Prediction model for additional procedure requirement in flexible ureterorenoscopy using explainable artificial intelligence

Comprehensive Summary

This study developed machine learning models to predict which patients will need additional intervention after flexible ureterorenoscopy (f-URS). The analysis included 656 patients, with about 27% requiring further treatment such as repeat f-URS, ureteroscopy, or shock wave lithotripsy. The authors tested multiple machine learning models and feature selection methods, then evaluated performance using both accuracy metrics and explainability tools like SHAP and LIME. Among all variables, the ureteropelvic junction–pelvis angle (UPJ–PA) stood out as the strongest predictor. Patients with a UPJ–PA below 110° had a dramatically higher risk of needing additional intervention compared to those above this threshold (84.3% vs 2.8%). Logistic regression achieved very high predictive performance (AUC ≈ 0.99), and explainability analyses consistently ranked UPJ–PA as the most important feature, followed by access sheath diameter and use of FANS-UAS. The results were stable across different models and sensitivity analyses, suggesting that the findings are not model-dependent.

Outcomes and Implications

The strong association between a low UPJ–PA and higher intervention rates suggests that this measurement could help identify patients who are less likely to benefit from the procedure. This is important because a significant number of patients still require repeat treatment, which adds risk and cost. The use of explainable machine learning also helps clarify which factors actually matter, rather than relying on unclear predictions. If validated in larger studies, this could support better patient selection and reduce unnecessary repeat procedures. However, since the study is retrospective, further testing in independent cohorts will be needed before applying these findings in routine practice.