Abstract 4369695: Unsupervised Clustering of Nocturnal Heart Rate Variability Reveals Autonomic Subtypes in Dementia Caregivers

Abstract

Background: Dementia caregivers face elevated cardiovascular risk due to chronic stress and poor sleep. Nocturnal RMSSD, a time-domain HRV measure of parasympathetic activity, offers a non-invasive marker of early autonomic dysfunction and has been linked to adverse CVD outcomes. Despite its promise, few studies have examined RMSSD trajectories to detect autonomic risk in high-stress groups like dementia caregivers. Research Questions: 1) Can unsupervised clustering of sleep-derived RMSSD features uncover distinct autonomic profiles among dementia caregivers? 2) Are these RMSSD-derived profiles associated with differences in perceived stress levels? Methods: We analyzed 14 nights of RMSSD data from dementia caregivers (N = 141) using validated wearables. For each participant, we computed three features: mean, intra-individual standard deviation (SD), and slope. Unsupervised clustering (K-means and Gaussian Mixture Models) identified RMSSD profiles. Levene’s test assessed between-group variability. Within the high-risk cluster (Cluster 0), RMSSD features were correlated with baseline perceived stress (PSS-10) to explore links between nocturnal autonomic function and subjective stress. Results: Among the 141 dementia caregivers, most were spouses (57.1%) or adult children (40.7%), and 88.5% co-resided with the care recipient. The sample was predominantly female (74.3%), with 58.7% reporting at least one chronic condition. The mean age was 66 years (SD = 13.1). Based on RMSSD trajectories during sleep, three distinct autonomic subtypes emerged (Figure 1 and 2). Cluster 0 (n = 103) exhibited the lowest average RMSSD (μ = 25.43) and the least day-to-day variability (SD = 4.44), reflecting a low-variability autonomic profile compared to Cluster 1 (n = 20; μ = 40.20, SD = 5.27) and Cluster 2 (n = 18; μ = 51.39, SD = 8.86). Levene’s test confirmed significant between-group differences in RMSSD variability (F = 17.97, p < .001). In the low-RMSSD cluster (n = 103), a non-significant inverse trend suggested lower RMSSD was associated with higher perceived stress (r = –0.22, p = 0.071; Figure 3). Conclusions: Real-time analysis of sleep-derived RMSSD identified a large subgroup with low day-to-day variability—suggesting reduced autonomic flexibility and early cardiovascular risk. Wearable-based RMSSD trajectories may help detect early autonomic imbalance; future research should examine whether this profile predicts CVD and is modifiable.