Machine learning models for screening clinically significant nephrolithiasis in overweight and obese populations

Area of Science:

• Nephrology
• Medical Informatics
• Machine Learning

Background:

• Overweight and obesity are risk factors for nephrolithiasis (kidney stones).
• Early detection of clinically significant nephrolithiasis is crucial for preventing complications.
• Current screening methods may not be optimal for these populations.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for screening clinically significant nephrolithiasis.
• To utilize easily obtainable clinical and urine parameters for the screening models.
• To target overweight and obesity populations (BMI ≥ 25 kg/m²).

Main Methods:

• Developed ML models using parameters: gender, age, BMI, gout, diabetes mellitus, estimated glomerular filtration rate, bacteriuria, urine pH, urine red blood cell counts, and urine specific gravity.
• Data collected from 2928 subjects in Kaohsiung, Taiwan (2012-2021).
• Validated models on a testing dataset of 574 subjects.

Main Results:

• The study included 2928 subjects; 39.21% had clinically significant nephrolithiasis.
• The best ML model achieved an area under the curve of 0.965 on the testing dataset.
• High performance metrics: sensitivity 0.860, specificity 0.947, positive predictive value 0.918, negative predictive value 0.907.

Conclusions:

• The ML-based model effectively distinguishes overweight/obese individuals with clinically significant nephrolithiasis.
• This model can serve as an accessible and reliable screening tool.
• Facilitates early intervention, lifestyle modifications, and medication to prevent stone complications.