Machine learning-assisted decision-support models to better predict patients with calculous pyonephrosis.

Area of Science:

• Urology
• Medical Informatics
• Machine Learning

Background:

• Developing accurate predictive models for calculous pyonephrosis is crucial for effective treatment planning.
• Integrating diverse clinical characteristics can enhance diagnostic capabilities.

Purpose of the Study:

• To develop and validate machine learning (ML)-assisted models for identifying patients with calculous pyonephrosis.
• To improve pre-treatment decision-making by accurately assessing pyonephrosis risk.

Main Methods:

• Retrospective data collection from 322 patients with obstructed hydronephrosis.
• Development of five ML models (LR, Lasso-LR, SVM, XGBoost, RF) using 22 clinical features.
• Model performance evaluated using Area Under the Curve (AUC) and Decision Curve Analysis (DCA).

Main Results:

• The XGBoost model demonstrated high discrimination in the training set (AUC=0.981).
• In the testing set, the Support Vector Machine (SVM) model achieved the highest AUC (0.977), followed closely by Lasso-LR (0.959) and XGBoost (0.958).
• All developed models showed good predictive performance, with high accuracy, sensitivity, and specificity.

Conclusions:

• ML-based models effectively predict obstructed hydronephrosis patients at high risk of pyonephrosis.
• These models offer significant benefits for urologists in treatment planning, patient selection, and clinical decision-making.