Predicting postoperative fever in culture-negative patients undergoing mini-PCNL using MAP score-augmented machine learning: a retrospective cohort study.

Abstract

Purpose: Postoperative fever is a common complication following percutaneous nephrolithotomy (PCNL) that occurs even in patients with sterile urine cultures. Traditional risk-assessment tools are insufficient in this subset of patients. This study aims to develop a risk prediction model for detecting postoperative fever in patients with sterile preoperative urine cultures by integrating the Mayo Adhesive Probability (MAP) score with machine learning (ML) techniques.

Methods: This retrospective cohort study included 730 patients with sterile urine cultures who underwent mini-percutaneous nephrolithotomy (mini-PCNL) at Taizhou Hospital from March 2022 to March 2025. The least absolute shrinkage and selection operator (LASSO) regression was employed to identify key variables. Ten ML models were built, and Shapley Additive exPlanations (SHAP) analysis was used to enhance model interpretability, and the optimal model was selected. An online tool was deployed for clinical use.

Results: Postoperative fever occurred in 17.4% of the patients. Logistic regression (LR) demonstrated the best predictive performance (AUC = 0.914, accuracy = 92.1%, specificity = 97.3%). Major predictive factors for fever risk included MAP score ≥ 3, diabetes mellitus, female sex, positive urine leukocytes, and low lymphocyte-monocyte ratio (LMR). SHAP analysis confirmed MAP score and urine leukocytes as the most influential variables. Limitations of the study include its single-centre design.

Conclusions: The proposed tool, combining MAP scores with ML, significantly enhances the accuracy of predicting the risk of postoperative fever following mini-PCNL in patients with sterile urine cultures. The LR model demonstrates strong utility and interpretability, offers personalised risk assessments preoperatively and constitutes a practical and accessible online tool for individualised clinical decision-making.