Prediction of Sepsis after Endourologic Kidney Stone Surgery: A Machine Learning Approach.

Area of Science:

• Urology
• Infectious Disease
• Machine Learning in Medicine

Background:

• Sepsis following urinary tract infections after kidney stone surgery presents significant morbidity.
• Limited research exists on using hemoglobin A1c (HbA1c) to predict postoperative sepsis in endourologic procedures.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting postoperative sepsis after kidney stone surgery.
• To identify key clinical predictors, including HbA1c, for preoperative optimization and risk stratification.

Main Methods:

• Retrospective analysis of patients undergoing ureteroscopy, shockwave lithotripsy, or percutaneous nephrolithotomy.
• Development of five supervised ML models (logistic regression, random forest, neural network, SVM, naïve Bayes) using demographic and clinical data.
• Model performance assessed using accuracy, AUCROC, calibration, and Brier score via cross-validation and a hold-out test set.

Main Results:

• The random forest model demonstrated superior performance with 91% accuracy and 0.88 AUCROC on the test set.
• Key predictors for urosepsis included preoperative hemoglobin, HbA1c, stone size, surgery duration, and BMI.
• The developed random forest model is accessible online for potential clinical application.

Conclusions:

• A random forest ML model effectively predicts sepsis risk following kidney stone surgery.
• This predictive tool can aid in preoperative surgical planning, patient optimization, and postoperative monitoring.
• Further validation is recommended before widespread clinical implementation.