• autonomic nervous system
• chronotype
• circadian rhythms
• heart rate
• sleep
• tda

This study aimed to identify distinct sleep phenotypes based on heart rate (HR) dynamics during sleep and to investigate their associations with demographic and sleep-related factors using a causal inference framework. We analyzed 5040 sleep episodes from 870 healthy adults (18-75 years), collected via the Sleepacta platform using Fitbit devices provided of hr. The analytical pipeline involved preprocessing of HR time series, feature extraction including Topological Data Analysis (TDA) of HR slopes (persistence entropy) and HR nadir timing characteristics, dimensionality reduction using UMAP, and KMeans clustering with optimal cluster number selection via Silhouette analysis. Subject-level cluster consistency and Bayesian multinomial mixed-effects models, guided by a Directed Acyclic Graph (DAG) for confounder adjustment, were employed for statistical analysis.
The clustering approach, primarily driven by weighted HR nadir timing, identified three distinct phenotypes: Cluster 0 ("Seahorse-like," 35.0%, mid-period nadir), Cluster 1 ("Sliding Slope," 38.2%, late nadir), and Cluster 2 ("Symmetric Hammock," 26.9%, early nadir). These phenotypes significantly differed in nadir timing percentage (p < 0.0001), mean/nadir HR, age, sex distribution, Sleep Fragmentation Index (SFI), individual chronotype, absolute chronotype desynchronization, and Wake After Sleep Onset (WASO) (all p < 0.05). However, intra-individual consistency in exhibiting a specific phenotype was low (mean consistency score = 0.17). After causal adjustment, Bayesian modeling revealed significant predictors of phenotype membership, including individual chronotype, age, sex, absolute chronotype desynchronization, Sleep Regularity Index (SRI), and Sleep Efficiency (SE).
This research demonstrates that HR nadir timing is a key differentiator of nocturnal HR patterns, yielding distinct phenotypes associated with various demographic and sleep characteristics. The low intra-individual consistency highlights substantial nightly variability in HR dynamics. These findings underscore the complexity of HR regulation during sleep and provide a robust methodological framework for future investigations into HR-based sleep phenotyping, particularly regarding the interplay between trait-like predispositions and state-dependent nightly variations.