Prediction model for postoperative urinary retention in patients undergoing totally extraperitoneal groin hernia repair.

Abstract

Postoperative urinary retention remains a common complication after totally extraperitoneal groin hernia repair, often prolonging hospitalization and increasing patient discomfort. This study aimed to develop a prediction model using machine learning for postoperative urinary retention risk stratification. A retrospective analysis was conducted on 1,125 patients who underwent elective totally extraperitoneal between April 2017 and September 2024. Postoperative urinary retention was defined as postoperative residual urine volume ≥400 mL confirmed by catheterization. Logistic regression was used as a preliminary screening to identify candidate predictors. Subsequently, a Gradient Boosting analysis was applied to these candidate predictors, and based on Gradient Boosting-derived feature importance, a simplified scoring system was established. Postoperative urinary retention occurred in 58 (5.16%) patients. Gradient Boosting analysis, on the basis of variables initially screened by logistic regression, identified advanced age, lower body mass index, longer operative time, and increased intraoperative fluid volume as the 4 most important predictors, collectively accounting for 89% of the predictive contribution. On the basis of these predictors, a simplified scoring system was developed, demonstrating good discrimination (area under the receiver-operating characteristic curve = 0.7803, 95% confidence interval, 0.667-0.797), satisfactory calibration (Brier score = 0.0465, 95% confidence interval, 0.036-0.057), and event-balanced performance that exceeded baseline prevalence (area under the precision-recall curve = 0.160, 95% confidence interval, 0.099-0.251). A 3-tier clinical recommendation was established: low-risk (0-3 points, discharge without monitoring), intermediate-risk (4-6 points, outpatient monitoring), and high-risk (≥7 points, inpatient observation); decision-curve analysis demonstrated a positive net benefit for this scheme across threshold probabilities between 4% and 15%, supporting its use for selective postoperative management after totally extraperitoneal repair. Prospective external validation is warranted.