Health System—said it aims to categorize patients at either high or low risk for recurrence, while turning around results within two days and at a much cheaper cost compared to DNA sequencing and expression assays.

For the retrospective study, COTA was tapped to facilitate a data set curated from 425 patients at Baptist Health South Florida’s Miami Cancer Institute—with digitized samples gathered specifically outside of the New York area to help validate the AI’s performance in health systems elsewhere in the country.

The two companies found similar results between PreciseBreast and Oncotype DX, with comparable results in helping patients with less aggressive cancer avoid potentially unnecessary chemotherapy. According to PreciseDx, its test characterized 79% of samples as low-risk, while Oncotype DX found 81% were at low or intermediate risk.

Of the patients who did eventually see their breast cancer return, PreciseBreast had labeled 38% of them as high-risk, compared to 35% for Oncotype DX. The study’s results are scheduled for a poster presentation at the San Antonio Breast Cancer Symposium on Dec. 11.

“It’s critical that providers understand more about their patient’s breast cancer as early as possible to ensure timely treatment and improve their outcomes,” PreciseDx CEO Eric Converse said in a statement. “With the ultimate goal of providing care decision support data that works seamlessly in the pathology workflow, returns results in less than 48 hours and costs a fraction of the current technology, PreciseBreast optimizes, and could ultimately replace, gene-expression testing.”

And when it comes to testing the AI itself, COTA’s chief medical officer, C.K. Wang, M.D., said using research- and regulatory-grade real-world evidence is critical to making sure an algorithm is ready for prime time.

“There's some studies out there saying that a lot of AI and large language models have now run out of data to train on—but that’s the data that's publicly available, which is not healthcare or patient-level data,” Wang said in an interview with Fierce Medtech. “All of that data is controlled, to ensure security and privacy.”

Wang said this digital pathology collaboration marks the first of its type for COTA, which has previously focused on partnerships with drug developers to provide longitudinal patient data to their clinical R&D programs.

He added that many companies and health systems are sitting on untapped digital resources that could be employed in training—and, specifically, externally validating—AI programs aimed at augmenting clinical practice.

“The way I look at it is that, in the future, we're not just going to see this explosion of AI limited to the diagnostic space—I suspect that we will see penetration of AI into every facet of the patient care journey,” Wang said. “And so every step, or every tool that’s developed, will need to be validated on real-world data to make sure that it’s actually performing correctly.”