Study On The Predictive Modeling Of Non-small Cell Lung Cancer Survival Analysis

According to the World Health Organization, lung cancer is a malignant tumor with the highest incidence and mortality rates, 1.35 million people throughout the world died of lung cancer each year, and the lung cancer five-year survival rate was only 15% to 30%. Macmillan Cancer Supports shows that the survival time of patients with lung cancer increased from 8 weeks in 1997 to 20 weeks in 2011, which indicates the survival time of patients with lung cancer did not significantly increase due to the development of medicine. The study on patients’ tumor progression modeling has become a new research topic for the majority of health researchers, which helps doctors understand the impact of clinical treatment and the clinical symptoms on tumor progression. Thus, this study has important scientific and clinical significance for lung cancer.In this paper, non-small lung cancer patient is the research object, and study the modeling methods of establishing Cox model to predict lung cancer tumor progression by dealing with time-dependent covariates and variable selection in the Cox model. The main content of this paper is as follows:(1) For time-dependent covariates in longitudinal clinical trial data, this paper proposed the modeling method of establishing Cox model based on the changing characteristic of individual symptoms, to deal with time-dependent covariates. The changing characteristic of patient’s symptoms can reflect the functional form of time-dependent covariates, which avoids the difficulty of estimating the functional form of time-dependent covariates and uses the changing information of time-dependent covariates in longitudinal data. Furthermore, C-Index and calibration curve were used to verify the prediction accuracy of Cox model.(2) This paper studied the variable selection method based on adaptive penalty function to establish varying coefficient model, and this method was verified by using simulation data. The verification results of the simulation data show this method can well identify the varying coefficients, constant coefficients and zero coefficients, which indicates the results of variable selection have Oracle property, and the estimation results were more stable than the present existing methods. In addition, this method was used to identify the type of clinical examination indicators in affecting the tumor size, and then obtain which clinical examination indicator has a varying effect on tumor size, which clinical examination indicator has a constant effect on tumor size, and which clinical examination indicator has no effect on tumor size.(3) To avoid the cumulative errors and over fitting when using classical variable selection method to deal with a large number of covariates, the variable selection method based on adaptive penalty function was used to establish Cox model. In this paper, the adaptive penalty function was improved according to the characteristics of survival time data, which makes the application of adaptive penalty method to the survival time data more suitable. The simulation results show that the improved method still has Oracle property and the coefficient estimates are stable when using this method to establish Cox model. Also, this method was applied to survival time data for patients with non-small cell lung cancer, and the prediction accuracy of the established Cox model increased than the Cox model established by traditional variable selection method 9.8%.