Development and Validation of Machine Learning Models for Predicting Initiation of Emergency Dialysis in Advanced Chronic Kidney Disease

The study aimed to develop machine learning models to predict the initiation of emergency dialysis in patients with advanced chronic kidney disease. The research included 3,062 individuals (2,449 in the derivation cohort and 613 in the validation cohort) from a Japanese medical database from 2014 to 2022. Emergency dialysis was initiated in 237 participants (7.7%), with 185 patients (7.6%) in the derivation cohort and 52 patients (8.5%) in the validation cohort. Four models were tested including logistic regression, support vector machine, XGBoost, and random forest. The random forest model achieved the highest value with an area under the receiver operating characteristic curve of 0.799. Key predictors were hemoglobin, proteinuria, baseline eGFR, diabetes history, and diuretic use. The models demonstrated moderate discriminatory potential and support their potential use for risk stratification, although further external validation is needed.