Sleep and cognitive aging in the eighth decade of life

Abstract

We examined associations between self-reported sleep measures and cognitive level and change (age 70-76 years) in a longitudinal, same-year-of-birth cohort study (baseline N = 1091; longitudinal N = 664). We also leveraged GWAS summary data to ascertain whether polygenic scores (PGS) of chronotype and sleep duration related to self-reported sleep, and to cognitive level and change. Shorter sleep latency was associated with significantly higher levels of visuospatial ability, processing speed, and verbal memory (β ≥ |0.184|, SE ≤ 0.075, p ≤ 0.003). Longer daytime sleep duration was significantly associated slower processing speed (β = -0.085, SE = 0.027, p = 0.001), and with steeper 6-year decline in visuospatial reasoning (β = -0.009, SE = 0.003, p = 0.008), and processing speed (β = -0.009, SE = 0.002, p < 0.001). Only longitudinal associations between longer daytime sleeping and steeper cognitive declines survived correction for important health covariates and false discovery rate (FDR). PGS of chronotype and sleep duration were nominally associated with specific self-reported sleep characteristics for most SNP thresholds (standardized β range = |0.123 to 0.082|, p range = 0.003 to 0.046), but neither PGS predicted cognitive level or change following FDR. Daytime sleep duration is a potentially important correlate of cognitive decline in visuospatial reasoning and processing speed in older age, whereas cross-sectional associations are partially confounded by important health factors. A genetic propensity toward morningness and sleep duration were weakly, but consistently, related to self-reported sleep characteristics, and did not relate to cognitive level or change.

Keywords: cognitive aging; daytime sleep; polygenic scores.

Figure 1.

A schematic latent growth curve model, in which self-reported sleep is associated with the intercept and linear slope of a latent factor of cognitive function (F) at wave j based on test scores A_j, B_j …, whose intercepts, and factor loadings (a–d) on the latent cognitive factor are constrained to equality across the three waves. Residual correlations between the same tests across waves were allowed, and manifest cognitive variables were corrected for sex and mean-centered age_j within the model (paths not shown—see Statistical Analysis). The model was centered on wave 3 (age 76) when sleep data were collected. The regressions of Sleep (predictor) on cognitive intercept (i) and slope (s) were the associations of interest.

Figure 2.

Significant associations between greater daytime sleep duration at age 76 and 6-year cognitive decline. Cognitive factor scores at each wave were extracted from the structural equation models in which both outcome and predictor variables are corrected for sex, intra-wave age variation, and time-varying health measures (HADS, BMI, hypertension, diabetes, cardiovascular history, and arthritis), and were normalized for wave 1 score to illustrate individual differences in trajectories of change. Top row: individual cognitive trajectories as modeled (continuous), and are shaded to indicate less-more (red-yellow) daytime sleep duration. Bottom row: regression lines for four equally sized groups of daytime sleep duration in the same data, for illustration purposes only.

Figure 3.

Associations between self-reported sleep characteristics and polygenic scores for chronotype (left) and sleep duration (right) at all thresholds. Red vertical dashed lines are indicative of nominal significance (p < 0.05). None survived FDR correction. Magnitude of associations (x-axis) are standardized regression coefficients, controlling for the four MDS components.