Development and verification of a prediction model using serum tumor markers to predict the response to chemotherapy of patients with metastatic or recurrent breast cancer

Purpose The aim of this study was to develop a prediction model using serum tumor markers to predict the response to chemotherapy of patients with metastatic or recurrent breast cancer. Methods We retrospectively analyzed a training set of 105 patients with metastatic or recurrent breast cancer. Their chemotherapeutic response had been evaluated according to the World Health Organization (WHO)'s response criteria. Our model for predicting response using carcinoembryonic antigen (CEA), carbohydrate antigen (CA) 15-3, and NCC-ST-439 was determined using the area under the receiver operating characteristic curve (ROC-AUC) and the overall misclassification rate (OMR) in a random cross-validation. The prediction model was then verified in a consecutive set of 64 patients. Their response had been evaluated using the response evaluation criteria in solid tumors (RECIST) criteria. Results The best prediction model consisted of the serum CEA, CA15-3, and NCC-ST-439 levels, but the prediction formula varied according to the baseline CA15-3 level (elevated or normal). The overall ROC-AUC and OMR in the training set were 0.83 and 0.19, respectively. The overall ROC-AUC and OMR in the verification set were 0.72 and 0.28, respectively. When the verification set was stratified according to either the objective response or the predicted response, the time-to-progression, but not the overall survival, was significantly different. Conclusion Our model for predicting the response to first-line chemotherapy of patients with metastatic or recurrent breast cancer may be valid because it predicted the outcome of more than 70% of the patients in an independent verification set.