Risk Assessment for Upper Urinary Tract Deterioration in Patients with Neurogenic Bladder: A Retrospective Study...

Upper urinary tract deterioration (UUTD) is a burdensome complication in neurogenic bladder (NB) patients. In this study, we aim to develop a machine learning-based practical tool for UUTD risk assessment.

A total of 2393 NB patients receiving clinical and urodynamic examinations were included in the present study. 2001 patients from northern regions of our country were divided into training and validation sets, 392 patients from southern regions were allocated into the test set for internal-external validation. A series of clinical and urodynamic parameters were analyzed and screened through Boruta, Lasso regression, and decision tree algorithms; the features consistently identified by all three algorithms were used for machine learning model development. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis. Shapley Additive Interpretation analysis was used as a visual interpretation for individual patient.

A set of six feature variables, notably bladder management method and detrusor sphincter dyssynergia (DSD), were determined as the risk factors of UUTD. Among the four machine learning algorithms, the gaussian naive bayes (GNB) model showed the best overall performance, with an AUC (95% CI) of 0.884 (0.867-0.901) in the training set, 0.868 (0.829-0.907) in the validation set, and 0.878 (0.839-0.916) in the test set. Subgroup analyses according to imaging outcome, NB etiology, and age showed the GNB model had consistent performance, with AUC ranging from 0.815 to 0.895. The GNB model incorporating the core variables was presented as a web-based practical tool for individualized risk assessment.

Our study developed a machine learning model for UUTD risk assessment in NB patients. These findings highlight the importance of combining bladder management method and urodynamic evaluation in UUTD risk assessment and provide a practical tool for individualized assessment. Independent external validation is warranted for the model.