Unsupervised Learning Identifies Sleep Disturbance Subtypes Among Dementia Caregivers.

Abstract

Background: Sleep disturbance is highly prevalent among caregivers of persons with dementia (PWD), affecting approximately 67% of the more than 10 million adults in this role in the United States. Accurately characterizing these disturbances is essential to identify individuals at increased risk for chronic sleep problems and related health consequences. However, few studies have examined how sleep disturbance manifests across diverse caregiver characteristics and contexts.

Objectives: To address this gap, we applied clustering analyses using wearable sleep data to identify distinct subtypes of sleep disturbance among dementia caregivers.

Methods: Clustering analyses were conducted using sleep parameters-total sleep time (TST), sleep efficiency (SE), wake after sleep onset (WASO), and sleep onset latency (SOL)-derived from 14 days of Oura Ring data collected from 143 caregivers of PWD. For each participant, the minimum, maximum, mean, and standard deviation of each sleep metric were computed during the 14-day period. Subsequent one-way ANOVA and chi-square tests were performed to explore differences in caregiver characteristics, caregiving demands, and caregiving support availability across the identified subgroups.

Results: The analytic sample included 143 adults. Three distinct sleep disturbance subtypes emerged. Cluster 1 (Optimal Sleep) demonstrated the least nocturnal wakefulness, the most efficient sleep, and ease in falling asleep relative to the other clusters. Cluster 2 (Disturbed Onset & Maintenance) exhibited the greatest difficulty both falling and staying asleep, whereas Cluster 3 (Insufficient Sleep) was characterized by markedly reduced sleep duration. Key contextual characteristics differentiated the clusters. Cluster 2 had the highest prevalence of comorbid conditions, including hypertension, diabetes, and other chronic illnesses, whereas Cluster 3 was predominantly male and had the lowest support availability.

Discussion: Findings highlight the potential of wearable-derived nocturnal data to characterize distinct sleep disturbance subtypes using clustering analyses. This work underscores the value of building a risk-profiling framework that integrates sleep disturbance subtypes, caregiver characteristics, caregiving demands, and caregiving support availability to inform the development of more precise and effective sleep interventions for dementia caregivers.