Abstract P4-21-24: Development of mathematical prediction models to identify disease-free survival events for HER2-positive primary breast cancer patients treated by neoadjuvant chemotherapy and trastuzumab

Abstract

Abstract Background) The addition of trastuzumab to standard neoadjuvant chemotherapy (NAC) doubles the pathological complete response (pCR) rate in patients with HER2-positive primary breast cancer. Patients who achieved pCR after NAC with trastuzumab showed a better prognosis compared to those without pCR. However, it is still difficult to predict the likelihood of recurrence after surgery at an individual patient-level. The aim of this study was to develop a mathematical model to predict disease-free survival (DFS) events such as recurrence for patients treated with NAC and trastuzumab. Because brain metastasis (BM) often occurs in HER2-positive cancer patients and it is a particular event for those, we planned to develop a specific model for BM as well. Patients and Methods) Data of 776 HER2-positive primary breast cancer patients from the multicenter cohort study (JBCRG-C03) were used in the analysis. All patients had received NAC plus trastuzumab between 2001 and 2010. Two prediction models using a machine learning method (alternating decision tree algorithm) were developed using age, body-mass index, menopausal status, clinical stage, histological type, ER/PgR status, histological/nuclear grade, type of surgery, pathological response, adjuvant radiation therapy, and adjuvant hormonal therapy. The model A (DFS) predicted the probability of any disease recurrence, death by any cause, or secondary malignancy within 5 years after starting treatment. The model B (BM) predicted the probability of occurrence of BM within the 5 years. First, bias-controlled virtual datasets were generated for the training of the models using a resampling method. Second, the models were optimized by cross-validation (CV). Finally, the developed models were validated using the original dataset. The area under the receiver operating characteristics curve (AUC) was calculated to assess the discrimination ability of the models. Results) The DFS and BM event was observed in 118 and 30 patients, respectively. The AUC values for the model A and model B were 0.833 (95% CI, 0.798–0.868, P < 0.001) and 0.927 (95% CI, 0.905–0.949, P < 0.001), respectively. The sensitivity and specificity at the cut-off value of 50% were 72.0% and 78.4% for the model A, and 100% and 83.7% for the model B, respectively. Patients predicted as “low-risk” by the model A showed a significantly better 5-year DFS rate than “high-risk” patients (91.2% vs 53.8%, P < 0.001). Patients predicted as “low-risk” by the model B showed a significantly better 5-year BM-free survival rate than “high-risk” patients (100% vs 76.1%, P < 0.001). The discrimination ability of these models were maintained for both ER/PgR-positive and ER/PgR-negative subgroups, and also for both pCR and non-pCR subgroups. Conclusions) Our models showed high accuracy for predicting DFS events and BM in HER2-positive primary breast cancer patients treated with NAC and trastuzumab. These two models would help to realize accurate prediction of DFS events and to optimize the postoperative surveillance plan. The identification of high-risk patients for recurrence including BM may be useful for selecting a patient-subpopulation who requires new therapeutic approach. Citation Format: Takada M, Sugimoto M, Masuda N, Iwata H, Kuroi K, Yamashiro H, Ohno S, Ishiguro H, Inamoto T, Toi M. Development of mathematical prediction models to identify disease-free survival events for HER2-positive primary breast cancer patients treated by neoadjuvant chemotherapy and trastuzumab [abstract]. In: Proceedings of the 2016 San Antonio Breast Cancer Symposium; 2016 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2017;77(4 Suppl):Abstract nr P4-21-24.