Machine Learning Approaches to Predict 24-Hour Urine Collection Results Based on Self-Reported Beverage Intake.

Area of Science:

• Nephrology
• Data Science
• Biomedical Informatics

Background:

• Kidney stone recurrence evaluation relies on 24-hour urine collections, which are resource-intensive and burdensome for patients.
• Predictive models for urine volume could streamline risk assessment and personalize fluid intake recommendations for kidney stone prevention.

Purpose of the Study:

• To develop and validate machine learning models for predicting 24-hour urine volume using patient-reported beverage intake.
• To classify patient compliance with fluid intake guidelines for kidney stone recurrence prevention.

Main Methods:

• Utilized data from two clinical trials (n=380 development, n=142 validation).
• Trained regression models (Random Forest, SVM, etc.) to predict continuous 24-hour urine volume.
• Trained classification models (Random Forest, SVM, etc.) to identify low urine volume (<2L).

Main Results:

• The Random Forest model demonstrated superior performance in predicting 24-hour urine volume on both training and validation datasets.
• Classification models exhibited high negative predictive values, effectively identifying individuals with adequate urine volume (>2 L/d).
• Models showed limitations in accurately predicting low urine output volumes.

Conclusions:

• Machine learning models integrating beverage intake and demographics can predict 24-hour urine volume in kidney stone patients.
• Current models effectively identify patients with sufficient urine output but require enhancement for identifying those with low output.
• Further research incorporating additional patient data is recommended to improve prediction accuracy and target fluid intake interventions for stone prevention.