Longitudinal Sleep-Health Phenotypes Identified by Hierarchical Clustering in the Sleep Heart Health Study

Sleep health reflects interacting demographic, clinical, micro- and macroarchitectural, and neurophysiological factors that may not be captured by single metrics or diagnostic categories. We applied hierarchical clustering to longitudinal Sleep Heart Health Study data from 1,468 adults with complete polysomnographic, demographic/clinical, and pre-sleep electroencephalographic data at two visits separated by 5.19 ± 0.27 years. Forty nonredundant features selected from candidate demographic/clinical, sleep-stage, and pre-sleep spectral measures were clustered independently at each visit. Four reproducible sleep-health phenotypes emerged: a group with preserved deep sleep and favorable mental-health ratings; a large light-sleep group with low N3 and high N1; an older, physically unhealthy group with shorter total and rapid-eye-movement sleep; and a younger, physically healthy group with longer total and rapid-eye-movement sleep. The same population-level structure was evident at both visits, although only 37.7% of participants retained the same cluster assignment, with transitions most directed toward the light-sleep phenotype. An independent analysis of slow-wave morphology, excluded from cluster construction, differentiated all four phenotypes after false-discovery-rate correction. Groups with preserved or healthier sleep showed more numerous, higher-amplitude, steeper, and shorter slow waves, whereas the light-sleep and physically unhealthy groups showed weaker and more prolonged slow waves. Pre-sleep spectral features did not differ significantly across clusters after correction. These findings identify reproducible but individually dynamic sleep-health phenotypes and demonstrate that macro-architectural cluster structure is reflected in independent measures of NREM sleep microarchitecture.