Development and validation of a predictive model for postoperative urinary retention following pelvic organ prolapse surgery: a retrospective study

This is the authors' abstract. We don't add key points for this paper.

Objective Postoperative urinary retention (POUR) is a common complication following pelvic organ prolapse (POP) repair surgery, significantly impacting patient recovery. Establishing an effective predictive model facilitates individualized risk assessment and postoperative management. Methods This retrospective study included 698 patients who underwent POP surgery. The dataset was chronologically divided: patients from January 2020 to June 2024 formed the training group ( n = 505 ), and those from July 2024 to December 2025 formed the temporal external validation group ( n = 193 ). Variables were determined using univariate analysis and least absolute shrinkage and selection operator (LASSO) regression, and six machine learning models were developed and compared. Results Among 698 patients, 176 (25.21%) developed POUR. Following feature screening, 10 predictors were included in the models: degree of anterior pelvic prolapse (point Ba), bladder neck mobility, Charlson Comorbidity Index (CCI), preoperative venous thromboembolism (VTE) risk score, preoperative blood glucose (GLU), postoperative analgesia, Parity, and three types of surgical procedures. In the temporal validation cohort, the gradient boosting decision tree (GBDT) model achieved an AUC of 0.848 and showed a modest but statistically significant improvement over logistic regression (ΔAUC = 0.0397; 95% CI: 0.0165–0.0629; Holm-adjusted P = 0.0032), with stable bootstrap performance and higher net benefit across thresholds of 0.10–0.50. Conclusion These findings suggest that the GBDT model has potential for POUR prediction in this single-center setting, but multicenter prospective validation is required before clinical implementation.