Wrist-Worn Devices Detect Sleep Architecture Changes Throughout Alzheimer's Disease Progression But May Underestimate Their Magnitude

BackgroundThe relationship between sleep and Alzheimer’s Disease (AD) is bi-directional. Sleep may contribute to AD pathogenesis, but spreading AD pathology across the brain may also de-regulate sleep. Assessment of sleep architecture change is therefore an important contextual factor for symptom severity in AD. However, polysomnography is impractical for use at scale. The goal of this study was to explore wearable-based estimation of sleep stages across the AD spectrum and compare these results to the gold-standard.MethodsRADAR-AD is a cross-sectional study (n=200), including biomarker-confirmed healthy controls (n=51), preclinical AD (n=37), prodromal AD (n=59), and mild-to-moderate AD patients (n=53), in which participants wore a Fitbit Charge 3 for eight weeks. Sleep features included total sleep time, time awake, sleep efficiency, light, deep, and REM sleep durations and proportions. Linear regression models were used to examine the differences between disease stages across features, and these were compared to polysomnography studies in literature using standardised mean differences. Test-retest reliability was assessed using intraclass correlation coefficients.FindingsBy the prodromal AD stage sleep architecture had changed, with higher proportions of light sleep and lower proportions of REM and deep sleep compared to healthy controls. Lower sleep efficiency and higher waking after sleep onset were seen even earlier, from the preclinical stage onward. However, as disease severity increased, the wearable tended to underestimate the magnitude of disease-related change. Test-retest reliability suggests a recommended study duration of at least three weeks.InterpretationWearables can detect sleep architecture changes throughout the Alzheimer’s spectrum, even before cognitive symptoms emerge, but long study durations are essential to overcome natural variability and measurement errors. Comfortable, inexpensive wearables designed for healthy participants have the potential to facilitate larger, more powerful studies than are possible using polysomnography, but their disease-specific performance must always be considered.