Sleep quality modifies the association between nocturnal sleep, daytime napping, and mild cognitive impairment: A population-based dose-response study

Abstract

Objectives

This cross-sectional study, based on the two-process model of sleep regulation, explored the nonlinear relationships and interactions among nocturnal sleep duration, daytime napping, sleep quality, and mild cognitive impairment (MCI) in middle-aged and older Chinese adults.

Methods

Data from 7480 participants in the 2018 wave of the China Health and Retirement Longitudinal Study (CHARLS) were analyzed. Restricted cubic splines modeled dose-response relationships, and stratified logistic regression assessed subgroup variations. Models were adjusted for sociodemographic, lifestyle, and clinical covariates.

Results

The lowest odds of MCI were observed among participants reporting 6–8 h of nocturnal sleep and 30–60 min of daytime napping. Short (<6 h) and long (>8 h) sleep durations, along with prolonged napping (≥90 min), were linked to higher odds of MCI. Sleep quality modified these associations: poor sleep quality elevated the odds of MCI, especially among long sleepers (OR = 1.88 vs. 1.33 for good sleep quality). Among short sleepers (<6 h), 30–60 min naps were associated with lower odds of MCI (OR = 0.72).

Conclusions

These findings indicate that, in this cross-sectional sample, 7–8 h of nocturnal sleep combined with 30–60 min of daytime napping were associated with better cognitive status, whereas long sleep with poor sleep quality was associated with higher odds of impairment. These observational associations, considered within a circadian–homeostatic framework, are hypothesis-generating and warrant confirmation in future longitudinal and interventional studies.

Highlights

• •7–8 h nocturnal sleep + 30–60 min nap associated with lower MCI odds.
• •Long sleep (>8 h) with poor quality increases MCI odds (OR=1.88).
• •Napping benefits short sleepers, but not long sleepers.

1. Introduction

As the global population ages, dementia has emerged as a major public health challenge [1,2]. In China, approximately 6.0% of the elderly population is affected by dementia, and the number of affected individuals is projected to reach 27.29 million by 2030 [3,4]. Mild cognitive impairment (MCI) is characterized by declines in memory, attention, and other cognitive functions that exceed age-related expectations, representing a transitional state between normal aging and dementia [5]. With an annual conversion rate of 10–15%, MCI presents a critical opportunity for early intervention to slow disease progression [6].

Sleep, as a modifiable lifestyle factor, has increasingly been recognized as a promising target for interventions aimed at delaying cognitive decline [[7], [8], [9]]. Borbély's two-process model provides a neurobiological framework for understanding sleep-wake patterns, positing that sleep is regulated by two interacting processes: a homeostatic process (Process S) that reflects the increasing need for sleep during wakefulness, and a circadian process (Process C) that governs sleep-wake timing via the suprachiasmatic nucleus [10]. This model serves as a valuable framework for examining the relationship between sleep patterns and cognitive health.

Extensive research has investigated the associations between individual sleep parameters and cognitive function. Nocturnal sleep duration often exhibits a U-shaped relationship with cognitive impairment, where both short (<6 h) and long (>8 h) sleep durations are linked to increased risks of MCI and cognitive decline, while optimal cognitive performance is consistently associated with 6 to 8 h of sleep [[11], [12], [13], [14]]. In China, napping is considered an integral component of a healthy lifestyle. Data from the Chinese Elderly Cohort indicate a high prevalence of napping at 57.7% (averaging 63.3 min) [15], and studies suggest a non-linear association between napping duration and MCI, with both short and prolonged daytime naps linked to an increased risk of cognitive decline [[16], [17], [18]]. However, meta-analytic evidence remains inconsistent: some studies report significant adverse effects of prolonged napping [19], whereas others find no robust associations [20]. Beyond sleep duration, sleep quality—characterized by reduced sleep efficiency, increased fragmentation, and restlessness—is increasingly recognized as an even stronger predictor of dementia risk [21,22], potentially mediated by neurobiological mechanisms such as amyloid-β clearance and synaptic homeostasis [23,24]. Notably, emerging evidence indicates that the combination of long nocturnal sleep (>8 h) and poor sleep quality is associated with the most pronounced cognitive deficits [9,25]. Despite these insights, most prior studies have examined these sleep parameters in isolation, overlooking the inherent interactions among nocturnal sleep, daytime napping, and sleep quality as posited by the two-process model. Consequently, there is a critical need to investigate their combined effects on cognitive health, particularly in populations where daytime napping is culturally prevalent.

Grounded in Borbély’s two-process model, this study proposes two hypotheses: first, that sleep quality interacts with nocturnal sleep duration to influence cognitive function; and second, that nocturnal sleep duration moderates the relationship between daytime napping and cognitive outcomes. This study aims to characterise dose–response relationships and interaction patterns across 24-h sleep dimensions in relation to cognitive status. Ultimately, the goal is to generate hypotheses that may inform the design of tailored sleep interventions and the incorporation of circadian–homeostatic principles into future geriatric care research.

2. Methods

2.1. Study population

This study used data from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative survey of adults aged 45 and older [26]. CHARLS employs a multistage, stratified sampling design across 150 counties and 450 communities, collecting data from 10,257 households and 17,708 individuals. Structured interviews were conducted to collect comprehensive information on demographics, health status, cognitive function, mental health, retirement, income, family structure, and social support. All survey waves of CHARLS received approval from the Biomedical Ethics Committee of Peking University (IRB00001052-11015). Written informed consent was obtained from all participants, with signed copies securely stored and digitized by the CHARLS team. The current analysis included participants from the fourth wave of the survey. Individuals were excluded if they did not participate in or failed to complete the cognitive function assessment; had missing or unreliable data on sleep duration, nap duration, or sleep quality; reported extreme values—defined as nighttime sleep exceeding 15 h or napping longer than 240 min, which are outside the plausible range of habitual sleep patterns and likely to reflect reporting or recording errors; self-reported a diagnosis of mental disorders or memory-related illnesses; or had missing data on key covariates (Fig. 1).

Fig. 1.

Participant screening and inclusion flowchart.

2.2. Main variables

2.2.1. Cognitive impairment

Cognitive impairment was assessed in the CHARLS study following the protocol of the Health and Retirement Study (HRS), using the Telephone Interview for Cognitive Status-10 (TICS-10) to evaluate cognitive function across four domains: orientation, calculation, memory, and drawing [27,28]. Orientation was evaluated through questions about the current date, weekday, and season, with responses scored on a 0–5 scale. Calculation ability was assessed via five serial subtractions of 7 from 100, scored from 0 to 5. Memory was evaluated through three trials of immediate recall of ten words (with scores averaged to a 0–10 scale) and one trial of delayed recall (scored 0–10). Drawing ability was assessed by figure copying tasks, with 1 point assigned for accurate reproduction. The total TICS-10 score ranged from 0 to 31. The TICS-10 has been extensively validated and is widely used in China for the early detection of MCI and dementia [29]. Research indicates that cognitive performance generally declines gradually with advancing age, with particularly marked reductions occurring at approximately 57, 70, and 78 years of age [30]. Prior studies have defined MCI according to the Age-Associated Cognitive Decline (AACD) criteria [31,32]. In this study, MCI was operationally defined as scoring at least one standard deviation below the mean TICS-10 score of one’s respective five-year age group, indicating significant cognitive decline relative to age peers. This age-stratified definition, adapted from AACD criteria and previous population-based studies, improves the precision of identifying individuals who exhibit accelerated cognitive deterioration compared with age-matched peers.

2.2.2. Sleep variables

Guided by Borbély's two-process model, sleep was analyzed through three variables: nocturnal sleep duration, nap duration, and sleep quality. Nocturnal sleep was assessed using the question: “How many hours did you typically sleep each night during the past month?”. Based on previous research results [33,34], this study classified the duration of nighttime sleep into short sleep (<6 h), normal sleep (6–8 h; reference group), and long sleep (>8 h). Nap duration, assessed by the question “How long did you typically nap during the past month?”, was recorded in minutes. Given the widespread practice of napping among middle-aged and elderly individuals in China, nap duration was categorized as no nap (0 min), normal nap (1–30 min, reference group), long nap (31–89 min), or very long nap (≥90 min) [35].

Sleep quality was assessed with the question: “During the past week, how often did you have trouble with sleep?” with response options ranging from “Rarely or none of the time (<1 day)” to “Most or all of the time (5–7 days)”. In this study, this item was used as an index of subjective sleep quality, rather than objective sleep fragmentation. Based on the distribution of cognitive scores across the four response categories (Fig. 2), and to ensure sufficient sample sizes and model stability, we dichotomised sleep quality into two groups: “better sleep quality” (rarely or none of the time, or 1–2 days) and “poorer sleep quality” (3–4 days or 5–7 days), as participants in the first two and last two categories had broadly similar cognitive performance.

Fig. 2.

Cognitive performance across different frequencies of poor sleep during the past week.

2.2.3. Covariates

The covariates included demographic characteristics, lifestyle factors, and chronic disease status. Demographic variables comprised age, sex, marital status, region of residence, and educational level. Lifestyle factors encompassed smoking status, alcohol consumption, and life satisfaction. The analysis adjusted for the presence of key chronic conditions, including hypertension, diabetes, and dyslipidemia.

2.2.4. Statistical analysis

All statistical analyses were conducted using R software (version 4.2.0; R Foundation for Statistical Computing, Vienna, Austria). Continuous variables were assessed for normality via the Kolmogorov–Smirnov test and are reported as means ± standard deviations (SD); between-group differences were examined using independent samples t-tests or one-way analysis of variance (ANOVA), as appropriate. Categorical variables were summarised as frequencies and percentages (n, %) and compared using chi-square or Fisher’s exact tests based on expected cell counts. Multicollinearity among independent variables was evaluated using variance inflation factor (VIF) calculations, with VIF <5 indicating acceptable levels of collinearity. Baseline characteristics were stratified by categories of nocturnal sleep duration, nap duration, and sleep quality.

To examine potential non-linear dose–response relationships, we fitted restricted cubic spline (RCS) models with three knots to assess associations between sleep duration (nocturnal sleep and nap duration) and the odds of MCI. The use of three knots is a commonly applied specification in epidemiological research, providing adequate flexibility to capture non-linear patterns while limiting model overfitting. We additionally conducted stratified analyses to examine whether these non-linear associations differed across predefined subgroups.

For multivariable modelling, two sets of logistic regression models were implemented: an unadjusted model including only sleep-related variables (nocturnal sleep duration, nap duration, and sleep quality) and a fully adjusted model controlling for all sociodemographic, lifestyle, and clinical covariates. Additional stratified logistic regression analyses were used to investigate associations between nap duration and MCI within categories of nocturnal sleep duration, and between nocturnal sleep duration and MCI within levels of sleep quality. Linear regression models were also developed for each sleep variable–cognitive score pair, with Model 1 including main effects only and Model 2 adding interaction terms. Overall interaction significance was evaluated using ANOVA, and marginal effect plots were generated to visualise these interactions and to elucidate the interrelationships among sleep-related factors. All analyses accounted for the complex survey design of CHARLS through the application of appropriate sampling weights.

3. Results

3.1. Participant characteristics by MCI

Among the 7480 participants, 1186 (15.9%) fulfilled the criteria for MCI. No significant difference in age was observed between the groups (normal cognition: 59.65 ± 8.81 years; MCI: 59.68 ± 8.88 years; P = 0.916). Significant between-group differences were found in region, education level, marital status, smoking behavior, alcohol consumption, and life satisfaction (all P < 0.05). There was a statistically significant difference in sleep patterns between the groups (P < 0.001). Compared with the normal cognition group, the MCI group presented greater proportions of both short (<6 h: 34.5% vs. 29.6%) and long (>8 h: 8.2% vs. 5.7%) sleep durations and a lower proportion of normal sleep durations (6–8 h: 57.3% vs. 64.7%). For napping, the MCI group had higher rates of no napping (42.2% vs. 36.1%), prolonged napping (≥90 min: 21.6% vs. 16.9%), and a lower proportion reporting of good sleep quality (61.0% vs. 69.4%) (Table 1).

Table 1.

Baseline characteristics of participants stratified by MCI (n = 7480).

Variable	No MCI (%)	MCI (%)	P
Participants	6294	1186
Age (mean ± SD)	59.65(8.81)	59.68(8.88)	0.916
Region n (%)			<0.001
Rural	4202(66.8)	1000(84.3)
Urban	2092(33.2)	186(15.7)
Education Level n (%)			<0.001
illiteracy	144(2.3)	177(14.9)	<0.001
Primary school dropout	873(13.9)	326(27.5)
Junior high school education	1591(25.3)	352(29.7)
High school education	2174(34.5)	250(21.1)
Bachelor degree or above	1512(24.0)	81(6.8)
Gender n (%)			0.072
Male	3488(55.4)	623(52.5)
Female	2806(44.6)	563(47.5)
Smoking n (%)			0.002
Never	3449(54.8)	629(53.0)
Quit	1042(16.6)	164(13.8)
Still	1803(28.6)	393(33.1)
Alcohol Consumption n (%)			0.001
More than Once a Month	2034(32.3)	355(29.9)
Drinks less than once a month	616(9.8)	84(7.1)
None of These	3644(57.9)	747(63.0)
Marriage Status n (%)			<0.001
Married	5341(84.9)	955(80.5)
Separated	426(6.8)	89(7.5)
Alone	527(8.4)	142(12.0)
Life Satisfaction n (%)			<0.001
Completely Satisfied	342(5.4)	61(5.1)
Very Satisfied	1684(26.8)	351(29.6)
Somewhat Satisfied	3795(60.3)	628(53.0)
Not Very Satisfied	386(6.1)	118(9.9)
Not at All Satisfied	87(1.4)	28(2.4)
Cognitive Score (mean (SD))	21.11(3.63)	12.70(2.94)	<0.001
nocturnal sleep duration n (%)
Short(<6 h)	1865(29.6)	409(34.5)	<0.001
Normal(6∼8 h)	4070(64.7)	680(57.3)
Long(>8 h)	359(5.7)	97(8.2)
Nap Duration n (%)
No(0 min)	2269(36.1)	500(42.2)	<0.001
1∼30 min	1013(16.1)	122(10.3)
31∼89 min	1947(30.9)	308(26.0)
≥90 min	1065(16.9)	256(21.6)
Sleep Quality n (%)			<0.001
Good	4366(69.4)	723(61.0)
Poor	1928(30.6)	463(39.0)
Hypertension n (%)	2228(35.4)	399(33.6)	0.259
Dyslipidemia n (%)	1605(25.5)	255(21.5)	0.004
Diabetes n (%)	833(13.2)	139(11.7)	0.169

Note:Data are presented as n(%) or mean ± SD. P-values are from chi-square or t-tests.

3.2. Dose-response relationships between nocturnal sleep duration and MCI

A nonlinear association was observed between nocturnal sleep duration and MCI (Fig. 3). Overall, the restricted cubic spline curves suggested a U-shaped pattern, and tests for nonlinearity were statistically significant (P < 0.05). In the total sample, the lowest odds of MCI were observed at around 6–7 h of nocturnal sleep. Among participants with good sleep quality, the lowest odds of MCI were observed between approximately 6 and 7.5 h, whereas among those with poor sleep quality, higher odds were seen at both very short durations (around 4 h) and at durations longer than about 6 h.

Fig. 3.

Dose-response relationships between nocturnal sleep duration and MCI stratified by sleep quality. The solid red lines represent ORs derived from restricted cubic spline models, and the pink shading indicates 95% confidence intervals. Tests for overall association and nonlinearity were both significant (P < 0.001). Analyses are shown for (a) all participants, (b) participants with good sleep quality, and (c) participants with poor sleep quality.

3.3. Dose-response relationships between nap duration and MCI

Further analyses explored the nonlinear associations between nap duration and the odds of MCI across categories of nocturnal sleep duration (Fig. 4). U-shaped patterns were observed in the overall sample (Fig. 4a), in short sleepers (<6 h, Fig. 4b) and in normal sleepers (6–8 h, Fig. 4c), with significant tests for nonlinearity (P < 0.05). Among short sleepers, the lowest odds of MCI were seen for nap durations of roughly 30–60 min, and a similar range was suggested in normal sleepers. In contrast, no clear nonlinear association between nap duration and MCI was observed in the long-sleep group (>8 h; P for nonlinearity = 0.266).

Fig. 4.

Dose-response relationships between nap duration and MCI stratified by nocturnal sleep duration. The solid red lines represent OR derived from restricted cubic spline models, and the pink shading indicates 95% confidence intervals. Tests for nonlinearity were significant (P < 0.05) in short and normal sleepers but not in long sleepers (>8 h; P-nonlinear = 0.266). Analyses are shown for (a) all participants, (b) short sleepers (<6 h), (c) normal sleepers (6–8 h), and (d) long sleepers (>8 h).

3.4. Logistic regression analysis of the relationships between sleep parameters and MCI

Multivariate logistic regression analyses revealed significant associations between sleep parameters and MCI (Table 2). According to the adjusted models, a long nocturnal sleep duration (>8 h) was associated with greater odds of MCI than a normal sleep duration (6–8 h) was (OR = 1.333, 95% CI: 1.027–1.716). For nap duration, both no napping (OR = 1.480, 95% CI: 1.187–1.857) and prolonged napping (≥90 min; OR = 1.709, 95% CI: 1.335–2.195) were associated with greater odds of MCI than short napping (1–30 min). Poor sleep quality was also associated with increased odds of MCI (OR = 1.288, 95% CI: 1.107–1.499).

Table 2.

Multivariate logistic regression analysis of sleep parameters associated with MCI.

Variable	Unadjusted model	Adjusted model
	OR (95% CI)	OR (95% CI)
nocturnal sleep duration group
Short(<6 h)	1.135(0.980, 1.313)	1.075(0.920, 1.254)
Normal(6∼8 h)	1.00 (Reference)	1.00 (Reference)
Long(>8 h)	1.595(1.248, 2.022) ∗∗∗	1.333(1.027, 1.716) ∗
Nap Duration group
No(0 min)	1.813(1.472, 2.249) ∗∗∗	1.48(1.187, 1.857) ∗∗
1∼30 min	1.00 (Reference)	1.00 (Reference)
31∼89 min	1.367(1.095, 1.716) ∗∗	1.206(0.954, 1.531) ∗
≥90 min	2.063(1.636, 2.614) ∗∗∗	1.709(1.335, 2.195) ∗∗∗
Sleep Quality group
Good	1.00 (Reference)	1.00 (Reference)
Poor	1.445(1.254, 1.664) ∗∗∗	1.288(1.107, 1.499) ∗∗

Note: ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001. The unadjusted model only includes the sleep variable, whereas the adjusted model incorporates all covariates.

Stratified analyses suggested that the associations between nap duration and MCI differed by nocturnal sleep duration (Table 3). The association between prolonged napping (≥90 min) and MCI was strongest among normal-duration sleepers (6–8 h; OR = 1.881, 95% CI: 1.360–2.629) and remained significant among short sleepers (<6 h; OR = 1.661, 95% CI: 1.082–2.562). No significant associations were detected between nap duration and MCI among long sleepers (>8 h).

Table 3.

Stratified logistic regression analysis of nap duration and MCI stratified by nocturnal sleep duration.

Nighttime Sleep Duration	Nap Duration	Unadjusted model	adjusted model
		OR (95% CI)	OR (95% CI)
ALL	No	1.830 (1.486, 2.269) ∗∗∗	1.481 (1.188, 1.858) ∗∗∗
	1-30 min	1.00 (Reference)	1.00 (Reference)
	31-89 min	1.314 (1.054, 1.647) ∗	1.176 (0.932, 1.492)
	≥90 min	1.996 (1.586, 2.524) ∗∗∗	1.663 (1.302, 2.134) ∗∗∗
Short (<6 h)	No	1.670 (1.209, 2.343) ∗∗	1.446 (1.016, 2.086) ∗
	1-30 min	1.00 (Reference)	1.00 (Reference)
	31-89 min	1.213 (0.839, 1.767)	1.074 (0.719, 1.62)
	≥90 min	1.701 (1.149, 2.531) ∗∗	1.661 (1.082, 2.562) ∗
Normal (6∼8 h)	No	1.990 (1.493, 2.692) ∗∗∗	1.557 (1.152, 2.131) ∗
	1-30 min	1.00 (Reference)	1.00 (Reference)
	31-89 min	1.496 (1.113, 2.038) ∗∗	1.305 (0.958, 1.80)
	≥90 min	2.426 (1.782, 3.341) ∗∗∗	1.881(1.360, 2.629) ∗∗∗
Long (>8 h)	No	1.333 (0.608, 3.173)	0.905 (0.373, 2.347)
	1-30 min	1.00 (Reference)
	31-89 min	0.941 (0.411, 2.310)	0.649 (0.259, 1.720)
	≥90 min	0.972 (0.428, 2.370)	0.628 (0.252, 1.657)

Note: ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001. The unadjusted model only includes the sleep variable, whereas the adjusted model incorporates all covariates.

Additional stratification by sleep quality indicated that the associations between nocturnal sleep duration and MCI varied by sleep-quality status (Table 4). Among those with good sleep quality, both short (<6 h; OR = 1.234, p < 0.05) and long (>8 h; OR = 1.334, p < 0.05) sleep durations were associated with increased odds of MCI. However, among participants with poor sleep quality, only long sleep duration had elevated odds in the unadjusted model (OR = 2.077, p < 0.05), which was attenuated after adjustment (OR = 1.875, 95% CI: 0.94–3.62, p = 0.066).

Table 4.

Stratified logistic regression analysis of nocturnal sleep duration and MCI stratified by sleep quality status.

Sleep Quality	nocturnal sleep duration	Unadjusted model	adjusted model
		OR (95%CI)	OR (95%CI)
ALL	Short(<6 h)	1.313 (1.147, 1.501) ∗∗∗	1.167 (1.009, 1.349) ∗
	Normal(6∼8 h)	1.00 (Reference)	1.00 (Reference)
	Long(>8 h)	1.617 (1.268, 2.045) ∗∗∗	1.335 (1.030, 1.71) ∗
Good	Short(<6 h)	1.357 (1.113, 1.648) ∗∗	1.234 (0.999, 1.518) ∗
	Normal(6∼8 h)	1.00 (Reference)	1.00 (Reference)
	Long(>8 h)	1.653 (1.265, 2.139) ∗∗∗	1.334 (1.004, 1.754) ∗
Poor	Short(<6 h)	0.962 (0.782, 1.186)	0.936 (0.746, 1.174)
	Normal(6∼8 h)	1.00 (Reference)	1.00 (Reference)
	Long(>8 h)	2.077 (1.113, 3.745) ∗	1.875 (0.939, 3.623)

Note: ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001. The unadjusted model only includes the sleep variable, whereas the adjusted model incorporates all covariates.

3.5. Interactions between sleep parameters and cognitive scores

Marginal effect plots with 95% confidence intervals were used to examine the interaction effects between sleep quality, nocturnal sleep duration, and nap duration on cognitive scores (Fig. 5). A marginally significant interaction was observed between sleep quality and nocturnal sleep duration (Fig. 5a, P = 0.061). The participants with good sleep quality and normal nocturnal sleep (6–8 h) had significantly higher cognitive scores than did those with short (<6 h; P < 0.0001) or long (>8 h; p < 0.0001) sleep durations. Short sleepers also scored higher than long sleepers did (p = 0.0039). Among those with poor sleep quality, both short and normal sleep durations were associated with higher scores than long sleep durations were (P < 0.01), but no significant difference was found between short and normal durations. A significant interaction was found between sleep quality and nap duration (Fig. 5b, P = 0.046). Under good sleep quality conditions, 1–30 min naps were associated with significantly higher cognitive scores than no naps, 31–89 min, and ≥90 min nap durations (all p < 0.0001). With poor sleep quality, 1–30 min naps remained associated with higher scores than no naps (P = 0.0102) and ≥90 min naps (P = 0.0001). The interaction effect between nocturnal sleep duration and nap duration did not reach statistical significance (Fig. 5c, P = 0.096). Overall, participants with normal nocturnal sleep duration (6–8 h) and brief daytime naps (1–30 min) tended to have higher cognitive scores than those with shorter or longer sleep and with longer or no naps, particularly among individuals reporting good sleep quality.

Fig. 5.

Marginal effect plots of sleep variables and cognitive scores. (a) Interaction effect between sleep quality and nocturnal sleep duration. (b) Interaction effect between sleep quality and nap duration. (c) Interaction effect between nocturnal sleep duration and nap duration. The error bars represent 95% confidence intervals.

4. Discussion

This cross-sectional study identified non-linear associations between 24-h sleep patterns and cognitive status among middle-aged and older Chinese adults. In particular, 6–8 h of nocturnal sleep combined with approximately 30–60 min of daytime napping was the pattern associated with the lowest odds of MCI, whereas both shorter and longer nocturnal sleep and prolonged napping (≥90 min) were associated with higher odds. Sleep quality modified these associations, with poor sleep quality among long sleepers (>8 h) linked to especially elevated odds of MCI.

Our population-based analyses revealed U-shaped associations between sleep duration and the odds of MCI, with the lowest odds observed among individuals reporting 6–8 h of nocturnal sleep and 30–60 min of daytime napping. Short nocturnal sleep (<6 h) may impair cognitive function by reducing slow-wave sleep, thereby limiting restorative processes essential for synaptic homeostasis and memory consolidation [35,36]. Conversely, long nocturnal sleep (>8 h) was associated with higher odds of MCI, a finding consistent with international cohort studies and potentially reflecting circadian misalignment, subclinical neuropathology, or inflammatory processes linked to prolonged time in bed [12,37,38]. These nonlinear associations align with previous research reporting similar patterns [11,13,16], while extending prior findings by characterising population-specific dose–response profiles in middle-aged and older adults. Comparative evidence suggests that these associations may be underpinned by sleep-related neurobiological mechanisms. Notably, the associations involving daytime napping were context-dependent: short naps (around 30–60 min) were associated with reduced odds of MCI primarily among individuals with insufficient nocturnal sleep, in a pattern that is consistent with the role of sleep homeostasis. In these individuals, napping may help to alleviate accumulated sleep pressure and support attentional recovery [39]. In contrast, the absence of an association among long sleepers may indicate saturation of homeostatic regulation or underlying circadian disruption [10].

Sleep quality emerged as a critical effect modifier, particularly among long sleepers. In our study, the subgroup with extended nocturnal sleep duration and poor sleep quality had the highest odds of MCI, suggesting that the combined adverse effects may be greater than the impact of either factor considered alone. Previous studies have consistently shown that both long sleep duration and poor sleep quality are linked to cognitive impairment [40]. This pattern is consistent with evidence that poor sleep quality contributes to neurobiological disruptions, including impaired glymphatic clearance of neurotoxic proteins and elevated levels of inflammatory markers such as interleukin-6 (IL-6) and C-reactive protein (CRP) [38,41]. Neuropathological research further indicates that sleep fragmentation is associated with reduced efficiency in clearing pathological biomarkers such as amyloid-β, independent of total sleep duration [24,42]. A recent meta-analysis of 72 studies using objective sleep measures, including actigraphy and polysomnography, found small but significant associations between sleep continuity indices (such as sleep efficiency, wake after sleep onset, and night-time restlessness) and cognitive performance in healthy older adults, particularly in memory and executive functions, whereas total sleep time itself was not consistently related to cognition [43]. These meta-analytic findings are broadly consistent with the pattern observed in our study and raise the possibility that subjective reports of “poor sleep” may partly reflect difficulties in maintaining consolidated sleep, although this hypothesis would need to be tested in future work using combined subjective and objective sleep assessments. From the perspective of geriatric health research, individuals who exhibit prolonged sleep duration coupled with poor sleep quality may represent a subgroup at elevated risk who warrant closer monitoring in future studies. In community-based settings, brief assessments of sleep quality could potentially serve as pragmatic tools for flagging individuals at higher risk, but this possibility requires confirmation in longitudinal and interventional research.

Nocturnal sleep duration appeared to modulate the association between daytime napping and cognitive status. Daytime napping was inversely associated with the odds of cognitive impairment, but only among individuals with nocturnal sleep duration ≤8 h. Specifically, among those with short nocturnal sleep (<6 h), naps lasting 30–60 min were linked to a 28% lower odds of cognitive impairment. In contrast, no significant association was observed among long sleepers (>8 h). Rather than indicating a strict threshold, this pattern is compatible with Borbély’s dual-process model of sleep regulation [10,38], which proposes two interacting mechanisms: first, homeostatic sleep pressure (Process S), whereby daytime napping may reduce accumulated sleep debt and support synaptic recovery [39]; and second, circadian regulation (Process C), where pre-existing circadian misalignment in long sleepers may constrain the restorative potential of napping [44]. These observational findings highlight potential targets and subgroups for future intervention studies on cognitive ageing. For individuals with shorter nighttime sleep duration (<6 h), it would be informative for future trials to test whether encouraging daytime naps of around 30–60 min improves cognitive outcomes. Among those with longer sleep duration (>8 h), our results suggest that sleep quality may be a more relevant focus than promoting additional napping, although this needs to be confirmed prospectively. Individuals with normal sleep duration (6–8 h) might not require changes to their current napping habits from a cognitive perspective, but this too should be evaluated in longitudinal research. Rather than providing direct recommendations for clinical practice or community programmes, the present findings should be viewed as hypothesis-generating and useful for guiding the design of future studies. However, because nap timing was not recorded in CHARLS, we were unable to determine whether the associations with napping differed according to when the nap occurred in relation to an individual’s circadian phase.

However, this study has several limitations. First, sleep parameters were assessed using self-report questionnaires, which are susceptible to recall bias and measurement inaccuracy, particularly among individuals with mild cognitive impairment. This population may experience greater difficulty in accurately recalling sleep patterns, potentially compromising the validity of the observed associations. Future research should incorporate objective measures such as actigraphy or polysomnography to corroborate subjective reports. Second, CHARLS did not assess the timing of daytime naps (e.g. morning vs. afternoon, proximity to nocturnal sleep). Thus, we could not examine whether nap timing relative to the individual’s circadian phase contributed to the observed associations, particularly among long sleepers. Future studies incorporating more detailed information on nap timing and circadian phase markers would help clarify the potential role of circadian misalignment in the relationship between napping and cognitive outcomes. Third, although core sleep dimensions—including nighttime sleep duration, nap duration, and sleep quality—were evaluated, the assessment did not capture a comprehensive sleep architecture or diagnose specific sleep disorders. This omission limits the ability to examine the contribution of particular sleep disorders to cognitive function. Fourth, the cross-sectional design precludes causal inference regarding the relationship between sleep patterns and cognitive status. It is possible that early neurodegenerative changes concurrently impair both sleep regulation and cognitive performance, suggesting a bidirectional association that can only be clarified through longitudinal investigations. To advance understanding in this area, future studies should adopt longitudinal cohort designs that integrate objective sleep monitoring with standardised neuropsychological assessments, thereby helping to strengthen causal inference, minimise bias, and elucidate the mechanisms linking sleep and cognition in ageing populations.

5. Conclusion

This population-based study identified specific 24-h sleep patterns associated with better cognitive status among Chinese middle-aged and older adults: 7–8 h of nocturnal sleep combined with 30–60 min of daytime napping. Sleep quality significantly modified these associations, with particularly elevated odds of MCI observed among individuals reporting long sleep duration (>8 h) and poor sleep quality. These cross-sectional findings, interpreted within a circadian–homeostatic framework, suggest that both sleep duration and sleep quality, together with culturally embedded napping practices, are important dimensions to consider in future longitudinal and interventional studies of cognitive ageing. However, our design does not permit causal inference, and firm recommendations for clinical practice or public health strategies should await confirmation from prospective and experimental research.

CRediT authorship contribution statement

Xi Luo: Writing – review & editing, Writing – original draft, Methodology, Funding acquisition, Data curation, Conceptualization. Zhonghua Yin: Writing – review & editing, Writing – original draft, Methodology, Data curation, Conceptualization. Xueli Sun: Conceptualization. Ke Li: Writing – review & editing, Methodology, Funding acquisition, Conceptualization.

Informed consent statement

Informed consent was obtained from all participants as part of the CHARLS protocol.

Ethics approval and consent to participate

All methods concerning human participants in our study were conducted in accordance with the ethical standards laid out in the 1964 Declaration of Helsinki and its subsequent amendments. CHARLS data collection was approved by the Biomedical Ethics Committee of Peking University (IRB00001052-11015). Informed consent was obtained from all participants, with signed copies stored and digitized by CHARLS.

Clinical trial number

Not applicable.

Funding

This work was supported by the Sichuan Science and Technology Program (grant numbers 2023YFQ0057, 2022YFQ0008 and 2024ZHCG0018), the Nursing Research Project of Sichuan Province (grant number H21031), and the 2025 Sichuan Provincial Key Research and Development Program (grant number 25QYCX0359). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

Publicly available datasets were analyzed in this study. These data can be found at https://charls.charlsdata.com/pages/Data/2018-charls-wave4/zh-cn.html (Authorized on September 19, 2024).