Abstract
INTRODUCTION: Disturbed sleep patterns are common in dementia but have not been objectively quantified over long periods. METHODS: We compared a cohort of 83 Alzheimer's disease (AD) patients to 13,588 individuals from the general population. Sleep patterns, heart rate, and breathing rate data were acquired using a zero-burden contactless, under-mattress pressure sensor. Data reduction and explainable machine learning approaches were used to identify sleep phenotypes. RESULTS: AD was characterized by longer time in bed, more bed exits, less snoring, and changes in estimated sleep states. We derived the Dementia Research Institute Sleep Index for Alzheimer's Disease (DRI-SI-AD), a digital biomarker quantifying sleep disturbances. DRI-SI-AD detected the effects of acute clinical events and dementia progression at the individual level. DISCUSSION: Our approach may help bridge a gap in dementia care by providing a zero-burden method for longitudinal monitoring of health events, disease progression, and dementia risk. HIGHLIGHTS: Continuous monitoring reveals dementia-specific nocturnal sleep disturbances. We developed a novel sleep biomarker, Dementia Research Institute Sleep Index (AD), for tracking Alzheimer's disease (AD) progression. We used contactless under-mattress sensors for low-burden, long-term data collection. Prolonged bedtimes and frequent exits were identified as key dementia-related sleep traits. We demonstrated the feasibility of in-home monitoring for dementia care and risk assessment.