70P Development and validation of a machine learning (ML) nomogram to predict RSClin results and guide adjuvant treatment of node-negative (N0) hormone receptor-positive (HR+)/human epidermal growth factor receptor-negative (HER2-) early breast cancer (eBC) in Europe

RSClin is a proprietary algorithm that integrates clinical-pathological (CP) factors and genomic risk in patients with N0 HR+/HER2- eBC. RSClin refines the prognosis of distant recurrence (DR) and chemotherapy (CT) benefit more accurately than CP and Recurrence Score (RS) alone. Since RSClin is not available in Europe, we aimed to validate an automated ML-based nomogram able to predict RSClin outcomes with potential clinical impact in European countries. We retrospectively collected CP and genomic characteristics of 290 patients with N0 HR+/HER2- eBC from 3 hospitals in Italy and Belgium from 2020 to 2022. Patients were randomly assigned with a 3:1 ratio to either the training or validation cohort. Features with a Pearson correlation over 0.2 were selected for model development. The ML-nomogram development and validation were based both on classification and regression models and included linear and logistic regression to predict DR and CT benefit. Compared to RSClin outcomes, classification model for DR achieved a ROC AUC of 0.97, while for CT benefit reached a score of 0.99. The regression analyses for DR and CT benefit yielded significant R2 scores of 0.84 and 0.72, respectively. A web-based tool was then implemented to increase ML nomogram accessibility worldwide. Within the study population, the use of this tool was able to refine the RS alone estimates of CT benefit converting the advice of CT sparing to CT recommendation in 27/290 (9.8%) patients, assuming a CT benefit>3% as clinically relevant.Table: 70PPerformance of ML tool for RSClin anticipationClassifierNumber of featuresROC AUCF1DR110.970.93CT benefit120.990.97RegressorNumber of featuresR2RMSEDR100.844.5CT benefit110.725.2Abbreviations: CT: chemotherapy benefit; DR: distant relapse; ML: machine learning; R2: R-squared; RMSE: Root mean squared error; ROC AUC: Receiver operating characteristic, area under the curve. Open table in a new tab Abbreviations: CT: chemotherapy benefit; DR: distant relapse; ML: machine learning; R2: R-squared; RMSE: Root mean squared error; ROC AUC: Receiver operating characteristic, area under the curve. Our ML-based nomogram can accurately reproduce RSClin results to support treatment decisions for patients with N0 HR+/HER2- eBC, identifying high-risk patients who may benefit from treatment intensification compared to RS alone. It can also be freely used in Europe where RSClin is not available.