Recurrence risk prediction model in HER2-positive early breast cancer after HER2-targeted therapy.

HER2-positive breast cancer patients treated with adjuvant targeted therapy, including trastuzumab or pertuzumab have demonstrated improved outcomes. However, a part of patients still experience recurrence despite targeted therapy. This study aims to develop a time-dependent model to predict recurrence risk in HER2-positive early breast cancer patients following targeted therapy, utilizing data from the APHINITY trial. The APHINITY trial included two arms: trastuzumab-only arm (n = 2,400) and trastuzumab + pertuzumab dual-target therapy arm (n = 2,404). Each group was randomly divided into training (70%) and validation (30%) cohorts, resulting in 3,363 patients in the training set and 1,441 patients in the validation set. The Cox proportional hazards model and two machine learning models, Random Survival Forest (RSF) and XGBoost (XGB), were used to predict invasive disease-free survival. Model performance was evaluated using Harrell`s C-index and area under the curve (AUC). After selecting clinical variables provided by the APHINITY trial, 12 variables were included in the model training. The predictive performance of Cox model, RSF and XGB machine learning models was assessed. Among them, the RSF model demonstrated the best predictive effectiveness. In the training set, the RSF model achieved a C-index of 0.66, with AUCs of 0.78 for 1-year recurrence risk, 0.70 for 3-year recurrence risk, and 0.66 for 5-year recurrence risk. In the validation set, the RSF model achieved a C-index of 0.68, with AUCs of 0.79 for 1-year recurrence risk, 0.73 for 3-year recurrence risk, and 0.71 for 5-year recurrence risk. The XGB model performed slightly worse than RSF, and the machine learning methods significantly outperformed the Cox model. In this study, a time-dependent recurrence prediction model was established based on large-sample randomized controlled trial, demonstrating a favourable short-term recurrence prediction effect, which can serve as a clinical decision assistant for screening patients at high risk of recurrence for intensified adjuvant therapy or follow-up monitoring.

Duke Scholars

Author Haotian Zou