A Predictive Model for the Success of Endoscopic Combined Intrarenal Surgery by Machine Learning Using Medical Record Information and Diagnostic Image Findings

Abstract

Introduction: There is a difficulty to predict the success rate in endoscopic combined intrarenal surgery (ECIRS). This study aimed to create a prediction model for successful ECIRS using machine learning from medical record information and diagnostic imaging data, and verify its area under the curve (AUC). Methods: Patients who underwent ECIRS for urinary tract stones at Meirikai Tokyo Yamato Hospital were recruited. Patients were excluded if the surgical position was other than the modified Valdivia position, if the patient had been treated multiple times, or if the urinary tract stone had never been evaluated by computed tomography. Collected data included clinical information, blood biochemical findings, urinary findings, and imaging findings of the patients. To assess the performance of our model, we used 10-fold cross-validation where 90% of the data were used for training and 10% for validation. All possible input variables were used to train the model and validate its AUC and accuracy. Results: A total of 441 patients who underwent ECIRS were included. Logistic regression, which had the highest AUC, was used as the machine-learning model. The learning accuracy was adjusted by calculating the importance of the features and selecting 18 items. LDH, stone shape, TP, age, 3rd largest stone exist distal ureter, double renal pelvis or ureter, height, 1st largest stone exists mid-ureter, length diameter of the 1st largest stone and the 2nd largest stone, WBC count of 30-49/HPF and ≥ 100/HPF in urine sediment, MCHC, horseshoe kidney, presence of stones or the number of stones in the renal calyx, and short diameter of the 2nd largest stone were predictors of high success in ECIRS. AUC, accuracy, sensitivity, and specificity were 0.71, 71.9, 77.7, and 54.5, respectively. Conclusion:The machine-learning model obtained using medical record information accurately predicted the success of the ECIRS.