To investigate interactions between high and low amounts of sleep and other predictors of cognitive performance.
We used four cognitive tests to determine whether sleep time interacted with age, personal history of a memory problem, parental history of a memory problem, or personal concerns about memory and were associated with cognitive performance. Data were collected from an internet-based cohort study. We used an ordinary least squares regression with restricted cubic splines, controlling for demographic variables and comorbidities.
We found significant nonlinear interactions between (1) total sleep time and age and (2) total sleep time and personal history of a memory problem and cognitive performance. Short and long sleep durations and self-reported memory complaints were associated with poorer performance on a test of attention and this was true to a greater degree in younger and older adults. A repeat analysis excluding subjects reporting dementia was significant only for the test of attention.
These results extend existing data on sleep duration and cognition across the lifespan by combining in a single study the results from four specific cognitive tests, both younger and older adults, and four self-reported risk factors for cognitive impairment. Longitudinal studies with biomarkers should be undertaken to determine whether causal mechanisms, such as inflammation or amyloid buildup, account for these associations.
Mohlenhoff BS, Insel PS, Mackin RS, Neylan TC, Flenniken D, Nosheny R, Richards A, Maruff P, Weiner MW. Total sleep time interacts with age to predict cognitive performance among adults. J Clin Sleep Med. 2018;14(9):1587–1594.
Current Knowledge/Study Rationale: Sleep durations that are comparatively short and long have been associated with poorer cognitive performance but little is known about how total sleep time interacts with other risk factors for poorer cognitive performance. Furthermore, most studies to date have considered only short or long sleep durations, not both.
Study Impact: For the first time, we show that individuals who were concerned about their memory and reported short and long sleep durations performed significantly worse on standardized tests of memory and attention than their peers who reported mid-range sleep durations. Additionally, after excluding subjects with dementia, older individuals who reported short and long sleep durations performed significantly worse on a standardized test of attention than their peers who reported mid-range sleep durations, raising the possibility that older adults are especially susceptible to the effects of irregular sleep durations.
Studies have shown an inverted U-shaped relationship between total sleep time (TST) and cognition such that individuals with TSTs closer to the mean demonstrated better performance than their peers with shorter or longer TST.1,2 The same U-shaped relationship between cognition and TST has been observed in older adults as well.3,4 In addition to TST, three other correlates of cognitive performance are age (independent of TST), having a subjective memory complaint or a memory problem, and having a family history of memory problems or dementia.5–7 No published study has tested whether the association between long and short TST and cognitive dysfunction interacts with these other risk factors (subjective memory complaint, or personal or family history of memory problems) for poorer cognitive performance. Moreover, most studies of TST and cognition have used linear regression analyses using a quadratic term, without evaluating other potential nonlinear shapes of the curve.8
The purpose of this study was to test for the presence of a relationship between TST and cognition and determine whether family history, memory concerns, or memory problems show significant interactions with TST in relationship to cognitive function. We hypothesized that self-reported TST would be correlated, in an inverted U-shaped manner, with cognition in adults with other risk factors for poorer cognition, namely (1) a personal history of a memory problem or concern or (2) a family history of a memory problem or dementia. A secondary goal was to determine whether any such relationships depended on age. We used a large internet-based sample that included at-home cognitive assessments and extensive self-reported behavioral and health information. We used a cubic spline analysis for continuous data to allow for the possibility of nonlinear relationships.
Data were drawn from a large internet-based cohort study called the Brain Health Registry (BHR). Launched in January 2014, the goal of the BHR is to facilitate the development of treatments for Alzheimer disease and other brain diseases. The BHR collects detailed health information and cognitive data online from a large pool of registrants interested in participating in neuroscience research in order to create a well-characterized cohort that can be accessed for future clinical trials. After registration and consent (the BHR is endorsed by the institutional review board at the University of California, San Francisco), participants complete questionnaires asking about demographics, overall health, medication use, memory, family history of dementia, mood, sleep, diet, and exercise.9 Early advertising was accomplished through networking presentations at scientific conferences and presence at dementia-related events open to the public. Participation was voluntary and participants received no compensation for completing online questionnaires and tests.
At the time of this study, 5,483 individuals were enrolled in the BHR who had answered a question about sleep— “During the past month, how many hours of actual sleep did you get at night? (This may be different than the number of hours you spend in bed.)”—in addition to the three questions pertaining to the other parameters of interest for this study, including questions related to dementia risk. Of this number, cognitive measures were available for 4,012 individuals. The sample used for this study consisted of 3,926 individuals who had provided age and sleep data, underwent cognitive tests, and for whom demographic data were also available. Subjects provided written consent to participate and all work has been approved by the institutional review board at the University of California, San Francisco.
Respondents listed up to nine medications using write-in fields. We identified patients taking either benzodiazepine or nonbenzodiazepine sedative-hypnotic medications. We used the restricted Damerau-Levenshtein distance from the search terms10 to identify correct and incorrect spellings of medications available in the United States.11
To test our hypothesis using the BHR data, we identified two items on the set of initial screening questions that might indicate increased risk of dementia. These were: (1) having personal concerns about their own memory (OMC) or (2) having had a parent with a memory problem including dementia (PMP).12,13 We also considered respondents who had reported having a personal history of having a memory problem including dementia (OMP) being diagnosed. We controlled for age, number of years of education, sex, scores on the Geriatric Depression Scale, use of sleep medications, and presence or absence of obstructive sleep apnea (with responses to the question, “Have you ever been diagnosed with sleep apnea?”). This information was gathered at the time of initial baseline questionnaire administration. We included subjects between the ages of 18 and 90 years. We ran all analyses first including and then excluding respondents who indicated on their medical history questionnaire that they had dementia (n = 21), Parkinson disease (n = 62), schizophrenia (n = 5), or stroke (n = 63).
Participants also complete online neuropsychological tests including the Cogstate Brief Battery in an unsupervised setting. The Cogstate Brief Battery is a computer-based collection of cognitive tests that have been validated under supervised and unsupervised conditions in a variety of populations.14–19 Cogstate data included results from four tests, representing four distinct cognitive domains.
The One-Back (ONB) task is a test of working memory.20 For this task participants are asked “is this card the same as that on the immediately previous trial?” The task has 30 trials. The primary outcome variable for this test is the mean of log-transformed response times for correct answers.
The Identification Test (IDN) is a measure of visual information processing speed (attention). For this test the participant is asked “Is the card red?” and must press either “yes” or “no” as cards are displayed. The primary outcome for this test is reaction time in milliseconds normalized using a logarithmic base 10 (Log 10 transformation).
The One Card Learning (OCL) test is a measure of visual learning and memory based on a recognition paradigm previously described.21 For this task participants are asked, “Have you seen this card before?” The test contains 80 trials and the primary outcome variable is the proportion of correct responses (accuracy) normalized using an arcsine transformation.
The Detection Task (DET) is a simple reaction time measuring psychomotor speed. In this test the participant must answer the question “has this card turned over?” Participants are to press the spacebar as soon as the card flips over and the test ends after completion of 35 trials. The primary outcome variable for this test is reaction time in milliseconds normalized using a logarithmic base 10 (Log 10 transformation).
The relationships between performance on Cogstate tests and other variables were modeled using ordinary least-squares regression. Performance on each test was modeled separately and included age, sex, education, score on the Geriatric Depression Scale, presence of sleep apnea, sleep medication, and number of hours slept as predictors. Interactions between sleep and (1) family history, (2) memory concern, (3) memory problem, and (4) age were also evaluated, separately. Continuous predictors (age and hours slept) were assessed for potential nonlinear associations with cognition using restricted cubic splines. Answer categories allowed subjects to report TST between 4 and 12 hours in half-hour increments and extremely low and high sleep durations as “ < 4” and “ > 12” hours. The Akaike Information Criterion (AIC) was used to select the optimal number of spline knots.22 Both age and hours slept consistently demonstrated significant nonlinear associations with all cognitive responses and were modeled with the two parameters resulting from three-knot splines. AIC was also used to test for the significance of the association between sleep and cognition. Associations were considered significant if the inclusion of the variable or interaction of interest in the model resulted in a drop of AIC of 2 or more. Two-sided P values from F tests are reported in Table 1. Model fits were inspected by an analysis of residuals. P values were not corrected for multiple comparisons. All analyses were done in R v3.1.1 (www.r-project.org).
Associations between hours of sleep and potential covariates were evaluated using the Wilcoxon rank-sum test for categorical variables (sex, sleep apnea, sleep medication, family history of dementia, diagnosed memory problem, memory concern) and Spearman correlation for either continuous or ordered categorical variables (age, education). Education was parameterized as an ordered factor with three groups: some college or less, college graduate, and graduate school.
The demographic data in Table 1 include the 3,926 subjects in BHR who reported on their age, sleep, and demographics, and completed at least one cognitive task (54 subjects completed one task and the remainder completed all 4 tasks). The average age was 60 years (standard deviation [SD] 13.4). Of this number, 70.4% were female. A total of 80% of respondents (n = 3,103) reported having either a bachelor's degree or higher, 21% (n = 823) reported having earned a 2-year college degree or less, and 3% (n = 121) reporting having taken no college courses at all. A total of 11.5% of respondents endorsed having obstructive sleep apnea. About half of respondents (49.4%) reported having a parent with a memory problem or dementia. The mean age of respondents endorsing a parent with a memory problem or dementia was 62 (SD 10.5) years, with 21.3% of this group reporting less than a college education. A total of 4.2% of subjects reported taking benzodiazepine or nonbenzodiazepine sedative-hypnotic sleep medications. There was a statistically significant but small difference between the age of subjects who did not report memory problems in their parents (average 59 years) and those who did (62 years). The correlation between age and personal memory problems did not reach statistical significance (P = .38).
Table 2 shows that most respondents (88.5%) reported that they usually slept between 6 and 9 hours nightly. TST showed positive correlation with age (r = .085, P < .001). Those respondents reporting a personal or parental history of memory problem or dementia had no difference in TST compared to their peers who did not report a personal or parental or history of memory problem or dementia. Respondents who indicated that they were concerned about their memory reported a TST that was 15 minutes shorter than those of their peers (6.99 versus 7.14, P < .001).
All four cognitive domains tested showed significant nonlinear relationships with hours of sleep (Table 2 lists statistical test values and Figure 1 and Figure 2 demonstrate the nonlinear relationship between cognition and sleep). Performance was best at a TST in a midrange—approximately 8.0 hours, depending on the particular test—whereas performance was significantly lower at higher and lower TST.
Cognitive performance and TST by age.
Graphs of cognitive performance in two tested domains based on self-reported TST and age. Shown are tests of working memory (ONB; top) and processing speed (IDN, attention; bottom). Age was modeled continuously, but for graphical purposes, three age groups are plotted. Solid lines indicate means and dashed lines indicate 95% confidence intervals. TST = total sleep time.
Cognitive performance and TST by age.
Cognitive performance and TST by reported personal memory problem.
Graphs of cognitive performance in two tested domains based on self-reported history of having a memory problem/dementia/Alzheimer's disease. Shown are tests of working memory (ONB; top) and processing speed (IDN, attention; bottom). Solid lines indicate means and dashed lines indicate 95% confidence intervals. TST = total sleep time.
Cognitive performance and TST by reported personal memory problem.
Sleep and Age
We found a nonlinear interaction between age and TST that was associated with better cognitive performance at mid-range TST (Table 2 and Figure 1) but only in the cognitive domains of working memory and attention. Subjects in the middle range of our cohort in terms of age demonstrated better cognitive performance at high and low sleep durations compared with both their younger and older peers. In other words, the U-shaped curves describing cognitive performance in relation to TST were less pronounced. The results of tests for combined effects of age and TST on measures of learning and psychomotor speed did not achieve significance.
Sleep and Other Variables
We found significant nonlinear interactions between TST and OMP on memory and attention (Table 2 and Figure 2). Subjects reporting a memory problem performed worse on measures of memory and attention at both higher and lower TST than their peers who did not report personal memory problems (Figure 2). The results of tests for effects of OMP and TST on measures of learning and psychomotor speed did not achieve significance. All tests related to reported history of having a PMP or a OMC did not achieve significance.
Post Hoc Analyses
We conducted a post hoc analyses excluding subjects who endorsed “dementia” on a health questionnaire. After these subjects were removed, TST was associated with performance on tasks of working memory, attention, and psychomotor speed but not visual learning. An interaction between age and sleep was associated with attention only (Figure S1 in the supplemental material). In each case, subjects reporting middle-range TST performed the best. Other tests showed a similar trend but did not reach the level of statistical significance (Figure S2 in the supplemental material).
Our data show that concern about memory problems and age altered the relationship between sleep cognitive performance. In this study of adults, in each scenario tested, subjects in the middle range of our cohort in terms of age demonstrated better cognitive performance at high and low sleep durations compared with both their younger and older peers who reported similar sleep times (importantly, the same U-shaped relationship between sleep and cognitive performance existed at every age). Our work expands on previous findings of this same relationship between TST and cognition using four specific tests of cognitive performance. In the realms of memory and attention, we found that individuals who were concerned about their memory and reported long and short TST demonstrated cognitive performance that was significantly poorer than that of their peers who reported mid-range sleep durations. Additionally, TST interacted with age such that younger and older subjects reporting relatively short and long TST demonstrated the poorest performance on tests. Inversely, subjects in the middle age range who reported middle-range TST demonstrated the best cognitive performance in a group that was similar demographically and in terms of dementia risk factors.
When we repeated our analyses excluding individuals who had reported dementia on the medical screening portion of their baseline questionnaires, the only interaction that still reached statistical significance was for the relationship between TST and age and poorer performance on tests of attention. Thus, although these data are only partially supportive of our initial hypothesis, they demonstrate that TST interacts with at least one other specific predictor of cognitive performance, even in subjects denying dementia. Although previous studies have shown relationships between TST and cognitive performance and between age and cognitive performance, our findings demonstrate the complexity of the relationships between age, TST and cognitive performance.
The association between middle-range TST and better cognitive performance may be causal: it may be that short and long amounts of sleep have detrimental effects on cognition. This could mean that, in older individuals and in those in whom a diagnosis has been made who suspect that they have memory problems, it is more important to get “the right amount” of sleep—that is, an optimal average TST that is closer to the population mean. These individuals may be especially susceptible to subtractions from performance by biological factors associated with insufficient sleep. It is difficult to imagine a single disease process that would have the same effect on cognition at both short and long TST.
Studies of insomnia, acute sleep deprivation or chronic partial sleep deprivation all show poorer cognition.23–25 Short sleep duration is associated with a wide range of poorer health outcomes including diabetes and heart disease as well as all-cause mortality.26 We have controlled for a variety of factors that have been previously described as leading to both cognitive and sleep problems. These include age,27 presence of depression,28–30 education level,31 use of sleep medications,32–34 and presence or absence of obstructive sleep apnea.35 Aside from indirectly affecting cognition via these mediating health factors, multiple possible mechanisms could account for the association between shorter sleep duration and poorer cognition, including impaired neural plasticity and inflammation.36,37 Less is known about associations between poor health and cognition and longer sleep durations. There are very few studies of experimental increases in sleep in humans who sleep normally, as this is a problematic intervention. One mechanism by which increased sleep duration might affect cognition is via increased inflammatory mediators such as C-reactive protein.38,39 It is possible that inflammation causes direct harm to memory networks or inhibits consolidation of new memories.40 It is also possible that longer sleep durations associated with poorer cognitive performance do not promote a distinct disease process but instead represent incomplete compensation for impaired sleep quality by increased TST.
Another possibility is that the relationship between cognition and short and long TST seen in our study are both caused by some neurological insult or early disease process that could not be identified in our sample. If a neurological disease leads to deviations in TST as well as cognitive dysfunction, then TST in certain populations may be a useful predictor of poorer cognitive performance. Both short41 and long42,43 TST have been shown to be associated with cognitive decline or subsequent cognitive impairment,44 though this has been less extensively studied and some studies4,45,46 have not found this association. It would be interesting to conduct a similar study while controlling for other known predictors of cognitive decline, disease, and dementia, such as amyloid burden. It is quite possible that short and long sleep durations are associated with poorer cognitive performance for different reasons.
Because all cognitive tests considered depended on attention, it is possible that this cognitive domain drove all other results prior to removing from analyses those subjects reporting dementia. Alternatively, it is possible that only the most robust finding remained true after removing subjects who were most impaired. Previous studies have shown measures of attention to be particularly sensitive to sleep loss.23 Our results suggest that this is true even in the setting of dementia. Importantly, our data showed that performance on a measure of attention is also lower among subjects reporting longer sleep durations.
This study has several important limitations. Volunteer studies are inherently at risk of sampling bias. Our sample was older, more highly educated, and contained more female subjects than other previous studies of sleep duration and cognition and the generalizability of our results may be limited. Additionally, self-reported sleep duration has been shown in some studies to have poor correlation with objectively measured sleep duration.47 It is possible that subjects who had memory concerns may have estimated their sleep times more or less accurately than their peers without memory concerns, introducing bias into our results. Reassuringly, in a study of 78 older adults, accuracy of individuals' estimations of sleep time was not correlated with cognitive performance as measured by the Montreal Cognitive Assessment.48 It is also possible that lack of an objective assessment for sleep apnea resulted in an underappreciation of the prevalence of that condition in our sample, though our finding of 11.5% reporting sleep apnea is higher than other estimates ranging from 2% to 7% in adults.49 However, given that the average age of subjects in this study was 60 years and 70% female, our sample might be expected to have a somewhat higher prevalence of sleep apnea, though perhaps not as high as 28% in men and 20% in women (using an apnea-hypopnea index of greater than or equal to 5) as found in another study of adults of average age 72.5 years.50 Another potential limitation is that asking subjects to complete Cogstate tests under unsupervised conditions introduced a bias into our assessments. The Cogstate tests have been validated in a variety of populations but administration of the tests under unsupervised conditions has been validated only in a sample of younger adults, whereas our sample contained predominantly older adults.14–18
Important next steps in research on cognitive performance and sleep time might include conducting studies similar to ours in which biological data, such as inflammatory markers and brain and cerebrospinal fluid amyloid beta burden, could be considered. Furthermore, outcomes such as mortality and incident dementia will be available for our study subjects, allowing for the consideration of actual risk of cognitive decline in relation to TST and other variables of interest.
Dr. Weiner receives support for his work from the following grants; 2U01AG024904 (NIH/NIA), W81XWH-13-1-0259 (DOD), W81XWH-12-2-0012 (DOD), R01 AG010897 (NIA/NIH), P01 AG19724 (NIH/NIA), R01A G038791 (NIA/NIH), ADNI 2-12-233036 (Alzheimer's Association), R01 MH098062-01 (NIH/NIMH), 1I01CX000798-01A2 (Veterans Administration), W81XWH-14-1-0462 (DOD), W81XWH-09-2-0044 (DOD), 12-12004 (CA Department of Public Health), 5U19AG010482-23 (NIH/NIA) and P50 AG23501 (NIH/NIA). Dr. Weiner is a full-time Professor of Medicine, Radiology, Psychiatry, and Neurology for the University of California San Francisco (UCSF), and Principal Investigator of the Brain Health Registry, and many projects with the aforementioned grant funding, including Alzheimer's Disease Neuroimaging Initiative (ADNI). He has served on the Scientific Advisory Boards for Pfizer, BOLT International, Neurotrope Bioscience, Alzheon, Inc., Alzheimer's Therapeutic Research Institute (ATRI), Eli Lilly, U. of Penn's Neuroscience of Behavior Initiative, National Brain Research Centre (NBRC), India, Dolby Family Ventures, LP, and ADNI. He served on the Editorial Boards for Alzheimer's & Dementia and MRI. He has provided consulting to Synarc, Pfizer, Janssen, Alzheimer's Drug Discovery Foundation (ADDF), Neurotrope Bioscience, Avid Radiopharmaceuticals, Clearview Healthcare Partners, Perceptive Informatics, Smartfish AS, Araclon, Merck, Biogen Idec, BioClinica, and Genentech. He holds stock options with Alzheon, Inc. The following entities have provided funding for academic travel; Pfizer, Neuroscience School of Advanced Studies (NSAS), Kenes, Intl., ADRC, UCLA, UCSD; ADCS, Sanofi-Aventis Groupe, University Center Hospital, Toulouse, Araclon, AC Immune, Nutricia, Eli Lilly, New York Academy of Sciences (NYAS), National Brain Research Center, India for Johns Hopkins Medicine, Consortium for Multiple Sclerosis Centers (CMSC), Northwestern University, Fidelity Biosciences Research Initiative, University of Pennsylvania, The Alzheimer's Association, Merck, ADPD, Alzheimer's Drug Discovery Foundation (ADDF), Tokyo University, Kyoto University, Weill-Cornell University, Rockefeller University, Memorial Sloan-Kettering Cancer Center, and Biogen Idec. Paul Maruff is a full-time employee of Cogstate. Brian S. Mohlenhoff, Philip S. Insel, R. Scott Mackin, Thomas C. Neylan, Derek Flenniken, Rachel Nosheny, and Anne Richards report no conflicts of interest.