Sleep disturbances, changes in sleep, and cognitive function in low-income, African Americans

Abstract

BACKGROUND:

Sleep problems are modifiable factors that may contribute to the disproportionate burden of Alzheimer’s disease and related dementias (ADRD) among African Americans (AAs).

METHODS:

This study (n=216, 78.8% female; mean age=67.7 years) examined associations between 1) the level (i.e., measured in 2018) and 2) change over time (from 2013 to 2018; n=168) in actigraphy-assessed sleep with domain-specific cognitive function and clinically adjudicated cognitive impairment (2018) in a community-dwelling, predominantly AA (96.9%) sample. Assessors administered a comprehensive cognitive battery to assess global cognitive function (3MS) and domain-specific cognitive function (attention, visuo-spatial ability, language, delayed recall, immediate recall, and executive function) in 2018. Sleep was measured in 2013 and 2018 via wrist-worn actigraphy.

RESULTS:

Higher sleep efficiency and less wakefulness after sleep onset (WASO; measured in 2018) were associated with greater attention, executive function, and visuospatial ability. Increases in sleep efficiency between 2013 and 2018 were associated with better executive function, language, immediate recall, and visuospatial ability, whereas increases in WASO (2013-2018) were associated with poorer attention, executive function, and visuospatial ability. Level or change in sleep duration were not associated with domain-specific cognitive function. Also, none of the three sleep measures were associated with clinically adjudicated cognitive impairment.

CONCLUSION:

In a predominantly African American sample of older adults, sleep efficiency and WASO measured concurrently with cognitive assessments, were associated with domain-specific cognitive outcomes. In addition, worsening of sleep efficiency and WASO were associated with poorer cognitive function. Improving sleep health may support ADRD prevention and reduce health disparities.

INTRODUCTION

African Americans (AAs) are disproportionately affected by higher rates and earlier onset of Alzheimer’s disease and related dementias (ADRD) relative to White Americans [1, 2]. Despite the disproportionate burden faced by AAs, they remain under-represented in epidemiologic studies of cognitive outcomes. Therefore, there is a critical need to identify potential modifiable factors that influence risk for ADRD among AAs.

Sleep disruption and insufficient sleep are key, modifiable factors that may potentiate risk for poor cognitive function and cognitive decline. Upwards of 50% of older adults report chronic sleep problems, and over 70% of patients in early-stage dementia have sleep disturbances [3]. Further, emerging epidemiologic research has also shown links between specific sleep disturbances, including poor sleep quality, reduced sleep efficiency, and both short and long sleep duration with increased amyloid in brain tissue, and decreased amyloid in cerebrospinal fluid (CSF) [4]. Sleep disruption or deficiency is also related to increased risk of Alzheimer’s Disease (AD) symptoms, suggesting the possibility of bidirectional associations and for improving cognitive outcomes by targeting improvements in sleep[5, 6].

Importantly, much of the extant literature (as reviewed in [7, 8]) has relied on self-reported sleep characteristics or physician diagnoses of sleep disorders, and have utilized global measures of cognitive function, primarily using the Mini Mental Status Exam (MMSE) [3, 9]. However, the MMSE is not a comprehensive cognitive evaluation and has a high misclassification rate for cognitive impairment [10]. Assessing relations between objectively measured sleep disturbances and specific cognitive domains may provide evidence towards mechanisms that underlie the associations of sleep with cognition. Moreover, there has been limited focus on specific populations of community-dwelling adults that are known to be at high-risk for both sleep disturbances and poor cognitive outcomes.

In particular, there are well-documented racial/ethnic disparities in sleep across the lifespan, with AAs having higher rates of OSA and sleep disturbances, including short and long sleep durations and poorer sleep efficiency [11-14]. Such sleep health disparities may be driven by a confluence of risk factors operating at multiple levels, including both structural and individual racism, which impact where individuals live, their access to health-promoting resources and socioeconomic status (SES), and their exposure to stress [15, 16]. However, community-dwelling AAs remain under-represented in ADRD research in general, and specifically, in research focused on the association between sleep and cognitive function.

The current study seeks to address several important gaps in the literature by utilizing data from a community-based research study of urban, AA adults (aged 50 and older) and examining the association between level and changes in objectively-measured sleep disturbances over two assessment periods (2013 and 2018) with cognitive function (assessed in 2018). We hypothesized that poorer objectively measured sleep quality and decline in objectively measured sleep over time would be related to poorer cognitive function, specifically in the domains of memory, executive function, and attention, and to greater risk of cognitive impairment.

Materials and Methods

Study population and participants

Participants were enrolled in the Pittsburgh Hill/Homewood Research on Neighborhood Change and Sleep study (PHRESH Zzz) [17], part of an ongoing series of longitudinal studies that build upon the Pittsburgh Hill/Homewood Eating, Shopping, and Health (PHRESH) study [18, 19]. Beginning in 2011, the original PHRESH study recruited a random sample of households from two low-income, predominantly African-American neighborhoods (the Hill District and Homewood) in Pittsburgh, Pennsylvania, with subsequent data collection occurring in 2013, 2014, 2016, and 2018. Actigraphy-assessed sleep data used in the present analyses were first collected in 2013 and again in 2018.

The primary goal of PHRESH was to examine how changes in the neighborhood environment affect a range of health outcomes including sleep. In 2018, a subsample (N=256) of PHRESH Zzz participants ages 50 or older were invited to participate in an ancillary study called “Think PHRESH,” which added assessments of cognitive functioning and clinical adjudication of cognitive impairment to a subsample of eligible participants in 2018. To be included in the current analyses, participants had to be included in the Think PHRESH sample and have sleep data from a prior wave. Specifically, for concurrent (i.e., both sleep and cognitive outcomes collected in 2018) analyses, the analytic sample includes 216 participants with both assessments of both sleep actigraphy and cognitive assessments in 2018; and 209 participants with clinically adjudicated cognitive impairment. Of the 216, a subset of 168 had sleep data in both 2013 and 2018, so this is the sample size for analyses which examined change in sleep measures and cognitive outcomes. The analytic sample size for analyses which examined change in sleep measures and clinically adjudicated cognitive impairment in 2018 is 163.

Study covariates (described below), including sociodemographics, measured height and weight for calculation of body mass index (BMI), measured blood pressure, and high blood sugar (HbA1c), were assessed in 2018, concurrent with cognitive outcomes. Covariates were selected a priori based on prior literature, to account for independent associations between sleep and cognitive outcomes. All participants provided informed consent. The study protocol was approved by RAND and University of Pittsburgh Institutional Review Board.

Sleep Assessment

The Actigraph GT3x+ is a wrist-worn device that has been validated to measure sleep/wake rhythms relative to both polysomnography and Actiwatch. Actigraphy outcomes for the current study included sleep duration, wakefulness after sleep onset (WASO) and sleep efficiency, averaged across all available nights of data [20, 21]. Participants with fewer than 4 nights of actigraphy data were excluded from analyses, to establish reliable sleep-wake patterns via actigraphy [22]. The average number of nights of actigraphy for the analytic sample was 5.9 (SD=0.8, range=4-7) in 2013 and 5.9 (SD=0.7, range=4-7) in 2018. Sleep diary-reported bedtimes and waketimes were used to define the sleep interval, which was further verified by visual inspection of the actigraphy tracings. Actigraphy data were scored using GGIR in R.

For all statistical models and to facilitate interpretation of the coefficients in clinically meaningful units, sleep duration and WASO were standardized such that coefficients reflect associations with a 10-minute increase of these predictors. Efficiency was standardized such that coefficients reflect associations with a 1 percentage point increase.

Cognitive Assessment and Cognitive Domains

Participants enrolled in Think PHRESH were administered a comprehensive neuropsychological battery [23] assessing global cognitive function with the Modified Mini-Mental State Exam (3MS) [24] as well as multiple measures of language, visuospatial ability, attention, memory, and executive functions (summarized in Table 1). Subjective concern about change in cognitive function was assessed via the Everyday Cognition Scale (ECog) [25] by the participant and by an informant if one was available. The test battery is harmonized with several previous studies of older AAs with similar demographic characteristics to provide appropriate norms and demonstrated reliability and validity [26-33]. Neuropsychological evaluations took approximately 2 hours and were conducted in the participants’ homes or nearby community centers.

Table 1:

PHRESH Cognitive Assessment Battery

Domains	Neuropsychological Assessment(s)*
Attention	WAIS-III Digit Span forward and backward Total WAIS-III Digit Symbol Substitution Test (90 seconds) WAIS-III Digit Symbol Substitution Test Copy Trial Cronbach’s alpha=0.68
Visuo-Spatial Ability	WMS-III Visual Reproduction Copy Raw score
Language	Boston Naming Test Total Score Controlled Oral Word Association Test Total Score Category Fluency Animals Total Score Cronbach’s alpha=0.72
Immediate Recall	WMS-III Logical Memory I Sum of Stories A&B CERAD Memory Trials 1-3 Total Score WMS-III Visual Reproduction Immediate Recall Cronbach’s alpha=0.65
Delayed Recall	WMS-III Logical Memory II Sum of Stories A&B CERAD Long-Delay Free Recall WMS-III Visual Reproduction Delayed Recall Cronbach’s alpha=0.68
Executive Function	Clock drawing total Trail Making Test Ratio of Part B (seconds per connection) to Part A (seconds per connection)) Golden Stroop Test Ratio of Color-Word Interference (seconds) to Color X (seconds) Digit Ordering Test Raw score Cronbach’s alpha=0.51

WAIS-III = Wechsler Adult Intelligence Scale-III

WMS-III = Wechsler Memory Scale-III

Cognitive test scores were Z-scored to have a within-sample mean of 0 and standard deviation of 1 using a regression-based norming approach [34] that adjusted for gender, age in years, and years of education. After standardization, cognitive tests were assessed for inclusion in domains of cognitive ability. Domains were calculated as the average of the standardized test scores for cases that had non-missing values for at least half of the tests in the domain. All domains demonstrated adequate internal reliability, indicated by Cronbach’s alpha (see Table 1).

Cognitive Diagnostic Adjudication.

All available cognitive assessments and medical data were reviewed together by the study’s clinical team, including neuropsychologists, a geriatric psychiatrist, and study staff involved in the assessments to determine cognitive diagnosis (i.e., clinically adjudicated impairment). University of Pittsburgh ADRC procedures[35, 36] and 2011 NIA-AA criteria [37, 38] were used to determine whether each participant exhibited normal cognitive function or met diagnostic criteria for mild cognitive impairment (MCI) or dementia. These ADRC procedures are a gold standard method that has clinicians such as neurologists, psychiatrists, and neuropsychologists review medical history, medications, neurologic and psychiatric exams, neuropsychological test results, and neuroimaging where available to determine classification of cognitive function, including a diagnosis of cognitive impairment when appropriate, by using NIA-AA research recommendations (36, 37). The procedures were adapted for the current study as follows: adjudications were based on an individual’s performance on cognitive assessments relative to their expected performance, accounting for estimated literacy (Wide Range Achievement Test 3-Reading Subtest; WRAT3-Reading) [39], educational and occupational history, available medical information (self-reported health history, medications, and psychiatric history), and available relevant clinical information. For adjudications, raw neuropsychological test scores were standardized using external normative sources adjusted for sociodemographic factors (age, education, race, and/or gender) [40]. A diagnosis of MCI was made based on 1) concern regarding a change in cognitive function as reported by the participant, an informant, or clinician, 2) objective evidence of cognitive impairment in one or more domains, as indicated by performance at 1-1.5 SDs below expectation on either two tests within a single domain (with the exception of visuospatial) or three tests scattered across domains, and 3) relatively preserved functional independence as demonstrated by self-report and performance on subtests of the Performance Assessment of Self-care Skills (PASS) [41] (shopping, medication management, critical information retrieval) that are shown to discriminate between normal cognition and MCI [42]. A diagnosis of dementia was made based on cognitive performance at 1.5+ SDs below expectation, evidence of decline, and functional dependence. Diagnostic consensus by all study clinicians and staff had to be reached for all adjudications.

Seven cases were omitted from the analyses due to invalid assessment (e.g., incomplete) or due to the presence of additional diagnoses that may reflect life-long cognitive impairment as opposed to age-related cognitive decline (e.g., developmental disability). Due to a low number of cases diagnosed with dementia (n=3), analyses combined individuals with any cognitive impairment, defined as either MCI or dementia.

Covariates

Cognitive test scores were adjusted for gender, age in years, and years of education; therefore, these variables were not included as covariates in models of 3MS or domain scores. They were included for models predicting clinically adjudicated cognitive impairment as adjusted z-scores were not used for adjudication. Neighborhood was included in all models given the original design and sampling frame of the study. We additionally controlled for variables known to be related to sleep and/or cognitive function, including marital status, income, presence of children in the home, years of residence in neighborhood, hypertension, high blood sugar, BMI, current smoking status, and presence or absence of self-reported snoring[43]. Covariates were selected a priori based on prior literature and were limited to avoid over-controlling the models, particularly given the relatively small sample size. As a robustness check, we conducted sensitivity analyses (see Supplemental Tables 2 and 3), additionally controlling for depression, alcohol use, and sleep medication use. These results were substantively similar to the models using the a priori set of covariates. Therefore, herein, we present the more parsimonious models with a priori selected covariates.

Participants were coded as having hypertension if they had high measured blood pressure (≥90 mmHg for diastolic or ≥140 mmHg for systolic), reported taking medication for high blood pressure, or reported a prior diagnosis of hypertension. Participants were coded as having high blood sugar if they had high measured Hemoglobin A1c ≥ 6.5%, reported taking medication for diabetes for high blood sugar, or reported a prior diagnosis of diabetes. Height was measured to the nearest eighth of an inch using a carpenter’s square and an 8-foot folding wooden ruler marked in inches. The weight of each participant was measured to the nearest tenth of a pound using the SECA Robusta 813 digital scale. BMI was calculated as the ratio of measured weight (kg) divided by squared height (m²).

Analytic strategy

The analytic sample differed from otherwise eligible participants (i.e., 50 or older) from the 2018 PHRESH data collection on age, education, and sex. Therefore, we used analytic weights to weight Think PHRESH participants to resemble eligible 2018 PHRESH participants. Weights were used when reporting descriptive statistics and in all modeling. Supplemental Table 1 provides descriptive statistics for all study covariates for the analytic sample compared to eligible participants who were not included in Think PHRESH.

First, we modeled the concurrent associations between sleep measures and cognitive outcomes, both measured in 2018. Separate models were fit regressing each cognitive outcome on each sleep measure, with all models adjusted for the full set of covariates described above. Linear models were used for 3MS and the cognitive domains, and logistic models were used for clinically adjudicated cognitive impairment.

Next, we modeled the association between change in each sleep measure (difference between 2013 and 2018) and cognitive outcomes measured in 2018 for participants who had sleep data measured at both timepoints. In addition to the full set of covariates described above, we controlled for the given sleep measure from 2013, to account for regression to the mean.

As the majority of the sample was AA (97%) and female (79%), we were unable to test heterogeneity of associations by race or gender in either the cross-sectional or change score models. However, sensitivity analyses that restricted the sample to only female or only AA participants demonstrated similar results as with the full sample. Therefore, analyses are presented in the full sample.

Results

Table 2 presents sample characteristics in 2018. On average, participants were 67.7 years old (SD=9.3) and lived in their neighborhood for 37.1 years (SD=22.5). The sample was mostly of low socioeconomic status, unmarried/ unpartnered, and without children in the home. Participants had high rates of comorbidities (e.g., 82.5% had hypertension, 39.4% had high blood sugar) and poor sleep across all sleep measures in 2018. Sleep also tended to get worse between 2013 and 2018 for all sleep outcomes. Roughly half of the sample had some level of clinically adjudicated cognitive impairment.

Table 2.

Study characteristics for the analytic sample (n=216)

	Weighted Mean (SD) or Weighted Percent
Covariates
Female	78.8%
Mean Age (years)	67.7 (9.3)
Education (years)	12.6 (2.2)
Black/African American	96.9%
Annual household income in $1000s	20.8 (20.5)
Married or living with partner	13.7%
Household with children	8.2%
Years in neighborhood	37.1 (22.5)
Hypertension	82.5%
High blood sugar	39.4%
BMI (kg/m2)	30.5 (7.9)
Current smoker	41.8%
Habitual snoring	29.9%
2018 Sleep Variables
Duration (minutes)	324.9 (75.4)
Efficiency (%)	70.5 (12.8)
WASO (minutes)	126.4 (79.1)
Change in Sleep Variables from 2013 to 2018
Duration (minutes)	−10.9 (85.7)
Efficiency (%)	−7.7 (13.1)
WASO (minutes)	41.7 (81.2)
Cognitive Domains a
3MS	0.0 (1.0)
Attention	0.0 (1.0)
Executive function	0.0 (1.0)
Language	0.0 (1.0)
Immediate recall	0.0 (1.0)
Delayed recall	0.0 (1.0)
Visuospatial	0.0 (1.0)
Clinically-Adjudicated Cognitive Impairment a
Mild cognitive impairment or dementia	49.6%

Notes: ^a Due to missingness on individual cognitive tests, sample sizes for the cognitive domains range slightly by measure, from 206 to 216. The sample size for the clinically-adjudicated cognitive outcome is 209.

When examined concurrently in 2018, sleep duration was not associated with any of the cognitive outcomes (Table 3). However, higher sleep efficiency was significantly associated with better performance on 3MS and three of the six cognitive domains (attention, executive function, and visuospatial ability). For example, a one percentage point higher sleep efficiency was associated with a 0.014 SD higher 3MS (CI=0.003, 0.025). Sleep efficiency was not associated with odds of having clinically adjudicated cognitive impairment. Higher WASO was associated with worse cognitive function on four of the six cognitive domains (attention, executive function, immediate recall, and visuospatial ability). For example, a 10-minute increase in WASO was associated with a 0.027 SD decrease in executive function (CI=−0.043, −0.011). WASO was not associated with 3MS or the odds of having cognitive impairment.

Table 3.

Results of cross-sectional linear and logistic regression models of the concurrent association between objective sleep measures and cognitive function (2018)^a

	Sleep Duration	Sleep Efficiency	Wakefulness after sleep onset
Linear Regression B (95% CI)
3MS	0.001 (−0.018, 0.021)	0.014 (0.003, 0.025) *	−0.023 (−0.045, 0.000)
Attention	0.004 (−0.013, 0.021)	0.012 (0.002, 0.022) *	−0.022 (−0.038, −0.007) **
Executive function	−0.007 (−0.027, 0.012)	0.014 (0.003, 0.024) *	−0.027 (−0.043, −0.011) **
Language	0.000 (−0.020, 0.019)	0.011 (−0.000, 0.022)	−0.017 (−0.039, 0.005)
Immediate recall	0.002 (−0.017, 0.021)	0.011 (−0.001, 0.022)	−0.020 (−0.039, −0.001) *
Delayed recall	−0.001(−0.019, 0.018)	0.010 (−0.002, 0.021)	−0.013 (−0.034, 0.007)
Visuospatial	0.008 (−0.013, 0.029)	0.021 (0.007, 0.035) **	−0.042 (−0.068, −0.016) **
Logistic Regression OR (95% CI)
Clinically-Adjudicated Cognitive Impairment	1.02 (0.98, 1.06)	0.98 (0.96, 1.01)	1.05 (1.00, 1.12)

Note: Sleep and cognitive function measures are both assessed in 2018. Each of the sleep variables are entered into separate models. 3MS and the cognitive domains are adjusted for education, age, and gender. Each model includes the following covariates: neighborhood, marital status, income, child in home, years in neighborhood, hypertension, high blood sugar, BMI, current smoking status, and an indicator for habitual snoring. Models of clinically-adjudicated cognitive impairment are additionally adjusted for education, age, and gender. Sample sizes range by outcome from 206 to 216. The sample size for the clinically-adjudicated cognitive outcome is 209.B (95% CI) for linear regression; OR (95% CI) for logistic regression. ***p <0.001; **p <0.01; *p <0.05.

^aDuration and WASO have been standardized such that coefficients reflect associations with a 10-minute increase of these predictors. Efficiency has been standardized such that coefficients reflect associations with a 1 percentage point increase.

For analyses of the change in sleep from 2013 to 2018 in association with cognitive outcomes, change in sleep duration was not statistically significantly associated with any of the cognitive outcomes in 2018 (Table 4). However, an increase in sleep efficiency (i.e., improvement) was associated with better scores for four of the six cognitive domains (executive function, language, immediate recall, and visuospatial ability). For instance, a 1 percentage point increase in sleep efficiency was associated with a 0.017 SD higher score on the executive function domain (CI=0.004, 0.029). Increased WASO (i.e., worsening) was associated with lower scores for three of the six cognitive domains in 2018 (attention, executive function, and visuospatial ability). For example, a 10-minute increase in WASO was associated with a 0.021 SD lower score on the executive function domain (CI=−0.038, −0.005). Change in efficiency or WASO in 2013-2018 was not associated with either 3MS or risk of clinically adjudicated cognitive impairment in 2018.

Table 4.

Results of linear and logistic regression models examining the association between changes in sleep (2013 to 2018) and cognitive function in 2018 ^a

	Change in Duration	Change in Efficiency	Change in WASO
Linear Regression B (95% CI)
3MS	0.006 (−0.019, 0.030)	0.013 (0.000, 0.026)	−0.010 (−0.028, 0.007)
Attention	0.012 (−0.008, 0.032)	0.011 (0.000, 0.022)	−0.015 (−0.029, −0.001) *
Executive function	0.005 (−0.018, 0.028)	0.017 (0.004, 0.029) **	−0.021 (−0.038, −0.005) *
Language	0.000 (−0.025, 0.025)	0.013 (0.000, 0.025) *	−0.010 (−0.029, 0.010)
Immediate recall	0.006 (−0.018, 0.031)	0.015 (0.001, 0.029) *	−0.017 (−0.035, 0.001)
Delayed recall	0.001 (−0.024, 0.025)	0.013 (−0.000, 0.026)	−0.009 (−0.029, 0.011)
Visuospatial	0.017 (−0.008, 0.042)	0.026 (0.012, 0.040) ***	−0.033 (−0.049, −0.016) ***
Logistic Regression OR (95% CI)
Cognitive impairment	1.02 (0.97, 1.07)	0.98 (0.95, 1.01)	1.06 (0.98, 1.14)

Note: Each of the sleep variables are entered into separate models. 3MS and the cognitive domains are adjusted for education, age, and gender. Each model includes the following covariates: neighborhood, marital status, income, child in home, years in neighborhood, hypertension, high blood sugar, BMI, current smoking status, and an indicator for habitual snoring. Models of cognitive impairment are additionally adjusted for education, age, and gender. Models control for baseline values from 2013 for Duration, Efficiency, and WASO, and each model is subset to participants who have the sleep measure at both time points. Sample sizes range by outcome from 161 to 168. The sample size for the clinically-adjudicated cognitive outcome is 163. B (95% CI) for linear regression; OR (95% CI) for logistic regression. ***p <0.001; **p <0.01; *p <0.05.

^aDuration and WASO have been standardized such that coefficients reflect associations with a 10-minute increase of these predictors. Efficiency has been standardized such that coefficients reflect associations with a 1 percentage point increase.

Discussion

The current study leveraged an ongoing cohort of predominantly AA older adults residing in urban, socioeconomically disadvantaged neighborhoods to study the cross-sectional association between objectively measured sleep duration and fragmentation (efficiency and WASO) with cognitive function and clinically adjudicated cognitive impairment. Further, we extend the current literature by examining changes in sleep over a 5-year period with global cognition, cognitive domains, and clinically-adjudicated cognitive impairment. We found that both level and changes over time in sleep efficiency and WASO were significantly associated with several cognitive domains, including attention, executive function, visuospatial ability, and immediate recall. Sleep duration was not associated with any cognitive outcomes. Neither the level nor change in any of the sleep measures were associated with clinically adjudicated cognitive impairment.

Poor sleep health is an emerging risk factor for ADRD that may be an important and understudied biobehavioral pathway contributing to the disproportionate burden of ADRD experienced by AA and socioeconomically disadvantaged individuals [3]. Importantly, sleep health is socially patterned, with AAs showing higher rates of specific sleep problems and disorders, including short sleep duration, sleep fragmentation, lesser slow wave sleep, and higher prevalence of obstructive sleep apnea, than their White counterparts [12, 16, 44]. However, the role of sleep in contributing to cognitive outcomes in older adults, remains under-investigated, particularly among community-based samples.

Despite the fact that short sleep duration has been associated with increased morbidity and mortality [45], including increased risk for cognitive decline [46-48], we did not find evidence for associations between sleep duration and global cognitive function, cognitive performance, or cognitive impairment. This may be due to differences in the sample or the measurement of sleep or cognitive function. Most prior studies have utilized self-report measures of sleep duration, whereas we used actigraphy assessments of sleep[6]. The lack of significant associations may also be due to insufficient variability in sleep duration in this sample, as the majority of the sample were short sleepers (<6 hours/night). Notably, while some evidence also indicates that excessively long sleep (>10 hours/night) is also associated with increased risk of cognitive decline [46], there were insufficient numbers of long sleepers to investigate the association in this sample.

In contrast, we found relatively consistent associations between indicators of sleep fragmentation, including sleep efficiency and WASO, and specific cognitive domains. In support of our hypotheses and based on prior literature [3, 49, 50], we found evidence that these indicators of sleep fragmentation were significantly associated with cognitive domains primarily assessing attention, executive function, visuospatial ability, and immediate memory. These results are consistent with other cross-sectional analyses [51, 52] as well as limited evidence suggesting that sleep fragmentation longitudinally predicts cognitive outcomes in older adults. For example, in a study of 737 older adults (mean age 81.6 at baseline), higher levels of actigraphy-assessed sleep fragmentation predicted a 1.5 times greater risk of incident Alzheimer's disease (AD) and accelerated cognitive decline over a 3-year follow-up period.[53] Finally, we found that changes in sleep efficiency and WASO over a 5-year period were associated with domain-specific cognitive outcomes, particularly for executive function and visuospatial domains.

These results are consistent with prior neuroimaging results indicating associations of sleep disruption specifically with frontal, parietal and temporal integrity [54]. However, we found no associations between sleep measures and clinically adjudicated cognitive impairment. This may be due to our small sample size as results trended in the hypothesized direction, or due to loss of precision by utilizing a dichotomous rather than continuous outcome. In addition, these results may indicate that in this population detectable associations of sleep disruption with poor cognitive function precede the onset of cognitive impairment, though we lacked sufficient sample size to directly test this hypothesis. Further work with a larger sample, including examining change in cognitive status, is needed.

These findings add to the limited existing data on changes in sleep and cognitive outcomes [55], by focusing on multiple sleep dimensions, as opposed to sleep duration exclusively, utilizing actigraphy-assessed sleep measures, and focusing on a relatively younger population (mean age= 65), comprised primarily of AA women. These findings are also consistent with a growing body of evidence suggesting that objectively measured poor sleep quality and continuity is associated with key biomarkers of ADRD pathology, including higher amyloid in brain tissue and lower amyloid in cerebrospinal fluid (CSF) [56, 57].

Findings should be interpreted in the context of study limitations, including relatively small sample size and focus on a relatively homogenous sample of predominantly AA women. Although we statistically adjusted for a host of relevant sociodemographic and health-related variables, other unmeasured variables, in particular, objectively measured OSA, may account for observed associations. Although we included an objective assessment of OSA using a home sleep apnea testing device in a subset of PHRESH participants [58], we did not have sufficient sample size for participants who had both sleep apnea testing and cognitive testing. However, based on our own prior work in this sample [58] and others’ [59, 60] demonstrating that older age, male sex, self-reported snoring, hypertension, and BMI are among the strongest correlates of sleep apnea; all of these variables were included as covariates in analyses. Nevertheless, future work is needed that statistically control for OSA as well as other sleep disorders that may account for observed associations.

Regarding cognitive assessments, we lacked objective data on change in cognitive function, did not have informants for most participants, and utilized protocols developed in more highly educated and socioeconomically advantaged individuals, although we used norms based on sociodemographically similar populations. Given that cognitive outcomes were only assessed at one time point, we cannot make inferences regarding causality or directionality; however, our future work, with repeated cognitive assessments, will allow us to do so. Finally, we did not examine upstream determinants that might contribute to poor sleep health as well as cognitive function. This will be the focus of our future work, as identifying the multi-level determinants of sleep health and cognitive function are critical to reduce health disparities.

Despite these limitations, the study has important strengths. First, we examined several actigraphy-measured sleep characteristics, which provides an ecologically valid assessment of habitual sleep patterns. Second, we examined global cognitive function and specific cognitive domains via neuropsychological evaluation, to explore specificity of associations between sleep disturbances and domains of cognitive function, and to assess clinically meaningful outcomes. Finally, by focusing on both level and changes in sleep in relation to cognitive outcomes in a sample of low-income, AA adults, these results have important clinical and policy implications for prevention and intervention, as sleep is a modifiable risk factor that is influenced by individual factors, including behavior, as well as upstream social determinants including historic and contemporary forms of systemic racism [16]. For example, at the individual level, behavioral treatments, such as cognitive behavioral therapy for insomnia have demonstrated efficacy and are particularly effective at improving sleep continuity. At the systemic level, policies that target the byproducts of systemic racism, such as the legacy of discriminatory housing policies which influence where AAs live today, including their access to safe and walkable neighborhoods, exposure to environmental pollutants and adequate housing conditions, can also have an impact on sleep and potentially downstream outcomes, including cognitive function and decline. For example, our prior work in this cohort has shown that residents who live in closer proximity to neighborhood investments show improvements in sleep profiles over time as compared to those living further from investments [61]. Focusing on multi-level determinants, including sleep is critical to address the disproportionate burden of ADRD risk in AAs.