Identifying recurrent stone formers with machine learning: A single-centre observational study

Area of Science:

• Nephrology
• Medical Informatics
• Data Science

Background:

• Kidney stones affect a significant portion of the population, leading to high healthcare costs.
• Recurrent kidney stone formation necessitates effective risk identification strategies.
• Current methods lack accuracy in predicting high-risk patients for recurrent kidney stones.

Purpose of the Study:

• To investigate the efficacy of machine learning (ML) in identifying patients prone to recurrent kidney stone formation.
• To develop and validate an ML model using routinely collected clinical and laboratory data.
• To improve early identification of recurrent kidney stone formers.

Main Methods:

• An observational study utilizing the Bern Kidney Stone Registry data.
• Evaluation of data imputation techniques (KDE, median, KNN) within a logistic regression model.
• Application of recursive feature elimination for feature selection and fivefold cross-validation.

Main Results:

• The study included 706 patients, with 79.7% experiencing recurrent stone events.
• Median imputation provided the best model performance, achieving a mean AUC of 0.71 ± 0.03.
• Key predictive features included estimated glomerular filtration rate, age at first stone, oxalate, and pH levels.

Conclusions:

• Routinely collected clinical and laboratory variables are valuable for identifying recurrent stone formers.
• The developed ML approach demonstrates superior performance compared to previous methods.
• Further validation could enable clinical decision support for personalized stone management strategies.