The Impact of REM Sleep Fragmentation on Depression: A Cross-Sectional Analysis Using Machine Learning

This study explores the predictive relationship between REM sleep fragmentation and depressive symptoms across a heterogeneous participant sample, including both obstructive sleep apnea (OSA) patients and non-OSA individuals. Previous research has highlighted the impact of REM fragmentation on emotional regulation and its association with increased risks of depression. However, these studies primarily focus on OSA populations, leaving the effects on non-OSA individuals underexplored. This research aims to bridge that gap by evaluating whether REM fragmentation, combined with physiological factors such as neck circumference, BMI, and age, can predict Hospital Anxiety and Depression Scale - Depression Subscale (HADSD) scores. Using machine learning models, including Support Vector Regression (SVR) and K-Nearest Neighbors (KNN), we assessed the predictive performance of these variables across 144 polysomnography (PSG) recordings. The results showed that REM fragmentation emerged as the most significant predictor of HADSD scores, with SVR outperforming KNN in terms of overall accuracy and specificity. These findings highlight the importance of REM fragmentation as a critical marker for psychological health risk, extending its relevance beyond OSA populations.