Abstract
Background
As the global population ages, the preservation of cognitive function in middle-aged and older adults has become a critical public health issue. Sleep duration represents a modifiable factor associated with cognitive performance, yet its nonlinear relationship across diverse socioeconomic and cultural contexts remains underexplored.
Methods
This multinational cross-sectional study harmonized data from three representative aging cohorts: the China Health and Retirement Longitudinal Study (CHARLS), the English Longitudinal Study of Ageing (ELSA), and the Longitudinal Aging Study in India (LASI), comprising 25,783 adults aged 50 years or older. Linear regression, piecewise linear models, and random-effects meta-analysis were employed to examine associations between self-reported sleep duration and global cognitive function, adjusting for sociodemographic, lifestyle, and health-related covariates.
Results
A consistent inverted U-shaped association was observed between sleep duration and cognitive function across all populations. Both short (≤ 4 h) and long (≥ 9 h) sleep durations were significantly associated with poorer cognitive performance compared to 7 h of sleep (reference). Age and sex significantly moderated this relationship: middle-aged adults (50–64 years) showed greater vulnerability to extreme sleep durations, whereas adults aged 75 years and older were predominantly affected by long sleep. Men were more susceptible to cognitive impairment from short sleep, whereas women exhibited greater sensitivity to long sleep.
Conclusion
This study provides the first cross-national evidence of a robust nonlinear sleep-cognition relationship in aging populations across high- and middle-income countries. Our findings suggest that maintaining approximately 7 h of sleep may be optimal for cognitive health in later life. Public health strategies should consider age- and sex-specific interventions to promote healthy sleep patterns and mitigate cognitive decline.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12889-026-26749-y.
Keywords: Sleep duration, Cognitive function, Cross-national study, Middle-aged and older adults
Introduction
The current era is characterized by an unprecedented acceleration of global population aging, rendering the preservation of cognitive function in middle-aged and older adults a critical public health priority. Cognitive decline not only detrimentally impacts quality of life at the individual level but also imposes substantial healthcare and socioeconomic burdens. Dementia—a progressive neurodegenerative disorder with core manifestations of cognitive impairment—exhibits a globally increasing prevalence. Recent findings from the Global Burden of Disease Study revealed approximately 57.4 million prevalent dementia cases worldwide in 2019, with projections indicating a surge to 152 million cases by 2050, representing a 169% growth trajectory [1]. This exponential trend underscores the pivotal importance of identifying modifiable risk factors to delay cognitive decline and improve health outcomes in middle-aged and older adults. Among the myriad factors influencing cognitive function, sleep—a fundamental and universally observed health behavior—has increasingly emerged as a key focus of research, given its established association with cognitive performance [2, 3].
Neuroscientific evidence indicates the crucial role of sleep in preserving neurocognitive integrity. Specifically, sleep facilitates cerebral homeostasis through accelerated clearance of metabolic byproducts while concurrently promoting neural network reorganization via oscillatory synchronization during non-rapid eye movement stages [4].
These neurophysiological processes are indispensable for maintaining neuronal viability and stabilizing cognitive circuitry. Accumulating evidence further suggests a nonlinear, inverted U-shaped association between sleep duration and cognitive performance, where both sleep deprivation (< 6 h) and hypersomnia (> 9 h) demonstrate deleterious neurocognitive consequences [5–8]. Nevertheless, the generalizability of this relationship across diverse populations and the underlying mechanistic pathways remain contentious. Prevailing research remains constrained by single-nation designs with limited statistical power, while multinational investigations encompassing diverse socioeconomic contexts are notably scarce. Moreover, whether age and sex moderate this relationship across different populations has not been systematically examined in a cross-national framework.
Against this backdrop, three key research gaps have become apparent. First, most studies focus on single-country populations, lacking cross-cultural validation across middle-high-income (e.g., China, England) and low-income (e.g., India) regions, making it impossible to confirm the universality of the sleep-cognition relationship. Second, the specific turning points of the nonlinear sleep-cognition association (i.e., the optimal sleep duration threshold for peak cognitive function) have not been clearly defined in aging populations across different countries. Third, age and sex, as important influencing factors of both sleep and cognitive function, have only been sporadically studied for their moderating effects on the sleep-cognition relationship, with no systematic exploration in multinational cohorts to clarify vulnerability differences across populations.
To address these knowledge gaps, we harmonized data from three longitudinal aging studies: the China Health and Retirement Longitudinal Study (CHARLS) [9], English Longitudinal Study of Ageing (ELSA) [10], and Longitudinal Aging Study in India (LASI) [11]. These surveys encompass divergent economic contexts (emerging versus developed economies), cultural frameworks (Eastern versus Western traditions), and healthcare infrastructures (universal versus developing systems). The present study has three specific aims:
to test whether an inverted U-shaped association between sleep duration and global cognitive function is consistent across China, England, and India;
to examine whether age moderates this association, hypothesizing that middle-aged adults (50–64 years) would be more vulnerable to both short and long sleep, whereas older adults (≥75 years) would be primarily affected by long sleep; and
to investigate whether sex moderates the relationship, with the expectation that men would show greater cognitive sensitivity to short sleep and women to long sleep.
By addressing these questions, this study provides the first cross-national evidence on the nonlinear sleep-cognition relationship and its modifiers, thereby informing the development of tailored sleep-health strategies in aging populations worldwide.
Materials and methods
Study population
This cross-sectional study integrated data from three nationally representative aging surveys: the fourth wave of CHARLS in 2015, the eighth wave of ELSA in 2016–2017, and the first wave of LASI in 2017–2019. Considering that LASI 2017–2019 represents India’s first publicly available survey data release, and to ensure temporal alignment across the three cohorts, the present study selected these specific data waves as the basis for analysis. As all three studies are international partner projects of the Health and Retirement Study (HRS), they follow standardized protocols and measurement methodologies. Each study adopts a household survey design: eligible main participants are first recruited, followed by invitations to their spouses or partners (regardless of age). ELSA targets adults aged 50 and over and their spouses in England, whereas CHARLS and LASI enroll adults aged 45 and over and their spouses in China and India, respectively. Consequently, while all primary respondents are middle-aged or elderly, their spouses may be younger adults. To enhance the physiological comparability of aging trajectories, we restricted analyses to participants aged ≥ 50 years across all cohorts. In LASI, sleep data collection followed a randomized module approach where participants were randomly assigned to one of four experimental modules, with sleep duration assessed only in the “Time Use” module. Our analysis therefore included only LASI participants assigned to this module. Furthermore, while sleep duration in ELSA was constrained within a specific range (1–15 h), no such restrictions were applied in the other two databases. To ensure cross-national comparability of sleep duration and exclude pathological sleep states, participants with sleep durations less than 1 h or more than 15 h were excluded from the analysis. The final analytical sample comprised 25,783 eligible participants; Fig. 1 details the participant flow diagram.
Fig. 1.
The flow chart of study population
Assessment of cognitive function
Cognitive function was assessed across three core domains: memory, executive function, and orientation. To ensure cross-national comparability, we selected analogous assessment components. The memory assessment task entailed both immediate and delayed recall of 10 unrelated words. Immediate recall required participants to retrieve the words immediately after the word list was read aloud, whereas delayed recall took place after they had answered other unrelated survey questions. The memory score was calculated as the total number of words successfully recalled across both the immediate and delayed tasks, with a possible range of 0 to 20. The orientation test consists of four questions, where respondents are asked to report the current day of the week, month, date, and year at the time of the interview. Each correct response is awarded 1 point, resulting in a total score ranging from 0 to 4. Executive function was evaluated via the serial 7’s test, in which participants count backward from 100 by sequentially subtracting 7 (with a total of five calculations; each correct subtraction yields 1 point, resulting in a score range of 0–5). These assessments have been validated as effective measures of cognitive function [12–14].
To ensure cross-database comparability of cognitive metrics, domain-specific z-standardisation was applied. For each cognitive subdomain—memory score, orientation score, and executive function score—the pooled mean and standard deviation (SD) were computed across the harmonized CHARLS-ELSA-LASI dataset. Standardized z-scores were generated by subtracting the mean from test scores and dividing by the pooled SD. The composite cognitive z-score was derived by averaging the z-scores of the three cognitive test domains (each weighted equally at 1/3) [6, 15, 16], and reflects an individual’s comprehensive cognitive level.
Sleep duration
Habitual sleep duration was self-reported during face-to-face interviews without predefined categories. In CHARLS, sleep duration was assessed by asking, ‘During the past month, how many hours of actual sleep did you get at night (average hours for one night)? (This may be shorter than the number of hours you spend in bed.)’. In ELSA, it was evaluated through the question ‘How many hours of sleep do you have on an average week night?‘; in LASI, it was assessed based on participants’ reported bedtime and wake-up time from the previous night. The continuous sleep duration variable was subsequently categorized into seven mutually exclusive intervals: ≤4 h, 5 h, 6 h, 7 h (reference), 8 h, 9 h, and ≥ 10 h to facilitate dose-response analysis.
Covariates
To examine the association between sleep duration and cognitive function, we incorporated a range of potential confounding variables for adjustment, including sex, age (years), educational attainment, wealth status, smoking and alcohol consumption behaviors, body mass index (BMI, kg/m²), cohabitation status, depressive symptoms, and chronic disease history (hypertension, diabetes, cancer, heart disease, lung disease, and stroke). Educational levels were categorized into three groups based on the International Standard Classification of Education (ISCED 2011): low (below junior high school), secondary (high school or vocational training), and higher (university or above) [17]. Household per capita consumption was used as a proxy for wealth status. To account for cross-national differences in purchasing power, household per capita consumption was quintiled within each database and then collapsed into three tiers: low (bottom 40%), medium (middle 20%), and high (top 40%) [18]. Smoking status was categorized into three groups based on self-reported smoking history and current behavior: never smoked, former smoker, and current smoker. Alcohol consumption was grouped into three categories based on frequency in the past year: never drank, less than once per month, and at least once per month. Body mass index (BMI) was derived from anthropometric measurements and was calculated as weight in kilograms divided by height in meters squared. Depressive symptoms were assessed via the Center for Epidemiologic Studies Depression Scale (CES-D) and operationalized as a binary variable (1 = presence; 0 = absence). Given the variations in scale versions across datasets, study-specific screening thresholds were applied for consistency: the 10-item CES-D (range: 0–30) was used in CHARLS and LASI, with a total score ≥ 10 indicating positive depressive symptoms; the 8-item CES-D (range: 0–8) was utilized in ELSA, with a total score ≥ 3 denoting positivity [19, 20]. Chronic disease indicators included self-reported physician-diagnosed hypertension, diabetes (or current use of anti-diabetic treatment), cancer, heart disease, lung disease, and stroke.
Statistical analysis
Descriptive statistics characterized the analytical sample, with categorical variables (e.g., sex and education level) summarized as frequencies (percentages) and continuous variables (e.g., age, BMI, and cognitive scores) summarized as means ± standard deviations; missing data proportions were systematically reported, and detailed classification of missing value codes, counts, and corresponding reasons for each variable is presented in Supplementary Table S1 for the CHARLS, ELSA, and LASI datasets.
To address missing data, we implemented multiple imputation via the R mice package (version 4.4.2), generating 20 imputed datasets per country using the random forest algorithm (method = “rf”). The predictor matrix for the imputation model included all covariates specified for the final analytical model—sex, age, educational attainment, wealth status, smoking and alcohol consumption, BMI, depressive symptoms, and history of chronic diseases (hypertension, diabetes, cancer, lung disease, heart disease, and stroke)—serving as auxiliary variables to strengthen the plausibility of the missing at random (MAR) assumption. The random forest method automatically accommodates complex relationships and non-linear patterns among variables, ensuring compatibility between the imputation and analysis models. On each imputed dataset, a separate multiple linear regression model was fitted. The global cognitive z-score served as the dependent variable, with sleep duration categories (using 7 h as the reference) as the primary independent variable, adjusted for the aforementioned covariates. HC3 robust standard errors were employed in each model to mitigate potential heteroskedasticity. Parameter estimates and standard errors from the 20 datasets were pooled using Rubin’s rules to obtain final regression coefficients, 95% confidence intervals, and p-values.To control the false discovery rate (FDR) arising from multiple comparisons, we applied the Benjamini-Hochberg correction to the p-values for all sleep-category associations. An FDR-adjusted p-value < 0.05 was considered statistically significant. Multicollinearity was assessed via variance inflation factors (VIFs), with a VIF < 5 indicating acceptable collinearity thresholds.
Next, we synthesized results across different datasets using random-effects meta-analyses via the metafor package. The model synthesized the β coefficients for each sleep duration category (compared with 7 h) using inverse-variance weighting, which automatically assigns greater weight to estimates with greater precision (typically from larger samples). We calculated the pooled effect size, 95% confidence interval, and assessed heterogeneity using the I² statistic.
To address multiple testing across sleep categories in the meta-analyses, we further applied the Benjamini-Hochberg false discovery rate correction to the resulting p-values. Generalized additive models (GAM) were utilized to explore nonlinear associations between sleep duration and cognitive function. Using piecewise linear regression models (two-segment) and likelihood ratio tests, we identified turning points (k-values) in sleep duration and calculated changes in slopes along with their significance before and after these turning points.
We then investigated the moderating effects of sex and age on sleep-cognition associations through stratified subgroup analyses. Cross-sectional associations between sleep duration and global cognitive function were examined separately within age strata (50–64, 65–74, and ≥ 75 years) and sex subgroups, which were defined as the stratification variables. All analyses utilized linear regression models, adjusted for all covariates except age and sex (the stratification variables). The heteroscedasticity was corrected using HC3-robust standard errors, and the results from multiple imputed datasets were pooled via Rubin’s rules. Multicollinearity among independent variables was evaluated via VIFs. We also applied GAMs within sex and age strata to visually assess the consistency of the sleep-cognition dose-response curves across key subgroups. The resulting plots are provided in Supplementary Figures SX-SY.
Finally, three sensitivity analyses were performed to assess the robustness of the models: the first was a complete-case analysis excluding participants with any missing covariates (n = 16,726) to verify consistency with the multiple imputation results; the second was an Inverse Probability Weighting (IPW) analysis (n = 31832) to mitigate potential selection bias from excluding participants with missing sleep or cognitive data; the third utilized unstandardized raw cognitive function scores instead of z-scores to rule out standardization artifacts. All analyses were conducted via Stata version 18.0 and R version 4.4.2, with statistical significance defined as a two-sided p-value < 0.05.
Results
Baseline characteristics and sample size
A total of 25,783 eligible participants were included in the analysis: 12,543 from CHARLS, 7,245 from ELSA, and 5,995 from LASI. Significant inter-cohort differences were observed in demographic profiles, lifestyle behaviors, health status, and cognitive function (Table 1 is available at the end of the manuscript.). Participants in the ELSA cohort were significantly older (mean age: 68.74 years) than those in both the CHARLS (mean age: 61.59 years) and LASI (mean age: 62.84 years) cohorts. This cohort also demonstrated higher educational attainment (21.8% with tertiary education), a greater prevalence of living alone (33.8%), more frequent alcohol consumption (70.1% drinking at least once per month), and significantly better cognitive performance across all assessed domains (executive function, orientation, and memory) than the other two cohorts. The CHARLS cohort had the highest proportion of participants with low educational attainment (85.7% below lower secondary) and the highest current smoking rate (30.2%). This cohort also reported the highest prevalence of short sleep duration (≤ 4 h; 15.3%). The LASI cohort had the lowest mean BMI (22.68 kg/m²) and the lowest prevalence of major chronic conditions, including cancer (0.7%), hypertension (32.1%), lung disease (2.5%), heart disease (4.4%), and stroke (2.3%). However, this cohort had the highest proportion of individuals with depressive symptoms (40.8%) and the highest rate of long sleep duration (≥ 10 h; 13.7%). Despite these differences, all three cohorts exhibited similar distributions in terms of household per capita consumption stratification (approximately 40% in low, 20% in medium, and 40% in high tiers respectively).
Table 1.
Baseline characteristics of the participants
| Variable | CHARLS(n = 12,543) | ELSA(n = 7,245) | LASI(n = 5,995) |
|---|---|---|---|
| Age (years), mean ± SD | 61.59 ± 8.22 | 68.74 ± 9.06 | 62.84 ± 9.36 |
| BMI, mean ± SD | 24.63 ± 26.98 | 28.13 ± 5.45 | 22.68 ± 4.76 |
| Missing, n (%) | 2,155 (17.2) | 1,119 (15.4) | 504 (8.4) |
| Educational attainment, n (%) | |||
| Less than lower secondary | 10,745 (85.7) | 1,589 (24.1) | 4,419 (73.7) |
| Upper secondary and vocational training | 1,524 (12.2) | 3,568 (54.1) | 1,266 (21.1) |
| Tertiary | 272 (2.2) | 1,437 (21.8) | 310 (5.2) |
| Missing, n (%) | 2 (0.0) | 651 (9.0) | 0 (0.0) |
| Total household per capita consumption, n (%) | |||
| Low | 3,635 (40.8) | 2,524 (39.1) | 2,398 (40.0) |
| Medium | 1,758 (19.7) | 1,301 (20.1) | 1,211 (20.2) |
| High | 3,526 (39.5) | 2,635 (40.8) | 2,386 (39.8) |
| Missing, n (%) | 3,624 (28.9) | 785 (10.8) | 0 (0.0) |
| Sex(male), n (%) | 6,670 (53.2) | 3,258 (45.0) | 2,804 (46.8) |
| Drinking, n (%) | |||
| Never drink | 7,852 (62.8) | 855 (13.0) | 5,343 (89.2) |
| Less than once a month | 1,080 (8.6) | 1,112 (16.9) | 170 (2.8) |
| At least once a month | 3,570 (28.6) | 4,618 (70.1) | 480 (8.0) |
| Missing, n (%) | 41 (0.3) | 660 (9.1) | 2 (0.0) |
| Smoking, n (%) | |||
| Never | 6,500 (51.8) | 2,737 (37.9) | 4,796 (80.0) |
| Former | 2,256 (18.0) | 3,812 (52.8) | 313 (5.2) |
| Current | 3,781 (30.2) | 665 (9.2) | 883 (14.7) |
| Missing, n (%) | 6 (0.0) | 31 (0.4) | 3 (0.1) |
| Sleep duration, h | |||
| ≤ 4 | 1,922 (15.3) | 351 (4.8) | 415 (6.9) |
| 5 | 1794 (14.3) | 677 (9.3) | 252 (4.2) |
| 6 | 2759 (22.0) | 1,548 (21.4) | 549 (9.2) |
| 7 | 2296 (18.3) | 2,262 (31.2) | 1,202 (20.1) |
| 8 | 2695 (21.5) | 1,874 (25.9) | 1,550 (25.9) |
| 9 | 554 (4.4) | 407 (5.6) | 1,206 (20.1) |
| ≥ 10 | 523 (4.2) | 126 (1.7) | 821 (13.7) |
| Living alone, n (%) | 1,550 (12.4) | 2,447 (33.8) | 1,631 (27.2) |
| Missing, n (%) | 1 (0.0) | 4 (0.1) | 0 (0.0) |
| Depressive symptoms, n (%) | 3,939 (31.4) | 1298 (17.9) | 2446 (40.8) |
| Missing, n (%) | 1 (0.0) | 5 (0.1) | 4 (0.1) |
| Hypertension, n (%) | 3,997 (34.8) | 3,155 (43.5) | 1,923 (32.1) |
| Missing, n (%) | 1,064 (8.5) | 0 (0.0) | 1 (0.0) |
| Diabetes, n (%) | 1,224 (10.7) | 872 (12.0) | 853 (14.2) |
| Missing, n (%) | 1,135 (9.0) | 0 (0.0) | 1 (0.0) |
| Cancer, n (%) | 191 (1.7) | 947 (13.1) | 39 (0.7) |
| Missing, n (%) | 1,049 (8.4) | 0 (0.0) | 1 (0.0) |
| Lung disease, n (%) | 1,674 (14.6) | 509 (7.0) | 149 (2.5) |
| Missing, n (%) | 1,040 (8.3) | 0 (0.0) | 1 (0.0) |
| Heart disease, n (%) | 2,203 (19.2) | 1,706 (23.5) | 262 (4.4) |
| Missing, n (%) | 1,094 (8.7) | 0 (0.0) | 1 (0.0) |
| Stroke, n (%) | 417 (3.6) | 339 (4.7) | 139 (2.3) |
| Missing, n (%) | 1,028 (8.2) | 0 (0.0) | 1 (0.0) |
| Executive score, mean ± SD | 3.69 (1.47) | 4.38 (1.01) | 2.38 (1.81) |
| Orientation score, mean ± SD | 3.12 (1.06) | 3.80 (0.52) | 2.99 (1.20) |
| Memory score, mean ± SD | 6.88 (3.49) | 10.93 (3.60) | 8.83 (3.43) |
Abbreviations: CHARLS China Health and Retirement Longitudinal Study, ELSA English Longitudinal Study of Ageing, LASI Longitudinal Aging Study in India, BMI Body mass index, SD Standard deviation
Cross-sectional association between sleep duration and global cognitive function
Linear regression analyses revealed that, with 7 h as the reference, both short sleep durations (≤ 4 h) and long sleep durations (≥ 8 h, particularly ≥ 9 h and ≥ 10 h) were significantly associated with lower global cognitive function. However, this association was heterogeneous across the three cohorts (Table 2). Specifically, a nonsignificant relationship between sleep duration ≤ 4 h and global cognitive function was only observed in the ELSA cohort, whereas long sleep durations (9 h and ≥ 10 h) were consistently associated with lower global cognitive function in all cohorts and in the pooled analysis (all p < 0.05). These findings indicate that extreme sleep durations (≤ 4 h or ≥ 9 h) are significantly and cross-cohort consistently associated with lower global cognitive function, but with varying effect sizes across cohorts. The CHARLS cohort exhibited the strongest effect sizes, particularly for sleep durations ≥ 10 h (β = -0.34, 95% CI: -0.42 to -0.26, p < 0.001). In comparison, the ELSA cohort demonstrated the weakest associations for both short and long sleep durations. Significant heterogeneity was confirmed in the pooled estimates for extreme sleep durations (≤ 4 and ≥ 10 h), underscoring the variation in effect sizes across populations.
Table 2.
Regression analysis of sleep duration and global cognitive function(Reference: 7 h)
| CHARLS(n = 12,543) | ELSA(n = 7,245) | LASI(n = 5,995) | Pooled analysis (N = 25, 783) |
||||||
|---|---|---|---|---|---|---|---|---|---|
| β (95% CI) | p | β (95% CI) | p | β (95% CI) | p | β (95% CI) | I2, % | p | |
| Sleep duration, h | |||||||||
| ≤ 4 | -0.18 (-0.23, -0.13) | < 0.001*** | -0.04 (-0.11, 0.02) | 0.287 | -0.15 (-0.25, -0.06) | 0.006** | -0.13 (-0.21, -0.04) | 78.8 | 0.005** |
| 5 | 0.01 (-0.04, 0.06) | 0.797 | -0.04 (-0.09, 0.01) | 0.182 | 0.01 (-0.11, 0.12) | 0.952 | -0.01 (-0.05, 0.03) | 23.7 | 0.533 |
| 6 | 0.02 (-0.02, 0.07) | 0.440 | 0.00 (-0.03, 0.03) | 0.990 | -0.02 (-0.10, 0.07) | 0.792 | 0.01 (-0.02, 0.03) | 0.0 | 0.637 |
| 7 | Ref | ||||||||
| 8 | -0.06 (-0.10, -0.02) | 0.014* | -0.05 (-0.08, -0.01) | 0.012* | -0.03 (-0.09, 0.04) | 0.522 | -0.05 (-0.07, -0.02) | 0.0 | < 0.001*** |
| 9 | -0.19 (-0.27, -0.11) | < 0.001*** | -0.10 (-0.17, -0.04) | 0.006** | -0.18 (-0.25, -0.11) | < 0.001*** | -0.16 (-0.22, -0.10) | 50.0 | < 0.001*** |
| ≥ 10 | -0.34 (-0.42, -0.26) | < 0.001*** | -0.19 (-0.31, -0.07) | 0.004** | -0.22 (-0.30, -0.14) | < 0.001*** | -0.26 (-0.34, -0.17) | 64.3 | < 0.001*** |
Abbreviations: CHARLS China Health and Retirement Longitudinal Study, ELSA English Longitudinal Study of Ageing, LASI Longitudinal Aging Study in India, CI Confidence interval, I² Heterogeneity statistic
Covariates: age, sex, educational attainment, wealth level, smoking/alcohol consumption status, BMI, depressive status, and history of chronic disorders (hypertension, diabetes, cancer, heart disease, pulmonary disease, stroke). The heteroscedasticity was corrected via robust HC3 standard errors (via the `vcovHC` function) to ensure the robustness of parameter estimates. P-values for sleep duration categories were adjusted for multiple testing using the Benjamini-Hochberg false discovery rate (FDR) correction
Analysis of specific cognitive domains indicated that long sleep durations (≥ 9 h) were adversely associated with all three domains assessed: memory, executive function, and orientation. Orientation appeared to be the most sensitive to prolonged sleep, showing the largest negative effect size in the ≥ 10-hour sleep group. Short sleep (≤ 4 h) was consistently associated with poorer memory function across cohorts but showed weaker and often nonsignificant associations with executive function and orientation (Supplementary Tables 2–4. Multicollinearity diagnostics confirmed that all VIFs were less than 10, indicating that multicollinearity did not confound the model estimates.
Analysis of nonlinear relationship between sleep duration and cognitive function
After adjusting for covariates, smoothed curve fitting revealed significant inverted U-shaped associations between sleep duration and global cognitive function across all three cohorts (CHARLS, ELSA, and LASI) (Fig. 2). The inflection points for these curves consistently ranged between 5.5 and 7.0 h across cohorts. To formally test this nonlinearity, we conducted a piecewise linear regression analysis with threshold effects. A likelihood ratio test confirmed that the piecewise model provided a significantly better fit than did a simple linear model for all cohorts (p < 0.001). Using a two-step recursive method, we identified the optimal sleep duration thresholds associated with peak cognitive function: 5.5 h for CHARLS, 6.8 h for ELSA, and 6.4 h for LASI.
Fig. 2.
Smoothed curves: sleep duration vs. cognitive function. This model was adjusted for various variables, including age, sex, educational attainment, wealth level, smoking/alcohol consumption status, BMI, depressive status, and history of chronic disorders (hypertension, diabetes, cancer, heart disease, pulmonary disease, and stroke)
Below these thresholds, cognitive function scores increased significantly with each additional hour of sleep in all three cohorts: CHARLS (β = 0.13, 95% CI: 0.11 to 0.15, p < 0.001), ELSA (β = 0.02, 95% CI: 0.00 to 0.04, p = 0.040), and LASI (β = 0.05, 95% CI: 0.02 to 0.07, p < 0.001). Conversely, beyond these thresholds, longer sleep durations were associated with a significant decline in cognitive performance per hour in all three cohorts: CHARLS (β = -0.07, 95% CI: -0.08 to -0.06, p < 0.001), ELSA (β = -0.05, 95% CI: -0.07 to -0.03, p < 0.001), and LASI (β = -0.05, 95% CI: -0.07 to -0.04, p < 0.001). The detailed threshold effect results of sleep duration on cognitive function across the three cohorts are presented in Table 3. Other analyses of sleep duration and specific cognitive domains (memory, executive function, orientation) are provided in the Supplementary Materials (Tables S5–S7, Figures S1–S3). Overall, threshold effects on cognitive functions followed an inverted U-shaped trend; however, this trend was not statistically significant for executive function in ELSA (sleep duration < 6.8 h: β = 0.00, 95% CI: -0.02 to 0.02, p = 0.751) and orientation in ELSA (sleep duration < 7.5 h: β = -0.01, 95% CI: -0.02 to 0.00, p = 0.230), or for memory in LASI (sleep duration < 7.0 h: β = 0.01, 95% CI: -0.00 to 0.03, p = 0.125).
Table 3.
Analysis of the threshold effects of sleep duration on cognitive function
| CHARLS(n = 12,543) | ELSA(n = 7,245) | LASI(n = 5,995) | ||||
|---|---|---|---|---|---|---|
| β (95% CI) | P | β (95% CI) | P | β (95% CI) | P | |
| Multiple linear regression model | -0.00 (-0.01, 0.01) | 0.518 | -0.01 (-0.02, 0.00) | 0.004 | -0.02 (-0.03, -0.01) | < 0.001 |
| Piecewise linear regression model | ||||||
| k, h | 5.5 | 6.8 | 6.4 | |||
| Sleep duration < k (hours/day) | 0.13 (0.11, 0.15) | < 0.001 | 0.02 (0.00, 0.04) | 0.040 | 0.05 (0.02, 0.07) | < 0.001 |
| Sleep duration ≥ k (hours/day) | -0.07 (-0.08, -0.06) | < 0.001 | -0.05 (-0.07, -0.03) | < 0.001 | -0.05 (-0.07, -0.04) | < 0.001 |
| Log-likelihood ratio test | < 0.001 | < 0.001 | < 0.001 | |||
Abbreviations: CHARLS China Health and Retirement Longitudinal Study, ELSA English Longitudinal Study of Ageing, LASI Longitudinal Aging Study in India, BMI Body mass index, CI Confidence interval
This model was adjusted for various variables, including age, sex, educational attainment, wealth level, smoking/alcohol consumption status, BMI, depressive status, and history of chronic disorders (hypertension, diabetes, cancer, heart disease, pulmonary disease, stroke)
Subgroup analysis
To assess the potential moderating role of age, we stratified the analysis by age group (50–64, 65–74, and ≥ 75 years). Significant age-related heterogeneity was observed in the association between sleep duration and cognitive function (Fig. 3).
Fig. 3.
Association of sleep duration with cognitive function: age stratification analysis. Covariates: sex, educational attainment, wealth level, smoking/alcohol consumption status, body mass index (BMI), depressive status, and history of chronic disorders (hypertension, diabetes, cancer, heart disease, pulmonary disease, and stroke). P-values for sleep duration categories were adjusted for multiple testing using the Benjamini-Hochberg false discovery rate (FDR) correction
A pronounced inverted U-shaped relationship was evident in the youngest cohort (50–64 years), as shown in Figure S5. In pooled analyses, both short (≤ 4 h) and long (≥ 8 h) sleep durations were significantly associated with poorer cognitive function relative to 7 h of sleep (reference), with the most substantial detrimental effect observed in the ≥ 10-hour sleep group. These findings were consistent with the primary results. This association attenuated with increasing age. In the 65–74-year-old group, only sleep durations of 9 h were significantly associated with cognitive impairment. Among the oldest adults (≥ 75 years), only long sleep duration (≥ 9 h) demonstrated a significant negative association with cognitive function, and the magnitude of impairment increased progressively with longer sleep hours (P for trend < 0.001). In contrast, short sleep durations (≤ 6 h) were not significantly associated with cognitive outcomes in this age group.
Sex-stratified analyses were conducted to evaluate effect modification by sex (Fig. 4). The results revealed substantial sex disparities in the effects of sleep duration on cognitive function. Men exhibited greater vulnerability to short sleep durations. In the pooled analysis, a sleep duration of ≤ 4 h was associated with a significant decline in cognitive function in men (β = -0.16, p < 0.001). This association was consistently seen in men from the CHARLS and LASI cohorts, but no statistically significant link was detected among ELSA men after multiple comparison correction. Conversely, women were more susceptible to long sleep durations. Compared with males, females with sleep durations ≥ 9 h presented significantly lower cognitive function. This sex-specific effect was most pronounced among women in the CHARLS cohort, who presented the largest effect size (β = -0.43 for ≥ 10 h in women vs. β = -0.27 in men). For both sexes, long sleep-related cognitive impairment worsened with increasing sleep duration. Figure S6 visualizes the sex-specific nonlinear dose-response relationships between sleep duration and cognitive function.
Fig. 4.
Association of sleep duration with cognitive function: sex stratification analysis. Covariates: age, educational attainment, wealth level, smoking/alcohol consumption status, body mass index (BMI), depressive status, and history of chronic disorders (hypertension, diabetes, cancer, heart disease, pulmonary disease, and stroke). P-values for sleep duration categories were adjusted for multiple testing using the Benjamini-Hochberg false discovery rate (FDR) correction
Sensitivity analysis
To assess the robustness of our findings to missing data imputation, we performed a complete-case analysis excluding participants with any missing covariates (n = 16,726). The results were highly consistent with those of the primary analysis (Supplementary Table S7). Specifically, short sleep durations (≤ 4 h) remained significantly associated with poorer cognition according to CHARLS (β = -0.17, p < 0.001) and LASI (β = -0.15, p = 0.009), with a marginally significant association in the pooled analysis (β = -0.14, p = 0.047). Similarly, long sleep durations (≥ 10 h) were significantly associated with cognitive decline in all three cohorts and in the pooled analysis (all p < 0.05). The inverted U-shaped association was preserved in the complete-case smoothed curve analysis (Supplementary Figure S6). Piecewise linear regression confirmed a significant nonlinear relationship (LRT p < 0.001), identifying optimal thresholds between 5.5 and 6.5 h—slightly lower than those in the primary analysis (Supplementary Table S8). The core pattern of initial improvement followed by a decline beyond the threshold remained unchanged.
An Inverse Probability Weighting (IPW) sensitivity analysis was conducted to address potential selection bias from excluding participants with missing sleep or cognitive data (n = 6,049, 19.0%). After weighting, covariate balance improved substantially (Supplementary Table S9), and the association between sleep duration and cognitive function remained consistent with the primary analysis across all cohorts (Supplementary Table S10). The meta-analysis of IPW-adjusted estimates confirmed the inverted U-shaped pattern, with 8 h sleep significantly associated with poorer cognition (β = -0.05, 95% CI: -0.07 to -0.03, P = 0.008), while 9 h and ≥ 10 h showed negative associations with marginal significance after FDR correction (both P = 0.062). The point estimate for short sleep (≤ 4 h) was negative but non-significant (β = -0.12, 95% CI: -0.32 to 0.08, P = 0.124).
Finally, we repeated the primary analysis using raw cognitive scores instead of z-scores to rule out standardization artifacts. The results aligned substantially with the core findings (Supplementary Tables S11–12, Figure S7), confirming that the nonlinear association between sleep duration and cognitive function is robust to the use of standardized or raw cognitive metrics.
Discussion
By integrating large-scale aging datasets from China, England, and India, this study provides the first cross-national evidence of a nonlinear association between sleep duration and cognitive function. We demonstrated that both short (≤ 4 h) and long (≥ 9 h) sleep durations are significantly associated with poorer global cognitive function. Strikingly, this inverted U-shaped relationship remained consistent across populations from middle- to low-income and high-income countries, despite substantial socioeconomic, healthcare, and lifestyle disparities. This robustness suggests that the biological mechanisms linking extreme sleep durations to cognitive impairment might be universal, transcending geographical and cultural boundaries. Our findings substantiate and extend previous conclusions from single-country studies, offering robust, generalizable evidence for formulating sleep-cognitive health management strategies in the context of global aging.
The potential detrimental link of short sleep duration to cognitive function may be related to multiple interconnected physiological pathways. At the neurotransmitter level, sleep deprivation is significantly associated with dysregulated GABAergic signaling. Consequently, overactivated GABAergic signaling may induce hyperexcitability of GABAergic neurons in the hippocampal CA1 region, ultimately suppressing overall network excitability and being linked to impaired neuroplasticity. This dysregulated GABAergic signaling provides a plausible neurobiological mechanism for the cognitive deficits observed with extremely short sleep durations [21]. In addition to synaptic regulation, short sleep may be associated with compromised cognitive function through impaired clearance of neurotoxic waste products. Sleep deprivation is known to disrupt the glymphatic system and cerebrospinal fluid-blood barrier function, which may impede the efficient removal of amyloid-β (Aβ) [22]. In addition, the gut–brain axis has been hypothesized to be a potential mediator in animal and biomarker-based studies; however, this remains speculative in the context of our observational data and warrants further investigation [23].
In contrast to short sleep, the mechanisms underlying the association between long sleep duration and cognitive impairment are less well understood but may involve distinct pathological processes. The potential underlying mechanisms may include the following interrelated pathological processes: elevated plasma concentrations of Aβ40 and total tau protein (t-tau), alongside a decreased Aβ42/Aβ40 ratio in long sleepers [24]. Given that Aβ42 promotes the clearance of Aβ40, a decreased ratio may indicate impaired Aβ clearance mechanisms in the brain. Impaired brain clearance may be associated with abnormal deposition of Aβ (particularly Aβ40) within the brain parenchyma, which is linked to the formation of amyloid plaques. These plaques may disrupt neuronal synaptic structure and function, interfere with interneuronal signaling [25], and activate glial cells to induce neuroinflammation [26]. Neurostructural alterations further compound this pathology; long sleepers exhibit significant gray matter atrophy, most pronounced in the occipital lobe. Such structural deficits may be associated with impaired visual processing and higher-order network integration, which is hypothesized to be linked to accelerated cognitive decline [27]. These pathological processes collectively may be associated with neurodegeneration, forming the potential pathological basis for the link between long sleep duration and cognitive outcomes.
The relationship between sleep duration and cognitive function demonstrated considerable heterogeneity across age groups. Our analyses, which are consistent with large-scale studies such as the UK Biobank, identify the middle-aged population (50–64 years) as particularly vulnerable, exhibiting a pronounced inverted U-shaped association where both short and long sleep durations are associated with cognitive decline [28]. In contrast, the relationship appears to shift in the oldest-old (≥ 75 years), where only long sleep duration (≥ 9 h) emerges as a significant risk factor for cognitive impairment, with stronger associations observed with increased sleep time. This alteration may be associated with the deterioration of sleep microstructure in older adults during aging. Studies have shown that electrophysiological sleep markers closely related to cognitive function—such as slow-wave activity [29] and spindle power [30]—begin to decline as early as midlife. This degradation in microstructure is often accompanied by alterations in macroscopic sleep architecture, along with structural degeneration in several brain regions critical for sleep–cognition regulation, including volume reduction and cortical thinning in the medial prefrontal cortex, lateral orbitofrontal cortex, and hippocampus [7]. Collectively, these structural and functional changes may contribute to a gradual decline in sleep-dependent memory consolidation and cognitive regulation.
Sex further moderates the sleep-cognition relationship. Our stratified analyses revealed that women exhibit greater cognitive vulnerability to long sleep durations, whereas men are more susceptible to the potential detrimental effects of short sleep durations. This sexual dimorphism is supported by a study of older adults in Taiwan, which revealed that the association between long sleep duration and cognitive impairment was significantly stronger in women [31]. The underpinnings of this disparity are likely multifactorial, potentially involving physiological differences such as generally longer total sleep time [32] and higher levels of slow-wave activity (SWA) in women [33], which might render them more sensitive to the pathological processes associated with prolonged time in bed.
A key strength of this study lies in its utilization of diverse multinational samples. This study is the first to simultaneously include cohorts from China (upper-middle-income), England (high-income), and India (low-income), which represent populations with stark socioeconomic differences. Our novel finding demonstrates that the potential detrimental association of extreme sleep duration with cognition is stable across populations with varying healthcare resources, lifestyle factors, and cultural backgrounds. This conclusion not only aligns with findings from single-nation studies but also strengthens the scientific hypothesis of a ‘universal sleep-cognition association’ through cross-population evidence. This study provides critical evidence for the formulation of standardized sleep intervention strategies within the context of global aging. However, this study has several limitations. Several methodological limitations should be acknowledged. First, sleep duration was assessed inconsistently across cohorts: CHARLS and ELSA captured habitual sleep patterns, whereas LASI relied on a single-night report. This discrepancy may reduce direct comparability and introduce measurement bias. Second, sleep data were self-reported and thus susceptible to recall bias, as objective measures such as actigraphy were not available. Third, important sleep-related confounders—including obstructive sleep apnoea, hypnotic medication use, and sleep quality indicators such as fragmentation—were not adjusted for and may represent sources of residual confounding. Taken together, these limitations imply that some of the observed variation in sleep–cognition associations could reflect methodological heterogeneity rather than true biological or behavioural differences. Finally, although cognitive scores were harmonised using z-standardisation, subtle variations in assessment tools across surveys—such as the use of the 8-item CES-D in ELSA—may have introduced additional measurement bias.
Conclusion
This study, through the analysis of data from middle-aged and older adults in China, England and India, confirms an inverted U-shaped association between sleep duration and cognitive function across diverse cultural contexts. Prolonged/extreme sleep durations (≤ 4 h or ≥ 9 h) significantly impair cognitive function, with the detrimental effects of long sleep duration being more pronounced. This association is moderated by age and sex: the middle-aged population (50–64 years) demonstrates greater sensitivity to extreme sleep durations, whereas the older population (≥ 75 years) is affected primarily by long sleep durations; males are more susceptible to cognitive decline from short sleep durations, whereas females exhibit heightened sensitivity to long sleep durations. These findings suggest that maintaining approximately 7 h of sleep may be a crucial measure for preserving cognitive function in middle-aged and older adults, necessitating the development of differentiated sleep health strategies on the basis of age and sex. Future research should first establish causal relationships through longitudinal, cross-national studies that incorporate objective sleep measurements (e.g., actigraphy)—this will address the limitations of inconsistent and self-reported sleep assessment in the current cross-sectional design. Second, to reinforce the generalizability of our findings, future cross-national investigations should explore the consistency of biological mechanisms (e.g., amyloid-β, tau protein, GABAergic signaling) across these diverse populations, which will help verify whether the sleep-cognition association is driven by universal neurobiological pathways. Additionally, further research should continue to clarify the roles of sleep quality, genetics, and other potential contributing elements.
Supplementary Information
Acknowledgements
We would like to thank the research teams and participants of CHARLS, ELSA and LASI for their contributions to this study.
Abbreviations
- Aβ
Amyloid-β
- BMI
Body Mass Index
- CHARLS
China Health and Retirement Longitudinal Study
- CES-D
Center for Epidemiologic Studies Depression Scale
- CI
Confidence Interval
- ELSA
English Longitudinal Study of Ageing
- FDR
False Discovery Rate
- GAM
Generalized Additive Model
- HRS
Health and Retirement Study
- ISCED
International Standard Classification of Education
- IPW
Inverse probability weighting
- LASI
Longitudinal Aging Study in India
- LRT
Likelihood Ratio Test
- SD
Standard Deviation
- SWA
Slow-Wave Activity
- t-tau
Total Tau Protein
- VIF
Variance Inflation Factor
Authors’ contributions
LS designed the study, collected and analyzed core data, conducted statistical analysis, and drafted the manuscript. RL participated in data collection, assisted with statistical analysis, and revised the manuscript critically. CZ extracted and verified the dataset, contributed to data interpretation, and reviewed the manuscript. YY supported data collation, participated in the discussion of research results, and helped with manuscript proofreading. LL and PC supervised the entire study design, reviewed and edited the manuscript, and ensured the accuracy and integrity of the research content. All authors contributed to the intellectual content of the article during manuscript drafting, reviewed the final version, and approved the submitted manuscript.
Funding
This work was supported by the Science and Technology Development Plan Project of Jilin Province, China (Grant No. YDZJ202401278ZYTS ).
Data availability
The documentation and code for the harmonized datasets used here are provided on the Gateway to Global Aging Data website, along with links to the parent cohort data: https://g2aging.org/downloads. The analysis code can be found at https://g2aging.org/harmonized-data/get-data. The CHARLS data are freely accessible at: https://charls.charlsdata.com/pages/data/111/zh-cn.html. The ELSA data are freely available and can be accessed via the UK Data Service (SN 5050):https://discover.ukdataservice.ac.uk. The LASI data can be obtained at: (https://iipsindia.ac.in/content/LASI-data).
Declarations
Ethics approval and consent to participate
We utilized de-identified data from publicly available databases: CHARLS, ELSA and LASI. The ethical approval for this secondary analysis was waived as the original surveys obtained ethical clearance from their respective institutional review boards and informed consent from all participants. This secondary analysis strictly adheres to the ethical principles of the Declaration of Helsinki.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Contributor Information
Peng Chen, Email: cpeng@jlu.edu.cn.
Longyun Li, Email: longyun@jlu.edu.cn.
References
- 1.Nichols E, Steinmetz JD, Vollset SE, Fukutaki K, Chalek J, Abd-Allah F, Abdoli A, Abualhasan A, Abu-Gharbieh E, Akram TT, et al. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the Global Burden of Disease Study 2019. Lancet Public Health. 2022;7(2):e105–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Leong RLF, Chee MWL. Understanding the Need for Sleep to Improve Cognition. Ann Rev Psychol. 2023;74(1):27–57. [DOI] [PubMed] [Google Scholar]
- 3.Gottesman RF, Lutsey PL, Benveniste H, Brown DL, Full KM, Lee J-M, Osorio RS, Pase MP, Redeker NS, Redline S et al. Impact of Sleep Disorders and Disturbed Sleep on Brain Health: A Scientific Statement From the American Heart Association. Stroke. 2024;55(3):e61–e76. [DOI] [PubMed]
- 4.Laura DL. The interconnected causes and consequences of sleep in the brain. Science. 2021;374(6567):564–8. [DOI] [PMC free article] [PubMed]
- 5.Peng Q, Cheng D, Aifen L, Juanjuan X, Junyu W. Exploring the relationship of sleep duration on cognitive function among the elderly: a combined NHANES 2011–2014 and mendelian randomization analysis. BMC Geriatr. 2024;24(1):935. [DOI] [PMC free article] [PubMed]
- 6.Ma Y, Liang L, Zheng F, Shi L, Zhong B, Xie W. Association Between Sleep Duration and Cognitive Decline. JAMA Netw Open. 2020;3(9):e2013573. [DOI] [PMC free article] [PubMed]
- 7.Yuzhu L, Barbara JS, Jujiao K, Christelle L, Wei Z, Chao X, Shitong X, Jintai Y, Wei C, Jianfeng F. The brain structure and genetic mechanisms underlying the nonlinear association between sleep duration, cognition and mental health. Nat Aging. 2023;2(5):425–37. [DOI] [PubMed]
- 8.Yanliqing S, Haoqiang L, KeNan G, Yue L. Association between nighttime sleep duration, nap time, and mild cognitive impairment in Chinese older adults: a cross-sectional study. BMC Public Health. 2024;24(1):2381. [DOI] [PMC free article] [PubMed]
- 9.Zhao Y, Hu Y, Smith JP, Strauss J, Yang G. Cohort Profile: The China Health and Retirement Longitudinal Study (CHARLS). Int J Epidemiol. 2014;43(1):61–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Steptoe A, Breeze E, Banks J, Nazroo J. Cohort Profile: The English Longitudinal Study of Ageing. Int J Epidemiol. 2013;42(6):1640–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Perianayagam A, Bloom D, Lee J, Parasuraman S, Sekher TV, Mohanty SK, Chattopadhyay A, Govil D, Pedgaonkar S, Gupta S, et al. Cohort Profile: The Longitudinal Ageing Study in India (LASI). Int J Epidemiol. 2022;51(4):e167–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Cadar D, Brocklebank L, Yan L, Zhao Y, Steptoe A, Raymo JM. Socioeconomic and Contextual Differentials in Memory Decline: A Cross-Country Investigation Between England and China. Journals Gerontology: Ser B. 2023;78(3):544–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Zheng G, Zhou B, Fang Z, Chen X, Liu M, He F, Zhang H, Zhu H, Dong Y, Hao G. Long-Term Visit‐to‐Visit Blood Pressure Variability and Cognitive Decline Among Patients With Hypertension: A Pooled Analysis of 3 National Prospective Cohorts. Journal of the American Heart Association. 2024;13(13):e035504. [DOI] [PMC free article] [PubMed]
- 14.Wang Y, Wu Z, Duan L, Liu S, Chen R, Sun T, Wang J, Zhou J, Wang H, Huang P. Digital exclusion and cognitive impairment in older people: findings from five longitudinal studies. BMC Geriatr. 2024;24(1):406. [DOI] [PMC free article] [PubMed]
- 15.Ma Y, Hua R, Yang Z, Zhong B, Yan L, Xie W. Different hypertension thresholds and cognitive decline: a pooled analysis of three ageing cohorts. BMC Med. 2021;19(1):287. [DOI] [PMC free article] [PubMed]
- 16.Zhao X, Xu X, Yan Y, Lipnicki DM, Pang T, Crawford JD, Chen C, Cheng C-Y, Venketasubramanian N, Chong E et al. Independent and joint associations of cardiometabolic multimorbidity and depression on cognitive function: findings from multi-regional cohorts and generalisation from community to clinic. Lancet Reg Health - Western Pac. 2024;51:101198. [DOI] [PMC free article] [PubMed]
- 17.Cho T-C, Yu X, Gross AL, Zhang YS, Lee J, Langa KM, Kobayashi LC. Negative wealth shocks in later life and subsequent cognitive function in older adults in China, England, Mexico, and the USA, 2012–18: a population-based, cross-nationally harmonised, longitudinal study. Lancet Healthy Longev. 2023;4(9):e461–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Hong M, Thind A, Zaric GS, Sarma S. Emergency department use following incentives to provide after-hours primary care: a retrospective cohort study. Can Med Assoc J. 2021;193(3):E85–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Angel G, Hossein Z, Wehmah J, Manshu Y, Chidinma AI, Yidan C, Michele B, Marilyn G, Gayle P, Denise LW et al. Evaluating Depressive Symptoms Among Low-Socioeconomic-Status African American Women Aged 40 to 75 Years With Uncontrolled Hypertension: A Secondary Analysis of a Randomized Clinical Trial. JAMA Psychiatry. 2021;78(4):426–32. [DOI] [PMC free article] [PubMed]
- 20.Zheng F, Li C, Hua R, Liang J, Gao D, Xie W. Sex differences in changes of depressive symptoms among older adults before and during the COVID-19 pandemic: evidence from two longitudinal cohorts. BMC Geriatr. 2023;23(1):64. [DOI] [PMC free article] [PubMed]
- 21.Yun L, Siwen L, Jiafeng Y, Jingyu F, Shuqi M, Yize L, Lina Z, Yonghao Y. Preoperative Sleep Deprivation Exacerbates Anesthesia/Surgery-induced Abnormal GABAergic Neurotransmission and Neuronal Damage in the Hippocampus in Aged Mice. Mol Neurobiol. 2025;62(7):9385-9398. [DOI] [PubMed]
- 22.Liu H, Barthélemy NR, Ovod V, Bollinger JG, He Y, Chahin SL, Androff B, Bateman RJ, Lucey BP. Acute sleep loss decreases CSF-to‐blood clearance of Alzheimer’s disease biomarkers. Alzheimer’s Dement. 2023;19(7):3055–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Wang Z, Chen W-H, Li S-X, He Z-M, Zhu W-L, Ji Y-B, Wang Z, Zhu X-M, Yuan K, Bao Y-P, et al. Gut microbiota modulates the inflammatory response and cognitive impairment induced by sleep deprivation. Mol Psychiatry. 2021;26(11):6277–92. [DOI] [PubMed] [Google Scholar]
- 24.Tang S, Liu R, Ren J, Song L, Dong L, Qin Y, Zhao M, Wang Y, Dong Y, Zhao T et al. Association of objective sleep duration with cognition and brain aging biomarkers in older adults. Brain Commun. 2024;6(3):fcae144. [DOI] [PMC free article] [PubMed]
- 25.Peng L, Bestard-Lorigados I, Song W. The synapse as a treatment avenue for Alzheimer’s Disease. Mol Psychiatry. 2022;27(7):2940–9. [DOI] [PubMed] [Google Scholar]
- 26.Zhu Y, Guo X, Li S, Wu Y, Zhu F, Qin C, Zhang Q, Yang Y. Naringenin ameliorates amyloid-β pathology and neuroinflammation in Alzheimer’s disease. Commun Biology. 2024;7(1):912. [DOI] [PMC free article] [PubMed]
- 27.González KA, Tarraf W, Stickel AM, Kaur S, Agudelo C, Redline S, Gallo LC, Isasi CR, Cai J, Daviglus ML, et al. Sleep duration and brain MRI measures: Results from the SOL-INCA MRI study. Alzheimer’s Dement. 2023;20(1):641–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Tai XY, Chen C, Manohar S, Husain M. Impact of sleep duration on executive function and brain structure. Commun Biology. 2022;5(1):201. [DOI] [PMC free article] [PubMed]
- 29.Münch M, Knoblauch V, Blatter K, Schröder C, Schnitzler C, Kräuchi K, Wirz-Justice A, Cajochen C. The frontal predominance in human EEG delta activity after sleep loss decreases with age. Eur J Neurosci. 2004;20(5):1402–10. [DOI] [PubMed] [Google Scholar]
- 30.Luigi DG, Michele F. Sleep spindles: an overview. Sleep Med Rev. 2003;7(5):423–40. [DOI] [PubMed]
- 31.Chiu HY, Lai FC, Chen PY, Tsai PS. Differences Between Men and Women Aged 65 and Older in the Relationship Between Self-Reported Sleep and Cognitive Impairment: A Nationwide Survey in Taiwan. J Am Geriatr Soc. 2016;64(10):2051–8. [DOI] [PubMed] [Google Scholar]
- 32.Gildner TE, Liebert MA, Kowal P, Chatterji S, Snodgrass JJ. 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). J Clin Sleep Med. 2014;10(06):613–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Hajali V, Andersen ML, Negah SS, Sheibani V. Sex differences in sleep and sleep loss-induced cognitive deficits: The influence of gonadal hormones. Horm Behav. 2019;108:50–61. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The documentation and code for the harmonized datasets used here are provided on the Gateway to Global Aging Data website, along with links to the parent cohort data: https://g2aging.org/downloads. The analysis code can be found at https://g2aging.org/harmonized-data/get-data. The CHARLS data are freely accessible at: https://charls.charlsdata.com/pages/data/111/zh-cn.html. The ELSA data are freely available and can be accessed via the UK Data Service (SN 5050):https://discover.ukdataservice.ac.uk. The LASI data can be obtained at: (https://iipsindia.ac.in/content/LASI-data).