Sleep efficiency in community-dwelling persons living with dementia: exploratory analysis using machine learning.

Abstract

Study objectives: Sleep disturbances lead to negative health outcomes and caregiver burden, particularly in community settings. This study aimed to investigate a predictive model for sleep efficiency and its associated features in older adults living with dementia in their own homes.

Methods: This was an exploratory, observational study. A total of 69 older adults diagnosed with dementia were included in this study. Data were collected via actigraphy for sleep and physical activity for 14 days, a sweat patch for cytokines for 2-3 days, and a survey of diseases, medications, psychological and behavioral symptoms, functional status, and demographics at baseline. Using 730 days of actigraphy, sweat patches, and baseline data, the best prediction model for sleep efficiency was selected and further investigated to explore its associated top 10 features using machine learning analysis.

Results: The CatBoost model was selected as the best predictive model for sleep efficiency. In order of importance, the most important features were sleep regularity, number of medications, dementia medication, daytime activity count, instrumental activities of daily living, neuropsychiatric inventory, hypnotics, occupation, tumor necrosis factor-alpha, and waking hour lux.

Conclusions: This study established the best sleep efficiency predictive model among community-dwelling older adults with dementia and its associated features using machine learning and various sources, such as the Internet of Things. This study highlights the importance of individualized sleep interventions for community-dwelling older adults with dementia based on associated features.