Ensemble learning approaches for early prediction of chronic kidney disease based on polysomnographic phenotype...

This study presents an ensemble learning approach for automated screening and severity classification of chronic kidney disease (CKD) using polysomnographic (PSG) phenotypes. We analyzed PSG data from 358 subjects (179 CKD, 179 early-CKD) in the Cleveland Family Study using four ensemble algorithms: Random Forest, XGBoost, LightGBM, and CatBoost. A total of 1210 sleep-related variables were extracted, covering respiration, sleep stages, movement, and cardiovascular features. The models achieved high multiclass classification performance, with AUCs exceeding 89% across CKD stages. Feature importance analysis revealed that disruptions in oxygen saturation, sleep architecture, and heart rate variability were closely associated with CKD severity. These findings highlight the potential of PSG-derived phenotypes combined with ensemble learning for early CKD detection and risk stratification, supporting timely intervention and improved patient management.