The predictive value of modified-DeepSurv in overall survivals of patients with lung cancer
Jie Lei,
Xin Xu,
Junrui Xu,
Jia Liu,
Yi Wang,
Chao Wu,
Renquan Zhang,
Zhemin Zhang,
Tao Jiang
Affiliations
Jie Lei
Department of Thoracic Surgery, The Second Affiliated Hospital, Air Force Medical University, Xi’an 710038, China
Xin Xu
Department of Respiratory Medicine, Shanghai Pulmonary Hospital, School of Medicine, Tongji University, Shanghai 200433, China
Junrui Xu
Department of Thoracic Surgery, First Affiliated Hospital of Anhui Medical University, Hefei 230022, China
Jia Liu
Operations Management Department, People’s Hospital of Xinjiang Uygur Autonomous Region, No. 91 Tianchi Road, Tianshan District, Urumqi 830001, China
Yi Wang
Department of Epidemiology and Biostatistics, School of Public Health and Management, Wenzhou Medical University, Wenzhou 325035, China
Chao Wu
Department of Respiratory and Critical Care Medicine, People’s Hospital of Xinjiang Uygur Autonomous Region, No. 91 Tianchi Road, Tianshan District, Urumqi 830001, China; Xinjiang Clinical Research Center for Interstitial Lung Diseases, People’s Hospital of Xinjiang Uygur Autonomous Region, No. 91 Tianchi Road, Tianshan District, Urumqi 830001, China
Renquan Zhang
Department of Thoracic Surgery, First Affiliated Hospital of Anhui Medical University, Hefei 230022, China; Corresponding author
Zhemin Zhang
Department of Respiratory Medicine, Shanghai Pulmonary Hospital, School of Medicine, Tongji University, Shanghai 200433, China; Corresponding author
Tao Jiang
Department of Thoracic Surgery, The Second Affiliated Hospital, Air Force Medical University, Xi’an 710038, China; Corresponding author
Summary: The traditional prognostic model may induce the possibility of incorrect assessment of mortality risk under the assumption of linearity. It is urgent to develop a non-linearity precise prognostic model for achieving personalized medicine in lung cancer. In our study, we develop and validate a prognostic model “Modified-DeepSurv” for patients with lung carcinoma based on deep learning and evaluate its value for prognosis, while Cox proportional hazard regression was used to develop another model “CPH.” The C-index of the Modified-DeepSurv and CPH was 0.956 (95% confidence interval [CI]: 0.946–0.974) and 0.836 (95% CI: 0.774–0.896), respectively, in the training cohort, while the C-index of the Modified-DeepSurv and CPH was 0.932 (95%CI: 0.908–0.964) and 0.777 (95%CI: 0.633–0.919), respectively, in the test dataset. The Modified-DeepSurv model visualization was realized by a user-friendly graphic interface. Modified-DeepSurv can effectively predict the survival of lung cancer patients and is superior to the conventional CPH model.
