Comprehensive Summary
Postoperative risk stratification is important for identifying which patients may develop fever after percutaneous nephrolithotomy (PCNL), even when preoperative urine is sterile. In this study, Mayo Adhesive Probability (MAP) score was integrated with machine learning (ML) techniques to improve prediction of postoperative fever. After variable selection and testing models, logistic regression showed the strongest predictors for postoperative fever were MAP ≥ 3, diabetes, female sex, positive urine leukocytes, and a low lymphocyte-monocyte ratio. Within these indicators, MAP score and urine leukocytes were highlighted as the most influential contributors. Consequently, this work presents an interpretable, online tool that can deliver personalized, preoperative risk estimates.
Outcomes and Implications
This machine learning tool provides a data-driven method to identify high-risk patients who might otherwise be overlooked by traditional assessments. This can support earlier interventions such as modified antibiotic prophylaxis or closer postoperative monitoring. Second, the study demonstrates the growing practicality of applying interpretable machine learning—especially logistic regression coupled with SHAP analysis—in routine urologic care. Finally, by offering an accessible online interface, the model bridges research and clinical practice, helping physicians make individualized, evidence-based decisions that can reduce complications and improve recovery in PCNL patients.