Prediction of postoperative disease-free survival and brain metastasis for HER2-positive breast cancer patients treated with neoadjuvant chemotherapy plus trastuzumab using a machine learning algorithm

Purpose: This study aimed to develop mathematical tools to predict the likelihood of recurrence after neoadjuvant chemotherapy (NAC) plus trastuzumab in patients with human epidermal growth factor receptor 2 (HER2)-positive breast cancer. Methods: Data of 776 patients from a multicenter retrospective cohort study were collected. All patients had HER2-positive breast cancer and received NAC plus trastuzumab between 2001 and 2010. Two mathematical tools using a machine learning method were developed to predict the likelihood of disease-free survival (DFS) (DFS model) and brain metastasis (BM) (BM model) within 5 years after surgery. For validation, bootstrap analyses were conducted. The area under the receiver operating characteristics curve (AUC) was calculated to examine the discrimination. Results: The AUC values were 0.785 (95% CI 0.740–0.831, P < 0.001) for the DFS model and 0.871 (95% CI 0.830–0.912, P < 0.001) for the BM model. Patients with low-risk DFS or BM events, as predicted by the models, showed better 5-year DFS and BM rates than those with high-risk DFS or BM events (89% vs. 61% for the DFS model, P < 0.001; 99% vs. 87% for the BM model, P < 0.001). These models maintained discrimination abilities in both luminal and non-luminal subtypes, providing prognostic information independent of pathological response. Bootstrap validation confirmed the high generalization abilities of the models. Conclusions: The DFS and BM models have a high accuracy to predict prognosis among HER2-positive patients treated with NAC plus trastuzumab. Our models can help optimize adjuvant therapy and postoperative surveillance.