关键词: 加速失效时间模型, 核机器, 风险预测, 正则化, 再生核希尔伯特空间

Abstract: Accelerated failure time model is a widely used survival analysis model. In this paper, we combine with the LASSO penalty to eliminate the redundant predictors, and construct a kernel-based AFT model to capture the complex relationship between predictors and the response. In addition, we propose a new Regularized Garrotized Kernel Machine estimation method. It can better describe the potential nonlinear relationship between predictors and response, realize the automatic modeling of the interactive effects between predictors in nonparametric parts, and hence improve the predictive accuracy of the model. The simulation studies show that compared with the state-of-the-art methods, the method proposed in this paper has higher predictive accuracy for survival time, especially in the situation of complex relationships. Finally, we apply this method to a gastric cancer data analysis and use the clinical and genetic information to predict survival time and risk score. The empirical results reveal the proposed method can provide a helpful reference for the design of clinical accurate diagnosis and treatment scheme based on risk stratification of patients.

Key words: Accelerated Failure Time Model, Kernel Machine, Risk Prediction, Regularization, Reproducing Kernel Hilbert Space