# Screening for Clinically Significant Nephrolithiasis Based on Simple Health Checkup Clinical and Urine Parameters in General Populations: Multicenter Machine Learning Study

## TL;DR

This study developed a machine learning model to screen for kidney stones using simple health data, offering a low-cost and non-invasive alternative to traditional imaging methods.

## Contribution

A novel machine learning model for kidney stone screening using routine clinical and urine parameters in general populations.

## Key findings

- The best-performing model achieved an AUC-ROC of 0.968 and AUC-PR of 0.936 for predicting clinically significant nephrolithiasis.
- Urine red blood cell count, estimated glomerular filtration rate, and urine specific gravity were the top predictors identified by Shapley value analysis.
- The model offers a non-invasive, scalable solution for kidney stone screening, suitable for integration into health checkups or telemedicine.

## Abstract

Nephrolithiasis affects approximately 15% of the population and often remains undetected in asymptomatic individuals. Current diagnostic approaches rely on imaging tools, such as ultrasound or computed tomography, which are costly, operator dependent, or involve radiation, making them unsuitable for large-scale screening. A standardized, practical, and low-cost screening strategy for early identification of clinically significant kidney stones is still lacking.

This study aimed to develop a low-cost, rapid screening model for clinically significant nephrolithiasis using machine learning (ML) and simple clinical parameters.

We conducted a multihospital retrospective study using data from 3 hospitals in Kaohsiung, Taiwan (2012‐2021). Adults without renal colic were included. ML models were trained and tested using 10 routine variables: sex, age, BMI, gout, diabetes, estimated glomerular filtration rate, urine pH, red blood cell count, specific gravity, and bacteriuria. Multiple ML algorithms were trained and evaluated, and the best-performing model was selected based on the area under the receiver operating characteristic curve and the area under the precision-recall curve. To assess model interpretability, Shapley value analysis was performed to determine the relative importance and contribution of each variable to the model’s predictive performance.

Among 6528 participants, the best-performing model achieved an area under the receiver operating characteristic curve of 0.968 (95% CI 0.956‐0.980), an area under the precision-recall curve of 0.936 (95% CI 0.918‐0.953), a sensitivity of 0.873 (95% CI 0.841‐0.904), and a specificity of 0.947 (95% CI 0.935‐0.959). Shapley value analysis identified urine red blood cell count, estimated glomerular filtration rate, and urine specific gravity as the 3 most influential predictors.

This ML-based model enables efficient, noninvasive, and large-scale kidney stone screening using routine health data. It can be integrated into health checkups or telemedicine platforms to facilitate early detection and proactive management. Although the model was developed using an Asian population, future validation in diverse cohorts is warranted to confirm its generalizability.