445 Arousability from sleep as a multidimensional construct: Cross-sectional analysis and machine-learning based feature selection

Objectives/Goals: To develop and assess a multidimensional measure of arousability by integrating respiratory, autonomic, cortical, and movement-related events and to identify sex-specific arousal phenotypes using clustering analyses and LASSO-based machine learning for feature selection and model optimization. Methods/Study Population: We will conduct a retrospective cross-sectional analysis using sleep and clinical data from >10,000 adults in three different cohorts. Arousability will be defined by event-level metrics across respiratory, autonomic, cortical, and movement domains. Variable clustering will be used to construct a multidimensional composite index of arousability based on those domains. Generalized linear models will test sex differences in individual metrics composite indices. LASSO regression will be applied to select arousability features most predictive of sex, mortality, and cardiovascular disease (CVD). Results/Anticipated Results: We anticipate identifying distinct arousability dimensions and sex-specific arousal profiles. Women are expected to show greater non-respiratory arousals, while men will show higher respiratory-related arousals. The resulting multidimensional index will capture arousal burden more comprehensively than traditional metrics like the apnea–hypopnea index. This will aid in understanding sex-based difference of sleep fragmentation, as well as understanding the sleep mechanisms most associated with mortality and CVD. Discussion/Significance of Impact: This study will advance precision sleep medicine by developing a multidimensional, data-driven arousability index and identifying sex-specific phenotypes. These findings may improve diagnostic accuracy, influence tailored interventions, and inform risk stratification beyond traditional metrics.