Several studies presented at the recent 2026 ASCO Annual Meeting evaluated how digital pathology and artificial intelligence (AI)–based models may improve recurrence risk assessment in HR+/HER2- early-stage breast cancer. Joseph A. Sparano, MD, FACP, FASCO, discusses how these tools may complement gene expression assays and potentially expand access to risk stratification worldwide.

Following these presentations, featured expert Joseph A. Sparano, MD, FACP, FASCO, was interviewed by Conference Reporter Associate Editor-in-Chief Christopher Ontiveros, PhD. Clinical perspectives from Dr Sparano on these findings are presented here.

For HR+/HER2- early-stage breast cancer, what we are trying to assess at diagnosis is an individual’s risk of developing distant recurrence, as distant recurrence is potentially preventable with adjuvant therapeutic interventions. If distant recurrence does occur, although it is currently treatable, it is not curable and is associated with significant morbidity. Having prognostic information at baseline helps us assess whether a patient is at high or low risk of recurrence. If they are at high risk, then predictive information becomes important to identify which therapies would be useful for reducing that recurrence risk.

I think that there is tremendous potential for the use of digital pathology for risk assessment, either to augment existing technologies or, perhaps in some cases, to replace them. Gene expression assays are relatively expensive and are not available in all parts of the world. So, if we can develop a digital pathology signature that correlates with a specific gene expression pattern (eg, a high Oncotype DX Breast Recurrence Score [Exact Sciences Corporation], which indicates a high risk of progression), then we could democratize the availability of these digital pathology technologies. Certainly, everyone generates a hematoxylin and eosin slide in routine clinical care, even in low- and middle-income countries, and that image may be transmitted or photographed and sent centrally for digital analysis of pathology to provide information regarding a surrogate for the recurrence score. Digital pathology may also uncover biology that is not reflected by simply doing immunohistochemistry (IHC) for a protein marker such as ER, PR, or HER2. It may more accurately capture the biology of the disease, and it may be more reliably prognostic for recurrence and potentially predictive for benefit from a specific therapeutic intervention.

Several presentations at ASCO 2026 demonstrated that using multimodal AI-based platforms integrating clinical, genomic, and pathomic data provided more robust prognostic information for distant recurrence than any one of those data sets when used individually. A few of these studies, including those by Magali Lacroix-Triki, MD, PhD, and colleagues (abstract 554), Arkadi Piven et al (abstract 602), and Gil Shamai, PhD, and colleagues (abstract 3005), used the TAILORx image data set for the validation of risk assessment tools in early-stage breast cancer. The study by Dr Lacroix-Triki et al evaluated the RlapsRisk BC tool (Waiv), an AI pathology–based test that integrates digital pathology and clinical pathologic features, and showed that it provided important prognostic information that added to the recurrence score. Additionally, the interesting oral abstract presentation by Dr Shamai used both the TAILORx and the FINHER data sets.

When we see that a digital pathology platform does not correctly identify a tumor as, say, ER+ or HER2+, we assume that it is a false-negative test. Or, if we see it identify ER- tumors as ER+, we assume that it is a false-positive test. The study by Dr Shamai and colleagues presented at ASCO 2026 demonstrated that digital pathology may actually be more accurate at identifying the biology of the tumor. Using the FINHER data set, investigators found that digital pathology–predicted HER2 status added prognostic value and predicted trastuzumab benefit in HER2+ patients beyond HER2 IHC status alone. In HER2+ patients, trastuzumab benefit was observed in those predicted as being HER2+, but not in those predicted as being HER2-. The study suggested that these histomorphology-based platforms may not be giving false negatives, but rather they may more accurately reflect the biology of the disease that is not apparent by simple IHC alone.

Some of these AI-based platforms for assessing breast cancer recurrence risk are commercially available and are supported primarily by retrospective model development and validation studies. There have been no prospective trials yet with these platforms, and we are at the earliest stages of understanding how to clinically implement them.

I would like to see more studies focused on digital pathology as a surrogate to replace gene expression assays for both prognosis and prediction so that we can use these technologies in low- and middle-income countries to facilitate providing better medical care for patients who are diagnosed with breast cancer and live in areas of the world where they do not have access to standard molecular tests. The second thing that I would like to see is more development of multimodal models that integrate clinical, genomic, and pathomic information to better risk stratify patients and to provide better predictive information for treatment benefit.

Although HR+/HER2- early-stage breast cancer has the most favorable prognosis among the breast cancer subtypes, early-stage breast cancers account for more than 60% of current breast cancer–specific deaths, mainly because of their high prevalence. This is particularly relevant to HR+/HER2- disease because it is the most common subtype and the recurrence risk continues beyond 5 years.