Predicting the Stone-Free Status of Percutaneous Nephrolithotomy With the Machine Learning System: Comparative...

Area of Science:

• Urology
• Medical Informatics
• Artificial Intelligence

Background:

• Percutaneous nephrolithotomy (PCNL) is a key procedure for kidney stone removal.
• Predicting stone-free status post-PCNL is crucial for patient management and outcomes.
• Existing scoring systems (Guy's stone score, S.T.O.N.E. score) have limitations in predictive accuracy.

Purpose of the Study:

• To develop and evaluate machine learning methods (MLMs) for predicting stone-free status after PCNL.
• To compare the performance of MLMs against established scoring systems (Guy's stone score, S.T.O.N.E. score).

Main Methods:

• Retrospective analysis of data from 222 PCNL patients.
• Utilized 26 parameters including patient, renal, stone, and surgical factors as input for MLMs.
• Evaluated four MLMs: Lasso-logistic (LL), random forest (RF), support vector machine (SVM), and Naive Bayes, assessing performance via Area Under the Curve (AUC).

Main Results:

• The overall stone-free rate was 50%.
• All four MLMs demonstrated superior predictive performance (AUCs ranging from 0.803 to 0.879) compared to Guy's stone score (AUC 0.800).
• MLMs achieved higher accuracies (0.803%–0.818%) than the S.T.O.N.E. score system (0.788%), with Lasso-logistic showing the highest AUC.

Conclusions:

• Machine learning methods offer a robust approach to predicting stone-free status post-PCNL.
• MLMs demonstrate comparable or superior predictive performance to current clinical scoring systems.
• Lasso-logistic regression emerged as a highly effective MLM for this prediction task.