Application Of Decision Tree Model In Predicting 5-year Survival Of Breast Cancer

Objective:By constructing the decision tree(DT)model to predict the prognosis of female breast cancer within 5 years,and provide a reference for clinicians to predict the prognosis of patients and adjust the individualized follow-up strategy.Methods:In this study,405 breast cancer patients diagnosed from January 2010 to October 2014 were followed up to determine the prognosis(survival or death)of the patients within 5 years after the diagnosis of breast cancer.The Classification and Regression Tree(CART)algorithm of DT is selected to build the prediction model after a variety of model screenings and the model was identified by 10-fold cross-validation(CV).Finally,we evaluated the performance of the model through receiver operating characteristic(ROC)curve,precision – recall(PR)curve,learning curve and calibration curve.Result:After calculation,the decision tree model obtained ideal results: the CV of average recall =0.91 and standard deviation(Std)= 0.05.The prediction results of the model verified the results of the CV and stability of the model: recall = 0.88,average accuracy = 0.92,F1 value = 0.86,true positive rate(TPR)= 0.88.The area under curve(AUC)of ROC reached 0.91,and the average AUC of the PR curve was 0.882.The learning curve showed that the model was overfitting.The calibration curve indicated that the model was mainly overconfident and underestimated the risk of death.Among the 17 features,pathological stage,molecular typing,tumor size,HER-2 and other factors showed a significant correlation and importance in our prediction.Conclusion:The research results proved the accuracy and stability of the decision tree model in predicting the prognosis of breast cancer within 5 years and provided a reference for clinicians to predict the 5-year prognosis of female breast cancer patients and adjust the individualized follow-up strategy.