Longitudinal Sleep Patterns and Cognitive Impairment in Older Adults

Key Points Question Are longitudinal patterns of self-reported sleep duration associated with cognitive impairment in older adults? Findings In this cross-sectional study of 826 older adults, there was a significant association between cognitive impairment and short sleep duration and longitudinal variability in sleep duration. Meaning These findings suggest that longitudinal variability in sleep duration, rather than sleep duration in itself, may be an important factor in the development of cognitive decline in older adults.


Introduction
More than 60 million people globally currently have dementia, 1 with Alzheimer disease (AD) accounting for an estimated 60% to 80% of these cases. 2,3Observational clinical studies demonstrate that many events precede cognitive decline, including amyloid β deposition that appears at least 15 years before the onset of clinical cognitive impairment. 4Data from interventional trials 5,6 suggest that targeting pathogenic processes early in the course of disease may provide the most effective avenue for therapeutic intervention.Identifying the mechanisms preceding cognitive and functional decline may provide approaches to the treatment and prevention decades before dementia onset.
Sleep disruption has long been associated with dementia, with studies 7,8 indicating that up to 90% of patients experience disrupted sleep before the emergence of cardinal disease symptoms.
Although this association was initially believed to reflect the progressive degeneration of sleepregulatory centers within the brain, 9 more recent studies suggest that sleep disruption may influence the pathological processes underlying these dementing disorders.Short sleep duration has been associated with increased risk of cognitive impairment in healthy aging adults, [10][11][12][13] and AD biomarker studies 14,15 have found that poor sleep quality and short sleep duration are associated with greater AD-related pathological burden.Although these studies suggest a link between chronic sleep disruption and dementia diagnosis, they are limited by their reliance on relatively simple, generally cross-sectional, measures of sleep.
In this study, we evaluate how patterns of longitudinal sleep duration impact age-associated cognitive decline.Using both self-reported sleep duration and a neuropsychological battery regularly collected within the community-based Seattle Longitudinal Study (SLS), 16 we define the association of longitudinal sleep patterns (duration, change, and variability over time) with the incidence of cognitive impairment in an older population.

Sample
7][18][19][20] The present cross-sectional study evaluated a subsample within the SLS that underwent multiple rounds of both a Health Behavior Questionnaire (HBQ) and a neuropsychological battery.Of these individuals, only participants with complete demographic information were included in the analysis.Participants in this study were enrolled under an institutional review board-approved protocol through the University of Washington in accordance with guidelines from the Declaration of Helsinki. 21All participants provided written informed consent before participation in the Seattle Longitudinal Study.This report follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for cross-sectional studies.
Self-reported demographic information included sex (female or male), race and ethnicity (White or other, which combined participant responses of Asian or Pacific Islander, Black, Hispanic, Native American, or any other participant response), years of education (at or below high school, college, graduate school), and apolipoprotein E ε4 (APOE*E4) allele carrier status (carrier vs noncarrier).Race and ethnicity were analyzed in this study as demographic variables, in consideration that chronic sleep disorders may occur at disproportionate rates among certain racial and ethnic groups.

Participant Testing
The HBQ was provided at 3-to 5-year intervals for a total of 5 times between 1993 and 2012.Item 100 on the SLS HBQ asks, "In the PAST 7 DAYS, how many hours did you usually sleep per 24 hours (not counting naps)?" with Likert scale responses of 1 (Յ 5 hours), 2 (6 hours), 3 (7 hours), 4 (8 hours), and 5 (Ն 9 hours).From these data, we calculated sleep duration (median value of all measurements) and assigned a sleep duration phenotype based on these median values (short sleep, <7 hours; medium sleep, 7 hours; and long sleep, >7 hours).Because of the longitudinal assessments, it was possible to evaluate changes in self-reported sleep over time.We first evaluated whether for each participant's self-reported sleep durations, the slope of the best-fit line was significantly nonzero.We also measured the overall variability in participant self-reported sleep duration over time as the SD of all measurements, which we defined as sleep variability.
Within the original SLS, a neuropsychological battery was repeated every 5 to 7 years between 1997 and 2019.7][18][19][20] In the present analysis, to best capture distinct changes in global cognitive function, we defined cognitive impairment as performance below threshold on both the Mini-Mental State Examination [22][23][24] (MMSE; score <26) and Mattis Dementia Rating Scale [25][26][27] (DRS; score <129).The SLS neuropsychological battery also included the Center for Epidemiologic Studies-Depression Scale (CES-D). 28,29In this study, the median CES-D global score was used to define depression as a covariate throughout analysis.

Statistical Analysis
Data analysis was performed from September 2020 to May 2023.Statistical analyses were completed using Prism statistical software version 9.5 (GraphPad) and Python statistical software version 3.6 (Python Software Foundation).All tests were 2-sided, and P < .05 was considered statistically significant.Study characteristics for all demographic characteristics (sex, race and ethnicity, level of education, and APOE*E4 allele carrier status), depression (median CES-D total score), self-reported sleep measures (median sleep duration and sleep phenotype), and cognitive measures (MMSE and DRS) were evaluated stratified by ages younger than 65 years, 65 to 84 years, and 85 years and older.Statistical differences were assessed by χ 2 test and 1-way analysis of variance (ANOVA) with Šídák multiple comparison test used for post hoc analysis.
To assess the association of sleep parameters with the time to cognitive impairment, we used Cox proportional hazards regression.Hazard ratios (HRs) and 95% CIs were estimated using multivariable adjusted Cox models.Only participants who completed 3 rounds of testing were included in analysis, so that best-fit slopes and SDs of sleep duration values could be calculated.Thus, in this part of the study, participants had between 3 and 5 assessments of self-reported sleep duration).Model 1 controlled for demographic covariates of sex, race and ethnicity, years of education, and APOE*E4 allele carrier status.Model 2 added CES-D median depression score to model 1.The final model 3 added sleep factors: categorical assessment of sleep duration phenotype (short sleeper vs long sleeper), decreasing sleep duration (coded 0 if slope of sleep durations was not significantly decreasing and coded as 1 if slope was significantly decreasing), and sleep variability.All model assumptions were tested and met the criteria.Multicollinearity was assessed by variance inflation factor for covariates, resulting in no variance inflation factor greater than 5. Thus, no covariates were removed from the study.

Participant Demographics
We evaluated 1104 SLS participants and excluded 278 participants lacking demographic data, primarily APOE genotype, leaving an initial study sample of 826 participants (mean [SD] age, 76.3 [11.8]Global CES-D scores were significantly different (F 2,823 = 5.851; P = .004;1-way ANOVA), with 65-to 84-year-old participants reporting lower levels of depression compared with participants younger than 65 years and 85 years or older.

Self-Reported Sleep Duration
Although prior studies [10][11][12][13] have reported shorter sleep durations in older populations, other studies [30][31][32][33] in have suggested that this association may not be straightforward.Item 100 on the SLS HBQ assesses average nightly sleep duration over the past week (not including naps).From this, we calculated median sleep duration for each participant.We observed a significant age-related difference in sleep duration (F 2,823 = 9.592; P < .001;1-way ANOVA), with longer sleep durations for older participants.When participant median sleep duration phenotype was categorized as short (<7 hours), medium (7 hours), or long (>7 hours), we observed significant differences in the distribution of these phenotypes among age groups (χ 2 4 = 20.73;P < .001),with an increasing percentage of long sleepers and fewer short and medium sleepers with advancing age.

Longitudinal Changes in Sleep Duration
Although several studies [10][11][12][13]31 have sought to define the association of sleep duration assessed at a single time point with cognitive impairment, the repeated assessment of sleep parameters in the present study provides the opportunity to assess how changes in sleep behavior over time contributes to cognitive outcomes. Fo example, multiple studies [34][35][36][37] have shown that participants whose self-reported sleep duration declined between 2 assessments at intervals between 3 and 14 years exhibited a higher incidence of cognitive impairment.Because many SLS participants underwent 3 to 5 HBQ assessments, we were able to define generalized patterns of longitudinal sleep duration from these data.Excluding 212 participants from sample 1 who did not have at least 3 HBQ assessments resulted in a subsample of 614 participants (sample 2; Table 1).We evaluated whether each participant exhibited consistent changes in sleep duration over time, whether increasing or declining, by testing if the slope of the best-fit line for sleep durations was significantly positive or negative.Among all participants in sample 2, 41 (6.7%) exhibited a significant decrease in sleep duration, whereas only 4 (0.7%) exhibited a significant increase in sleep duration over time.In our analysis, because of the small number of participants with increasing sleep duration (0.7%), we did not evaluate this group separately.When comparing among the different age groups, we observed no differences in the proportion of participants exhibiting declining sleep duration over time (Table 1).This very low percentage of participants exhibiting a significant change in sleep duration at first seemed inconsistent with prior studies.[34][35][36][37][38] Figure 2A shows visit-level self-reported sleep duration data from 5 SLS participants who each participated in 5 study visits.It is noteworthy that when 3 data points are censored and only 2 study visits are considered (the first and last), 4 of the 5 participants appear to have clearly increasing or declining sleep durations (Figure 2B).However, when data from all 5 visits are considered, only 1 of the 4 participants exhibits a consistently increasing or declining sleep duration (Figure 2A).As shown for individual participant values in Figure 2, variability in self-reported sleep duration over several study visits was much more noticeable than were consistent increases or declines in these measures.To capture this feature, we measured sleep variability, defined as the SD of selfreported sleep duration across all participant visits.When compared among age groups, there was a significant association of age with sleep variability (F 2,611 = 3.378; P = .02;1-way ANOVA), with increased age associated with lower sleep variability across longitudinal sleep duration assessments.

Association of Sleep Parameters With Cognitive Decline
Finally, we evaluated whether these sleep features were associated with cognitive decline using Cox proportional hazards regression.For this analysis, we assessed data up to and including the study visit where participants converted to the status of cognitively impaired (subthreshold performance on MMSE and DRS).For participants who remained cognitively intact for the duration of the study, all data were included.
For the evaluation of cognitive performance within sample 2 based on the MMSE and the DRS (Table 1), as anticipated, we observed higher rates of cognitive impairment with increasing age.Of the 614 participants, 44 (7.2%) exhibited cognitive impairment during their final evaluation, with a higher prevalence of cognitive impairment among participants with older age (χ 2 2 = 84.57;P < .001).In model 1, we evaluated the contribution of sex, race and ethnicity, years of education, and APOE*E4 status on the incidence of cognitive impairment with age.Model 1 output is detailed in Table 2 and shows that years of education (HR, 1.16; 95% CI, 1.03-1.28)and possessing 1 or more copies of the APOE*E4 allele (HR, 2.10; 95% CI, 1.11-3.99)were significantly associated with cognitive impairment.We next developed model 2 (Table 2), which added the contribution of depression.
In a series of models based on model 1, we evaluated the association of different sleep parameters, sex, race and ethnicity, years of education, and APOE*E4 status with the incidence of  3 shows the associations between short sleep status (Figure 3A) and sleep variability (Figure 3B) and age on remaining cognitively intact among SLS participants.For both short sleeper status and sleep variability, increased risk of cognitive impairment is reflected in the leftward shift in the survival curve.

Discussion
In this cross-sectional study, analysis of longitudinal measures of self-reported sleep duration within the community-based SLS demonstrated that participants reporting being short sleepers (<7 hours per night) had significantly higher risk of cognitive impairment.1][12][13] Furthermore, and to our knowledge not previously reported, this study found that higher variability in self-reported sleep duration over the course of decades, rather than a more consistent decline in sleep duration, was significantly associated with cognitive impairment.
The present findings are consistent with previous studies [10][11][12][13] of healthy aging adults identifying an association of short sleep duration with the risk of cognitive impairment.0][41][42][43][44][45][46][47] Short sleep duration has been associated with greater amyloid β plaque burden 15 and faster ventricular expansion, 10 indicating its potential role in exacerbation of AD-related neurodegenerative processes.Rodent studies demonstrate that sleep-active glymphatic function contributes to the clearance of amyloid β, 48,49 tau, 50 and a-synuclein, 51 whereas acute sleep deprivation slows this clearance. 52In addition, recent studies 53,54 using intrathecal contrast-enhanced magnetic resonance imaging confirmed that solute clearance from the human brain was sleep dependent.Thus, the chronic impairment of sleep-active glymphatic function may underlie the observed clinical association of short sleep duration with cognitive impairment.
We observed that a higher median sleep duration and prevalence of long sleep phenotype was present among older participants in this study.Although this finding was somewhat surprising given the long-established association of aging with shorter sleep, [10][11][12][13] longer self-reported sleep duration may be observed with increasing age within the cohort for several reasons.It is possible that there is a selection bias within our study, representing an older population with increased sedentary behavior, increased opportunity for sleep with retirement, and changes in comorbidities.Similarly, it is possible that our older participants are overestimating sleep duration in their self-reports, which are known to become increasingly inaccurate with increasing age. 55,56Interestingly, although multiple meta-analyses [30][31][32]  The SLS offers the unique opportunity to examine longitudinal changes in self-reported sleep duration over an extensive follow-up period.With participants providing 3 to 5 HBQ responses over a period of up to 20 years, this allows for the evaluation of both general changes in sleep duration (whether increasing or decreasing) and the intraindividual variability in sleep duration (SD) over time.
Previous studies focusing on the association of changes in longitudinal sleep duration with cognition in aging have produced mixed results.Both declining 34,36 and increasing 35,37,38 sleep duration have been reported to be associated with cognitive impairment.We believe that one possible cause of these discrepancies stems from most studies quantifying longitudinal changes in sleep duration on the basis of only 2 measurements.As we show for 5 example participants in Figure 2 57 The observed variability may be associated with age-related comorbidities in the neurological or psychiatric domains, such as depression, chronic pain, frequent nocturia, or changes in social and behavioral factors, 58 such as shift work, retirement, or changes in marital status.Future studies will need to define the relationships between long-term variability in self-reported and objective sleep duration and their associations with these cognitive, psychological, social, and behavioral factors.
Our study emphasizes the importance of different experimental approaches when defining longitudinal changes in sleep patterns.Discrepancies in methodological approach, including approaches to assessing sleep duration and quality (eg, self-report, actigraphy, and electroencephalogram), frequency and duration of assessment (single time point assessment, multiday, and long-term ecological assessments), study timescales (from days to years), and synchrony of data collection can each substantially alter study findings.This highlights the need for a more comprehensive evaluation of longitudinal sleep behavior in future studies.

Limitations
This study has limitations that should be mentioned.The SLS, which was initiated in 1956, aimed to identify the combined impact of aging, birth cohorts, and lifestyle factors on downstream cognitive impairment. 16Although the present study did not evaluate birth cohorts, owing to the breadth of study duration, we were able to evaluate both self-reported sleep duration patterns and cognitive performance in 1104 participants assessed over decades.Critically, as a retrospective analysis of SLS data, this study solely relied on self-reported sleep duration as defined by a single question within the HBQ.Although this permits the assessment of longitudinal changes in self-reported sleep patterns, as discussed, the subjective nature of the measure and the infrequent sampling limit the conclusions that we can draw from these data.
In addition, the present study does not assess clinical sleep disorders, such as obstructive sleep apnea, 59,60 circadian dysregulation, 61 or insomnia, 62 which have previously been associated with cognitive impairment.These conditions often exist years before the onset of cognitive decline 7,8 and may contribute to unstable sleep durations among study participants.

Conclusions
To our knowledge, this study is the first to demonstrate that variability in longitudinal self-reported sleep duration is associated with cognitive impairment in the community-based SLS sample.It also supports previous findings that a short sleep phenotype is associated with impairment in cognitive performance.These findings suggest that longitudinal variability in sleep duration, in addition to average sleep duration alone, may be important contributors to the development of cognitive decline in older adults.They also argue that understanding the clinical relationships among age, sleep disruption, and cognitive impairment may require the assessment of sleep behavior over longer timescales than is presently common in research practice.

a
The Šídák multiple comparison test was used for all post hoc analyses.b Other race and ethnicity includes Asian or Pacific Islander, Black, Hispanic, Native American, and other participant responses.c Participants aged 65-84 y vs aged <65 y. d Participants aged Ն85 y vs aged 65-84 y. e Participants aged Ն85 y vs aged <65 y.

Figure 2 .
Figure 2. Representative Sleep Patterns Across the Cohort

Figure 3 .
Figure 3. Survival Associated With Sleep Patterns

in Figure 1. Table 1 presents the demographic data for
through calculation of median CES-D score for each participant and was compared across age groups.

Table 1 .
Baseline Characteristics of Study and Substudy Participants (continued)

Table 2 .
Cox Proportional Hazard Regression Models Evaluating Longitudinal Associations With Cognitive Impairment for Sample 2 (n = 614) impairment with age.Within this framework, we observed that neither median sleep duration nor status of declining sleep duration was significantly associated with cognitive impairment.Status as a short sleeper (HR, 2.79; 95% CI, 1.39-6.58)andincreasingsleep variability (HR, 2.22; 95% CI, 1.11-5.64)weresignificantlyassociated with cognitive impairment with age.We then developed a model (model 3) adding longitudinal sleep parameters, including short sleeper status, status of decreasing sleep duration, and sleep variability, to model 2. Within model 3 (Table2), several factors were found to be significantly associated with the risk of cognitive impairment, cognitive 33ve reported that total sleep time does decline with age, this does not extend into the oldest age brackets.In fact, a recent study by Coutrot and colleagues33of 730 187 participants across 63 countries found that shorter sleep durations occurred in persons up to age 50 years compared with young adults.However, these age-related differences plateaued, with an increase in nightly sleep duration being reported in those older than 70 years.Given the advanced age among SLS participants, it is possible that the present study is capturing the sleep patterns in older participants that are often unexplored in studies of healthy aging.
Although the present study does not provide clarity on the optimal number or frequency of longitudinal sleep assessments, it does demonstrate that increasing sampling numbers even modestly (such as from 2 to 3-5 instances) provides useful insights regarding sleep variability that would otherwise be missed.It is possible that, in line with the concept of the Nyquist frequency in signal processing, frequent longitudinal assessment of sleep features will be needed to arrive at a clear understanding of the association of age-related changes in sleep with cognitive decline.
, when longitudinal changes in sleep duration are derived from only 2 assessments, the consistency of change over time is exaggerated.When sampled more frequently, consistent trends in sleep duration become much less prevalent, and variability in sleep duration over the course of years emerges as a key feature of the data.Our observation that sleep variability is significantly associated with cognitive impairment but that general declines in sleep duration are not has important implications for the interpretation of longitudinal sleep data.It is possible that by assessing longitudinal sleep features at only a small number of instances, key biological relationships between changes in sleep and cognitive function are being missed, while spurious associations are arising from undersampled phenomena.