A machine learning approach using stone volume to predict stone-free status at ureteroscopy

Area of Science:

• Urology
• Medical Imaging
• Machine Learning

Background:

• Predicting stone-free (SF) status after ureteroscopy (URS) is crucial for patient management.
• Current methods often rely on stone size, which may not fully capture stone burden.

Purpose of the Study:

• To develop and validate a predictive model for SF status post-URS.
• Incorporate total stone volume alongside clinical and radiological factors.

Main Methods:

• Retrospective analysis of 330 patients undergoing URS for kidney stone disease (2012-2021).
• Stone volume measured by preprocedural CT; SF status defined by endoscopic and radiographic follow-up.
• A bagged trees machine learning model with cross-validation was employed.

Main Results:

• The model achieved 74.5% accuracy and 0.82 AUC.
• Total stone volume (17.7%) was the most significant predictor, followed by operation time and age.
• Stone volume demonstrated higher predictive importance than single or cumulative stone size.

Conclusions:

• Machine learning effectively predicts SF status in patients undergoing URS.
• Total stone volume is a more critical factor than stone size in predicting SF status.
• Findings can optimize patient counseling and guide endourological practice and guidelines.