http://dx.doi.org/10.5664/jcsm.

3782

Associations between Sleep Duration, Sleep Quality, and

Cognitive Test Performance among Older Adults from Six
Middle Income Countries: Results from the Study on Global
Ageing and Adult Health (SAGE)
Theresa E. Gildner, M.S.1; Melissa A. Liebert, M.S.1; Paul Kowal, Pharm.D.2,3; Somnath Chatterji, M.D.2; J. Josh Snodgrass, Ph.D.1
1
Department of Anthropology, University of Oregon, Eugene, OR; 2World Health Organization, Geneva, Switzerland; 3University of
Newcastle Research Centre on Gender, Health, and Ageing, Newcastle, NSW, Australia

Background: Alterations in sleep architecture are common (> 6-9 h/night) exhibited significantly higher cognitive scores
among older adults. Previous studies have documented than individuals with short sleep (0-6 h/night; p < 0.001) or
associations between sleep duration, sleep quality, and long sleep duration (> 9 h/night; p < 0.001). Self-reported
cognitive performance in older individuals, yet few studies sleep quality was positively correlated with cognitive z-score
S C I E N T I F I C I N V E S T I G AT I O N S

have examined these trends using population-based samples (p < 0.05). Significant sex differences were observed; men
from non-Western societies. The present cross-sectional study generally had higher sleep quality and cognitive scores, while
uses nationally representative datasets from six countries to women reported longer sleep durations.
test several hypotheses related to sleep patterns and cognitive Discussion: This study documented positive correlations
function. between cognitive scores and sleep quality, and between
Methods: Data were drawn from the first wave of the World cognitive z-scores and intermediate sleep duration. These
Health Organization’s study on global ageing and adult health findings are clinically important given the growing rates of
(SAGE), a longitudinal study using samples of older adults dementia and aging populations globally.
(≥ 50 years old) in 6 middle-income countries (China, Ghana, Keywords: cognitive function, aging, sleep duration, sleep
India, Russian Federation, South Africa, and Mexico). Self- quality, dementia
report data provided information on sleep quality and sleep Citation: Gildner TE, Liebert MA, Kowal P, Chatterji S,
duration over the previous 2 nights, and 5 cognitive tests Snodgrass JJ. Associations between sleep duration, sleep
(immediate and delayed verbal recall, forward and backward quality, and cognitive test performance among older adults
digit span, and verbal fluency) were used to create a composite from six middle income countries: results from the study on
z-score of cognitive performance. global ageing and adult health (SAGE). J Clin Sleep Med
Results: Individuals with intermediate sleep durations 2014;10(6):613-621.

D etrimental changes in sleep duration and quality are more

common with increasing age.1,2 These alterations include a
phase advance in circadian rhythms, thought to be the result of
BRIEF SUMMARY
Current Knowledge/Study Rationale: Few studies have assessed the
deterioration in the suprachiasmatic nuclei region of the brain.1 contribution of sleep quality and quantity to cognitive performance in
older adults, especially in non-Western populations. A comprehensive
Older adults also report an increased occurrence of sleep disor- understanding of all factors influencing mental acuity is essential in the
ders associated with disrupted and fragmented sleep patterns. design of disease-modifying therapies.
These conditions often reduce slow wave sleep (SWS) and REM Study Impact: This study evaluated associations between measures
sleep, the deepest and most restorative sleep stages. Disruption of sleep and cognitive performance using large nationally representa-
to SWS and REM sleep can lead to increased daytime sleepi- tive samples of older adults drawn from six middle-income countries,
providing a unique examination of this relationship. Enhanced cognitive
ness and a decreased ability to handle stress.2 functioning was significantly associated with intermediate sleep duration
Age-related changes in sleep may contribute to cognitive and higher sleep quality, suggesting that sleep measures may influence
decline among older individuals, yet this issue has not been cognitive performance in older individuals from different countries.
extensively studied. Both short and long sleep duration have
been linked to reduced cognitive performance.3,4 Sleep depriva-
tion dulls the senses, slows reaction times, and impedes memory of chronic problems with sleep latency (time to sleep onset) and
formation.3 Conversely, long sleep duration may reflect poor maintenance are often associated with adverse consequences,
sleep quality or disturbed sleep patterns, which may impair including drowsiness and reduced performance on cognitive
cognitive performance.4 Furthermore, studies on sleep quality tests.5 Unfortunately, research to date has largely been limited
in older adults indicate that the impairment of normal sleep to high income countries and has typically relied on data from
structure may detrimentally affect cognitive function.5 Reports small and non-representative samples.

613 Journal of Clinical Sleep Medicine, Vol. 10, No. 6, 2014

TE Gildner, MA Liebert, P Kowal et al.
Sociodemographic and lifestyle factors also appear to in the study. In China, strata were defined by eight provinces
strongly influence sleep patterns.6-9 Notably, low socioeco- and two localities (urban or rural), resulting in 16 total strata. In
nomic status (SES) and poor living conditions have been linked Ghana, the sample was stratified by 10 administrative regions
with reduced self-reported sleep quality, increased sleep latency, and two types of locality (urban or rural), resulting in 20 total
and increased time awake after sleep onset in American, British, strata. The Indian strata were defined by six states and two
and Finnish cohorts.7-9 Factors that may be more common as a localities (urban or rural), resulting in 12 total strata. The South
result of low SES (e.g., noise pollution, light exposure, uncom- African sample was stratified by nine providences, two localities
fortable sleeping surfaces, lack of temperature control, occu- (urban or rural), and predominant race/ethnic group (“African/
pational stress) may result in deficient sleep patterns.7-9 Sex Black,” “White,” “Colored,” and “Indian/Asian”); not all
differences also appear to influence sleep and cognitive test combinations of stratification variables were possible, resulting
performance in older individuals, with women reporting more in 50 total strata. The Russian Federation sample was stratified
sleep onset and maintenance problems than men.10 Cultural by the seven federal districts. Finally, in Mexico, strata were
differences also influence sex differences in cognitive perfor- defined by three localities: rural (< 25,000 inhabitants), urban
mance. Specifically, women in many countries receive less (< 100,000 and > 2,499 inhabitants), or metropolitan (> 99,999
schooling than men, likely contributing to observed differ- inhabitants). Analyses in the present study primarily focused
ences in cognitive test performance.4 Cross-cultural studies on older adults. Younger individuals (< 50 years old) were
in this area are needed to assess the extent to which these excluded from the analyses, except in a comparison between
changes represent global patterns or those strongly influenced older and younger (< 50 years old) participants for average
by cultural and environmental conditions. This information is cognitive function score by sex. Face-to-face interviews were
critical since the development of effective dementia treatments used to collect household and individual level data.13
and the ability to minimize age-related cognitive changes are
dependent on the identification of key contributing factors to Ethical Approval
cognitive decline. These clinical interventions have important SAGE was approved by the World Health Organization’s
implications toward addressing the global burden of dementia. Ethical Review Committee. Additionally, each partner organi-
Despite considerable research, questions remain about zation implementing SAGE obtained ethical clearance through
the link between sleep patterns and cognitive performance. their respective institutional review bodies. Informed consent
Although a clear relationship between sleep and cognitive was obtained from all study participants.
decline has been demonstrated, it is difficult to determine which
factor precedes the other. Nonetheless, identifying sleep disor- Independent Variables
ders in older adults is critical since they may have implications Age, sex, and highest level of education completed were
for cognition.11,12 Treatments can then be directed to addressing collected. Age reporting overall was quite good across all coun-
both sleep problems and cognitive decline simultaneously. tries. The United Nations International Standard Classification
The present study examines links between sleep patterns and of Education 1997 (ISCED 97) scheme was used to harmonize
cognitive test performance using data from the first wave of education levels across countries.14
the World Health Organization’s study on global ageing and
adult health (SAGE).13 Data from six middle income countries Sleep Variables
(China, Ghana, India, Mexico, the Russian Federation, and Participants were asked their sleep duration on each of
South Africa) are utilized to construct a more comprehensive the preceding two nights and then asked to rate their sleep
picture of how sleep architecture and cognitive performance quality on a scale of 1 to 5 (1 = very good quality, 5 = very
among older adults varies across these countries. Four hypoth- poor quality) for each night; these measures did not include
eses were tested: (1) age would be inversely correlated with daytime sleep. These results were then reverse coded, so that
cognitive scores; (2) short and long sleep duration would be a higher rating corresponded with good quality sleep and a
significantly associated with poor cognitive test performance; lower rating represented poor quality sleep. In accordance with
(3) good sleep quality would be positively correlated with other sleep studies, the duration values across two nights were
cognitive scores; and (4) women would report poorer sleep averaged together to create a summary measure of sleep length,
quality and longer sleep duration than men, and these tenden- so that mean sleep durations could be compared.4,15 Similarly,
cies would be correlated with reduced performance on cogni- the quality values were also averaged together to compute the
tive tests. sleep quality typical for each individual. Following a standard
approach,4,16,17 individuals in the six countries combined were
divided into three sleep duration categories: short sleep (0-6 h),
METHODS intermediate sleep (> 6-9 h), and long sleep (> 9 h). The total
pooled sample was also sorted into three generated average
Study Design and Participants sleep quality categories: low sleep quality (average rating of
Nationally representative samples of older adults (≥ 50 years 1-2); intermediate sleep quality (average rating of 2.5-3.5), and
old) and comparative samples of younger adults (18-49 years high sleep quality (average rating of 4-5).
old) were drawn from each SAGE country.13 Sampling in each
country was based on a stratified multistage cluster sample Cognition Variables
design. Strata were uniquely defined for each country to ensure Five cognitive performance tests were used to create a summary
the full range of living conditions in that nation was represented variable of cognitive function for each participant. These tests

Journal of Clinical Sleep Medicine, Vol. 10, No. 6, 2014 614

 Sleep and Cognitive Function among Older Adults
included immediate and delayed verbal recall, forward and back- Hypothesis Two: A third linear regression examined if
ward digit span, and verbal fluency. For the immediate verbal average sleep duration categories contributed to variation
recall test, interviewers first read a list of 10 words aloud and in cognitive test performance among older adults. Because
asked the participants to immediately recall as many words as intermediate sleep was expected to be associated with higher
they could in one minute. Three trials of this test were performed. cognitive z-scores, these categories were dummy coded using
Following the third trial, the interviewer administered the other intermediate sleep duration as the reference group and entered
cognitive tests, after which delayed recall ability was assessed in the second step of the regression.
by asking participants to recall the list of words. The digit span Hypothesis Three: Similarly, a fourth linear regression was
tests required participants to repeat back progressively longer conducted to estimate the relative contribution of sleep quality
series of numbers, with the total score the longest digit span categories to cognitive score variation. The sleep quality cate-
repeated without error. This process was then performed with gories were dummy coded and entered in the second step of the
the respondent repeating new sets of progressively longer digit regression; because cognitive scores were expected to increase
spans in reverse (backwards). The verbal fluency test consisted as sleep quality improved, low sleep quality served as the refer-
of naming as many animals as possible in one minute. The ence group.
final score was correct responses minus errors. In accordance Hypothesis Four: A final linear regression was conducted to
with other cognitive studies, composite z-scores were calcu- examine the relative association of country, sex, the interaction
lated to facilitate the comparison of cognitive test performance between country and sex to cognitive z-scores. The six country
between individuals.18,19 Z-scores for each cognitive test were categories were dummy coded using China as the reference
first computed; these five z-scores were then summed for each group (because this country had the highest z-scores for the
individual, resulting in a final composite z-score. set of cognitive tests used in the present study). The country
dummy codes were entered in the second step of the regression,
Statistical Analyses sex in the third step, and the interactions between sex and each
Tests for normality were performed, and no violations were country dummy code were entered in the fourth step.
observed. Parametric tests were subsequently conducted using
SPSS version 20 to test the hypotheses; statistically significant Examination of Relationship between Sleep Duration and
results were those with p < 0.05. Sleep Quality
Spearman correlations (ρ) were used to assess the association
Examination of Country and Sex Differences between average sleep duration and sleep quality. All correla-
Mean sleep duration, quality, and cognitive performance tions were conducted separately for men and women by country.
scores were calculated for males and females in each popula-
tion. Three factorial ANOVAs were used to test if cognitive RESULTS
test performance, sleep duration, and sleep quality varied by
sex and population. Simple effects analyses (with a Bonferroni Descriptive statistics are presented in Table 1. The total mean
adjustment) were utilized to compare males and females within composite cognition z-scores (sexes combined, n = 34,203)
and between countries on average cognitive, sleep duration, ranged from -1.88 in India to 0.93 in China. With the excep-
and sleep quality scores. tion of Russia and Mexico, a significant sex difference was
observed; average composite z-scores were significantly lower
Contribution of Age, Education Level, Sleep Quality, and Sleep for women than men (p < 0.001). Mean sleep duration (sexes
Duration to Cognitive Test Score Variation combined, n = 32,142) ranged from 7.1 h/night in India to 8.6
A series of linear regressions were conducted to test the h/night in South Africa. Women in all countries reported longer
hypotheses and evaluate the relative contribution of age and average sleep duration than men (p < 0.001); these differences
sleep patterns to cognitive score variation, while controlling were significant in all countries except Russia and Mexico.
for education level. Education level was included as dummy Average subjective sleep quality ratings (sexes combined,
codes in first step of the linear regression analyses to control n = 35,552) ranged from 3.3 in Russia to 4.0 in Mexico. In each
for the contribution of education on variation in cognitive test country the average sleep quality rating was significantly lower
performance. for women than men (p < 0.05)
Hypothesis One: The first linear regression estimated the
relative association of age on cognitive z-scores. Participants Cognitive Test Performance by Country and Sex
from all countries were pooled and divided into younger indi- Country (p < 0.001) and sex (p < 0.001) had strong main
viduals (18-49 years old) and older individuals (≥ 50 years effects on cognitive test performance in the present study. A
old). These two groups were coded into a dichotomous variable significant interaction (p < 0.001) was also apparent between
(younger = 1, older = 2), which was entered in the second step country and sex. Pairwise comparisons with a Bonferroni
of the regression, and sex was entered in the third step. A second adjustment indicated that cognition z-scores differed signifi-
linear regression was performed to determine if categories of cantly for each country pair (p < 0.001) except between
age by decade (50-59; 60-69; 70-79; 80+ years old) contributed China and Russia (D = 0.10, p = 1.00). A comparison of mean
significantly to the model. This variable was dummy coded composite cognition scores between sexes indicated that men
using the youngest age set (50-59 years old) as the reference exhibit significantly higher scores compared to women in all
group and entered in the second step of the regression, while countries (p < 0.001), except Russia (D = 0.20, p = 0.09) and
sex was entered in the third step. Mexico (D = 0.12, p = 0.22; Table 1).

615 Journal of Clinical Sleep Medicine, Vol. 10, No. 6, 2014

TE Gildner, MA Liebert, P Kowal et al.

Table 1—Mean cognition z-scores, sleep duration (h/night), and subjective sleep quality for men, women, and sexes combined
in each country with sample size (n).
Composite Cognitive Score Sleep Duration (h/night) Subjective Sleep Quality
China Total 0.93 (n = 12,518) 7.56 (n = 12,794) 3.51 (n = 12,887)
Men 1.30 (n = 5,839) 7.62 (n = 6,009) 3.61 (n = 6,040)
Women 0.61 (n = 6,679) 7.51 (n = 6,785) 3.42 (n = 6,847)
Mean Sex Difference 0.69 *** -0.11 *** 0.19 ***
Ghana Total -0.60 (n = 4,114) 7.73 (n = 3,679) 3.87 (n = 4,276)
Men 0.08 (n = 2,168) 7.53 (n = 1,938) 3.90 (n = 2,233)
Women -1.36 (n = 1,946) 7.95 (n = 1,741) 3.84 (n = 2,043)
Mean Sex Difference 1.44 *** -0.42 *** 0.07 **
India Total -1.88 (n = 6,291) 7.11 (n = 6,526) 3.63 (n = 6,551)
Men -1.16 (n = 3,212) 6.98 (n = 3,285) 3.71 (n = 3,298)
Women -2.64 (n = 3,079) 7.24 (n = 3,241) 3.54 (n = 3,253)
Mean Sex Difference 1.48 *** -0.26 *** 0.17 ***
South Africa Total -0.15 (n = 3,544) 8.58 (n = 3,507) 3.84 (n = 3,617)
Men 0.29 (n = 1,500) 8.48 (n = 1,492) 3.87 (n = 1,530)
Women -0.47 (n = 2,044) 8.66 (n = 2,015) 3.81 (n = 2,087)
Mean Sex Difference 0.76 *** -0.18 *** 0.06 *
Russia Total 0.82 (n = 3,374) 7.59 (n = 3,506) 3.28 (n = 3,813)
Men 0.95 (n = 1,190) 7.59 (n = 1,241) 3.43 (n = 1,354)
Women 0.75 (n = 2,184) 7.59 (n = 2,265) 3.20 (n = 2,459)
Mean Sex Difference 0.20 -0.01 0.24 ***
Mexico Total -0.90 (n = 4,362) 7.66 (n = 2,130) 3.98 (n = 4,408)
Men -0.82 (n = 1,698) 7.59 (n = 843) 4.00 (n = 1,744)
Women -0.95 (n = 2,664) 7.71 (n = 1,287) 3.96 (n = 2,664)
Mean Sex Difference 0.12 -0.12 0.05 *
The number of asterisks indicates the level of significance (* p < 0.05, ** p < 0.01, *** p < 0.001) in the factorial ANOVA.

Table 2—Pairwise comparisons of the difference in average sleep duration (h/night) for each country pair (separated by sex).
China Ghana India South Africa Russia Mexico
Males
China – 0.09* 0.64*** -0.86*** 0.03 0.03
Ghana -0.09* – 0.55*** -0.95*** -0.06 -0.06
India -0.64*** -0.55*** – -1.50*** -0.61*** -0.62***
South Africa 0.86*** 0.95*** 1.50*** – 0.89*** 0.89***
Russia -0.03 0.06 0.61*** -0.89*** – -0.004
Mexico -0.03 0.06 0.62*** -0.89* 0.00 –
Females
China – -0.44*** 0.27*** -1.15*** -0.09* -0.20***
Ghana 0.44*** – 0.71*** -0.71*** 0.35*** 0.24***
India -0.27*** -0.71*** – -1.42*** -0.36*** -0.47***
South Africa 1.16*** 0.71*** 1.42*** – 1.07*** 0.95***
Russia 0.09* -0.35*** 0.36*** -1.07*** – -0.11*
Mexico 0.20*** -0.24*** 0.47*** -0.95*** 0.11* –
Comparisons are statistically significant at: * p < 0.05, ** p < 0.01, *** p < 0.001. Each value represents the difference between the country in the column and
the country listed in row at the top of the table.

Sleep Duration by Country and Sex (p < 0.001). Conversely, men and women in India reported
Country (p < 0.001) and sex (p < 0.001) had strong main lower sleep duration compared to all other countries (p < 0.001;
effects on sleep duration. A significant interaction (p < 0.001) Table 2). Mean female sleep durations between countries were
was also identified between country and sex. Women reported all significant (p < 0.05; Table 2).
significantly higher mean sleep duration than men (p < 0.001) in
all countries except in Russia (D = -0.01, p = 0.89) and Mexico Sleep Quality by Country and Sex
(D = -0.12, p = 0.08; Table 1). Men and women in South Africa Country (p < 0.001) and sex (p < 0.001) had strong main
reported longer sleep duration compared to all other countries effects on sleep quality. A significant interaction (p < 0.001)

Journal of Clinical Sleep Medicine, Vol. 10, No. 6, 2014 616

 Sleep and Cognitive Function among Older Adults

Table 3—Pairwise comparisons of the difference in average sleep quality rating for each country pair (separated by sex).
China Ghana India South Africa Russia Mexico
Males
China – -0.29 ** -0.10 *** -0.27 *** 0.18 *** -0.40 ***
Ghana 0.29 *** – 0.19 *** 0.03 0.47 *** -0.10 ***
India 0.10 *** -0.19 *** – -0.17 *** 0.28 *** -0.30 ***
South Africa 0.27 *** -0.03 0.17 *** – 0.44 *** -0.13 ***
Russia -0.18 *** -0.47 *** -0.28 *** -0.44 *** – -0.57 ***
Mexico 0.40 *** 0.10 *** 0.30 *** 0.13 *** 0.57 *** –
Females
China – -0.41 *** -0.12 *** -0.39 *** 0.23 *** -0.53 ***
Ghana 0.41 *** – 0.29 *** 0.02 0.64 *** -0.12 ***
India 0.12 *** -0.29 *** – -0.27 *** 0.35 *** -0.41 ***
South Africa 0.39 *** -0.02 0.27 *** – 0.62 *** -0.15 ***
Russia -0.23 *** -0.64 *** -0.35 *** -0.62 *** – -0.76 ***
Mexico 0.53 *** 0.12 *** 0.41 *** 0.15 *** 0.76 *** –
Comparisons are statistically significant at: * p < 0.05, ** p < 0.01, *** p < 0.001. Each value represents the difference between the country in the column and
the country listed in row at the top of the table.

was also observed between country and sex. Men reported

Table 4—Spearman correlation coefficients of sleep duration
significantly higher sleep quality scores than women in all and sleep quality for each country (separated by sex).
countries (p < 0.05) (Table 1). Men and women in Mexico Men Women
reported significantly higher sleep quality values compared to China 0.32 *** 0.32 ***
all other countries (p < 0.001) (Table 3). Ghana 0.30 *** 0.18 ***
India 0.19 *** 0.25 ***
Age and Cognitive Test Performance Mexico 0.29 *** 0.24 ***
The results of the linear regression examining the contribution Russia 0.10 *** 0.28 ***
of age to variation in cognitive test performance indicated that South Africa 0.16 *** 0.18 ***
education level explained a significant amount of the variance in
cognitive scores (R2 = 0.127, p < 0.001). Increasing education rela- *** p < 0.001.
tive to the lowest schooling category was associated with linear
advances in cognitive test performance, highest at “completed
college/university” and then slightly lower for “completed Sleep Patterns and Cognitive Test Performance
post-grad degree.” Adding the dichotomous younger/older age A third linear regression was used to assess the contribution
variable explained a significant amount of additional variance of sleep duration to cognitive score variation while controlling
(R2 change = 0.013, p < 0.001). Older individuals (p < 0.001) for education. Education level explained a significant amount
exhibited lower cognitive test scores than their younger coun- of the variance in cognitive scores (R2 = 0.101, p < 0.001).
terparts. Adding sex to the model explained a marginal amount Linear advances in cognitive scores were evident as educa-
of additional variance (R2 change < 0.001, p < 0.001); this very tion level increased, peaking at “completed college/university”
small but statistically significant finding is likely due to the large and then decreasing slightly for “completed post-grad degree.”
sample size. Women exhibited lower cognitive scores relative to Adding sleep duration explained a significant amount of addi-
men on this battery of cognitive tests (p < 0.001; Table 5). tional variance (R2 change = 0.005, p < 0.001). Short sleepers
The results of a second linear regression examining the (< 6 h) exhibited significantly lower cognitive scores (p < 0.001)
contribution of age categories by decade to composite cognitive than intermediate sleepers (> 6-9 h). Similarly, long sleepers
z-scores indicated that education level explained a significant (> 9 h) displayed significantly lower cognitive test performance
10.6% of the variance in cognitive scores (R2 = 0.106, p < 0.001). (p < 0.001) than intermediate sleepers (see Table 5).
Increasing education relative to the lowest schooling category Significant differences were also observed in the linear
was associated with linear advances in cognitive scores, highest regression examining cognitive performance between the sleep
at “completed college/university” and then slightly lower quality categories. Education level explained a significant
for “completed post-grad degree.” Adding the age by decade amount of the variance in cognitive scores (R2 change = 0.106,
dummy codes explained a significant amount of additional vari- p < 0.001). Increasing education relative to the lowest schooling
ance (R2 change = 0.072, p < 0.001). Cognitive test performance category was associated with linear advances in cognitive
decreased linearly across age groups, in which individuals aged scores, peaking at “completed college/university” and then
60-69, 70-79, and 80+ years exhibited lower cognitive scores decreasing slightly for “completed post-grad degree.” Adding
than their younger (50-59 year old) counterparts (p < 0.001; sleep quality explains a significant amount of additional vari-
Table 5). Adding sex to the model explained a nonsignificant ance (R2 change = 0.007, p < 0.001). When compared to low
amount of variance (R2 change = 0.000, p = 0.145). quality sleepers (rating of 1-2), cognitive test performance

617 Journal of Clinical Sleep Medicine, Vol. 10, No. 6, 2014

TE Gildner, MA Liebert, P Kowal et al.

Table 5—Multiple regression models for prediction of cognitive scores (sexes and populations combined).
Variable Coefficients (SE) β p Model r 2 / p
Composite Cognitive Z-score 0.14 / < 0.001***
Constant 1.464 (0.112) < 0.001
Education Level : Less than primary vs. completed primary 0.827 (0.052) 0.107 < 0.001
Less than primary vs. completed secondary 1.840 (0.054) 0.229 < 0.001
Less than primary vs. completed high school 2.475 (0.053) 0.309 < 0.001
Less than primary vs. completed college/university 3.600 (0.070) 0.310 < 0.001
Less than primary vs. completed post-grad 3.080 (0.185) 0.091 < 0.001
Young-Old Category -0.940 (0.044) -0.117 < 0.001
Sex -0.139 (0.036) -0.021 < 0.001
Composite Cognitive Z-score 0.18 / < 0.001***
Constant 0.330 (0.045) < 0.001
Education Level : Less than primary vs. completed primary 0.723 (0.054) 0.096 < 0.001
Less than primary vs. completed secondary 1.557 (0.058) 0.192 < 0.001
Less than primary vs. completed high school 2.027 (0.057) 0.253 < 0.001
Less than primary vs. completed college/university 3.166 (0.076) 0.271 < 0.001
Less than primary vs. completed post-grad 2.837 (0.215) 0.079 < 0.001
Age Category: 50-59 vs. 60-69 -0.749 (0.045) -0.108 < 0.001
50-59 vs. 70-79 -1.825 (0.054) -0.218 < 0.001
50-59 vs. 80+ -3.197 (0.089) -0.219 < 0.001
Composite Cognitive Z-score 0.11 / < 0.001***
Constant -0.243 (0.045) < 0.001
Education Level : Less than primary vs. completed primary 0.747 (0.060) 0.099 < 0.001
Less than primary vs. completed secondary 1.683 (0.063) 0.212 < 0.001
Less than primary vs. completed high school 2.190 (0.062) 0.280 < 0.001
Less than primary vs. completed college/university 3.227 (0.085) 0.277 < 0.001
Less than primary vs. completed post-grad 2.813 (0.237) 0.078 < 0.001
Sleep Duration Category: Intermediate vs. short -0.367 (0.056) -0.044 < 0.001
Intermediate vs. long -0.657 (0.070) -0.062 < 0.001
Composite Cognitive Z-score 0.11 / < 0.001***
Constant -1.165 (0.086) < 0.001
Education Level : Less than primary vs. completed primary 0.874 (0.056) 0.116 < 0.001
Less than primary vs. completed secondary 1.881 (0.060) 0.232 < 0.001
Less than primary vs. completed high school 2.385 (0.059) 0.297 < 0.001
Less than primary vs. completed college/university 3.313 (0.079) 0.283 < 0.001
Less than primary vs. completed post-grad 2.841 (0.224) 0.079 < 0.001
Sleep Quality Category : Low vs. intermediate 0.295 (0.086) 0.043 0.001
Low vs. high 0.792 (0.083) 0.120 < 0.001
Composite Cognitive Z-score 0.18 / 0.038*
Constant 0.699 (0.125) < 0.001
Education Level : Less than primary vs. completed primary 0.554 (0.172) 0.073 0.001
Less than primary vs. completed secondary 1.471 (0.181) 0.181 < 0.001
Less than primary vs. completed high school 2.158 (0.180) 0.269 < 0.001
Less than primary vs. completed college/university 3.147 (0.240) 0.270 < 0.001
Less than primary vs. completed post-grad 4.332 (0.656) 0.120 < 0.001
Country: China vs. Ghana -1.414 (0.075) -0.119 < 0.001
China vs. India -2.394 (0.061) -0.251 < 0.001
China vs. Mexico -1.404 (0.061) -0.152 < 0.001
China vs. Russia -1.135 (0.069) -0.103 < 0.001
China vs. South Africa -2.058 (0.065) -0.222 < 0.001
Sex -0.202 (0.076) -0.031 0.008
Country*Sex Interaction : China vs. Ghana 0.181 (0.108) 0.038 0.094
China vs. India 0.159 (0.115) 0.031 0.167
China vs. Mexico 0.325 (0.114) 0.064 0.004
China vs. Russia 0.357 (0.152) 0.048 0.019
China vs. South Africa -0.404 (0.488) -0.015 0.408
Comparisons are statistically significant at: * p < 0.05, ** p < 0.01, *** p < 0.001. Sex: 1 = male, 2 = female. Reference groups used in the creation of dummy
codes for each categorical variable: educational levels = less that primary school; age by decade = 50-59 year olds; sleep duration = intermediate sleep
(> 6-9 h); sleep quality = low sleep quality (average rating of 1-2); country = China.

Journal of Clinical Sleep Medicine, Vol. 10, No. 6, 2014 618

 Sleep and Cognitive Function among Older Adults
increases linearly, in which individuals with intermediate and circadian drive for wakefulness over time and affects neural
high sleep quality exhibited significantly higher cognitive processing, resulting in the destabilization of the wake state and
z-scores (p = 0.001) than low quality sleepers (see Table 5). overall neurocognitive function. The negative impact of these
Spearman correlations indicated that average sleep duration and neurocognitive effects are particularly evident in the prefrontal
sleep quality were significantly positively correlated (p < 0.001) cortex, the area of the brain related to attention and working
for both sexes in all countries (Table 4). memory abilities.20 A recent study has directly linked structural
changes in prefrontal cortex with sleep-related memory prob-
Country, Sex, and Cognitive Test Performance lems.22 In a study by Mander and colleagues, natural atrophy in
A final linear regression was used to examine the effect the medial prefrontal cortex was observed in older adults; this
of country and sex on cognitive test performance variation area of the brain was roughly one-third smaller in older indi-
while controlling for schooling. Education level explained viduals than younger individuals. These changes are thought to
a significant amount of the variance in cognitive scores (R2 reduce the amount of restorative SWS, thus inhibiting knowl-
change = 0.106, p < 0.001). Cognitive scores increased linearly edge consolidation and memory formation.22
as education level increased, peaking at “completed post-grad Previous studies have also found that long sleep duration
degree.” Adding country explained a significant amount of may contribute to variation in cognitive test performance, as
additional variance (R2 change = 0.078, p < 0.001), suggesting was observed in the present study.4,21 Long sleep duration may
that there are significant country differences in test perfor- simply reflect the presence of sleep disorders or other illnesses.
mance on the set of cognitive measures used. Adding sex to These individuals may require more time in bed to feel rested
the model explained a nonsignificant amount of variance (R2 and therefore report longer sleep durations. However, longer
change = 0.000, p = 0.397), indicating that sex does not signifi- sleep duration may not completely compensate for underlying
cantly affect cognitive scores in this model. Adding the inter- health conditions, and cognitive decline may still occur.
action between country and sex explained a marginal amount An important finding in the present study is the sex differ-
of additional variance (R2 change < 0.0010, p = 0.038); this ence in all sleep and cognition variables. There are several
very small but statistically significant finding is likely due to possible reasons for these observed differences. First, older
the large sample size. women typically report more sleep disturbances than older men,
which in part is thought to be due to postmenopausal changes
DISCUSSION in hormone profile (for instance, estrogen deficiency) and
increased bladder instability.23 Second, longer sleep duration
This study provides a unique examination of the asso- and poorer subjective sleep scores in women have also been
ciations between sleep quality, duration, and cognitive test associated with feelings of isolation, which can result from
performance in older individuals using an extensive and widowhood or a perceived lack of social support.15 Third, poor
unparalleled collection of cross-cultural measurements. cognitive scores in women may be due to suboptimal sleep
Previous studies examining these patterns have been confined patterns. However, lower scores on cognitive performance tests
to high income countries and have generally relied on data may also result from women receiving less schooling than their
collected from small and non-representative population male counterparts.4 The present study provides evidence of this
samples. The SAGE sample is unique in that it is very large, idea and indicates that sex was not a significant predictor of
drawn from several diverse nations, and representative of the cognitive z-scores beyond what was accounted for by educa-
range of living conditions in each country. Furthermore, as tion and country. This supports the idea that schooling and other
it is a longitudinal study, factors affecting well-being during cultural factors may play an important role in structuring these
aging can be identified by analyzing health measures in the sex differences.
same individuals over time. The large, nationally representative samples from six coun-
The present study found support for all four hypoth- tries used in the present study provided an opportunity to
eses. Short (0-6 h/night) and long (> 9 h/night) sleep dura- examine the variation in sleep patterns and associations with
tions were significantly associated with poorer cognitive test cognition between countries. Significant differences in sleep
performance relative to intermediate sleep lengths (6.01-9 h/ duration and quality were observed between countries with
night). Higher average sleep quality scores (≥ 4) were signifi- different levels of economic development. The two African
cantly associated with increased cognitive test performance. countries (South Africa and Ghana) exhibited the longest sleep
Composite cognition z-scores were lower in older individuals. durations and, together with Mexico, also reported the highest
Sex differences were evident in all analyses, with women sleep quality scores compared to the other countries. This
generally exhibiting significantly longer sleep durations finding might be explained by differences in economic devel-
(except in Russia and Mexico), lower sleep quality ratings, opment (e.g., different work patterns and occupations, types
and lower composite cognition z-scores in the present study of housing, and living arrangements) between these countries
(except in Russia and Mexico) relative to their male counter- and the others included in the analyses. Sleep environment has
parts in each country. been shown to affect sleep quality and duration; factors typi-
These results are consistent with previous findings from cally associated with a more urban lifestyle (such as, noise,
research examining smaller samples drawn from high income light exposure, and non-private sleep quarters) often have
countries; these studies indicate that sleep deprivation nega- a detrimental impact on sleep quality and quantity.7-9 Social
tively impacts cognitive performance as well as mood and factors like financial stress, health worries, alcohol consump-
motor function.20,21 Sleep deprivation degrades the peak tion, and an unfavorable work schedule (e.g., working the night

619 Journal of Clinical Sleep Medicine, Vol. 10, No. 6, 2014

TE Gildner, MA Liebert, P Kowal et al.
shift) may also negatively affect sleep quality.7-9 Environmental
factors and economic infrastructure therefore play an impor- REFERENCES
tant role in shaping sleep patterns. It is possible that individuals 1. Bombois S, Derambure P, Pasquier F, Monaca C. Sleep disorders in aging and
living in more rural areas are less exposed to these adverse dementia. J Nutr Health Aging 2010;14:212-7.
social stressors and poor sleep environments, and subsequently 2. Descamps A, Cespuglio R. Influence of aging on the sleep rebound induced by
immobilization stress in the rat. Brain Res 2010;1335:14-23.
report longer sleep durations and higher sleep quality scores. 3. Crenshaw M, Edinger, J. Slow-wave sleep and waking cognitive performance
among older adults with and without insomnia complaints. Physiol Behav
Limitations 1999;66:485-92.
The present study has several important limitations. First, 4. Faubel R, López-García E, Guallar-Castillón P, et al. Usual sleep duration and
cognitive function in older adults in Spain. J Sleep Res 2009;18:427-35.
the battery of cognitive tests used was not developed in the 5. Nebes RD, Buysse DJ, Halligan EM, Houck PR, Monk TH. Self-reported sleep
countries studied. These tests only examined certain aspects of quality predicts poor cognitive performance in healthy older adults. J Gerontol
cognitive performance, but did not measure overall cognitive 2009;64:180-7.
function. Furthermore, differences in composite z-scores were 6. Matthews K, Gallo L. Psychological perspectives on pathways linking
socioeconomic status and physical health. Annu Rev Psychol 2011;62:501-30.
likely affected by cultural variation and extent of economic 7. Stamatakis KA, Kaplan GA, Roberts RE. Short sleep duration across income,
development, which structure differences in schooling. These education, and race/ethnic groups: population prevalence and growing
dissimilarities may have resulted in specific populations disparities over 34 years of follow-up. Ann Epidemiol 2007;17:948-55.
obtaining higher tests scores simply due to the nature of cogni- 8. Lallukka T, Ferrie JE, Kivimäki M, et al. Economic difficulties and subsequent
sleep problems: Evidence from British and Finnish occupational cohorts. Sleep
tive measures utilized. Med 2012;13:680-5.
A second limitation is that the sleep data used were reliant on 9. Mezick EJ, Matthews KA, Hall M, et al. Influence of race and socioeconomic
accurate participant responses in sleep duration, yet it is often status on sleep: Pittsburgh SleepSCORE project. Psychosom Med
difficult for individuals to discern between time in bed and time 2008;70:410-16.
10. Baldwin CM, Kapur VK, Holberg CJ, Rosen C, Nieto FJ. Associations between
asleep. Studies have shown that when asked how many hours gender and measures of daytime somnolence in the Sleep Heart Health Study.
they slept, participants often report time spent in bed; however, Sleep 2004;27:305-12.
this value may exceed time spent asleep.24 Objective measures 11. Haimov I, Shatil E. Cognitive training improves sleep quality and cognitive
of sleep based on polysomnography or actigraphy would more function among older adults with insomnia. PLoS One 2013;8:e61390.
12. Keage HA, Banks S, Yang KL, Morgan K, Brayne C, Matthews FE. What
accurately capture individual sleep patterns, including sleep sleep characteristics predict cognitive decline in the elderly? Sleep Med
duration, transitions between sleep states, and duration of night 2012;13:886-92.
awakenings.25 A third limitation is the reliance on data from 13. Kowal P, Chatterji S, Naidoo N, et al. Data resource profile: The WHO Study
only the two nights of sleep prior to the interview, which may on Global Ageing and Adult Health (SAGE). Int J Epidemiol 2012;41:1639-49.
14. United Nations Educational, Scientific and Cultural Organization. International
not accurately capture typical sleep patterns. Standard Classification of Education 1997; (ISCED 97), 2006. www.unesco.org/
Finally, because the data in the present study are cross- education/information/nfsunesco/doc/isced_1997.htm
sectional, it is impossible to determine the causality between 15. Patel SR, Malhotra ADJG, Gottlieb DJ, White DP, Hu FB. Correlates of long
the variables assessed. It is possible that the cognitive decline sleep duration. Sleep 2006;29:881-9.
16. Tamakoshi A, Ohno Y. Self-reported sleep duration as a predictor of all-cause
observed may not be due to deficient sleep patterns; instead, mortality: results from the JACC study, Japan. Sleep 2004;27:51-4.
cognitive decline in older individuals could result in subop- 17. Ohayon MM, Vecchierini M. Normative sleep data, cognitive function and daily
timal sleep duration and poor sleep quality. Longitudinal data living activities in older adults in the community. Sleep 2005;28:981-9.
following the progression of these trends over time is required 18. Grodstein F, Chen J, Willett WC. High-dose antioxidant supplements and
cognitive function in community-dwelling elderly women. Am J Clin Nutr
to further parse out these interactions. SAGE is in the process 2003;77:975-84.
of collecting the second wave of data on participants which 19. Scarmeas N, Albert SM, Manly JJ, Stern Y. Education and rates of cognitive
will help address this issue. These data will also facilitate the decline in incident Alzheimer’s disease. J Neurol Neurosurg Psychiatry
examination of other issues such as undiagnosed illness, which 2006;77:308-31
20. Goel N, Rao H, Durmer JS, Dinges DF. Neurocognitive consequences of sleep
could affect sleep and cognition variables, but would only deprivation. Semin Neurol 2009;29:320-39.
emerge over time. However, given the association between 21. Xu L, Jiang CQ, Lam TH, et al. Short or long sleep duration is associated with
sleep patterns and cognitive test performance, the diagnosis memory impairment in older Chinese: the Guangzhou Biobank Cohort Study.
of sleep disorders in older adults is critical for the design and Sleep 2011;34:575-80.
22. Mander BA, Rao V, Lu B, et al. Prefrontal atrophy, disrupted NREM slow
implementation of effective cognitive decline intervention waves and impaired hippocampal-dependent memory in aging. Nat Neurosci
programs. 2013;16:1-10.
23. Wolkove N, Elkholy O, Baltzan M, Palayew M. Sleep and aging: Sleep disorders
commonly found in older people. CMAJ 2007;176:1299-304.
CONCLUSION 24. Stenholm S, Kronholm E, Bandinelli S, Guralnik JM, Ferrucci L. Self-reported
sleep duration and time in bed as predictors of physical function decline: results
In conclusion, this study documented relationships between from the InCHIANTI Study. Sleep 2011;34:1583-93.
sleep quality and quantity and cognitive test performance 25. Lockley SW, Skene DJ, Arendt J. Comparison between subjective and actigraphic
among older individuals from six middle income countries. measurement of sleep and sleep rhythms. J Sleep Res 2002;8:175-83.
These results confirm previous findings in Western populations
and suggest that sleep patterns are associated with cognitive ACKNOWLEDGMENTS
test performance cross-culturally in diverse societies. Thus, The authors thank Nirmala Naidoo for assistance with data analysis, Larry Sugi-
optimizing sleep duration and quality are important consider- yama for discussions of the project, all the participants in the study, and the study
ations in future clinical studies aimed at mitigating cognitive PIs and teams.
decline in older individuals.

Journal of Clinical Sleep Medicine, Vol. 10, No. 6, 2014 620

 Sleep and Cognitive Function among Older Adults

SUBMISSION & CORRESPONDENCE INFORMATION DISCLOSURE STATEMENT

Submitted for publication August, 2013 This was not an industry supported study. Support for the research was provided
Submitted in final revised form January, 2014 by NIH NIA Interagency Agreement YA1323-08-CN-0020 with the World Health Or-
Accepted for publication January, 2014 ganization and grant NIH R01-AG034479. The funding organizations had no role in
Address correspondence to: Theresa E. Gildner, Department of Anthropology, study design, data collection and analysis, decision to publish, or preparation of the
University of Oregon, Eugene, OR 97403; Tel: (541) 346-5102; Fax: (541) 346-0668; manuscript. The authors have indicated no financial conflicts of interest.
E-mail: [email protected]

621 Journal of Clinical Sleep Medicine, Vol. 10, No. 6, 2014