Search

Background: This study aimed to develop a prognostic model to predict the breast cancer-specific survival and overall survival for breast cancer patients in Asia and to demonstrate a significant difference in clinical outcomes between Asian and non-Asian patients.Methods: We developed our prognostic models by applying a multivariate Cox proportional hazards model to Taiwan Cancer Registry (TCR) data. A data-splitting strategy was used for internal validation, and a multivariable fractional polynomial approach was adopted for prognostic continuous variables. Subjects who were Asian, black, or white in the US-based Surveillance, Epidemiology, and End Results (SEER) database were analyzed for external validation. Model discrimination and calibration were evaluated in both internal and external datasets.Results: In the internal validation, both training data and testing data calibrated well and generated good area under the ROC curves (AUC; 0.865 in training data and 0.846 in testing data). In the external validation, although the AUC values were larger than 0.85 in all populations, a lack of model calibration in non-Asian groups revealed that racial differences had a significant impact on the prediction of breast cancer mortality. For the calibration of breast cancer-specific mortality, P values < 0.001 at 1 year and 0.018 at 4 years in whites, and P values ≤ 0.001 at 1 and 2 years and 0.032 at 3 years in blacks, indicated that there were significant differences (P value < 0.05) between the predicted mortality and the observed mortality. Our model generally underestimated the mortality of the black population. In the white population, our model underestimated mortality at 1 year and overestimated it at 4 years. And in the Asian population, all P values > 0.05, indicating predicted mortality and actual mortality at 1 to 4 years were consistent.Conclusions: We developed and validated a pioneering prognostic model that especially benefits breast cancer patients in Asia. This study can serve as an important reference for breast cancer prediction in the future. [ABSTRACT FROM AUTHOR]