The association between sleep disorders and the risk of atherosclerotic cardiovascular disease: regression analysis and neural network prediction based on NHANES data.

BACKGROUND: Sleep quality and duration play a critical role in cardiovascular health. In recent years, sleep disorders have been identified as potential risk factors for atherosclerotic cardiovascular disease (ASCVD). However, their independent effects and underlying mechanisms remain unclear. METHODS: This study analyzed data from the National Health and Nutrition Examination Survey (NHANES) conducted between 2013 and 2018, encompassing information on sleep quality, sleep duration, and atherosclerotic cardiovascular disease (ASCVD)-related details from 4,719 participants. Spearman correlation analysis, restricted cubic spline regression, and multivariate logistic regression were employed to evaluate the associations. To enhance predictive modeling under class imbalance, we also developed and compared class-weighted neural networks, random forests, and XGBoost models. RESULTS: The results revealed a significant association between sleep disturbances and ASCVD risk (OR = 1.69, 95% CI: 1.20–2.37, P = 0.0079). A “U-shaped” nonlinear relationship was observed between sleep duration and ASCVD, with the lowest risk occurring at 6–8 h of sleep. Compared to 6–8 h of sleep, short sleep (< 6 h) and long sleep (> 8 h) increased ASCVD risk by 45% and 28%, respectively. Among the machine learning models, the neural network demonstrated the highest sensitivity (0.819) and area under the curve (AUC, 0.841) for ASCVD detection, outperforming random forests and XGBoost in identifying true positive cases. CONCLUSION: In a large, nationally representative U.S. cohort study, self-reported sleep disturbances and abnormal sleep durations were significantly and independently associated with an increased risk of ASCVD. These findings underscore the potential value of sleep health assessment in public health strategies for preventing ASCVD and highlight the utility of neural network models in improving screening sensitivity under conditions of imbalanced data.