A Novel Digital Phenotype of Physical Activity: Associations with Heart Rate Fragmentation

Background Heart rate fragmentation (HRF) has emerged as a novel marker of sinoatrial instability and neuroautonomic dysregulation and is a predictor of incident cardiovascular disease (CVD) events and cardiovascular mortality. Leveraging heart rate fragmentation as a novel measure may help us better understand the prognostic significance of physical activity. Physical activity (PA) patterns are multidimensional and difficult to study holistically, however. We sought to examine for novel PA clusters in a cohort of veteran men and evaluate their prognostic significance based on their associations with neuroautonomic dysregulation. Methods In this cross-sectional study, we examined 602 male veterans from the Vietnam Era Twins Study of Electrophysiology in Posttraumatic Stress Disorder (VETS-EP) who were enrolled between 2022 and 2026. Accelerometry data were used to derive physical activity parameters across 6 domains: volume, sedentary behavior, intensity, diurnal pattern, fragmentation, and variability. Heart rate fragmentation was assessed using percentage of inflection points (PIP), inverse average length of segments (IALS), percentage of short segments (PSS), and percentage of alternating segments (PAS). Gaussian mixture models were used to derive digital phenotypes of physical activity, and associations with heart rate fragmentation were assessed using linear mixed models. Results Participants had a mean age of 73.8 ± 2.9 years and 93.2% reported White race. Two overlapping phenotypes of physical activity patterns were identified: a compact-early phenotype, characterized by earlier activity timing and slightly greater overall activity, and a diffuse-late phenotype, characterized by later activity timing and greater heterogeneity in activity volume. In fully adjusted models, the diffuse-late phenotype was associated with higher PIP, the primary outcome (β: 1.199, 95% CI: 0.034 to 2.363, p-value: 0.044), and higher IALS (β: 0.01210, 95% CI: 0.00048 to 0.02373, p-value: 0.041), but not PSS or PAS, compared with the compact-early phenotype. Conclusions In older male veterans, Gaussian mixture model phenotyping identified two overlapping physical activity phenotypes, differing mainly in diurnal timing and in the dispersion of overall activity, that were associated with modest differences in heart rate fragmentation. These phenotypes provide a framework for capturing complex physical activity patterns. Future prospective studies representing diverse populations are needed to validate such phenotypes.