Associations Between Objectively Measured Sleep and Cognition: Main Effects and Interactions With Race in Adults Aged ≥50 Years

Abstract

Background

This study examined associations of actigraphy-estimated sleep parameters with concurrent and future cognitive performance in adults aged ≥ 50 years and explored interactions with race.

Methods

Participants were 435 cognitively normal adults in the Baltimore Longitudinal Study of Aging who completed wrist actigraphy at baseline (mean = 6.6 nights) and underwent longitudinal testing of memory, attention, executive function, language, and visuospatial ability. On average, participants with follow-up data were followed for 3.1 years. Primary predictors were baseline mean total sleep time, sleep onset latency, sleep efficiency (SE), and wake after sleep onset (WASO). Fully adjusted linear mixed-effects models included demographics, baseline health-related characteristics, smoking status, sleep medication use, APOE e4 carrier status, and interactions of each covariate with time.

Results

In adjusted models, higher SE (per 10%; B = 0.11, p = .012) and lower WASO (per 30 minutes; B = −0.12, p = .007) were associated with better memory cross-sectionally. In contrast, higher SE was associated with greater visuospatial ability decline longitudinally (B = −0.02, p = .004). Greater WASO was associated with poorer visuospatial ability cross-sectionally (B = −0.09, p = .019) but slower declines in visuospatial abilities longitudinally (B = 0.02, p = .002). Several sleep-cognition cross-sectional and longitudinal associations were stronger in, or limited to, Black participants (compared to White participants).

Conclusions

This study suggests cross-sectional sleep-cognition associations differ across distinct objective sleep parameters and cognitive domains. This study also provides preliminary evidence for racial differences across some sleep-cognition relationships. Unexpected directions of associations between baseline sleep and cognitive performance over time may be attributable to the significant proportion of participants without follow-up data and require further investigation.

From 2019 to 2050, the worldwide prevalence of dementia is expected to increase by 117% (1), with consequences including significant health care costs and mortality (2–4). Currently, no cure for dementia exists; therefore, it is essential to determine populations at higher risk for dementia and identify factors that could be preventative or slow its onset (5). Sleep disturbances commonly occur in older adults (6), and studies have found cross-sectional associations between self-reported disturbed sleep (ie, poorer quality, shorter and longer sleep duration) and poorer cognitive performance in older adults (7–9). Prospective research has also found that self-reported shorter sleep, longer sleep, and poorer sleep quality among middle-aged adults were linked with poorer cognitive function at 22-year follow-up (on average) (10).

Although numerous studies have provided evidence of a sleep-cognition link in older populations, most research to date has used self-report (ie, subjective) sleep measures. Given research has reported discrepancies between subjective and objective sleep characteristics in older adults (11,12), additional studies using objective sleep measures are needed. Whereas cohort studies have demonstrated a relationship of actigraphically measured poor sleep with worse cognitive performance, most are limited by a cross-sectional design (13–16), and assessment of 1 or 2 cognitive domains and/or global cognitive function (17–19).

A recent review found poor sleep quality (eg, short sleep duration, lower sleep efficiency) to be more common among Black adults than among non-Hispanic White adults, and racial differences have been observed prior to older adulthood (20). There is also evidence that dementia occurs more frequently in Black adults than non-Hispanic White adults (21). However, to date, little to no research has evaluated whether associations of sleep and cognition differ by race or whether racial and ethnic differences in sleep may partially explain or account for race-related differences in dementia.

Taken together, there are important gaps in the literature concerning the link between sleep and cognitive performance. More specifically, there is a need to better understand these associations using objective measures and longitudinal data. In light of research suggesting poor sleep may be a marker of dementia risk (22), improving our understanding of these relationships across multiple cognitive domains may have implications for the proper and more timely detection of cognitive impairment among older adults. Therefore, to clarify these relationships, we investigated associations between baseline actigraphically ascertained sleep parameters and performance and cognitive decline on neuropsychological tests across 5 cognitive domains using multiple cognitive measures in adults aged ≥ 50 years. As a secondary objective, we explored whether observed associations between sleep and cognition differed between Black and White participants.

Method

Data Source

The Baltimore Longitudinal Study of Aging (BLSA) began in 1958 and is an ongoing continuous enrollment observational study of physical and cognitive changes that occur with aging (23,24). The BLSA enrolls participants who are aged ≥ 20 years and healthy. At enrollment, BLSA participants were free of dementia, any major illnesses, functional limitations, or medical conditions that inhibit daily functioning. Once enrolled, participants are followed for life regardless of the development of diseases and chronic conditions. During baseline and follow-up visits, participants complete a battery of neuropsychological tests. In 2012, the BLSA began collecting objective sleep data using wrist actigraphy. In the present study, we analyzed data from BLSA participants aged ≥ 50 years with concurrent actigraphy and cognitive data from at least one visit. We used the earliest visit with both actigraphy and cognitive data as the study “baseline,” and analyzed all concurrent and subsequent cognitive performance data available. All participants provided informed consent. BLSA protocols are approved by institutional review boards of the Intramural Research Program of the National Institutes of Health and Johns Hopkins University.

Participants

In total, 449 participants aged ≥ 50 years had valid actigraphy data and ≥ 1 cognitive assessment at the current study baseline. We excluded 14 participants who, at our study baseline visit, received an adjudicated diagnosis of dementia, mild cognitive impairment (MCI), or cognitive impairment not meeting MCI criteria, leaving a final analytic sample of 435 participants (Figure 1). The process for adjudication of cognitive status has been described elsewhere (25). Briefly, consensus case conferences were performed to evaluate clinical and neuropsychological data among participants who scored ≥ 0.5 on the combined Clinical Dementia Rating scale (26). Participants with > 3 errors on the Blessed Information-Memory-Concentration Test were also evaluated at consensus case conferences (27). Established criteria for MCI (28), dementia (29), and Alzheimer’s disease (30), were used to determine participants’ adjudicated diagnoses.

Figure 1.

Participant flowchart.

Wrist Actigraphy

Participants were instructed to wear a wrist actigraph (Actiwatch-2, Philips-Respironics, Bend, OR) on their nondominant wrist for seven 24-hour periods. On wear days, participants were asked to complete a sleep diary covering the time they got into bed with the intention of sleeping (ie, “lights out”), and the time they got out of bed in the morning to start their day. They were also asked to press an event-marker button on the watch at these times. Diaries also queried naptimes and any watch removal. Diary and event-marker data were used to identify in-bed intervals to which a validated algorithm was applied to differentiate sleep from wake (31), and data were exported using Actiware 6.0.9 software (Philips-Respironics). The mean of multiple sleep parameters across nights with valid actigraphy data was computed for each participant: total sleep time (TST; total minutes spent asleep while in bed), sleep onset latency (SOL; minutes from “lights out” until sleep onset), wake after sleep onset (WASO; minutes spent awake after sleep onset); and sleep efficiency (SE; % of time in bed spent asleep).

Cognitive Assessment

Trained psychometric testers administered a comprehensive battery of neuropsychological tests to participants at baseline and follow-up visits. Verbal learning and memory were assessed using the California Verbal Learning Test immediate free recall and long delay free recall scores (32). Immediate free recall scores reflect the total number of words recalled across 5 learning trials and can range from 0 to 80. Long delayed recall scores reflect the total number of words recalled after an approximately 20-minute delay; scores can range from 0 to 16. The Trail Making Test, Part A and the Wechsler Adult Intelligence Scale-Revised (WAIS-R) digit span forwards tests were used to measure attention (33,34). Executive function was measured using the Trail Making Test, Part B and the WAIS-R digit span backward tests (33,34). We used verbal fluency—both letter (ie, F, A, S in 3 separate trials) and category (ie, fruits, vegetables, animals in 3 separate trials) fluency—to quantify fluent language production (ie, language) (35,36). Participants had 60 seconds to complete each trial, and the scores for letter and category fluency tests, respectively, were the average number of words named across the 3 trials. Finally, visuospatial ability was assessed with Clock Drawing Tests (showing 2 different clock times) (37) and a modified version (38) of the Educational Testing Service Card Rotations test (39). A z-score for each cognitive test was calculated using the baseline sample mean and standard deviation which were averaged within each cognitive domain to produce a composite score for that domain. For each cognitive domain, higher z-scores indicate better cognitive performance. Because the Trail Making Test, Parts A and B scores were non-normally distributed, we log-transformed them prior to calculating their corresponding z-scores and inverted their signs so that higher scores reflect better performance.

Other Measures

Participants reported demographic data, including age, sex, race and ethnicity, and years of education. Race and ethnicity were categorized as White, Black, and Other (ie, Chinese, Filipino, Japanese, Other Asian, Other non-White). Participants were asked how often they took “sleeping pills or other medications” to help them sleep during the past month. We dichotomized responses as no (“never”) or yes (“<1/wk”, “1–2/wk”, “3–4/wk”, “5+/wk”). During an examiner-administered interview, participants were asked whether they had hypertension, diabetes, congestive heart failure, stroke, peripheral artery disease, angina, and heart attack. Those who reported any of these conditions or behaviors were categorized as having cardiovascular disease or being at-risk (ie, “cardiovascular disease risk”). We created a dichotomous smoking status variable to reflect whether participants reported that they “never smoked” or “quit smoking at least 10 years prior” versus were a “current smoker” or “quit within the past 10 years.” Participants’ height and weight data were measured at study visits and used to calculate body mass index (BMI; kg/m²). Participants also completed the 20-item Center for Epidemiological Studies Depression scale (CES-D) (40), a validated measure of depressive symptoms. We excluded a sleep-related item (ie, “My sleep was restless”) when calculating CES-D scores. APOE e4 carrier status was determined using DNA extracted from blood samples (41,42).

Statistical Analysis

We calculated descriptive statistics for baseline participant characteristics for the entire sample and stratified by race. We used independent samples t-test and chi-squared or Fisher’s exact test to evaluate potential racial differences in participants’ baseline characteristics, as well as differences between those with baseline data only versus those with follow-up data. Because only 29 participants identified as “Other” racial or ethnic groups, they were excluded from race-stratified analyses.

To evaluate the associations of baseline actigraphic sleep parameters with cross-sectional and longitudinal cognitive performance, we fit minimally and fully adjusted linear mixed-effects models for each primary predictor and outcome. The primary predictors were baseline TST, SOL, SE, and WASO. The primary outcomes were composite scores of memory, attention, executive function, language, and visuospatial ability. Each set of minimally adjusted linear mixed-effects models included covariates (ie, baseline age, sex, race, and years of education), time, interactions between time and each of the covariates (eg, baseline age × time; sex × time), the primary predictor, and an interaction term between the primary predictor and time. In each model, the main effect estimated the cross-sectional association of baseline sleep parameters with baseline cognitive performance, whereas the primary predictor interaction with the time term (eg, TST × time) assessed the association of the baseline sleep parameters with change in cognitive performance over time. Interactions of time with each of the covariates were included to account for the potential longitudinal impact of the covariates on change in cognition.

The fully adjusted models included all covariates in the minimally adjusted models, as well as baseline body mass index (BMI), baseline CES-D score (minus the sleep item), baseline sleep medication use (yes vs no), baseline cardiovascular disease risk status (yes vs no), baseline smoking status, APOE e4 carrier status (carrier vs non-carrier), and interactions between time and each of these covariates.

In separate analyses, we explored race as a moderator by adding 2-way (ie, primary predictor × race) and 3-way interaction terms (ie, primary predictor × race × time) to the fully adjusted models. As noted above, there were relatively few individuals who identified as “Other” racial or ethnic groups; therefore, we excluded these individuals from the moderation analyses.

To examine whether incident cognitive impairment influenced observed associations, we conducted sensitivity analyses that excluded data points at or after which participants developed MCI, dementia, or cognitive impairment not meeting MCI criteria post-baseline. Previous data points, during which these participants were cognitively normal, were preserved in the sensitivity analyses.

For the exploratory analyses of race interactions, we report results for interaction terms with a p < .10. For all other analyses, a p < .05 was used to determine statistical significance. The random effects for all linear mixed-effects models included intercept and time with unstructured covariance. All the continuous primary predictors and covariates (except time) were mean centered. All analyses were performed in R Version 4.0.4 (R Core Team, 2020).

Results

Table 1 describes participant characteristics for the entire sample, as well as stratified by race. Overall, participants’ mean age at baseline was 72.7 (standard deviation [SD]: 10.2) years, 229 (52.6%) were female, and the mean educational attainment was 17.8 (SD: 2.4) years. At baseline, the mean BMI was 27.0 (SD: 4.4), the average CES-D score (excluding the sleep item) was 3.8 (SD: 4.4), 63 (14.7%) participants used sleep medications, 237 (54.0%) had, or were at risk, for cardiovascular disease, and 420 (96.6%) never smoked or quit smoking > 10 years ago. Additionally, 310 participants (71.3%) were White, and 104 (23.9%) were APOE e4 carriers. Of the 435 participants, 131 (30.1%) only had baseline data. Participants with follow-up data had an average of 3.2 (SD: 1.3) follow-up visits (ie, post-baseline; range: 2–7); additionally, their follow-up time averaged 3.1 (SD: 1.6) years (range: 1–7). Baseline average TST and SOL were 402.45 minutes (SD: 62.09) and 12.59 minutes (SD: 12.76), respectively. Average SE and WASO were 83.69% (SD: 7.28) and 47.78 minutes (SD: 22.28), respectively, at baseline.

Table 1.

Baseline Participant Characteristics, Mean ± Standard Deviation or n (%)

Participant Characteristics	Total Sample 100% (N = 435)	Race
		White 71.3% (n = 310)	Black 22.1% (n = 96)	Chi-squared	t-value	p-value
Baseline age, in years, M ± SD	72.7 ± 10.2	74.1 ± 9.9	69.2  ± 9.7	—	4.33	.0003
Sex, n (%)				9.39	—	.002
Male	206 (47.4%)	162 (52.3%)	33 (34.4%)	—	—	—
Female	229 (52.6%)	148 (47.7%)	63 (65.6%)	—	—	—
Educational attainment (in years), M ± SD	17.8 ± 2.4	17.8 ± 2.4	17.3 ± 2.4	—	—	—
BMI, M ± SD	27.0 ± 4.4	26.6 ± 4.2	28.7 ± 4.7	—	−3.81	.0002
CES-D Score*, M ± SD	3.8 ± 4.4	3.7 ± 4.3	4.5 ± 4.8	—	−1.54	.127
Use of sleep medication†, n (%)				5.69	—	.017
No	366 (85.3%)	252 (82.6%)	88 (92.6%)	—	—	—
Yes	63 (14.7%)	53 (17.4%)	7 (7.4%)	—	—	—
Cardiovascular disease risk, n (%)				24.72	—	.000007
No	198 (46%)	164 (52.9%)	23 (24.0%)	—	—	—
Yes	237 (54%)	146 (47.1%)	73 (76.0%)	—	—	—
Smoking status†, n (%)				—	—	.708‡
Never smoked or quit > 10 y ago	420 (96.6%)	298 (97.7%)	93 (96.9%)	—	—	—
Current smoker or quit < 10 y ago	10 (3.4%)	7 (2.3%)	3 (3.1%)	—	—	—
APOE carrier status†, n (%)				—	—	—
APOE e4 -	301 (69.9%)	—	—	—	—	—
APOE e4 +	104 (23.9%)	—	—	—	—
Follow-up time among participants with > 1 visit (in years), M ± SD	3.13  ± 1.58	3.31 ± 1.28	2.72 ± 1.11	1.71	—	.088
Number of visits, n (%)				—	—	—
1	131 (30.1%)	98 (31.6%)	27 (28.1%)	—	—	—
2	124 (28.5%)	72 (23.2%)	42 (43.8%)	—	—	—
3	73 (16.8%)	55 (17.7%)	11 (11.5%)	—	—	—
4	69 (15.9%)	53 (17.1%)	13 (13.5%)	—	—	—
5	19 (4.4%)	18 (5.8%)	0 (0%)	—	—	—
6	10 (2.3%)	8 (2.6%)	2 (2.1%)	—	—	—
7	9 (2.1%)	6 (1.9%)	1 (1.0%)	—	—	—
Sleep characteristics, M ± SD
TST (minutes)	402.45 ± 62.09	410.66 ± 58.54	383.32 ± 66.62	—	3.61	.0004
SOL (minutes)	12.59 ± 12.76	11.99 ± 13.21	14.10 ± 11.92	—	−1.47	.142
SE (%)	83.69 ± 7.28	84.4 ± 7.32	81.93 ± 7.31	—	2.90	.004
WASO (minutes)	47.78 ± 22.28	47.92 ± 22.70	46.31 ± 21.01	—	0.644	.521

Notes: BMI = body mass index; M = mean; SD = standard deviation; SE = sleep efficiency; SOL = sleep onset latency; TST = total sleep time; WASO = wake after sleep onset. Missing data: Years of education = 3; Depressive symptoms (minus the sleep item) = 3; Use of sleep medication = 6; Smoking status = 5; APOE e4 carrier status = 30.

*CES-D = Center for Epidemiologic Studies Depression scale score without the sleep item.

^†Percentages based on participants with complete data.

^‡Fisher’s test was used to determine differences in smoking status by race.

Compared to those with follow-up data, participants with baseline data only were younger (p = .00001), had a lower proportion of cardiovascular disease risk (p = .009), and had a lower average WASO (p = .001) at baseline (Supplementary Table S1).

Racial differences were also observed across several baseline participant characteristics (Table 1). Compared to Black adults, White adults were significantly older (p = .0003) and had lower BMI (p = .0002). In addition, a greater proportion of White adults identified as male (p = .002) and used sleep medications (p = .017) than Black adults. A greater proportion of Black adults had cardiovascular disease risk (p = .000007) than White adults. White adults also had significantly longer average TST (p = .0004) and higher SE (p = .004) relative to Black adults.

Cross-sectional Associations

Table 2 depicts the minimally adjusted associations of baseline sleep characteristics with cognitive performance across 5 domains, adjusting for age, sex, race, and years of education. Aligned with findings observed in minimally adjusted models, after further adjustment for baseline BMI, cardiovascular disease risk, smoking status, sleep medication use, depressive symptoms, and APOE e4 status, higher SE (per 10%) was cross-sectionally associated with better verbal memory (fully adjusted model, B = 0.11, 95% confidence interval [CI] = 0.02, 0.20; Table 3), whereas higher WASO (per 30 minutes) was associated with poorer memory (fully adjusted model, B = −0.12, 95% CI = −0.21, −0.03). In both minimally and fully adjusted models, greater WASO was cross sectionally associated with lower visuospatial performance (fully adjusted model B = −0.09, 95% CI = −0.16, −0.01; Table 3). There were no significant cross-sectional associations of TST or SOL with cognitive performance, and no sleep parameters were cross-sectionally associated with attention, executive function, or language measures cross-sectionally.

Table 2.

Minimally Adjusted Cross-sectional and Longitudinal Associations of Baseline Sleep Parameters With Cognitive Performance, B (95% CI)

	Memory		Attention		Executive Function		Language		Visuospatial Ability
	Cross-sectional	Longitudinal	Cross-sectional	Longitudinal	Cross-sectional	Longitudinal	Cross-sectional	Longitudinal	Cross-sectional	Longitudinal
TST (per 30 min)	0.03 (−0.06, 0.12)	0.01 (−0.01, 0.03)	0.02 (−0.05, 0.08)	0.003 (−0.01, 0.02)	0.01 (−0.06, 0.09)	0.01 (−0.01, 0.02)	−0.03 (−0.10, 0.05)	0.002 (−0.01, 0.01)	0.02 (−0.05, 0.09)	−0.01 (−0.03, 0.01)
SOL	−0.05 (−0.12, 0.03)	0.01 (−0.02, 0.03)	−0.03 (−0.09, 0.03)	−0.02 (−0.04, 0.0005)	−0.05 (−0.11, 0.02)	−0.01 (−0.03, 0.01)	−0.04 (−0.11, 0.02)	0.00001 (−0.02, 0.02)	−0.03 (−0.09, 0.04)	0.004 (−0.01, 0.02)
SE (per 10%)	0.10 * (0.02, 0.19)	−0.004 (−0.02, 0.01)	0.05 (−0.02, 0.11)	0.001 (−0.01, 0.02)	0.07 (−0.003, 0.14)	0.003 (−0.01, 0.02)	0.02 (−0.05, 0.09)	−0.001 (−0.01, 0.01)	0.07 (−0.01, 0.14)	−0.02** (−0.04, −0.01)
WASO (per 30 min)	−0.11* (−0.19, −0.02)	0.005 (−0.01, 0.02)	−0.05 (−0.11,0.02)	0.01 (−0.01, 0.02)	−0.07 (−0.14, 0.003)	0.004 (−0.01, 0.02)	0.002 (−0.07, 0.08)	0.003 (−0.01, 0.01)	−0.08* (−0.16, −0.01)	0.02 *** (0.01, 0.04)

Notes: SE = sleep efficiency; SOL = sleep onset latency; TST = total sleep time; WASO = wake after sleep onset. Cross-sectional = sleep parameter main effect. Longitudinal = time × sleep parameter interaction. Models adjusted for baseline age, sex, race, years of education, baseline age × time, sex × time, race × time, and years of education × time. Sample sizes by outcomes: Memory models n = 425; Visuospatial Ability and Language models = 430; Executive Function and Attention models n = 431.

*p < .05;

**p < .01;

***p < .001.

Table 3.

Fully Adjusted Cross-sectional and Longitudinal Associations of Baseline Sleep Parameters With Cognitive Domains, B (95% CI)

	Memory		Attention		Executive Function		Language		Visuospatial Ability
	Cross-sectional	Longitudinal	Cross-sectional	Longitudinal	Cross-sectional	Longitudinal	Cross-sectional	Longitudinal	Cross-sectional	Longitudinal
TST (per 30 min)	0.03 (−0.06, 0.12)	0.01 (−0.004, 0.03)	0.01 (−0.06, 0.08)	0.003 (−0.01, 0.02)	0.02 (−0.06, 0.09)	0.003 (−0.01, 0.02)	−0.04 (−0.12, 0.04)	0.01 (−0.01, 0.02)	0.02 (−0.06, 0.10)	−0.01 (−0.03, 0.01)
SOL	−0.05 (−0.13, 0.03)	0.01 −0.01, 0.03)	−0.02 (−0.08, 0.04)	−0.02 (−0.04, 0.003)	−0.03 (−0.10, 0.03)	−0.01 (−0.03, 0.01)	−0.03 (−0.10, 0.03)	−0.002 (−0.02, 0.01)	−0.02 (−0.09, 0.04)	0.004 (−0.02, 0.02)
SE (per 10%)	0.11 * (0.02, 0.20)	−0.01 (−0.02, 0.01)	0.03 (−0.04, 0.10)	0.001 (−0.01, 0.02)	0.06 (−0.01, 0.14)	−0.0001 (−0.02, 0.02)	0.01 (−0.07, 0.08)	0.003 (−0.01, 0.02)	0.06 (−0.01, 0.14)	−0.02** (−0.04, −0.01)
WASO (per 30 min)	−0.12** (−0.21, −0.03)	0.01 (−0.01, 0.02)	−0.03 (−0.10, 0.04)	0.01 (−0.01, 0.02)	−0.07 (−0.14, 0.01)	0.01 (−0.01, 0.02)	0.01 (−0.07, 0.08)	0.002 (−0.01, 0.01)	−0.09* (−0.16, −0.01)	0.02 ** (0.01, 0.04)

Notes: SE = sleep efficiency; SOL = sleep onset latency; TST = total sleep time; WASO = wake after sleep onset. Cross-sectional = sleep parameter main effect. Longitudinal = time × sleep parameter interaction. Models adjusted for baseline age, sex, race, years of education, BMI, cardiovascular disease risk, smoking status, sleep medication use, APOE e4 status, depressive symptoms (minus the sleep item), baseline age × time, sex × time, race × time, years of education × time, BMI × time, cardiovascular disease risk × time, smoking status × time, sleep medication use × time, APOE e4 status × time, and depressive symptoms (minus the sleep item) × time. Sample sizes by outcome: Memory models n = 394; Executive Function and Attention models n = 400, Language and Visuospatial Ability Models = 399.

*p < .05;

**p < .01.

Longitudinal Associations

Minimally adjusted longitudinal associations of baseline sleep characteristics with change in performance across 5 cognitive domains are displayed in Table 2. In minimally and fully adjusted models, higher baseline SE and lower WASO were unexpectedly associated with greater declines in visuospatial ability longitudinally (fully adjusted model SE [per 10%] × time B = −0.02, 95% CI = −0.04, −0.01; fully adjusted model WASO [per 30 minutes] × time B = 0.02, 95% CI = 0.01, 0.04; Table 3). No other significant associations of sleep parameters with change in cognitive performance over time were observed.

Exploratory Analysis

Race × sleep cross-sectional interactions

Overall, 96 (22.1%) participants self-identified as Black. Table 4 describes significant and near-significant (p < .10) race by sleep interactions. In fully adjusted models, when memory was the outcome, there was an interaction between race and WASO (interaction term p = .039), such that there was a statistically significant cross-sectional association of greater WASO with poorer memory performance among Black participants but not White participants. Race also moderated the cross-sectional association of SOL with visuospatial skills (interaction term p = .077). Greater SOL was cross-sectionally linked with lower visuospatial ability among Black participants but not among White participants.

Table 4.

Cross-sectional Associations of Sleep With Cognitive Performance by Race, B (95% CI)

	White	Black	Interaction Term p-value
	n = 310	n = 96
Memory
WASO (per 30 min)	−0.06 (−0.16, 0.04)	−0.29** (−0.49, −0.10)	.039
Visuospatial Ability
SOL	−0.01 (−0.08, 0.06)	−0.17* (−0.32, −0.01)	.077

Notes: SOL = sleep onset latency; WASO = wake after sleep onset. Cross-sectional = sleep parameter main effect.

Models adjusted for baseline age, sex, race, years of education, BMI, cardiovascular disease risk, smoking status, sleep medication use, APOE e4 status, depressive symptoms minus the sleep item, baseline age × time, sex × time, race × time, years of education × time, BMI × time, cardiovascular disease risk × time, smoking status × time, sleep medication use × time, APOE e4 status × time, depressive symptoms (minus the sleep item) × time, and sleep variable × time.

*p < .05;

**p < .01.

Race × sleep longitudinal interactions

As shown in Table 5, there was a significant interaction of race and WASO on change in memory over time (interaction term p = .046). Among Black participants, but not White participants, greater WASO was significantly associated with slower declines in memory.

Table 5.

Longitudinal Associations of Sleep With Cognitive Performance by Race, B (95% CI)

	White × Time	Black × Time	Interaction Term p-value
	n = 310	n = 96
Memory
WASO (per 30 min)	0.0002 (−0.02, 0.02)	0.05 * (0.01, 0.10)	.046
Language
TST (per 30 min)	−0.0005 (−0.02, 0.02)	0.03 * (0.001, 0.05)	.066
WASO (per 30 min)	−0.003 (−0.02, 0.01)	0.05 * (0.01, 0.08)	.013

Notes: TST = total sleep time; WASO = wake after sleep onset. Longitudinal = time × sleep parameter interaction.

Models adjusted for baseline age, sex, race, years of education, BMI, cardiovascular disease risk, smoking status, sleep medication use, APOE e4 status, depressive symptoms minus the sleep item, baseline age × time, sex × time, race × time, years of education × time, BMI × time, cardiovascular disease risk × time, smoking status × time, sleep medication use × time, APOE e4 status × time, depressive symptoms (minus the sleep item) × time, and sleep variable × time.

*p < .05.

Significant/near-significant interactions of race with TST (interaction term p = .066) and WASO (interaction term p = .013) were observed on longitudinal change in language. Greater WASO and TST were significantly associated with slower declines in language ability among Black participants but not White participants.

Sensitivity Analysis

Overall, 29 participants (6.7%) developed MCI, dementia, or cognitive impairment not meeting MCI criteria after the study baseline. After excluding time points at and after the development of cognitive impairment (number of observations excluded = 31), in fully adjusted models, cross-sectional associations of SE and WASO with memory remained statistically significant, as expected (Supplementary Table S2). Additionally, each 30-minute increase in WASO was still cross-sectionally associated with poorer visuospatial ability. In terms of longitudinal associations, greater SE and lower WASO were still associated with a greater decline in visuospatial ability over time. No other significant associations of sleep parameters with cognitive performance were observed.

In race × sleep interaction analyses, all statistically significant or near-significant associations remained after the removal of observations at or after participants developed cognitive impairment in fully adjusted models (Supplementary Tables S3 and S4).

Discussion

We examined the associations of baseline actigraphic sleep parameters with baseline cognitive performance and change in cognitive performance over time across 5 domains among adults aged ≥ 50 years who were cognitively normal at baseline. After adjustment for potential demographic and health-related confounders, we found that less efficient sleep and greater time spent awake after the initial onset of sleep were cross-sectionally associated with poorer memory performance. Greater time spent awake after initial sleep onset was also cross-sectionally linked with poorer visuospatial ability. However, unexpectedly, these baseline sleep characteristics (ie, greater WASO, lower SE) were also prospectively associated with a slower decline in visuospatial ability over time. Yet, it is possible unexpected prospective findings were spurious, given 30.1% of participants did not have follow-up data, and those with and without longitudinal data significantly differed across several baseline characteristics (ie, age, cardiovascular disease risk, WASO). Meanwhile, TST and SOL were not associated with any cognitive domains in the main analyses. These findings build on previous studies linking sleep disturbance to cognitive performance, most of which used self-report rather than actigraphic sleep, presented only cross-sectional results (13–16), or studied one or 2 and/or global cognitive outcomes (17–19).

Some of our findings are aligned with those of prior cross-sectional studies conducted in cohorts of older adults, including among older women in the Study of Osteoporotic Fractures (SOF) cohort and older men in the Osteoporotic Fractures in Men (MrOS) cohort. Similar to one SOF study (13), we found that baseline WASO was linked with poorer memory performance. However, unlike that study (13), we observed no links of actigraphic sleep with language or attention in our main analyses. We also did not find cross-sectional or longitudinal associations between TST and cognitive performance in our primary analyses; while this is aligned with research that has found small or no associations (14,16,19), other studies have reported cross-sectional associations between TST and cognitive performance in this population (13,43). Meanwhile, unlike findings from one SOF study, the results of our primary analyses did not find a link of SOL with cognitive performance (16). Our results also diverge from studies that found cross-sectional or longitudinal associations between sleep characteristics and executive function (15,16,19). It is unclear why our study found associations of sleep characteristics with some but not all cognitive domains. This may be due to demographic differences across the cohorts and a smaller sample size in the present study, warranting further investigation in future research.

Differences in our findings from those of other cohorts may be due to several factors. For example, MrOS and SOF studies included men and women only, respectively. In addition, unlike numerous studies (13,15,16), we limited all analyses to participants who were cognitively normal at baseline. Our study included adults aged ≥ 50 years, but MrOS and SOF studies were performed in adults aged ≥ 65 years at enrollment (13,15,16,19). There were also differences in baseline sleep characteristics across the cohorts. More specifically, whereas our study reported a similar average TST (6.7 hours) as SOF and MrOS studies (range across studies: 6.4–7.1 hours) (13,15,16,19), those studies reported higher average WASO (range across studies: 73–78 minutes) (13,15,16,19), higher average SOL (range of SOF studies: 31–41 minutes) (13,16), and lower average SE (range across studies: 77%–80.5%) (13,15,16,19), than our study (approximately 48 minutes, 12.6 minutes, 84%, respectively). Moreover, our study used different statistical methods, which allowed us to assess cross-sectional and longitudinal associations with cognitive performance at the same time.

In exploratory analyses, we also found evidence that race modifies some sleep-cognition associations. These associations were also observed in sensitivity analyses which excluded observations of incident cognitive impairments. Research suggests a higher dementia incidence in Black older adults, compared to White older adults (21). Additionally, Black adults have been shown to have longer objectively measured SOL and WASO and shorter sleep duration than White participants; however, it is of note that Black–White differences in WASO were partially mediated by neighborhood disadvantage (44). Sleep disparities may be an important factor of disparities in other areas of health such as cardiometabolic health (45). Our cross-sectional findings support the plausibility of poor sleep as a potential contributor to racial disparities in cognitive impairment and dementia, not only through poorer sleep in Black older adults, compared to White older adults, but perhaps because poor sleep has more detrimental effects on cognition in Black older adults. Much more research is needed in the area of sleep and dementia disparities, including studies investigating whether addressing sleep disparities reduces racial disparities in dementia, and studying the extent to which racism and neighborhood characteristics drive these associations.

In sensitivity analyses that excluded cognitive data collected after onset of cognitive impairment, all cross-sectional sleep-cognition associations remained. This is unlike a previous cross-sectional SOF study in which some sleep-cognition associations were weakened or nonsignificant after removing participants with adjudicated dementia diagnoses (13). That study proposed that a shared neuropathology underlying both poor sleep and poor cognitive function may explain some observed cross-sectional associations. Further studies are needed to examine how sleep and neurological changes affect cognitive performance over time.

The current study had multiple strengths. We investigated both cross-sectional and prospective links of several actigraphic sleep parameters at baseline with performance across 5 cognitive domains. In addition, we obtained objective sleep parameters across an average of 6.6 nights and were able to adjust for numerous important confounders, including sleep medication use, depressive symptoms, smoking status, and cardiovascular disease risk.

However, the current study also had several limitations. First, participants were a nonrepresentative sample of adults aged ≥ 50 years. Results may not extend to the general population of older people. In addition, we were unable to account for sleep-disordered breathing (eg, obstructive sleep apnea), which is common among older adults (46) and is associated with a greater risk of cognitive impairment (47). Studies with polysomnography are needed to investigate how sleep-disordered breathing affects links of non-respiratory sleep parameters, such as those measured by actigraphy, with performance in distinct cognitive domains. Moreover, although sleep medication usage was included as a covariate, this study did not adjust for types of sleep medications used. On the one hand, sleep medication use is independently associated with increased dementia risk (48), yet some sleep medications may have beneficial effects on cognitive health (eg, trazadone, melatonin) (49,50). Future research in this area should account for differential associations of sleep medication types on observed associations. Additionally, among participants with follow-up data, the mean follow-up time was less than 4 years, which may have limited our ability to detect more prospective associations. Despite this limited follow-up time, selection bias may have affected our outcomes. If poor sleep caused a significant cognitive decline in some participants that affected their ability to remain enrolled in the study, other participants with poor sleep who were not as affected cognitively would be more likely to have follow-up data. This could result in misleading associations of poorer sleep with better cognitive trajectories. Finally, although this study was prospective, we cannot rule out the possibility that neurodegeneration negatively affected sleep and cognitive performance.

In conclusion, among older adults, objectively measured poor sleep may be a cross-sectional marker of poorer cognition, particularly in the domain of memory. As stated above, prospective sleep-cognition associations are less clear. Studies using objective sleep measures with larger, more representative samples and longer follow-up time are needed to better understand the interactions we observed.

Funding

This work was supported in part by the National Institute on Aging (grant numbers R01AG050507, R01AG050507-02S1), by the Intramural Research Program, National Institute on Aging, National Institutes of Health, and by Research and Development Contract HHSN-260-2004-00012C. R.J. T. was supported by the National Institute on Aging (grant number P30AG059298).

Conflict of Interest

J.T.O. is an employee of, and has been granted equity in Lyra Health, Inc. A.P.S. received payment for serving as a consultant for Merck, and received honoraria from Springer Nature Switzerland AG for guest editing special issues of Current Sleep Medicine Reports.