Introduction
Age-related cognitive declines are prevalent in both pathological and nonpathological aging (Dzierzewski et al., Reference Dzierzewski, Dautovich and Ravyts2018; Keller, Reference Keller2006; Li et al., Reference Li, Huxhold and Schmiedek2004) and are difficult to reverse after declines begin. For instance, pharmaceutical trials have shown limited ability to improve cognitive outcomes in patients with Alzheimer’s disease (Alexander et al., Reference Alexander, Emerson and Kesselheim2021; Doody et al., Reference Doody2014). Thus, researchers are increasing attention to lifestyle modifications that may ameliorate age-related cognitive decline (Shatenstein et al., Reference Shatenstein, Barberger-Gateau and Mecocci2015) and Alzheimer’s disease progression (Bhatti et al., Reference Bhatti, Reddy, Reddy and Bhatti2019). Sleep is a variable in older adults’ cognitive functioning that may be modifiable through treatment (Taylor and Pruiksma, Reference Taylor and Pruiksma2014), given its association with various types of memory, including memory consolidation and retrieval (Bonnet, Reference Bonnet and Kryger2005; Nadel et al., Reference Nadel, Hupbach, Gomez and Newman-Smith2012; Plihal and Born, Reference Plihal and Born1997; Scullin and Bliwise, Reference Scullin and Bliwise2015); therefore, improving sleep presents a potential avenue to enhance cognition (Dzierzewski et al., Reference Dzierzewski, Dautovich and Ravyts2018).
Given changes in sleep architecture (e.g. reduced slow-wave sleep) and reductions in sleep need as individuals age (Hirshkowitz et al., Reference Hirshkowitz2015), it is hypothesized that sleep’s influence on cognition may vary depending on age (Wilckens et al., Reference Wilckens, Woo, Kirk, Erickson and Wheeler2014). Cognitive functioning also changes with advancing age; therefore, it is postulated that the cognitive impacts of various lifestyle and clinical factors also differ by age group.
Previous studies that compared the effects of sleep disturbance (e.g. sleep deprivation and poor self-reported sleep quality) in different age groups (e.g. 20–25 vs. 50–60; 50–64 vs. 65+) have found worse performance in the older adult group across multiple domains of cognitive function, such as word detection (distinguishing words from nonwords), word memory, processing speed, psychomotor vigilance task, executive functions (EFs), and verbal fluency (Miller et al., Reference Miller, Wright, Ji and Cappuccio2014; Webb, Reference Webb1985) (Bartolacci et al., Reference Bartolacci, Scarpelli, D'Atri, Gorgoni, Annarumma, Cloos, Giannini and De Gennaro2020; Kim et al., Reference Kim, Suh, Cho, Yang, Yun, Thomas, Lee and Shin2013), even though older adults reported lower levels of subjective level of sleepiness in some of these studies (Bartolacci et al., 2020; Kim et al., 2013).
On this note, it is postulated that more cognitively demanding tasks of executive functioning governed by the frontal network system, such as inhibitory control (Harrison and Horne, Reference Harrison and Horne1998; Wilckens et al., Reference Wilckens, Woo, Kirk, Erickson and Wheeler2014) and decision-making (Harrison and Horne, Reference Harrison and Horne1998, Reference Harrison and Horne1999; Muzur et al., Reference Muzur, Pace-Schott and Hobson2002), are particularly vulnerable to poor sleep. Sleep deprivation literatures also noted that sleep-related EF deficits remain, even though vigilance may be resilient via mechanisms such as stimulant use (Killgore, Reference Killgore2010). The frontal network, including the fronto-striatal systems, is implicated in age-related changes in attention and EF (Buckner, Reference Buckner2004; Hedden and Gabrieli, Reference Hedden and Gabrieli2004; Pace et al., Reference Pace, Treloar and Scott2011), and the combination of the aging process and sleep disturbance may lead older adults with sleep disturbance to be particularly susceptible to impairment in EF tasks.
Nonetheless, some studies have indicated no age effect or less sleep-related susceptibility in older adults, further contributing to the unclear understanding in this area. For instance, research has shown sleep may be less beneficial in older adults compared to younger adults in terms of improving memory performance, particularly declarative memory and slow-wave sleep-dependent memory consolidation (Scullin and Bliwise, Reference Scullin and Bliwise2015; Spencer et al., Reference Spencer, Gouw and Ivry2007; Stickgold, Reference Stickgold2005). In other studies, psychomotor vigilance was less impacted by sleep disturbance in older adults (Bliese et al., Reference Bliese, Wesensten and Balkin2006; Stenuit and Kerkhofs, Reference Stenuit and Kerkhofs2005), indicating that the effect of age on the sleep–cognition relationship may be dependent on the cognitive domain that is being examined.
In addition, previous studies were limited by their use of age as a categorical concept (by age groups, e.g. young, middle-aged, and older adults), which may have underestimated the changes that happen within each age group due to lower statistical power (Altman and Royston, Reference Altman and Royston2006). Examining age as a continuous variable may help to understand the relationship between sleep and cognition along with the broad continuum of the aging process and allow for the identification of specific age ranges during which the change in the relationship between sleep and cognition occurs. This is especially important in brain aging research, given that age-related changes in brain structures and networks may affect sleep quality, cognition, and the interaction between these two variables (Hukkelhoven et al., Reference Hukkelhoven2003; Scullin and Bliwise, Reference Scullin and Bliwise2015; Yagi et al., Reference Yagi, Nouchi, Murayama, Sakaki and Kawashima2020).
In this study, we examined the associations of age with the relationship between sleep quality and cognitive performance within an age-diverse adult sample using the Human Connectome Project-Aging (HCP-A) data. We hypothesized that there will be a significant age × sleep quality interaction in various cognitive domains, particularly in demanding executive functioning tasks, fluid cognition, and episodic memory. Specifically, we examined sleep’s interaction with age, with age as both a linear and nonlinear (quadratic) term, which allows us to identify the “sensitive age ranges” outside of traditionally defined age groups, where the relationship between sleep quality and cognitive performance is modified. We hypothesized that ages in midlife to older adulthood (ages 50–70 years) would be particularly vulnerable to sleep-related cognitive changes, given that significant changes in both sleep architecture and age-related neurodegenerative processes begin in midlife and continue with the aging process (Holanda and de Almondes, Reference Holanda and de Almondes2016). We also examined different cognitive domains included in the HCP-A study, such as executive function, episodic memory, and fluid and crystallized cognition that are part of the NIH Toolbox (TB) Cognition Battery (Weintraub et al., Reference Weintraub2013). Of note, it may be important to specify which specific EF tasks are implicated in sleep disturbance because EF tasks more broadly have shown mixed associations with sleep disturbance (Killgore, Reference Killgore2010). Therefore, we added the Trail Making Test (TMT), a measure of set-shifting abilities, given that sleep has a crucial role in tasks that require the ability to quickly change behaviors and adapt flexibly to modifying conditions (Couyoumdjian et al., Reference Couyoumdjian2010).
Methods
Participants
Data analyses were performed using the most recent HCP-A dataset release (available at https://www.humanconnectome.org/). A detailed description of the dataset has been previously published (Bookheimer et al., Reference Bookheimer2019). Briefly, the HCP-A participants were recruited to represent the current US population, with regard to age, gender, race, ethnicity, and socioeconomic metrics. Participants were recruited from four sites (Washington University St Louis, University of Minnesota, Massachusetts General Hospital, and University of California, Los Angeles), and all sites strived for a balance of participants with low, middle, and high socioeconomic status. Participants were recruited through advertisements and flyers, active senior centers, places of worship, public lectures and workshops on aging, and senior living centers.
Participant eligibility was determined through phone screens to identify exclusionary health conditions. For instance, individuals diagnosed and treated for major psychiatric disorders (e.g. schizophrenia, bipolar disorder, and severe depression) or neurological disorders (e.g. stroke, brain tumors, and Parkinson’s disease), scoring <30 on the Telephone Interview for Cognitive Status modified (TICS-M) (for participants aged 60–80 years) (de Jager et al., Reference de Jager, Budge and Clarke2003), not passing critical orientation items (day of the week, date, season, age, and phone number) of the TICS-M for participants over 80 years, and scoring below age-bracket thresholds (<20 for ages 36–79 years, <18 for ages >79 years) for the Montreal Cognitive Assessment (MoCA) were excluded (Nasreddine et al., Reference Nasreddine2005). For the current study, we used a dataset of 725 individuals from HCP-A who completed neuropsychological tests and had self-reported sleep data. A flow chart of sample selection is presented in Supplemental Figure 1.
Measures
The Pittsburgh Sleep Quality Index (PSQI) (Buysse et al., Reference Buysse, Reynolds, Monk, Berman and Kupfer1989) is a self-rated measure of sleep quality and disturbances over the past month. The PSQI contains 19 items, which produce a total score and 7 component scores: subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbance, use of sleeping medication, and daytime dysfunction. Each component score is rated on a 0–3 scale, with higher scores representing poorer sleep, and adds up to a 0–21 total score. The PSQI total score has acceptable internal homogeneity, consistency, and validity (Buysse et al., Reference Buysse, Reynolds, Monk, Berman and Kupfer1989).
The NIH Toolbox (TB) Cognition Battery (Weintraub et al., Reference Weintraub2013) consists of seven measures assessing subdomains of executive function, episodic memory, language, processing speed, working memory, and attention. Composite measures have been developed using factor analytic methods (Akshoomoff et al., Reference Akshoomoff2013) to represent overall cognition and/or certain categories of abilities that change across the lifespan. The Crystallized Cognition Composite score includes the Oral Reading Recognition Test and the Picture Vocabulary Test, and the Fluid Cognition Composite score includes the Dimensional Change Card Sort Test (cognitive flexibility), the List Sorting Working Memory Test, the Picture Sequence Memory Test (episodic memory), the Pattern Comparison Processing Speed Test, and the Flanker Inhibitory Control and Attention Test. Higher scores on the NIH TB Cognition Battery reflect better performance. The NIH TB Cognition Battery demonstrated strong test–retest reliabilities and adequate convergent and discriminant validities (Weintraub et al., Reference Weintraub2013). Full descriptions of the tasks are provided in Weintraub et al., Reference Weintraub2013. For the analysis, the uncorrected standard scores were used to model with age.
While the NIH Toolbox has the advantage of compiling crystallized and fluid cognition broadly, its construct validity with standard executive function measures is limited, and it does not encompass some key executive functions such as cognitive set-shifting abilities or learning and recall that may be impacted by sleep (Ott et al., Reference Ott2022; Scott et al., Reference Scott, Sorrell and Benitez2019). Therefore, we added the following individual neuropsychological tests included in the HCP-A dataset to supplement our cognitive outcome measures.
The Trail Making Test (TMT) A and B (Army Individual Test Battery, 1944; Reitan and Wolfson, Reference Reitan and Wolfson1985) consists of two tests of processing speed and working memory. In TMT-A, an individual draws lines connecting 25 circled numbers that are spread out on a sheet of paper, and in TMT-B the person does a similar task but alternates between numbers and letters (Tombaugh, Reference Tombaugh2004). Accordingly, TMT-B has been shown to measure cognitive set shifting (Olivera-Souza et al., Reference Olivera-Souza2000). In both TMT-A and TMT-B, higher scores reflect poorer performance. The TMT has strong interrater reliability (Strauss et al., Reference Strauss, Sherman and Spreen2006) and construct validity (Sanchez-Cubillo et al., Reference Sanchez-Cubillo2009).
The Rey Auditory Verbal Learning Test (RAVLT) (Rey, Reference Rey1964) is a test of verbal episodic memory where participants are asked to learn 15 words across 5 learning trials (Ivnick et al., Reference Ivnick, Malec, Tangalos, Petersen, Kokmen and Kurkland1990). The tester reads List A before each recall trial, and then the tester reads a separate list, List B, and asks the subject to freely recall the words. In trial 6, the tester then asks the participant to recall the words from List A. The HCP-A employs an alternate RAVLT administration, which does not include the additional 20-min delayed recall in the standard RAVLT (Bookheimer et al., Reference Bookheimer2019). This abbreviated version reduces the length of the testing battery, which may otherwise be overly burdensome for older participants, and is supported by findings that short-term delayed recall is equivalent to long-term delayed recall in identifying amnestic mild cognitive impairment and other neurological dysfunctions (Bookheimer et al., Reference Bookheimer2019; Schoenberg et al., Reference Schoenberg, Dawson, Duff, Patton, Scott and Adams2006; Zhao et al., Reference Zhao, Lv, Zhou, Hong and Guo2012). Higher scores on the RAVLT reflect better performance. The RAVLT has shown adequate test–retest reliability, internal consistency, and divergent and convergent validity (de Sousa Magalhães et al., Reference de Sousa Magalhães, Mallow-Diniz and Hamdan2012).
Statistical analyses
Residual diagnoses were performed to test for normality and homoscedasticity of cognitive outcomes. Multicollinearity was examined using the variance inflation factor (VIF). For our analyses, one individual with a TMT-A score <1 s was removed and two individuals with a TMT-B score <20 s were removed. In addition, TMT-B was truncated at 300 s, so three individuals’ scores were lowered to 300. The following numbers of individuals had cognitive outcomes for each of the measures indicated: TMT-A: 705, TMT-B: 706, NIH TB Fluid Cognition: 608, NIH TB Crystallized Cognition: 607, and RAVLT: 694.
Descriptive analysis was used to profile the characteristics of the participants (Table 1) using mean, standard deviation, and range for continuous variables and frequency and percentage for discrete variables. For the primary analysis, we tested the association of PSQI and age moderation using linear regression models. Since previous studies reported the quadratic association of age (Hukkelhoven et al., Reference Hukkelhoven2003), we added the quadratic term of age (age2). Age moderation was tested by adding age × PSQI and age2 × PSQI interaction terms. For the models with significant age × PSQI or age2 × PSQI interactions, we performed post hoc contrast analyses to identify age ranges with significant PSQI associations. All models included sex, race, education, hypertension, and sleep medication as covariates (Knutson et al., Reference Knutson2009; Lee et al., Reference Lee, Vigoureux, Hyer and Small2022), and age was centered on the mean age (59.66 years).
Table 1. Demographics of the study sample
SD, standard deviation; PSQI, Pittsburgh Sleep Quality Index; RAVLT SD TC, Rey Auditory Verbal Learning Test Short Delay Total Correct; TMT-A, Trails Making Test A; TMT-B, Trails Making Test B; RAVLT LB TC, Rey Auditory Verbal Learning Test Short Delay List B Total Correct; RAVLT Trials 1-5 SD TC, Rey Auditory Verbal Learning Test Short Delay Trials 1-5 Total Correct. Uncorrected scores were used for composite scores from the NIH Toolbox.
Since our models are hypothesis-driven and primarily aim to evaluate age moderation to examine the consistency of the existing literature, we reported regression coefficients (B) with their 95% confidence intervals and standardized regression coefficients (β) to inform future studies. We also performed multiple comparison corrections using the False Discovery Rate corrections. Additionally, we reran the models without quadratic age terms to test the robustness of the interactive association. When defining the sensitive period at which the association between sleep quality and cognitive performance is most strongly presented, we determined the areas of significance by inspecting when the confidence intervals for the PSQI beta coefficient stopped including zero. We also conducted analyses to examine whether there are other higher-order polynomial terms of age (e.g. cubic and quartic terms) that interact with sleep and found that the quadratic model performed the best for all cognitive outcomes based on the Bayesian information criterion (BIC) (e.g. For crystallized cognition, BIC for quadratic was the smallest [4508.10] followed by cubic [4511.80] and qartic [4523.88]. Similar patterns were evident for other cognitive outcomes). All analyses were performed using R 4 .1.2, and p values <0.05 were considered to indicate statistical significance.
Results
Demographic characteristics
Demographic characteristics of the study sample are presented in Table 1. The mean age of the study sample (N = 711) was 59.66 (SD = 14.91), and 55.7% were females. The sample consisted of predominately White (72.2%) participants. The mean years of education in the sample was 17.51 years (SD = 2.19).
The association between sleep quality and cognitive performance
The associations between sleep, age, and cognitive performance are presented in Table 2, with the main associations of PSQI and age presented in Model 1 and their interactive terms presented in Model 2. In Model 1, higher PSQI scores (i.e. poorer sleep) were associated with worse performance on TMT-B (β = 0.09; Std. CI = 0.02 – 0.17; p = 0.02), but not with other cognitive outcomes (Ps ≥ 0.42) after adjusting for race, sex, hypertension, use of sleep medication, and years of education (Table 2. Model 1). In Model 2, higher PSQI scores were associated with poorer performance on TMT-B (β = 0.09; Std. CI = 0.02 – 0.17; p = 0.001) and crystallized cognition tasks (β = −0.03; 95% CI = −0.11 to 0.05; p = 0.03), but not TMT-A, fluid cognition tasks, or RAVLT short-term delayed total recall (Table 2. Model 2).
Table 2. Regression of sleep quality and age on cognition
PSQI, Pittsburgh Sleep Quality Index; Crystalized Cognition, NIH-TB Crystalized Cognition Composite; Fluid Cognition, NIH-TB Fluid Cognition Composite.
Covariates: sex, race, use of sleep medication, and years of education.
*Higher scores on Trail Making Test-A, Trail Making Test-B, and PSQI indicate poorer cognitive performance/sleep. For all other metrics, a higher score indicates stronger cognitive performance.
NOTE: Separate regression models were conducted to examine the main effect of sleep (“Model 1”) and the interactive effect of sleep and age on cognitive performance (“Model 2”).
The association between age and cognitive performance
In Model 1, higher age was associated with poorer performance on all cognitive measures (linear term, p < 0.001) except for crystallized cognition tasks (Table 2. Model 1). When the nonlinear (quadratic) term for age was examined, age2 was only associated with TMT-A (β = 0.15; Std. CI = 0.08 – 0.22; p < 0.001) and TMT-B (β = 0.13; Std. CI = 0.06 – 0.19; p < 0.001) (Table 2. Model 1).
Age moderation on the association between sleep quality on cognitive performance
In Model 2, there was a significant interaction term between the PSQI and age (linear) on crystallized cognition (β = −0.09; Std. CI = −0.16 to −0.02; p = 0.01) (Table 2. Model 2). TMT-A, TMT-B, fluid cognition, and RAVLT short-term delayed recall were not significantly associated with the interaction of age (linear) and sleep. The interaction between the nonlinear age term (age2) and PSQI was significantly associated with crystallized cognition (β = 0.08; Std. CI = 0.01 – 0.15; p = 0.02) and TMT-B (β = −0.08; Std. CI = −0.15 to −0.02; p = 0.02) (Table 2. Model 2), and these findings remained significant after the false discovery rate correction.
Finally, when the estimates of PSQI’s association with cognitive performance (y-axis) were visualized with age (x-axis) for TMT-B and crystallized cognition (Figures 1 and 2, along with 95% confidence intervals), the sensitive age ranges were determined to be ages 50–75 years for TMT-B and ages 66–70 years for crystallized cognition, indicating that poorer sleep quality and worse cognitive performance were most strongly associated in these age ranges.
Figure 1. PSQI total interaction with age for TMT-B.
Figure 2. PSQI total interaction with age for crystallized cognition.
Discussion
The present study examined the relationship between sleep quality, age, and cognition. We first examined the association between sleep quality and cognitive performance and found that worse sleep quality was significantly associated with poorer performance on a measure of cognitive set shifting and speed (TMT-B). Furthermore, we tested the hypothesis that age would significantly modify the relationship between sleep quality and cognitive performance, using age as both linear and nonlinear quadratic terms. Our results showed that there was a significant interaction between the quadratic age term and sleep quality in cognitive set shifting and crystallized cognition. These findings suggest that worse sleep quality begins to negatively affect set-shifting abilities at the age of 50 years, with this effect peaking around the age of 62 years, and crystallized cognition may be most associated with sleep quality for individuals between 66 and 70 years. There was also a significant interaction between the linear age term and sleep quality on crystallized cognition. These findings suggest that age could modify the association between sleep quality and cognitive performance and that there are more sensitive age ranges, namely midlife and early late-life, for when sleep-related cognitive changes are evident.
Identification of these sensitive age ranges suggests that individuals in younger ages (e.g. under 50 years) may be able to preserve cognitive performance in certain domains even in the presence of sleep disturbance. In contrast, older individuals (e.g. aged >50 years for Trails B) may experience a more salient negative impact from sleep disturbance. This finding is somewhat consistent with a previous study that demonstrated the preservation of multiple aspects of frontal network functioning, such as sustained attention, inhibition, and decision-making in sleep-restricted young adults (ages 18–35 years) (Schaedler et al., Reference Schaedler, Santos, Vincenzi, Pereira and Louzada2018) and another study that indicated the impairment of similar cognitive domains in sleep-deprived middle-aged adults ranging from ages 50 to 60 years (Webb, Reference Webb1985). Nevertheless, our findings contradicted other studies which suggested that cognition in older adults (e.g. ages 55–65 years) may be resilient to sleep deprivation (Bliese et al., Reference Bliese, Wesensten and Balkin2006; Stenuit and Kerkhofs, Reference Stenuit and Kerkhofs2005). Our results also indicate that within older adulthood, sleep’s association with cognitive performance becomes notably reduced in the oldest old group (mid-70s and older).
The sensitive age ranges of 50–75 years (Trails B) and 66–70 (Crystalized Cognition) may be supported by age-related changes in sleep and sleep need. As adults may need less sleep as they age, the association between sleep-related neurobiological mechanisms (e.g. slow-wave-dependent cognitive performance) and cognition in older adults may be weaker compared to young and middle-aged adults (Scullin and Bliwise, Reference Scullin and Bliwise2015). Our findings are in line with previous literature that showed sleep-related executive function impairments may begin as early as midlife (Wilckens et al., Reference Wilckens, Woo, Kirk, Erickson and Wheeler2014). Furthermore, these data are consistent with the literature on modifiable risk factors for age-related cognitive impairments that highlight the importance of risk management in midlife (Barnes and Yaffe, Reference Barnes and Yaffe2011; Nishtala et al., Reference Nishtala2014).
While sleep quality and the interaction between sleep quality and age were not associated with poorer performance on NIH Toolbox fluid cognition tests, an additional test of fluid cognition, TMT-B, demonstrated such an association. The TMT-B assays different executive functioning subdomains, such as cognitive set shifting, and was demonstrated to measure a separate construct than the NIH fluid cognition measures (Ott et al., Reference Ott2022; Scott et al., Reference Scott, Sorrell and Benitez2019). To this point, TMT-B’s association with the sleep quality and age interaction term was consistent with findings suggesting that performance on executive functioning tasks is sensitive to sleep deficits in early older adulthood (Wilckens et al., Reference Wilckens, Woo, Kirk, Erickson and Wheeler2014). Declines in executive cognitive control tasks may be especially impacted by sleep deprivation due to reliance on the prefrontal cortex functioning (Harrison and Horne, Reference Harrison and Horne1998, Reference Harrison and Horne1999; Muzur et al., Reference Muzur, Pace-Schott and Hobson2002; Wilckens et al., Reference Wilckens, Woo, Kirk, Erickson and Wheeler2014), which may become evident when performing everyday tasks (Rana et al., Reference Rana2018; Waters and Bucks, Reference Waters and Bucks2011)
Though crystallized cognition has been shown to remain stable (or improve) as individuals age (Dzierzewski et al., Reference Dzierzewski, Dautovich and Ravyts2018), as was the case in our study, the interaction between sleep quality and age (both linear and quadratic) was significantly associated with poorer crystallized cognition performance. This finding could be due to older adults experiencing difficulties with retrieval (Craik and Bialystok, Reference Craik and Bialystok2006; Kurdziel et al., Reference Kurdziel, Mantua and Spencer2017). For instance, word reading and recognition tasks that make up “crystallized cognition” in the current neuropsychological battery may depend on the information that was acquired throughout the lifespan but may also reflect the ability to retrieve previously learned information. As there are few studies conducted on sleep and crystallized cognition, further studies are needed to elucidate the relationship and mechanism between sleep and crystallized cognition in the brain aging process.
A strength of this paper is the use of age as a continuous variable (both as a linear and quadratic term), which may help clarify previous contradictory findings when comparing sleep and cognition between two distinct age groups. For instance, while we found that older individuals’ cognition generally declines, poor sleep quality may be especially problematic for the maintenance of crystallized cognition and performance on cognitive set-shifting tasks during midlife and early late-life. Furthermore, examining quadratic age terms in sleep and cognition research is novel and allowed us to identify sensitive age ranges when sleep quality is most strongly associated with cognition. Finally, the current study includes multiple domains of cognitive functioning, which allows us to parse out the relationships between sleep quality with different cognitive domains. Our findings have important clinical implications in that they suggest that timely detection and intervention of sleep disturbance could benefit cognitive health, particularly in midlife to early older adulthood. Although prolonged use of pharmacological sleep treatments, such as benzodiazepines and non-benzodiazepine receptor agonists, are associated with poorer cognitive outcomes and adverse side effects (e.g. drowsiness and memory impairment) (Schroeck et al., Reference Schroeck2016), nonpharmacological treatments, such as cognitive-behavioral therapy for insomnia, could be effective without side effects and have superior long-term benefits compared with commonly prescribed sleep aids (Morin et al., Reference Morin, Colecchi, Stone, Sood and Brink1999).
A limitation of the paper is that it used a cross-sectional sample. Thus, we cannot make conclusions about the causal relationship between sleep, aging, and cognition. Another limitation of the paper is that we used a subjective rating of sleep quality (PSQI), which may capture sleep less accurately compared with polysomnography. Nonetheless, the PSQI is the most common measure of sleep quality, particularly in large-scale studies (Pilz et al., Reference Pilz, Keller, Lenssen and Roenneberg2018). The NIH Toolbox Fluid Cognition Composite is limited in the cognitive domains that it assesses and does not include a strong measure of memory and recall due to a lack of a delayed memory component. Similarly, it may have lower construct validity in domains such as attention, executive function, and processing speed and estimate different levels of performance compared to standard neuropsychological tests (Ott et al., Reference Ott2022; Scott et al., Reference Scott, Sorrell and Benitez2019). To address some of these limitations, we additionally examined TMT-B, which assesses cognitive set shifting. It is also notable that the effect sizes for our findings are small; nonetheless, our sample is adequately powered to answer our research question, and heterogeneity and large unexplained variance in cognition are expected in large-scale, multisite studies as those in the HCP-A project.
Lastly, our sample consisted of predominantly White and highly educated participants (i.e. average years of education equivalent to a master’s degree), which limits generalizability to a broader population.
Conclusion
In summary, findings from this study suggest the relationship between sleep quality and cognitive performance may be modified by age and related mechanisms. There may be a sensitive period, encompassing midlife to early late-life, that increases the risk of sleep-related cognitive performance decrements.
Conflict of interest
Daniel Cohen: Nothing to Disclose
Hyun Kim: Nothing to Disclose
Alina Levine: Nothing to Disclose
Davangere Devanand: Grant support: National Institute on Aging, Alzheimer’s Association.
DSMB Chair: BioExcel Therapeutics.
Scientific Advisory Board member: Eisai, Biogen, Corium, Genentech, Acadia, Jazz
Pharmaceuticals, Tau Rx.
Seonjoo Lee: Nothing to Disclose
Terry Goldberg: Grant support: National Institute on Aging
Description of authors’ roles
D. Cohen and H. Kim designed the study and oversaw the manuscript preparation. A. Levine conducted statistical analyses and assisted with manuscript preparation. D. Devanand assisted with manuscript preparation and revisions. S. Lee conducted statistical analyses and assisted with manuscript preparation. T. Goldberg assisted with study design, manuscript preparation, and revisions.
Acknowledgements
The authors thank all Human Connectome Project-Aging participants. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Dr Kim is supported by the National Institute on Aging (1K23AG080117-01). Dr Lee is supported by R01AG062578. Data and/or research tools used in the preparation of this manuscript were obtained from the National Institute of Mental Health (NIMH) Data Archive (NDA). NDA is a collaborative informatics system created by the National Institutes of Health to provide a national resource to support and accelerate research in mental health. Dataset identifier(s): 10.15154/1528398. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH or of the Submitters submitting original data to NDA.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1041610223000911.
