Ensemble Learning Approaches for Automatic Detection of Chronic Kidney Disease Stages during Sleep

This study investigates the use of ensemble learning methods for the automatic detection of chronic kidney disease (CKD) stages during sleep. We applied and evaluated four ensemble learning approaches-CatBoost, random forest, XGBoost, and LightGBM-to analyze polysomnography (PSG) data from 730 participants of the Cleveland Family Study (539 control subjects and 191 CKD patients). Our analysis involved extracting 1,210 phenotypic parameters from the PSG data, which enabled the classification of CKD into four stages: mild (stages 1 and 2), moderate (stage 3), severe (stage 4), and critical (stage 5). The most informative top 10 parameters were identified for effective stage detection. The models demonstrated high performance, with an area under the curve (AUC) exceeding 89% in detecting severe CKD stages. The proposed ensemble methods show promise as effective pre-screening tools for assessing CKD severity during sleep and could potentially enhance diagnostic capabilities of PSG by not only detecting sleep disorders but also identifying stages of CKD.