Integrating digital pathology with genomic and clinical data may enhance prediction of late recurrence and refine selection for adjuvant treatment intensification in HR-positive, HER2-negative disease
Digital pathology is, today, “an extremely busy field of research”, remarked Prof. Peter Dubsky, University of Lucerne, Switzerland, who discussed several abstracts during the Rapid Oral session 1 at the ESMO Breast Cancer 2026 congress. “We have had high-impact publications in the last three years, based on a very common methodology framework that starts with a scan of an H&E slide, and complement information contained there with other data. The result is a prognostic score which can be used, for example, to simply infer a very expensive gene expression test in breast cancer (Lancet Oncol. 2026 Apr;27(4):512-526) or predict distant recurrence and optimise patient selection for CDK 4/6 inhibition (J Clin Oncol. 2025 Oct;43(28):3090-3101).”
Adding to this growing body of research, data presented at the Congress showed that integrating digital pathology (AI-Path) with OncotypeDX21-gene recurrence score, which predicts 10-year distant recurrence risk in early-stage, HR-positive, HER2-negative breast cancer, and RSClin N0, which also incorporates clinical tumour features, outperforms existing risk-stratification tools and can improve the selection of those patients who may benefit from adjuvant treatment intensification (Abstract 3RO).
The study included 6,319 patients with HR-positive, HER2-negative, axillary node-negative breast cancer from the TAILORx trial (N Engl J Med. 2018 Jul 12;379(2):111-121) (4,233 for fitting, 2,086 for validation) as well as 979 patients from a Chicago cohort. AI-Path outperformed RS in TAILORx (C-index 0.678 versus 0.640, p < 0.001) driven by better prediction of late DR after 5 years (C-index 0.650 versus 0.568). The novel biomarker, AI-PathClinRS, consistently outperformed RSClin (p=0.0006)C-index 0.710/0.730/0.794 in the TAILORx fitting (p = 0.006) and validation cohorts (p = 0.002) and identified about twice as many high-risk patients as the NATALEE eligibility criteria (Figure).
Figure. A multimodal clinical, pathological, and genomic model (PathClinRS) outperforms current risk-stratification tools (ESMO Breast Cancer 2026, Abstract 3RO)
During the session, Dubsky highlighted the prognostic value of the multimodal biomarker: “What I find striking is the late recurrence time point, the prognostic performance, the C-index increasing to almost 0.7, which is rarely observed in conventional gene expression biomarkers,” he said. Beyond the advances in research, he also highlighted a shift in how digital biobanks are now perceived in oncology. “I believe fully digital biobanks will soon become common, and that is something we need to prepare for. As we continue integrating additional data sources, including circulating DNA, imaging modalities, and spatial biology, we will be able to generate truly individualised predictions of recurrence,” he concluded.
