Abstract PO4-01-10: Multi-modal artificial intelligence models from baseline histopathology predict prognosis in HR+ HER2- early breast cancer

Abstract

Abstract Background: Prognostic assessment in HR+ HER2- early breast cancer (EBC) remains challenging given relatively low rates of disease progression. Nuanced risk stratification is needed for decisions regarding systemic therapy. Modern artificial intelligence (AI)-based techniques have already provided substantial medical progress, particularly in prostate cancer. Here, we leverage multi-modal artificial intelligence (MMAI) trained using digital histopathology and clinical data to evaluate whether the MMAI technology can be expanded to other disease indications by applying the technology to the WSG PlanB and ADAPT trials in HR+ HER2- EBC. Methods: Pre-treatment breast biopsy and surgical hematoxylin and eosin (H&E) slides were digitized from the WSG PlanB and ADAPT trials for HR+ HER2- EBC patients receiving endocrine therapy +/- chemotherapy. Median follow-up in both trials exceeded 5 years. A multi-modal artificial intelligence architecture was developed and validated on predicting risk of distant recurrence (DR). Time-to-event endpoints were summarized using cumulative incidence curves. Univariable and multivariable Fine-Gray models were used to assess performance; hazard ratios were reported per standard deviation increase of the model score. Results: A total of 5539 patients from the two trials with H&E images and available follow-up data were used for development and validation of a multi-modal artificial intelligence architecture. The MMAI score derived using the training set was significantly associated with risk of DR in a validation cohort. To put the results into the clinical context in HR+ HER2- EBC, comprehensive validation analyses are currently ongoing and will be presented at the meeting. Conclusions: We have successfully developed a multiple-instances, learning-based deep neural network for outcome prediction using H&E-stained images. This study provides important evidence that MMAI technology can further personalize breast cancer management by adding standardizable information. Citation Format: Daniel Kates-Harbeck, Hans-Heinrich Kreipe, Oleg Gluz, Matthias Christgen, Sherko Kuemmel, Monika Graeser, Ulrike Nitz, Sven Mahner, Doris Mayr, Rachel Wuerstlein, Akinori Mitani, Jingbin Zhang, Hans Pinckaers, Yi Ren, Peter Wood, Jacqueline Griffin, Felix Feng, Andre Esteva, Jess Keim-Malpass, Ronald Kates, Nadia Harbeck. Multi-modal artificial intelligence models from baseline histopathology predict prognosis in HR+ HER2- early breast cancer [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO4-01-10.