| Objective: Development and validation the prediction model of objective response to transcatheter arterial chemoembolization(TACE)for intermediateand advanced-stage hepatocellular carcinoma using machine learning(ML)and R-Studio.Methods: 1.The data included in this study were from 425 patients with advanced hepatocellular carcinoma in two medical centers.Among the 394 patients from medical center A,314 patients constituted the derivation set of the model,and the remaining 80 patients constituted the internal validation set of the model.In addition,31 patients from medical center B constituted the external validation set.This study uses a two-stage ML model framework,The framework uses 8 feature selection methods(such as ANOVA F-value,Kendall’s tau,Wilcoxon signed-rank test,etc.)and 14 ML classifiers(such as Ada Boosting Classifier,Bernoulli Naive,etc.)Bayesian and Gradient Boost Classifier,etc.)were used to select predictive indicators,and fourfold cross validation method was used to determine the best combination of feature selection methods and classifiers in the derivation set.At the same time,the indicators with the best performance of the ML framework and the prediction weights of the indicators were determined.In model selection and performance evaluation,the area under the receiver operating characteristic curve(AUC),accuracy,sensitivity and specificity were calculated in the internal and external validation sets,respectively.2.On the basis of the first step ML method,the selected indicators are further verified by using Logistic multivariate regression of R-Studio.At the same time,394 patients from medical center A were randomly divided into the derivation group(n=275)and the internal validation set(n=119)according to7:3.The external validation set was also composed of 31 patients from medical center B.Subsequently,a nomogram was constructed using the selected indicators by multivariate Logistic regression analysis.In the model validation and performance evaluation,AUC,sensitivity and specificity were calculated in the derivation set,internal validation set and external validation set,and the consistency and clinical benefit rate of the model were verified by calibration curve and Decision curve analysis(DCA).Results: 1.In the derivation set of ML framework,the combination of the feature selection method(ANOVA F-value,FTEST)and the Support Vector Machine Classifier(SVMC)achieved the highest average AUC(0.8835)in four-fold cross validation.In the internal validation cohort,the selected combination had an AUC of 0.8444(95% CI [0.7849,0.9439]),an accuracy of0.7625,a sensitivity of 0.6000,and a specificity of 0.9250.The AUC in the external validation cohort was 0.8571(95%CI [0.7239,0.9903]),the accuracy was 0.6774,the sensitivity was 0.5714,and the specificity was 0.9000.The 9predictive indicators selected by the ML method are as follows.tumor diameter,tumor number,Alpha-fetoprotein(AFP),Portal vein tumor thrombus(PVTT),Activated partial thromboplastin time(APTT),Barcelona Clinic Liver Cancer stage(BCLC),Albumin(ALB),Thrombin time(TT)and Hepatitis B surface antigen(HBs Ag).2.In the R-Studio prediction model,multivariate Logistic analysis showed that AFP,PVTT grade,tumor number and tumor diameter had independent predictive value.Using the above four indicators to construct a nomogram,the AUC of the model in the derivation set was 0.870,the sensitivity was 0.692,and the specificity was 0.953.In the internal validation set,the AUC was 0.839,the sensitivity was 0.659 and the specificity was 1.000.In the external validation set,the AUC was 0.888,the sensitivity was 1.000,and the specificity was 0.762.The calibration curve and DCA curve further show that the model has good consistency and clinical benefit rate in the derivation set,internal validation set and external validation setConclusion: The prediction model based on ML method and R-Studio provides an effective method to predict the efficacy of TACE treatment for intermediate-and advanced-stage hepatocellular carcinoma. |
