Abstract

Info

Title:: Improved DeepSurv model for survival analysis in lung cancer and exploration of influencingfactors

Author(s):: ZHAO Qiyang1; ZHAO Xu1; ZHANG Ying1; KUANG Manman1; XI Qun1; 2; 1. School of Medical Information Engineering, Gansu University of Chinese Medicine, Lanzhou 730000, China; 2. Information Center,Lanzhou University Second Hospital, Lanzhou 730000, China

Abstract:: Objective To evaluate the performance of an improved DeepSurv model for survival analysis in lung cancerpatients, and investigate key factors affecting the prognosis of lung cancer. Methods The lung cancer data from the SEERdatabase (2018-2021) was used in the study, and the DeepSurv model was optimized by incorporating a self-attentionmechanism, a residual network, a LIME module and an entropy regularization term to enhance prediction accuracy andinterpretability. Model performance was assessed using C-index and Brier score, and the improved model was utilized toevaluate the effects of various features on the prognosis of lung cancer. Results The improved DeepSurv model achieved aC-index of 0.852 and a Brier score of 0.139. Feature importance analysis identified age as the primary determinant of thesurvival of lung cancer patients. Conclusion The improved DeepSurv model outperforms both the Cox proportional hazardsmodel and the original DeepSurv model in terms of accuracy, robustness and interpretability, offering a novel methodologyfor personalized medicine and survival analysis.