0280 Longitudinal Sleep–Health Phenotypes Identified by Hierarchical Clustering

Abstract Introduction Sleep health is multidimensional, yet most studies examine isolated metrics and overlook how combinations of physiological and behavioral features form stable sleep phenotypes. We utilized hierarchical clustering to identify reproducible sleep-health subgroups within the Sleep Heart Health Study (SHHS) and examined how these phenotypes evolved over 5 years. We additionally validated cluster differences using an independent slow-wave analysis. Methods We analyzed data from 1,468 adults who participated in both SHHS-1 and SHHS-2 (5.19 ± 0.27 years apart) who had complete datasets. Forty demographic, clinical, polysomnogram, and pre-sleep awake EEG spectral metrics were z-scored, outliers removed, and clustered using hierarchical agglomerative clustering with Ward’s linkage. Optimal solutions were determined via silhouette analysis and visualized using principal component analysis (PCA) and dendrograms. Longitudinal transitions in cluster membership were quantified across the two datasets. To independently validate the clusters, slow waves were analyzed during N2/N3 sleep and eight features (count, density, negative duration, total duration, negative amplitude, total amplitude, up slope, and down slope) were computed to assess physiological distinctions between groups. Results Four phenotypes emerged consistently: (1) Mentally Healthy—highest N3 and strongest subjective mental health (2) Light Sleep—lowest N3, highest N1 (3) Physically Unhealthy—oldest, least REM and total sleep (4) Physically Healthy—youngest, longest sleep and REM PCA revealed clear opposition between Clusters 1 vs. 2 and Clusters 3 vs. 4. Only 37.7% of participants remained in the same cluster at the SHHS-2 follow-up, with most cluster transitions shifting toward the Light Sleep phenotype, reflecting age-related deep sleep decline. Slow-wave metrics robustly validated the clusters: the healthiest groups showed the highest amplitude, steepest slopes, and shortest durations, while the Light Sleep and Physically Unhealthy groups exhibited reduced density, flatter slopes, and prolonged durations. Conclusion We identified 4 reproducible sleep–health phenotypes with distinct physiological profiles and longitudinal trajectories. Deep sleep characteristics were central in differentiating groups and drove the most common age-related cluster transitions. Independent slow-wave analysis confirmed that these clusters reflect genuine biological differences across phenotypes. Multidimensional phenotyping may improve personalized risk stratification and guide targeted sleep-health interventions during aging. Support (if any)