Independent, External, Blinded Retrospective Validation of an AI-based Prognostic Assay in ER-Positive/HER2-Negative Early Breast Cancer

ABCSG TR Spotlight: Validation of an Artificial Intelligence-based Prognostic Assay in ER+/HER2- Breast Cancer

This retrospective observational study evaluates the prognostic performance of a locked Artificial Intelligence (AI)-based assay in patients with Estrogen Receptor (ER)-positive, Human Epidermal Growth Factor Receptor 2 (HER2)-negative early breast cancer (EBC) from Austrian Breast & Colorectal Cancer Study Group (ABCSG)-8, with extended follow-up from ABCSG-16 where available. ABCSG will provide digitized hematoxylin and eosin (H&E) slides and required baseline clinicopathologic variables to Spotlight Medical without outcome data for blinded assay inference. ABCSG will then perform the prespecified statistical analyses linking assay outputs to clinical outcomes.

This is an independent, external, blinded, retrospective validation study of a locked AI-based prognostic assay in ER-positive/HER2-negative EBC.

The source population consists of eligible patients from ABCSG-8 with available archived tumor material suitable for H&E slide digitization, with extended follow-up from ABCSG-16 where available. Primary analyses will be conducted in patients with slides that pass prespecified quality control and with required baseline clinicopathologic variables available or derivable according to the prespecified statistical analysis plan.

The assay generates a continuous prognostic score and predefined risk categories using one digitized H&E-stained surgical resection slide together with routine baseline clinicopathologic variables. Spotlight Medical will generate assay predictions while blinded to clinical outcomes. ABCSG will conduct the statistical analyses according to a prespecified protocol and statistical analysis plan.

The primary objective is validation of the association of the assay with time to distant recurrence. Secondary objectives are validation of the association of the assay with disease-free survival and overall survival.

Inclusion Criteria:

• Female patients enrolled in ABCSG-8
• Postmenopausal patients with ER-positive/HER2-negative early invasive breast cancer
• Available archived primary tumor material suitable for H&E slide digitization
• Available required baseline clinicopathologic variables or variables derivable according to the prespecified statistical analysis plan
• Patients with permission for inclusion in this retrospective translational research study

Exclusion Criteria:

• No suitable archived primary tumor material for H&E slide digitization
• Missing required baseline clinicopathologic variables not recoverable according to the prespecified statistical analysis plan
• No analyzable follow-up for the endpoint of interest