Constructing a machine learning model for systemic infection after kidney stone surgery based on CT values

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• Post-operative Systemic Inflammatory Response Syndrome (SIRS) is a significant concern following endoscopic kidney stone surgery.
• Early identification of patients at risk for SIRS and subsequent urosepsis is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting SIRS after endoscopic kidney stone surgery.
• To utilize Computed Tomography (CT) values and other clinical variables for risk stratification.
• To provide urologists with a tool for early diagnosis and management of post-operative complications.

Main Methods:

• A retrospective study included 833 patients undergoing retrograde intrarenal surgery (RIRS) or percutaneous nephrolithotomy (PCNL).
• Five ML algorithms were trained using ten preoperative/intraoperative variables to predict SIRS.
• The SHapley Additive exPlanations (SHAP) method was employed to interpret feature importance.

Main Results:

• 15.1% of patients (126/833) developed postoperative SIRS.
• All five ML models showed strong predictive performance (AUC range: 0.690-0.858).
• The eXtreme Gradient Boosting (XGBoost) model achieved the highest AUC (0.858), with high sensitivity (0.877) and specificity (0.981).
• Key predictors identified by the ML model and SHAP analysis included Hounsfield Unit (HU), urinary protein, stone burden, and serum uric acid.

Conclusions:

• A novel ML model effectively predicts SIRS after endoscopic kidney stone surgery using CT values and clinical data.
• The developed model demonstrates high accuracy and can aid in assessing urosepsis risk in postoperative patients.
• This tool supports urologists in early risk identification and management strategies for post-surgical complications.