Simplified and Accurate Approach to Sleep Disorder Prediction Using Body Composition Metrics: A Development and Validation Study

Abstract Predicting the risk of sleep disorders such as insomnia, obstructive sleep apnea (OSA), and comorbid insomnia and sleep apnea (COMISA) typically requires costly and time-consuming assessments. The SLEEPS algorithm simplifies this process using only nine questions, including body mass index (BMI). However, BMI alone cannot capture differences in body composition, as individuals with the same BMI may have different muscle and fat distribution. This study aims to improve sleep disorder prediction by incorporating body composition metrics. To achieve this, we applied a tree-based machine learning algorithm to a dataset of 3,291 patients, evaluating demographic data, sleep-related questions, and body composition metrics as potential features for the model. The final feature selection was performed using Shapley additive explanations (SHAP) method. The resulting model, I-SLEEPS (InBody-based SimpLE quEstionnairE Predicting Sleep disorders), used a total of 10 features, including skeletal muscle index (SMI) and fat-free mass index (FFMI) instead of BMI, along with the original SLEEPS questionnaire items. I-SLEEPS achieved superior predictive accuracy (AUROC > 0.93 for insomnia, OSA, and COMISA) compared to SLEEPS (AUROC > 0.90). Additionally, our approach significantly enhanced area under the precision-recall curve (AUPRC) values, which is critical for addressing the imbalanced datasets of insomnia and COMISA. Furthermore, our analysis revealed distinct relationships between muscle mass indices (SMI and FFMI) and the risks of insomnia, OSA, and COMISA, providing new insights into the role of body composition in sleep disorders. By leveraging InBody analysis, I-SLEEPS offers a practical, non-invasive alternative to traditional screening methods such as polysomnography.