Survival analysis based on the Bayesian method

BACKGROUND: The increasing incidence and mortality of breast cancer in China is in contrast to the significant decline in European and American countries. Steady and dependable prognosis prediction models for Chinese breast cancer patients have not been established and tested. METHODS: This research adopts statistical data from the Shanghai Breast Cancer Survival Study (n = 5042) to establish 5-year overall survival (OS) prediction models. A prognostic modeling framework is proposed using the dynamic Cox model based on the Bayesian to estimate breast cancer patients’ survival rate. We then assess whether the Bayesian method improves the classic Cox models. All final models are validated internally by bootstrapping method. RESULTS: Compared with clinical expertise, the features selected by Elastic Net increase the accuracy of the prediction model. Tumor grade, Nodal status, TNM stage, Metastasis, ER status, and PR status constitute our final feature combination. The dynamic Cox model based on the Bayesian can obtain better prognostic results than the classic Cox model. The predictive performance of models demonstrates the superiority of the Bayesian method in predicting the survival rate of breast cancer. CONCLUSION: We find that Tumor grade, Nodal stasus, Metastasis, ER status, PR status, and TNM stage have significant value for predicting breast cancer patients’ 5-year (OS). The dynamic Cox model based on the Bayesian further improves the models’ predictable accuracy.