A Machine Learning Approach to Predict the Outcome of Urinary Calculi Treatment Using Shock Wave Lithotripsy: Model...

Area of Science:

• Urology
• Medical Informatics
• Machine Learning

Background:

• Shock wave lithotripsy (SWL) is a common, noninvasive treatment for renal stones <20 mm.
• SWL has a high failure rate (30-89%), necessitating better patient selection.
• Optimizing SWL candidate selection can improve treatment success and resource allocation.

Purpose of the Study:

• To develop a machine learning model for predicting SWL treatment outcomes.
• To aid clinicians in selecting appropriate patients for SWL.
• To enhance decision-making in renal stone management.

Main Methods:

• Utilized a dataset of 58,349 SWL procedures (1990-2016).
• Applied the AdaBoost algorithm with 17 patient/stone attributes.
• Compared AdaBoost performance against 5 other machine learning algorithms.

Main Results:

• The AdaBoost model significantly outperformed other algorithms.
• Achieved sensitivity 0.875, specificity 0.653, PPV 0.7159, NPV 0.839.
• Demonstrated excellent predictive ability with a C-statistic of 0.843 (ROC analysis).

Conclusions:

• A robust machine learning model accurately predicts SWL success based on patient/stone factors.
• The model assists in selecting optimal SWL candidates.
• Improves healthcare resource utilization and patient prognoses for renal stone treatment.