Sleep and longitudinal cognitive performance in preclinical and early symptomatic Alzheimer's disease

Abstract

Sleep monitoring may provide markers for future Alzheimer's disease; however, the relationship between sleep and cognitive function in preclinical and early symptomatic Alzheimer's disease is not well understood. Multiple studies have associated short and long sleep times with future cognitive impairment. Since sleep and the risk of Alzheimer's disease change with age, a greater understanding of how the relationship between sleep and cognition changes over time is needed. In this study, we hypothesized that longitudinal changes in cognitive function will have a non-linear relationship with total sleep time, time spent in non-REM and REM sleep, sleep efficiency and non-REM slow wave activity. To test this hypothesis, we monitored sleep-wake activity over 4-6 nights in 100 participants who underwent standardized cognitive testing longitudinally, APOE genotyping, and measurement of Alzheimer's disease biomarkers, total tau and amyloid-β42 in the CSF. To assess cognitive function, individuals completed a neuropsychological testing battery at each clinical visit that included the Free and Cued Selective Reminding test, the Logical Memory Delayed Recall assessment, the Digit Symbol Substitution test and the Mini-Mental State Examination. Performance on each of these four tests was Z-scored within the cohort and averaged to calculate a preclinical Alzheimer cognitive composite score. We estimated the effect of cross-sectional sleep parameters on longitudinal cognitive performance using generalized additive mixed effects models. Generalized additive models allow for non-parametric and non-linear model fitting and are simply generalized linear mixed effects models; however, the linear predictors are not constant values but rather a sum of spline fits. We found that longitudinal changes in cognitive function measured by the cognitive composite decreased at low and high values of total sleep time (P < 0.001), time in non-REM (P < 0.001) and REM sleep (P < 0.001), sleep efficiency (P < 0.01) and <1 Hz and 1-4.5 Hz non-REM slow wave activity (P < 0.001) even after adjusting for age, CSF total tau/amyloid-β42 ratio, APOE ε4 carrier status, years of education and sex. Cognitive function was stable over time within a middle range of total sleep time, time in non-REM and REM sleep and <1 Hz slow wave activity, suggesting that certain levels of sleep are important for maintaining cognitive function. Although longitudinal and interventional studies are needed, diagnosing and treating sleep disturbances to optimize sleep time and slow wave activity may have a stabilizing effect on cognition in preclinical or early symptomatic Alzheimer's disease.

Keywords: Alzheimer’s disease; EEG; biomarkers; memory; mild cognitive impairment.

Figure 1

Overview of data collection. Sleep monitoring was performed over 4–6 nights in all participants. CSF was collected within ≤1 year of sleep monitoring and CDR measured ≤2 years of sleep monitoring. Participants underwent annual cognitive assessments before and after sleep monitoring to generate PACC scores.

Figure 2

Distribution of longitudinal PACC scores by age. Spaghetti plots of the PACC scores are shown for each participant at the age when testing was performed. Overall, cognitive performance on the PACC was relatively stable between −1 and 1 for the majority of participants. A subset of participants who were >70 years of age at baseline showed more rapid decline in PACC performance.

Figure 3

Longitudinal PACC performance and sleep time are non-linearly related. In 100 participants, generalized additive models found that the association of longitudinal PACC performance with total sleep time, sleep efficiency, time in NREM stage 2 and stage 3 sleep, and REM sleep was non-linear after adjusting for APOE ε4-positive status, age at cognitive test, age at sleep monitoring, education, sex (male effect), CSF t-tau/amyloid-β₄₂ and CDR. For the models of total sleep time (A), sleep efficiency (C), time in NREM stage 2 and 3 sleep (E), and time in REM sleep (G), the estimated marginal effect on longitudinal PACC performance (i.e. change in PACC score over time) is shown for each of the covariates. The estimated smoothed spline function of total sleep time in the fully adjusted model shows that a total sleep time <4.5 h and >6.5 h was associated with worse PACC performance over time (B). For sleep efficiency (D), PACC performance was generally unchanged over time. Time spent in NREM stage 2 and 3 (F) and in REM sleep (H) showed inverse U-shaped relationships with longitudinal PACC performance.

Figure 4

Longitudinal PACC performance and NREM SWA are non-linearly related. In 100 participants, generalized additive models found that the association of longitudinal PACC performance with NREM SWA was non-linear after adjusting for APOE ε4-positive status, age at cognitive test, age at sleep monitoring, education, sex (male effect), CSF t-tau/amyloid-β₄₂ and CDR. For the models of 1–4.5 Hz NREM SWA (A) and <1 Hz NREM SWA (C), the estimated marginal effect on longitudinal PACC performance (i.e. change in PACC score over time) is shown for each of the covariates. The estimated smoothed spline function of ln (1–4.5 Hz NREM SWA) in the fully adjusted model shows a non-linear relationship with PACC performance generally unchanged over time (B). Ln (<1 Hz NREM SWA) showed inverse U-shaped relationships with longitudinal PACC performance (D).