Prognosis Prediction Model For Upper Ureteral Calculi Using Random Forest

Objective: The purpose of this study is to establish a prognostic prediction model of different surgical methods for patients with upper ureteral calculi by random forest,and to provide a reference for the surgical choice of upper ureteral calculi.Methods: Collected clinical information of patients with upper ureteral calculi who underwent surgery in the Department of Urology,Shengjing Hospital,China Medical University from January to August 2018 as variables.Sort the variables according to their importance by information gain rate,and filter out more valuable variables,use the SMOTE algorithm to deal with the problem of data imbalance.Use the random forest algorithm to build a prognostic prediction model,and compare the performance with the model obtained using the other three common machine learning algorithms(NB,SVM,ANNs).Results: After sorting by the information gain rate,it was found that 11 variables including the long diameter,short diameter,operation method,age,white blood cell count,stone location(Section A and B),number of stones,unilateral and bilateral,creatinine,and disease course as an effective modeling variable.The Precision of the prognosis prediction model constructed by the random forest algorithm is 87.3%,and the AUC is 0.902.Compared with the other three algorithms,the random forest algorithm has the highest AUC.Conclusion: Machine learning can select valuable variable models from the numerous clinical information of patients with upper ureteral stones.Compared with other common machine learning algorithms,we found that the prognosis prediction model established by the random forest algorithm is the best and can be used for the individual selection of preoperative surgery.When the prediction result under a certain surgery is positive,The technique is recommended,otherwise it is not recommended.