Survival Risk Prediction Of Esophageal Squamous Cell Carcinoma Based On Adaptive LASSO

Esophageal carcinoma,a malignancy of digestive tract,ranks next only to lung cancer,liver cancer and gastric cancer in China with high morbidity and mortality.Distinguished from western esophageal adenocarcinoma,esophageal squamous cell carcinoma is extraordinary common in China.Surgical treatment does not always prolong patient survival time commendably because of individual differences.According to the clinical phenotypic data of esophageal squamous cell carcinoma patients,survival risk can be predicted precisely,providing guidance for the selection of treatment schemes for patients.The main research contents are as follows:Aiming at the problem of possessing many variables in clinical phenotypic data and poor prediction effect of survival risk model based on full variables,a survival risk prediction model based on adaptive least absolute shrinkage and selection operator(Ada LASSO)is proposed.Combined with chi-square test and information entropy analyses,the early and mid-late decision classier is established by screening TNM-related pathological examination indexes.According to the blood indexes collected,further extracting strong-related makers by Ada LASSO,the early and mid-late linear prediction models are built by Logistic Regression(LR)analysis,respectively.The established risk prediction models can effectively predict the five-year survival condition and can provide credible clinical suggestions for patients.Aiming at the problem of low accuracy of risk prediction between TNM model and log odds of positive lymph nodes(LODDS)models,a survival risk prediction model based on Ada LASSO and nomogram is proposed.Univariate COX analysis is used firstly to screen pathological examination indicators,and Ada LASSO is applied to optimize reliable variables.According to the determined important variables,a probabilistic risk prediction model is constructed,with the advantages of visualizing weights of involved variables.Compared with the TNM and LODDS models,the established model obviously improves the precision of risk prediction,which provides an idea for the prognosis analysis of patients.Aiming at the problem of unsatisfactory accuracy of linear risk prediction model,a nonlinear risk prediction model based on Ada LASSO-Modify Density Peaks Clustering(MDPC)-Back Propagation(BP)neural network is proposed.Ada LASSO algorithm is employed to select meaningful clinical phenotypic features that are significantly associated with survival.Considering the individual differences among patients,DPC algorithm is improved with the help of cosine distance,which is contributed to realize effective cluster of patients.Furthermore,based on patients of different clusters,BP neural network is used to construct risk prediction models.The experimental results explain that the prediction accuracy of the model proposed is 5.2% higher than that of the linear model.According to the test report and risk assessment of each patient,the cluster category was automatically matched and the corresponding treatment scheme can be recommended.