Abstract PS3-10-22: An Online Tool to Assist Decisions on Chemotherapy for Men with Luminal Early Breast Cancer

Abstract Background: Due to the rarity of male breast cancer (MBC), prognosis prediction and chemotherapy indication remain understudied. We aimed to develop a generalized model to predict outcomes and identify candidates for chemotherapy omission in luminal early-stage MBC. Methods: Luminal MBC patients from the Surveillance, Epidemiology, and End Results (SEER) program (2010-2018, N=2996) and a Chinese cohort (2000-2020, N=599) were included. SEER-derived patients (2010-2016) were split into training (N=1356) and test (N=905) sets in a 6:4 ratio; others formed the US validation set (2017-2018, N=735). Chinese patients constituted the CHN validation set. The primary outcome was overall survival (OS). Four machine learning models and a Cox-based nomogram were evaluated. Risk stratification (low-, intermediate-, and high-risk groups) was based on the tertiles of 5-year risk scores in the training set predicted by the optimal model. Survival analysis employed Kaplan-Meier, log-rank tests, and Cox regression. Results: The optimal model showed robust performance (C-indices: training set=0.715, test set=0.713, US validation set=0.742, CHN validation set=0.748). OS of three risk groups differed significantly across all datasets (all P