基于8种算法的中晚期结直肠癌死亡风险预测模型的构建OA

Aim:To construct a mortality risk prediction model for advanced colorectal cancer.Methods:A multi-stage stratified random cluster sampling method was used to select 416 patients with advanced colorectal cancer from 11 tertiary hospitals across 5 cities in Shandong Province between September 2019 and December 2021.Among them,254 patients sur-vived and 162 died within 3 years.The patients were randomly divided into a training set and a testing set at a ratio of 7∶3.Based on the training set data,LASSO regression was employed to screen predictive factors,and then eight algorithms,inclu-ding decision tree(DT),random forest(RF),LightGBM(LGBM),AdaBoost,Logistic regression,CatBoost,XGBoost(XGB),and support vector machine(SVM),were used to construct mortality risk prediction models.Results:LASSO re-gression identified 4 predictive factors:radiotherapy,chemotherapy,surgery,and metastasis status.Using these 4 factors,the 8 models were constructed with AUC(95%CI)of ROC curve in testing set as follows:0.841(0.776-0.905),0.924(0.890-0.966),0.928(0.888-0.970),0.932(0.892-0.973),0.941(0.803-0.978),0.917(0.869-0.965),0.936(0.898-0.978),and 0.889(0.823-0.955).The Logistic model demonstrated the best performance,accuracy,precision,recall rate and F1 score were 0.864,0.878,0.976 and 0.835,and SHAP value of radiotherapy,chemotherapy,surgery,and metastasis status were 0.186,0.148,0.138 and 0.052,respectively.Conclusion:The mortality risk prediction model for advanced colorectal cancer constructed using Logistic regression exhibits optimal performance.