Predicting prolonged length of stay after day-case urolithiasis surgery: a machine learning approach

Abstract

Purpose

This study aimed to develop and validate a machine learning model to predict the risk of prolonged length of stay (PLOS) in patients undergoing day-case surgery for urolithiasis.

Methods

A retrospective study design was used to analyze data from 1,317 patients who underwent day-case surgery for urolithiasis. Demographic characteristics, comorbidities, stone features, surgical details, and preoperative blood test results were extracted from medical records. Feature selection was performed using the Boruta algorithm. Five machine learning algorithms—Tabular Prior-Data Fitted Network (TabPFN), Random Forest (RF), Gradient Boosting, eXtreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost)—were then trained to develop predictive models using 10-fold cross-validation. Model performance was evaluated using multiple metrics, including the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive and negative predictive values, F1-score, Cohen’s kappa coefficient, and confusion matrices. SHapley Additive exPlanations (SHAP) analysis was applied to interpret the best-performing model, explaining both global feature importance and individual prediction rationales. An interactive nomogram was constructed to present the final model’s predictions in an intuitive and clinically accessible format.

Results

Neutrophil count (NEUT), hemoglobin (Hb), surgical method, serum calcium (Ca), platelet count (PLT), and red blood cell count (RBC) were identified as significant predictors by the Boruta algorithm. Among the machine learning models evaluated, the Gradient Boosting model achieved the highest performance on the test set, with an accuracy of 95.96%, sensitivity of 78.57%, specificity of 99.69%, an F1-score of 87.32%, and an AUC of 0.727. Notably, SHAP analysis confirmed that Surgical Method, Ca, NEUT, RBC, PLT, and Hb were the dominant features driving the model’s predictive decisions.

Conclusion

Using demographic and clinical data, this study developed a machine learning-based model to predict PLOS after day-case urolithiasis surgery. The Gradient Boosting model achieved acceptable discriminative performance on an internal test set, with good sensitivity and good specificity. While this model shows potential for early risk stratification, prospective multicenter external validation is required before it can be recommended for clinical practice.