Self-report and actigraphy measures of sleep and domain-specific cognitive performance in older adults

Abstract

Poor sleep is associated with worse cognitive function in older adults. However, nuanced associations between sleep and cognition might be masked by the multidimensional nature of sleep which requires multiple approaches (e.g., self-report and actigraphy) to gain meaningful insight. We investigated associations of sleep with cognition and hypothesized that the most consistent association would be between self-reported sleep duration and actigraphy-measured wake after sleep onset (WASO). We utilized baseline data from the Investigating Gains in Neurocognition in an Intervention Trial of Exercise study. Cognitively unimpaired older adults (n=589, aged 65–80) completed a comprehensive cognitive assessment with generation of five domain-specific cognitive composite scores. Sleep was measured via the Pittsburgh Sleep Quality Index (PSQI) and 24-h actigraphy (GT9X Link). Greater actigraphy WASO and shorter self-reported sleep duration were associated with poorer performance in all five cognitive domains (β_{[range: WASO]} = −0.14 to −0.19, all p<0.05; β_{[range: duration]} = 0.08–0.15, all p<0.05). Shorter actigraphy sleep duration was also associated with poorer EF/attentional control (β=0.09, p=0.020) and processing speed (β=0.10, p=0.013). Actigraphy and self-reported sleep were more strongly associated with episodic memory in older (74 years) and younger (66 years) individuals, respectively. Actigraphy-derived WASO was consistently and robustly associated with cognitive performance. Additionally, our results suggest that self-reported sleep duration provides insight into sleep behaviors related to brain health (e.g., long periods of still wakefulness), beyond actigraphy-measured sleep duration. Thus, both self-report and actigraphy measures of sleep provide critical and unique information for interpreting relationships with cognitive performance.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11357-025-01665-y.

Introduction

Dementia is currently the seventh leading cause of death and a major cause of disability worldwide [1]. The prevalence of dementia is projected to double every 5–7 years [2], highlighting the need to identify health behaviors that can mitigate risk for dementia. Poor sleep is increasingly being viewed as a significant, modifiable dementia risk factor [3, 4] since it is associated with poorer cognitive performance, more rapid cognitive decline, and greater Alzheimer’s disease (AD) pathology (specifically brain beta-amyloid, Aβ) [5–8]. Although there is relatively consistent evidence for associations between sleep and cognitive function in older adults [9–12], more comprehensive assessments of sleep, cognition, and potential moderators are needed to gain a better understanding of sleep–cognition relationships. Such an understanding will inform interventions that target sleep improvement for reducing dementia risk in preventative stages.

Sleep and cognition are both multi-dimensional constructs. Yet, the existing literature on sleep–cognition associations has commonly focused on only a single dimension or assessment. For instance, many previous sleep–cognition studies reported only self-report or actigraphy-based sleep assessments [7, 13–17] or focused on a single sleep feature (e.g., sleep duration [14]). In terms of cognition, many studies have used a global cognitive screening measure (e.g., the Montreal Cognitive Assessment) [18, 19] or other brief assessments (e.g., Trails B) [20, 21] which provide limited information on the breadth or specificity of associations between sleep and cognition. These discrepancies may explain inconsistent findings in the literature. For instance, actigraphy-derived sleep measures are sensitive to periods of night-time wakefulness and restlessness, whereas these periods may be brief and go unreported in a sleep diary. However, actigraphy lacks sensitivity to long periods of still wakefulness that are common among individuals with insomnia [22]. These periods of awake time in bed as well as perceived sleep quality may be best captured with self-reported sleep [23]. Additionally, studies have demonstrated more consistent associations between self-reported sleep, compared to actigraphy-derived measures, with cognition [23, 24] and even brain Aβ [25], whereas other research has demonstrated the opposite pattern [19, 26]. Thus, utilizing actigraphy-measured and self-reported sleep measures in conjunction with each other and extensive, comprehensive cognitive assessments may yield the most complete insight into how multiple sleep dimensions support cognition [27].

Notably, a recent meta-analysis [12] indicated that studies including more representative samples and person-level moderators, such as age, are important future directions for the sleep–cognition field. For instance, greater age within older adults influences discrepancies between self-reported and device-measured sleep [28] as well as overall changes in sleep and cognition [29]. Even within older adults (> 60 years), sleep is shown to change into the 80s and beyond [29]. Examining the effects of age in a community-based sample of older adults with comprehensive assessments of sleep and cognition may explain heterogeneity in sleep–cognition associations and better inform sleep interventions tailored to improve cognition and reduce dementia risk.

This study investigated associations between sleep and cognitive performance in a well-characterized, community-based sample of older adults aged 65–80 years. We examined whether the strength of sleep–cognition associations depended on age, sleep features (duration, continuity, and quality), sleep assessment modalities (self-report and actigraphy), and cognitive domains (episodic memory, EF/attentional control, processing speed, working memory, and visuospatial performance). Based on the consistency of associations between wake after sleep onset (WASO) and other sleep continuity measures with cognition [12], we hypothesized that actigraphy-derived WASO and awakenings would be most consistently associated with cognitive performance. Similarly, based on previous studies demonstrating consistent associations between self-reported sleep and cognition [23], we hypothesized that self-reported sleep duration would consistently associate with cognitive performance. We hypothesized that the association between poor sleep and poorer cognition would be stronger with higher age since these individuals likely have declining sleep quality and greater pathology or age-related neural atrophy driving both sleep and cognitive declines.

Method

Participants

The current cross-sectional analyses used baseline data from the Investigating Gains in Neurocognition in an Intervention Trial of Exercise (IGNITE) study (NCT0287530, R01 AG053952) [30]. Participants were community-dwelling older adults from three sites (Boston, Pittsburgh, and Kansas) aged 65–80 years at baseline, free of neurological disease diagnoses, considered inactive (self-reported < 20 min moderate intensity physical activity 3 days/week), and classified as cognitively unimpaired within broad limits (Telephone Interview of Cognitive Status score > 25) [31]. Consensus adjudication was reached after a comprehensive neuropsychological assessment to exclude individuals with probable mild cognitive impairment (MCI) or dementia, but given known limitations of neuropsychological testing and the variable definitions of MCI, it is possible that some included participants were near the MCI range. Participants were recruited in proportion to the demographic representation of racial and ethnic minorities at each of the study sites. Exclusion criteria included diagnosis of psychiatric disorders or history of psychiatric illness, current cancer treatment (except for non-melanoma skin cancer), neurological conditions (e.g., Parkinson’s disease, multiple sclerosis), insulin-dependent diabetes—either type 1 or type 2, alcohol or substance use disorder within the last 5 years, current treatment for congestive heart failure, angina, or another cardiovascular event, myocardial infarction or other cardiac condition within the last year, engaged in > 20 min of moderate intensity physical activity 3 days/week (self-report), use of a walking device, or inability to complete any study-related procedures (e.g., MRI) [30]. For the current analyses, n= 59 participants were removed from the total of n= 648 enrolled in the IGNITE study due to missing actigraphy data and quality control procedures (e.g., calibration errors, protocol non-compliance, scoring artifacts; Supplementary Fig. 1).

Sleep measures

Self-reported sleep was measured using the Pittsburgh Sleep Quality Index (PSQI) [32]. The PSQI consists of 19 self-rated items evaluating sleep behaviors across the past month. For the current study, we used the following outcomes from the PSQI: sleep duration (h), sleep efficiency (%), and overall sleep quality (total PSQI score, where higher scores indicate poorer sleep quality).

Objective sleep was measured using triaxial accelerometers (ActiGraph GT9X Link; ActiGraph, Pensacola, FL, USA) worn on the non-dominant wrist. Devices were initialized at a sampling frequency of 70 Hz and asked to be worn at all times for 7 days. Raw data in the form of.gt3x files were processed using GGIR version 2.10 - 1 [33]. The configuration file is included in Supplementary Table 1. Sleep analyses were conducted using the vanHees2015 algorithm guided by the Heuristic algorithm looking at Distribution of Change in Z-Angle (HDCZA) [34, 35]. The vanHees2015 algorithm was used for sustained inactivity bout detection by defining periods where there was no change > 5 degrees in the arm angle (derived from the z-angle of the triaxial accelerometer) for at least 5 min [34]. The HDCZA algorithm was used to guide sleep period time window detection [35]. Inclusion criteria for a valid day required the sensor to be worn for at least 16 valid hours (or 2/3) of each day. Sensitivity analyses were conducted excluding any participants who did not have at least four valid days with one weekend day and three weekdays, or at least four valid nights. Quality control procedures were performed to confirm compliance with the protocol (i.e., wearing the device during sleep), device wear time, calibration errors, and any scoring artifacts (Supplementary Fig. 1).

Nightly data were extracted from part four of the GGIR output and additional sleep-specific quality control procedures were applied to identify any problematic nights that were not detected by GGIR. Namely, any nights with > 13-h or < 1-h sleep duration were removed. A difference variable was created for each participant: the maximum sleep duration across all measured nights minus the minimum sleep duration across all nights. If this difference variable was greater than the participants’ average sleep duration, i.e., their sleep variability was greater than their mean sleep duration, individuals were coded as a one. If participants were coded as a one and had sleep duration > 12 h or < 3 h, those nights were removed. After this process, if any individual had < 4 nights of valid actigraphy data, those participants were included in our sensitivity analysis. Variables of interest from the actigraphy included total sleep duration (h), WASO (h), and number of awakenings since these are commonly associated with cognitive performance in older adults [12, 14, 23] and can be reliably estimated in the absence of a sleep diary. Sleep efficiency and latency could not be calculated from the actigraphy due to a lack of sleep log data to accurately define the in-bed time.

Sleep disorder diagnoses were not formally assessed; however, participants had the opportunity to report diagnosed sleep disorders on a survey inquiring about their health history. Individuals reporting sleep apnea, insomnia, rapid eye movement sleep disorder, or restless leg syndrome were coded as one (n = 15) and otherwise coded as zero. This variable was used as a covariate in all analyses.

Cognitive assessment

Cognition was assessed at baseline using a comprehensive neuropsychological battery of verbal, paper-pencil, and computerized assessments. As previously published in this sample [36], raw scores were normalized and combined using a confirmatory factor analysis approach. The scores were loaded onto five latent domains representing episodic memory, processing speed, working memory, executive function (EF)/attentional control, and visuospatial function, with a good model fit using standard criteria [36]. The following tasks were included in each composite score: episodic memory (Brief Visuospatial Memory Test, Picture Sequencing Test, Hopkins Verbal Learning Test, Logical Memory Task, Verbal Paired Associates), processing speed (Letter Comparison Test, Digit Symbol Substitution Test, Trail Making Test, Part A), working memory (N-Back Working Memory Task, Spatial Working Memory Task, List Sorting Working Memory Task), visuospatial processing (Matrix Reasoning, Spatial Relations, Clock Draw), and EF/attentional control (Flanker Task, Stroop Task (incongruent trial), Dimensional Change Card Sort task, Trail Making Test, Part B).

Statistical analyses

All analyses were conducted using R statistical computing packages version 4.3.2. Visual inspection and skewness and kurtosis statistics were used to investigate the normality of the data. To test the strength of association between self-reported and actigraphy-measured variables, Spearman’s correlation coefficients (ρ) were evaluated. Linear regression models were used to assess associations between each sleep variable (actigraphy duration, WASO, and awakenings, PSQI duration, efficiency, and total score) and cognitive outcomes, with separate models for each sleep variable and cognitive outcome. Because of the potential U-shaped association between sleep duration and cognition [37], we also examined a non-linear (i.e., quadratic) term for sleep duration with cognitive function, as well as a categorized variable of sleep duration (< 6 h, n= 139; 6–8 h, n= 387; > 8 h, n= 63), with 6–8 h as the reference group. Finally, we wanted to examine whether self-reported and actigraphy-measured sleep variables explain unique variance in cognitive function, i.e., to test whether both measures are providing unique and important information about brain health. Thus, as an exploratory analysis we included actigraphy WASO and self-reported duration in the same linear model. We examined moderating effects of age on associations between sleep variables and cognition by adding an interaction term (age*respective sleep variable) to linear models. Simple slopes analyses were conducted to probe significant interaction terms, using the mean sample age (69.79 years), and ± 1 standard deviation (SD) from the mean (66.08 and 73.50 years). All models included age, sex, education, and chronic sleep conditions as covariates (when not included as an interaction term). These covariates were selected based on their associations with cognition and sleep [38–40], and to optimize model fit to the data. Apolipoprotein E (APOE) genotype and mood were also tested as covariates; however, their inclusion did not improve model fit, so they were excluded. The false discovery rate was applied with a threshold of 0.05 to correct for multiple comparisons for all sleep–cognition analyses [41].

Results

Descriptive data for the sample are presented in Table 1. The final sample size was n = 589 after quality control procedures were applied, as described above, and due to missing actigraphy data (Supplementary Fig. 1). Participants were aged 69.8 ± 3.7 years, were 70% female, had a mean sleep duration of approximately 7 h, and 45% were classified as “poor sleepers” based on the PSQI criteria (total score > 5; Table 1). The sample excluded due to missing actigraphy data did not vary from the included sample on age, sex, education, APOE ε4 carriership, or MoCA score, but did include a higher percentage of participants identifying as black and bi-racial compared to those included in this study.

Table 1.

Descriptive data for the IGNITE subsample with valid actigraphy data available

	Included subsample with available data (n = 589)	Excluded IGNITE sample (n = 59)	Test statistic
Age	69.8 (3.7)	70.8 (4.0)	t = − 1.78
Sex			χ2 = 1.13
Female	415 (70%)	46 (77%)
Male	174 (30%)	13 (22%)
Race			χ2 = 16.5*
Black	106 (18%)	17 (28%)
Asian	10 (1.7%)	0 (0%)
Bi-racial	12 (2%)	4 (7%)
White	455 (77%)	36 (61%)
Native Hawaiian or other Pacific Islander	2 (0.3%)	0 (0%)
Another Reported Race Identity	3 (0.5%)	1 (1.7%)
Unknown	1 (0.2%)	1 (1.7%)
Education (years)	16.3 (2.2)	16.2 (2.2)	t = 0.46
APOE ε4 carriers	161 (27%)	13 (22%)	χ2 = 0.61
MoCA score	25.8 (2.6)	25.5 (2.7)	t = 0.72
Actigraphy sleep duration (h)	6.7 (1.1)	–	–
Actigraphy WASO (h)	0.97 (0.4)	–	–
Actigraphy number of awakenings	12.9 (3.5)	–	–
PSQI sleep duration (h)	6.9 (1.2)	–	–
PSQI sleep efficiency (%)	86.9 (12.2)	–	–
Total PSQI score	5.6 (2.9)	–	–
“Poor” sleepers (total score > 5)	264 (45%)	–	–
“Good” sleepers (total score £5)	321 (55%)	–	–

Note. Excluded sample are those with missing or invalid actigraphy data

APOE apolipoprotein E, MoCA Montreal cognitive assessment, WASO wake after sleep onset, PSQI Pittsburgh Sleep Quality Index

^*p < 0.05, **p < 0.01, ***p < 0.001

Zero-order correlations between sleep variables

Spearman’s correlations evaluated the consistency between self-reported and actigraphy-measured sleep variables. Actigraphy-measured awakenings and WASO were the sleep variables most strongly correlated with each other (Table 2). As expected, all PSQI subdomains were significantly correlated with each other. The only directly comparable metric between actigraphy and self-report sleep measures was sleep duration, which demonstrated a moderate correlation between measures (ρ = 0.34; Table 2). However, actigraphy-measured WASO was also associated with PSQI sleep efficiency (ρ = − 0.09) and total PSQI score (higher scores indicate poorer sleep; ρ = 0.14; Table 2).

Table 2.

Zero-order correlations between actigraphy and self-report sleep variab

	PSQI sleep duration	PSQI sleep efficiency	PSQI total score	Actigraphy sleep duration	Actigraphy WASO
PSQI sleep efficiency	0.436***
PSQI total score	− 0.439***	− 0.483***
Actigraphy sleep duration	0.342***	0.017	− 0.173***
Actigraphy WASO	0.048	− 0.086*	0.140***	− 0.033
Actigraphy number of awakenings	0.206***	0.059	− 0.003	0.302***	0.647***

Note. Correlation calculated using Spearman’s method; boldface shows significant correlations

WASO wake after sleep onset, PSQI Pittsburgh Sleep Quality Index

^*p < 0.05, **p < 0.01, ***p £ < 0.001

Associations between sleep and cognition

Actigraphy

Greater actigraphy-measured WASO was associated with poorer performance for every cognitive domain after adjusting for covariates, with the strongest associations for visuospatial ability and working memory performance (Fig. 1; Table 3). Greater actigraphy-measured sleep duration was associated with better processing speed and EF/attentional control. Associations between number of awakenings and cognitive outcomes did not survive correction for multiple comparisons (Fig. 1; Table 3). Results remained similar in sensitivity analyses excluding those who did not have at least four valid days with one weekend day and three weekdays, or at least four valid nights.

Fig. 1.

Summary of significant associations between sleep variables and cognitive outcomes. Bubble size and color are indicative of association strength. Greater total PSQI score is indicative of poorer overall sleep quality. Covariates included age, sex, education, and sleep disorders. Abbreviations: WASO, wake after sleep onset; PSQI, Pittsburgh Sleep Quality Index. *Significant after correction for multiple comparisons

Table 3.

Linear regression of associations between sleep variables and cognition

	Episodic memory	Processing speed	Working memory	EF/attentional control	Visuospatial
Actigraphy sleep duration	0.05 (0.04)	0.10 (0.04)*†	0.06 (0.04)	0.09 (0.04)*†	0.04 (0.04)
Actigraphy WASO	− 0.18 (0.04)***†	− 0.14 (0.04)***†	− 0.19 (0.04)***†	− 0.16 (0.04)***†	− 0.19 (0.04)***†
Actigraphy awakenings	− 0.08 (0.04)*	− 0.04 (0.04)	− 0.08 (0.04)*	− 0.04 (0.04)	− 0.08 (0.04)*
PSQI duration	0.15 (0.04)***†	0.08 (0.04)*†	0.12 (0.04)**†	0.09 (0.04)*†	0.13 (0.04)**†
PSQI efficiency	0.06 (0.04)	0.08 (0.04)*	0.07 (0.04)	0.08 (0.04)*	0.08 (0.04)*
PSQI total	− 0.06 (0.04)	− 0.10 (0.04)*	− 0.07 (0.04)	− 0.08 (0.04)*	− 0.05 (0.04)

Note. Covariates include age, sex, education, and sleep disorders. Reported as standardized β (standard error). Boldface highlights significant slopes at p < 0.05

*Uncorrected p < 0.05, **uncorrected p < 0.01, ***uncorrected p < 0.001

WASO wake after sleep onset, PSQI Pittsburgh Sleep Quality Index

†Significant after correction for multiple comparisons; the false discovery rate correction was applied within each predictor variable

Pittsburgh Sleep Quality Index

Longer PSQI-assessed sleep duration was associated with better performance in every cognitive domain (Fig. 1; Table 3). Further investigation using a categorical sleep duration variable demonstrated that these associations were driven by poorer performance in short (< 6 h) versus optimal (6–8 h; reference group) sleepers (e.g., episodic memory: β = − 0.33, SE = 0.11, p = 0.005), as opposed to long (> 8 h) versus optimal sleepers (reference group; episodic memory: β = 0.18, SE = 0.12, p = 0.132). This pattern of results was consistent for all cognitive domains (Supplementary Table 2).

A quadratic model was tested for associations between both self-reported sleep duration and actigraphy measured sleep duration; however, linear models reported above indicated a better fit to the data (i.e., the quadratic term was non-significant; quadratic model not shown).

When controlling for actigraphy-measured WASO, self-reported sleep duration remained significantly associated with cognitive performance in each domain. Similarly, controlling for self-reported duration, WASO remained associated with all five cognitive variables with similar effect sizes (Supplementary Table 3).

Age as a moderator of the association between sleep and cognition

Age moderated the association between objectively measured WASO and episodic memory (β = − 0.11, SE = 0.04, p = 0.007), and number of awakenings and episodic memory (β = − 0.09, SE = 0.04, p = 0.020), such that the strength of these associations increased with greater age (Fig. 2, see Supplementary Table 4 for simple slopes analysis). Conversely, greater self-reported sleep efficiency was more strongly associated with better episodic memory as age decreased (β = − 0.10, SE = 0.04, p = 0.014; Fig. 2, Supplementary Table 4). Poorer self-reported sleep quality (higher PSQI score overall) was more strongly associated with poorer episodic memory (β = 0.09, SE = 0.04, p = 0.013) and visuospatial performance (β = 0.08, SE = 0.04, p = 0.039) as age decreased (Fig. 2; Supplementary Table 4).

Fig. 2.

Interactions between age and the respective sleep variable on cognitive outcomes. All interactions were significant at p < 0.05; only the strongest interactions are illustrated here (total PSQI*age on visuospatial performance was also significant). Greater total PSQI score is indicative of poorer overall sleep quality. Age is separated here for visualization purposes only and was treated as a continuous moderator in linear models. Abbreviations: WASO, wake after sleep onset; PSQI, Pittsburgh Sleep Quality Index

Discussion

The current study investigated associations of actigraphy-derived and self-reported sleep with domain-specific cognitive function in a large, representative, community-based sample of older adults. Consistent with our predictions, greater actigraphy-measured WASO and shorter self-reported sleep duration were associated with poorer performance in all five cognitive domains. Shorter actigraphy-measured sleep duration was associated with poorer EF/attentional control and processing speed. EF/attentional control and processing speed were the cognitive domains most consistently associated with a range of sleep measures, suggesting that sleep–cognition associations with processing speed and EF/attentional control are not sensitive to particular sleep features. However, the strongest sleep–cognition associations were found for working memory, visuospatial processing, and episodic memory, which demonstrated the largest effect sizes with WASO and sleep duration. Age influenced the association between sleep variables and episodic memory, the directionality of which was dependent on the sleep measure. Within our sample range of 65–80 years, those at the higher end of our age spectrum (70 and above) demonstrated the strongest associations between actigraphy-measured sleep and episodic memory, whereas those at the lower end (66 years) showed the strongest associations of self-reported sleep variables with episodic memory and visuospatial performance.

We extend previous work showing associations between sleep continuity measures and cognitive performance [12, 19] by demonstrating that actigraphy-measured WASO is most consistently and most strongly associated with multiple cognitive domains, particularly executive functions and memory performance. This contrasts with previous studies commonly utilizing a general cognitive measure (e.g., the Montreal Cognitive Assessment) [19] with limited sensitivity to domain-specific changes. Importantly, because actigraphy measures WASO based on movement, this measure may also detect periods of restlessness at night which are closely associated with cognitive functioning [12]. The consistency across cognitive domains indicates that WASO may be detecting a systematic pattern of how sleep fragmentation influences cognition, which cannot be easily compensated for with greater sleep duration. For example, greater WASO may disrupt progression through sleep stages that in combination support cognitive function. Additionally, fragmented sleep indicated by higher WASO may provide limited opportunity for clearance of beta-amyloid (Aβ; a pathological hallmark of AD) since clearance may be most efficient in deep sleep stages [42, 43] and in turn lead to preclinical AD-related cognitive deficits. Alternatively, greater WASO may be a consequence of the same pathological processes that disrupt both sleep and cognition [44]. Overall, our findings suggest that targeting WASO specifically, relative to sleep duration or even number of awakenings, may be important for preserving cognition and reducing dementia risk.

Self-reported sleep duration was more consistently associated with cognitive function compared to actigraphy-derived sleep duration. One previous study has demonstrated a similar trend, wherein self-reported sleep duration was more consistently associated with cognitive function [23], whereas others have demonstrated the opposite trend [26]. It is well recognized that objective and subjective sleep measures do not correlate well in older adults (as in the current study, ρ= 0.34) [27]. However, actigraphy, and accelerometers generally, may lack sensitivity to differentiate sleep and long periods of still wakefulness, particularly before sleep onset, since the measure is based on accelerations and movement [22]. In self-identified short sleepers, who drove the current self-reported findings, actigraphy may struggle to detect short sleep durations if individuals are lying still in bed for long periods trying to initiate sleep. Furthermore, when we included self-reported sleep duration and WASO in the same model, both sleep features remained significantly associated with cognitive performance across all domains, indicating that each variable is providing important and unique information about sleep–cognition associations. Thus, these results further support the utility of using both self-report and actigraphy measures of sleep in future studies in cognitively unimpaired individuals.

We found that age moderated the association between sleep and cognition in opposite directions for self-report and actigraphy-derived sleep measures. Specifically, consistent with our hypotheses, actigraphy measures (WASO and awakenings) were more strongly associated with episodic memory in those at the older end (≥ 70 years) of our included age spectrum (65–80 years). However, self-reported sleep measures were more strongly associated with episodic memory in those at the lower end of our age spectrum. These findings are inconsistent with a previous meta-analysis, which found no moderating influence of age (in individuals aged 50+ years) on the association between sleep macrostructure and cognition [12], and one study showing the discrepancy between self-reported and objectively measured sleep did not vary by age in older adults (aged 55+) [27]. The lack of association between self-reported sleep and episodic memory may be because older adults underestimate WASO, a pattern which intensifies with increasing age [28], likely due to modified sleep perception based on objective change with age (e.g., adjusted expectations of good quality sleep) [45]. Although further research is required to investigate possible explanations underlying this moderating effect of age, the current results indicate that in respect to cognition, self-report sleep measures may be less useful for assessing sleep–cognition relationships in older adults ≥ 70 years.

The cross-sectional nature of our study limits our ability to determine the direction of association between sleep and cognition. It is unclear whether specific sleep factors demonstrated here, i.e., WASO, awakenings, and duration, may contribute to increased risk for cognitive decline, or alternatively, whether these aspects of sleep are influenced by the same underlying neurobiological change causing declines in cognitive function. Indeed, sleep and AD pathology share a bi-directional relationship, whereby neurodegeneration in brain areas responsible for controlling the sleep–wake cycle may contribute to sleep disturbances, and sleep disturbances may also contribute to neurodegeneration [29, 46]. Further longitudinal and intervention studies are required to determine the directionality of these relationships. Participants were not specifically asked about the presence of clinical sleep disorders including insomnia and obstructive sleep apnea. We did aim to account for these conditions via our medical history questionnaire; however, the low reporting prevalence of these disorders in our sample (2.5%) likely means there were undiagnosed cases of such disorders, particularly sleep apnea. We did not consider other lifestyle factors which may influence sleep and cognition (e.g., body composition, dietary habits), nor did we exclude for some chronic conditions which may influence cognition (e.g., autoimmune or respiratory diseases).

Although many studies have investigated associations between sleep and cognition, few have demonstrated such consistent associations as the current study. Importantly, we utilized both self-report and actigraphy sleep measures which provide different, but important, insights into sleep [27]. Our sample, although cognitively unimpaired, included participants with a range of cognitive performance, increasing sensitivity to detect associations between sleep and cognition. Our sample was also racially representative of the population at each of the study sites, improving diversity and generalizability of our results [12]; however, generalizability may still be limited to community-dwelling older adults without major disability or severe health conditions. Finally, we utilized a comprehensive battery of cognitive tests and a factor analytic approach to generate composite scores to limit measurement error from individual tests [36].

In summary, both actigraphy-measured and self-reported sleep were consistently and robustly associated with cognitive function across a range of cognitive domains. Actigraphy-measured WASO and self-reported sleep duration were the two sleep dimensions most consistently and most strongly associated with cognitive performance. For the cognitive domains of episodic memory and visuospatial performance, these associations depended on age. These results support the importance of distinguishing sleep features, assessment modalities, cognitive domains, and age in characterizing the sleep–cognition association in older adults and highlight the importance of sleep in maintaining cognitive health in later life.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

This study was funded by the National Institutes of Health (R01 AG053952) awarded to KIE, JMB, AFK, EM, and to KIE (R35 AG072307). The study also received support from the Clinical and Translational Science Institute at the University of Pittsburgh (UL1-TR- 001857) and Kansas University infrastructure grants (P30 AG072973 and UL1 TR002366).

Data Availability

Data are available upon reasonable request to the executive committee of the IGNITE Trial.

Declarations

Competing interests

The authors declare no competing interests.