Machine Learning Models Decoding the Association Between Urinary Stone Diseases and Metabolic Urinary Profiles.

Area of Science:

• Urology
• Biochemistry
• Data Science

Background:

• Urolithiasis diagnosis and prevention often rely on identifying urinary metabolic abnormalities.
• Advanced machine learning models offer novel approaches to discover biomarkers from complex urinary data.

Purpose of the Study:

• To identify biomarkers for urolithiasis using 24-hour urinary metabolic profiles.
• To investigate the association between identified biomarkers and different types of urinary stone disease.

Main Methods:

• Retrospective analysis of 468 patients with diagnosed urinary stone disease (renal, ureteral, multiple locations).
• Application of machine learning algorithms, including Random Forest and Super Learner Ensemble Method, to urinary metabolite data.
• Multivariate logistic regression to identify significant predictive features for each stone type.

Main Results:

• Random Forest achieved high predictive accuracy (AUC 0.809 for kidney, 0.99 for ureter, 0.775 for multiple stones).
• 24-h urinary magnesium positively associated with kidney and multiple stones; 24-h urinary creatinine showed a protective effect for kidney and ureter stones.
• Estimated glomerular filtration rate (eGFR) was a risk factor for ureter and multiple location stones.

Conclusions:

• Machine learning effectively links 24-h urinary metabolic data to urological stone disease.
• Identified biomarkers can potentially enhance prediction accuracy for improved prevention strategies.
• Further research can refine dietary and pharmacological interventions based on these findings.