A Novel Machine-Learning Algorithm to Predict Stone Recurrence with 24-Hour Urine Data.

Area of Science:

• Nephrology
• Biostatistics
• Machine Learning

Background:

• Kidney stone recurrence lacks predictive markers, complicating management and clinical trials.
• Unpredictable stone events necessitate large patient cohorts in research.
• Novel algorithms are needed to forecast stone recurrence accurately.

Purpose of the Study:

• To develop and validate a machine learning algorithm for predicting kidney stone recurrence.
• To identify key predictors of stone recurrence from 24-hour urine tests.
• To improve clinical trial efficiency by enabling better patient stratification.

Main Methods:

• Utilized a training set of 423 nephrolithiasis patients from the Registry for Stones of the Kidney and Ureter (ReSKU).
• Employed seven prediction classification methods, including Logistic Regression with ElasticNet.
• Validated the model on a separate set of 172 patients with 24-hour urine data.

Main Results:

• The highest performing model achieved an area under the curve (AUC) of 0.65 in the training set.
• Restricting analysis to high-confidence predictions improved accuracy to AUC = 0.82.
• The model demonstrated moderate discriminative ability in the validation set (AUC = 0.64).

Conclusions:

• Machine learning models analyzing 24-hour urine composition can predict kidney stone recurrence.
• The developed algorithm offers moderate accuracy in forecasting stone events.
• This approach may enhance the management of kidney stone disease and streamline research.