Predicting the Efficacy of Repeated Shockwave Lithotripsy for Treating Patients with Upper Urinary Tract Calculi Using an Artificial Neural Network Model.

Abstract

Purpose: To establish a prediction model for repeated shockwave lithotripsy (SWL) efficacy to help choose an appropriate treatment plan for patients with a single failed lithotripsy, reducing their treatment burden.

Patients and methods: The clinical records and imaging data of 304 patients who underwent repeat SWL for upper urinary tract calculi (UUTC) at the Urology Centre of Shiyan People's Hospital between April 2019 and April 2023 were retrospectively collected. This dataset was divided into training (N = 217; 146 males [67.3%] and 71 females [32.7%]) and validation (N = 87; 66 males [75.9%] and 21 females [24.1%]) sets. The overall predictive accuracy of the models was calculated separately for the training and validation. Receiver operating characteristic (ROC) curves were plotted, and the area under the ROC curve (AUC) was calculated. The normalized importance of each independent variable (derived from the one-way analyses) in the input layer of the artificial neural network (ANN) model for the dependent variable (success or failure in repeat SWL) in the output layer was plotted as a bar chart.

Results: This study included 304 patients, of whom 154 (50.7%) underwent successful repeat SWL. Predictive models were constructed in the training set and assessed in the validation set. Fourteen influencing factors were selected as input variables to build an ANN model: age, alcohol, body mass index, sex, hydronephrosis, hematuria, mean stone density (MSD), skin-to-stone distance (SSD), stone heterogeneity index (SHI), stone volume (SV), stone retention time, smoking, stone location, and urinary irritation symptom. The model's AUC was 0.852 (95% confidence interval (CI): 0.8-0.9), and its predictive accuracy for stone clearance in the validation group was 83.3%. The order of importance of the independent variables was MSD > SV > SSD > stone retention time > SHI.

Conclusion: Establishing an ANN model for repeated SWL of UUTC is crucial for optimizing patient care. This model will be pivotal in providing accurate treatment plans for patients with an initial unsuccessful SWL treatment. Moreover, it can significantly enhance the success rate of subsequent SWL treatments, ultimately alleviating patients' treatment burden.