Predicting 90-day risk of urinary tract infections following urostomy in bladder cancer patients using machine learning and explainability.

This research aims to design and validate a machine learning model to predict the probability of urinary tract infections within 90 days post-urostomy in bladder cancer patients. Clinical and follow-up information from 317 patients who had urostomy procedures at the First Affiliated Hospital of Shanxi Medical University (May 2018-May 2024) were analyzed. The dataset were partitioned into training and testing sets, and feature selection was executed via the Least Absolute Shrinkage and Selection Operator regression technique. Seven machine learning algorithms were employed: Logistic Regression, K-Nearest Neighbors, LightGBM, Random Forest, XGBoost, Support Vector Machine, and Multi-Layer Perceptron. Performance metrics for the model were assessed using multiple evaluation indicators, including AUC, accuracy, sensitivity, specificity, Positive Predictive Value, Negative Predictive Value, and F1 score. SHapley Additive exPlanations and Local Interpretable Model-agnostic Explanations algorithms were applied for model interpretability. UTIs occurred in 22.08% of patients within 90 days after urostomy. The predictive model pinpointed eight important clinical features. Among the developed models, the SVM model demonstrated the best overall performance with AUC (0.835), accuracy (0.825), precision (0.583), recall (0.778), and F1 score (0.667). This model, designed to assess UTIs risk after urostomy in bladder cancer patients, has been deployed online for healthcare professionals at: https://zqmodel.shinyapps.io/shinydashboard_model/ .