Indexing the Predictive Utility of Real‐World Sleep to Discriminate Early Alzheimer’s Disease

Abstract

AbstractBackgroundSleep dysfunction is one of earliest symptoms of Alzheimer’s disease (AD) and can accelerate cognitive decline, increasing risk of AD. Emerging technologies, like actigraphy, allow for continuous monitoring of real‐world sleep. Our goal is to determine the predictive utility of real‐world sleep dysfunction features for indexing warning signs of incipient dementia like mild cognitive impairment (MCI).MethodSubjects were classed as cognitively neurotypical (N = 31) or impaired: MCI (n = 51) or AD (N = 7) based on NIA‐AA diagnostic criteria, as evaluated by a neurologist from clinical exam, neuropsychological, and demographic assessments. (Table 1) Each subject was tracked for 3 months with wrist‐worn actigraphy. Self‐reported sleep diaries validated real‐world sleep data. Sleep duration and quality were estimated across means and standard deviations (SD) of: Total Sleep Time (TST), Sleep Latency (SL), Sleep Efficiency (SE), Wakefulness After Sleep Onset (WASO), and Awakening Count (AC). An optimal logistic regression model, adjusting for demographics (age, gender, education), was quantified and used to rank the explanatory value of sleep covariates to discriminate impaired vs. typical aging using stepwise selection.ResultAcross 6,530 validated, nightly sleep periods (mean = 73.4 ± 17.0 sleep periods per subject), we found best discrimination of impaired vs. typical aging, ranked from best to least discrimination (Table 2), for average sleep efficiency [SE] (b = 3.58, p = 0.03), nightly wakefulness [WASO] (b = 3.42, p = 0.04), variability of daily sleep quality [SE] (b = 1.65, p = 0.001), and awakening frequency [AC] (b = ‐0.79, p = .03). TST did not discriminate impaired from typical aging.ConclusionResults align with prior literature, demonstrating that sleep quality is an optimal feature to track early warning signs of dementia from real‐world sleep. Critically, SE shows the most promise for discriminating early decline while TST did not show predictive utility in this study. Digital fingerprints of sleep behavior in the real world offer unique added value to neurobiological biomarker evidence of preclinical disease, improving early detection of AD risk to bolster interventions aimed at treating AD—‐even its earliest stages.