Suboptimal Sleep Duration Is Associated With Poorer Neuroimaging Brain Health Profiles in Middle‐Aged Individuals Without Stroke or Dementia

Abstract

Background

The American Heart Association's Life's Simple 7, a public health construct capturing key determinants of cardiovascular health, became the Life's Essential 8 after the addition of sleep duration. The authors tested the hypothesis that suboptimal sleep duration is associated with poorer neuroimaging brain health profiles in asymptomatic middle‐aged adults.

Methods and Results

The authors conducted a prospective magnetic resonance neuroimaging study in middle‐aged individuals without stroke or dementia enrolled in the UK Biobank. Self‐reported sleep duration was categorized as short (<7 hours), optimal (7–<9 hours), or long (≥9 hours). Evaluated neuroimaging markers included the presence of white matter hyperintensities (WMHs), volume of WMH, and fractional anisotropy, with the latter evaluated as the average of 48 white matter tracts. Multivariable logistic and linear regression models were used to test for an association between sleep duration and these neuroimaging markers. The authors evaluated 39 771 middle‐aged individuals. Of these, 28 912 (72.7%) had optimal, 8468 (21.3%) had short, and 2391 (6%) had long sleep duration. Compared with optimal sleep, short sleep was associated with higher risk of WMH presence (odds ratio, 1.11 [95% CI, 1.05–1.18]; P<0.001), larger WMH volume (beta=0.06 [95% CI, 0.04–0.08]; P<0.001), and worse fractional anisotropy profiles (beta=−0.04 [95% CI, −0.06 to −0.02]; P=0.001). Compared with optimal sleep, long sleep duration was associated with larger WMH volume (beta=0.04 [95% CI, 0.01–0.08]; P=0.02) and worse fractional anisotropy profiles (beta=−0.06 [95% CI, −0.1 to −0.02]; P=0.002), but not with WMH presence (P=0.6).

Conclusions

Among middle‐aged adults without stroke or dementia, suboptimal sleep duration is associated with poorer neuroimaging brain health profiles. Because these neuroimaging markers precede stroke and dementia by several years, these findings are consistent with other findings evaluating early interventions to improve this modifiable risk factor.

Nonstandard Abbreviations and Acronyms

BIANCA

Brain Intensity Abnormality Classification Algorithm

EPIC

European Prospective Investigation into Cancer

FA

fractional anisotropy

UKB

UK Biobank

WMH

white matter hyperintensity

Clinical Perspective.

What Is New?

• Among almost 40 000 middle‐aged adults with no history of stroke or dementia, suboptimal short and long sleep duration conferred an increased risk of brain magnetic resonance imaging markers of poor brain health (white matter hyperintensities and fractional anisotropy [FA]).

• These associations persisted despite accounting for hypertension, hyperlipidemia, diabetes, and smoking, which are cardiovascular risk factors that are known to associate with the same magnetic resonance imaging markers of poor brain health (white matter hyperintensities and FA).

• When evaluating the associations of short and long sleep duration and FA across 48 distinct white matter tracts, short sleep was associated with the biggest decrease of FA values in the bilateral posterior limbs of the internal capsule, while long sleep was associated with the biggest decrease of FA values in the right superior cerebellar peduncle and left fornix and stria terminalis.

What Are the Clinical Implications?

• Both short and long sleep durations may potentially be modifiable risk factors for magnetic resonance imaging markers of poor brain health.

• Because these magnetic resonance imaging markers of poor brain health (eg, white matter hyperintensities and FA) are known to precede the occurrence of stroke and dementia by many years, there may be merit in assessing and managing abnormal sleep duration among asymptomatic middle‐aged adults.

• Better understanding of the directionality and mechanisms of the association between suboptimal sleep and poor brain health are needed, as well as trials aimed at unveiling whether early interventions on sleep duration during middle age can potentially benefit brain health later in life.

Sleep, a crucial neurophysiological process, is associated with several important health outcomes. ¹ Both short and long sleep durations are associated with increased risk of all‐cause mortality, ² coronary heart disease, ³ and stroke. ⁴ In light of this evidence, the American Heart Association's (AHA's) Life's Simple 7, a public health and research construct that captures key determinants of cardiovascular health, recently became Life's Essential 8 after adding sleep duration as a new component. ⁵ Given that optimizing cardiovascular health during middle age can significantly benefit brain health in later life, ⁶ understanding the impact of sleep duration on brain health is critical. However, the association of sleep duration and brain health in middle‐aged individuals without stroke or dementia remains understudied and unclear. This is particularly important as these neuroimaging markers precede stroke and dementia by several years. ⁷ , ⁸ , ⁹

To address this knowledge gap, we leveraged one of the largest population health studies in the world to conduct a neuroimaging study based on standardized research brain magnetic resonance imaging (MRI) completed in thousands of middle‐aged adults. We hypothesized that suboptimal sleep durations, both short and long, lead to clinically silent brain injury in this age group. Because the MRI markers evaluated in this study precede, by many years, the occurrence of stroke and dementia, the main clinical manifestations of poor brain health, our goal is to generate the necessary evidence to support early interventions geared towards improving this modifiable risk factor.

Methods

Study Design

We conducted a prospective neuroimaging study nested within the UK Biobank (UKB), a large ongoing population study in the United Kingdom. ¹⁰ The UKB enrolled 502 480 individuals aged 40 to 69 years from March 2006 to October 2010. A randomly selected subset of these participants were invited to participate in a dedicated neuroimaging study and completed a standardized brain MRI. ¹¹ , ¹² For the present study, we excluded study participants with a history of stroke and dementia, the 2 most important clinical manifestations of poor brain health. We also excluded participants with a history of multiple sclerosis due to the known neuroimaging changes that usually accompany this disease. The UKB received approval from the National Information Governance Board for Health and Social Care and the North West Multi‐centre Research Ethics Committee (https://www.ukbiobank.ac.uk/learn‐more‐about‐uk‐biobank/about‐us/ethics). All UKB participants included in this study, or their legally designated surrogates, provided written informed consent.

Exposure Ascertainment and Modeling

In the primary analysis, our exposure of interest was sleep duration, initially assessed during the baseline interview conducted approximately 9 years prior to the neuroimaging assessments. During these baseline interviews, participants reported their average daily sleep duration, including daytime napping. ¹⁰ Sleep duration was again assessed in the same fashion as part of the neuroimaging assessment visit. For the purposes of our study, this latter measurement was used in a sensitivity analysis to account for potential fluctuations in sleep duration over the extended 9‐year follow‐up period. We classified sleep duration according to the AHA's Life Essential 8 scheme ⁵ : short (<7 hours), optimal (7–<9 hours), and long (≥9 hours).

Neuroimaging Measurements

Details about the MRI neuroimaging research protocol used by the UKB are available elsewhere. ¹³ Briefly, all brain MRI was performed using a Siemens Skyra 3T machine with a standard Siemens 32‐channel radio frequency receive head coil and the same software (VD13A SP4). The generated images were stored as DICOM images and later converted to Nifti format. Multishell diffusion scans were obtained at 2 b‐values (b=1000 s/mm² and b=2000 s/mm²) with 50 distinct diffusion‐encoding directions at each b‐value. The scan duration was 7 minutes. The UKB research team centrally segmented the evaluated radiological phenotypes, including white matter hyperintensities (WMHs) and white matter disintegrity, assessed via the diffusion tensor imaging metrics fractional anisotropy (FA) in 48 specific white matter tracts. WMH volume was first normalized to head size, then log‐transformed (to approach a normal distribution) and finally standardized (by subtracting the mean and dividing by the SD). In the context of this study, FA provides insights into the integrity of white matter tracts. Water tends to diffuse along the direction of the axonal tract in white matter. Therefore, lower FA values, indicating less directionality, suggest greater white matter disintegrity. WMH was ascertained via segmentation masks using the Brain Intensity Abnormality Classification Algorithm (BIANCA) segmentation tool. ¹⁴ The total volume of WMH was estimated using the number of affected voxels and voxel size. Diffusion tensor imaging metrics were calculated using DTIFIT, a tool from FSL software. Our primary analysis focused on the average FA for the 48 white matter tracts. FA was standardized by subtracting by the mean and dividing by the SD. Subsequently, we evaluated FA in each of the separate 48 white matter tracts as secondary analyses.

Covariates

Covariates used for adjusted models were universal covariates (age, sex assigned at birth, and race), cardiovascular risk factors (hypertension, hyperlipidemia, diabetes, body mass index, and smoking status), and history of myocardial infarction. Demographic information (age, sex assigned at birth, and race) and medical history (hypertension and smoking status) were directly collected from study participants during the baseline in‐person interview. Body mass index was constructed centrally by the UKB team using height and weight measurements calculated during the initial assessment. Diabetes, hyperlipidemia, and history of myocardial infarction were supplemented via validated International Statistical Classification of Diseases, Tenth Revision (ICD‐10) codes abstracted from admission data obtained from across electronic health records of the UK health system (Tables S1–S5).

Statistical Analysis

For WMH volume, we investigated 2 outcomes: ≤5 cm³ versus >5 cm³ (dichotomous) and overall volume burden (continuous). To minimize false positives due to the high sensitivity of the BIANCA algorithm, we defined WMH presence as having >5 cm³ of WMH. ¹⁴ We initially evaluated unadjusted associations between short/long sleep and WMH and FA using χ² and 1‐way ANOVA tests, as appropriate. Next, we tested for associations between sleep duration and WMH presence, WMH volume, and FA adjusting for relevant covariates using multivariable logistic and linear regression models, as appropriate. We constructed 3 adjusted models: Model 1 included age, sex, and race; model 2 added cardiovascular risk factors (hypertension, hyperlipidemia, diabetes, body mass index, and smoking status) to model 1; and model 3 incorporated model 2 variables plus myocardial infarction, the most relevant clinical end point related to cardiovascular health. We used a complete case approach for all analyses, where analyses are completed in study participants with available data for variables used in each model. We performed 2 different sensitivity analyses. First, because the primary analysis uses sleep duration data ascertained almost 9 years before the neuroimages were completed, we reevaluated the associations of interest using sleep duration evaluated at the time of the brain MRI (instead of study baseline), thus accounting for potential variations in sleep duration during the follow‐up period, and, second, we reevaluated the primary analysis excluding participants with a history of obstructive sleep apnea, a condition known to be associated with neuroimaging markers of poor brain health. ¹⁵ In secondary analyses, we tested for an association between sleep duration and FA separately in each of the 48 white matter tracts. The software used for all statistical analyses was R version 4.1.3 (R Project for Statistical Computing).

Data Availability

UK Biobank data are accessible through a procedure described at http://www.ukbiobank.ac.uk/using‐the‐resource/. Data were accessed using project application number 58743.

Results

Figure 1 describes the assembly of the analytic sample. Of the 502 480 UKB participants, 41 434 agreed to participate in a dedicated brain MRI neuroimaging study and had WMH data. After exclusions, the final analytical sample included 39 771 participants (Table 1). The mean time from enrollment to the neuroimaging assessment was 8.9 years (standard error, 1.8 years). Table S2 shows the comparison between the analytic sample, cohort of participants who did not undergo research brain MRI, and the whole UKB study population.

Figure 1. Flow chart summarizing the exclusions that led to the study population.

FA indicates fractional anisotropy.

Table 1.

Baseline Characteristics of the Studied Population

Variable	Optimal sleep (n=28 912)	Short sleep (n=8468)	Long sleep (n=2391)	P value
Age, mean (SD)	55 (7.6)	55 (7.2)	56.5 (7.8)	<0.001
Men, n (%)	13 506 (46.7)	4148 (49)	1018 (42.6)	<0.001
Race, n (%)
White	28 134 (97.5)	8051 (95.3)	2318 (97.1)	<0.001
Black	128 (0.4)	124 (1.5)	6 (0.3)
Asian	343 (1.2)	142 (1.7)	33 (1.4)
Other	243 (0.8)	127 (1.5)	31 (1.3)
Smoking status, n (%)
Never	17 819 (61.7)	5022 (59.5)	1424 (60)	<0.001
Former	9365 (32.4)	2818 (33.4)	810 (34)
Current	1680 (5.8)	606 (7.2)	151 (6.3)
Body mass index, mean (SD)	26.33 (4.09)	27.02 (4.42)	27 (4.44)	<0.001
Hyperlipidemia, n (%)	6473 (22.4)	2127 (25.1)	684 (28.6)	<0.001
Diabetes, n (%)	1457 (5)	598 (7.1)	196 (8.2)	<0.001
Hypertension, n (%)	4965 (17.2)	1655 (19.5)	522 (21.8)	<0.001
Myocardial infarction, n (%)	780 (2.7)	300 (3.5)	87 (3.6)	<0.001

White Matter Hyperintensities

In unadjusted analyses, the prevalence of WMH was 28.8%, 30.5%, and 33.6% for optimal, short, and long sleep, respectively (unadjusted P<0.001) (Table 2). In multivariable logistic and linear regression analyses, compared with optimal sleep, short sleep was associated with a higher likelihood of WMH presence (model 1: odds ratio, 1.11 [95% CI, 1.05–1.18]; P<0.001) and higher WMH volumes (model 1: beta=0.06 [95% CI, 0.04–0.08]; P<0.001), whereas long sleep was associated with higher WMH volumes (model 1: beta=0.04 [95% CI, 0.01–0.08]; P=0.02) but not with WMH presence (Table 3). When using different multivariable models, results were similar for short sleep but nonsignificant for long sleep (P>0.05, models 2 and 3) (Table 3). Sensitivity analyses using sleep duration ascertained at the time of the brain MRI (instead of baseline) (Tables S3 and S4) yielded similar results, as well as sensitivity analyses excluding participants with obstructive sleep apnea (Tables S3 and S5).

Table 2.

Unadjusted Results for Optimal, Short, and Long Sleep

MRI outcome	Optimal sleep	Short sleep	Long sleep	P value
WMH presence, n (%)	8323 (28.8)	2585 (30.5)	803 (33.6)	<0.001
WMH volume,* mean (SD)	−0.02 (0.99)	0.05 (1)	0.13 (1.01)	<0.001
Fractional anisotropy,† mean (SD)	0.02 (0.99)	−0.03 (1)	−0.13 (1.07)	<0.001

^*Normalized to head size, log‐transformed and standardized.

^†Standardized.

FA indicates fractional anisotropy; MRI, magnetic resonance imaging; and WMH, white matter hyperintensity.

Table 3.

Adjusted Logistic and Linear Regression Results for Short and Long Sleep, Compared With Optimal Sleep

Model	Sleep duration	WMH presence			WMH volume§			FA||
		OR (95% CI)	P value	N	Beta (95% CI)	P value	N	Beta (95% CI)	P value	N
Model 1*	Optimal	Reference			Reference			Reference
	Short	1.11 (1.05 to 1.18)	<0.001	39 680	0.06 (0.04 to 0.08)	<0.001	39 680	−0.04 (−0.06 to −0.02)	0.001	39 680
	Long	1.03 (0.93 to 1.13)	0.6	39 680	0.04 (0.01 to 0.08)	0.02	39 680	−0.06 (−0.10 to −0.02)	0.002	39 680
Model 2†	Optimal	Reference			Reference			Reference
	Short	1.07 (1.01 to 1.13)	0.02	39 559	0.04 (0.02 to 0.06)	<0.001	39 559	−0.02 (−0.05 to −0.00)	0.04	39 559
	Long	0.97 (0.88 to 1.07)	0.6	39 559	0.02 (−0.02 to 0.05)	0.3	39 559	−0.04 (−0.08 to −0.00)	0.03	39 559
Model 3‡	Optimal sleep	Reference			Reference			Reference
	Short	1.07 (1.01 to 1.12)	0.02	39 559	0.04 (0.02 to 0.06)	<0.001	39 559	−0.03 (−0.05 to −0.00)	0.03	39 559
	Long	0.97 (0.87 to 1.07)	0.6	39 559	0.02 (−0.02 to 0.05)	0.3	39 559	−0.04 (−0.08 to −0.00)	0.03	39 559

FA indicates fractional anisotropy; OR, odds ratio; and WMH, white matter hyperintensity.

^*Model 1: covariates=age, sex, and race.

^† Model 2: covariates=model 1 + hypertension, diabetes, smoking status, hyperlipidemia, and body mass index.

^‡ Model 3: covariates=model 2 + myocardial infarction.

^§ Normalized to head size, log‐transformed, and standardized.

^|| Standardized.

Fractional Anisotropy

In unadjusted analyses, the FA values across the 48 evaluated white matter tracts were 0.02, −0.03, and −0.13 for optimal, short, and long sleep, respectively (unadjusted P<0.001) (Table 2). Of note, lower FA values indicate more prominent white matter disintegrity. In multivariable linear regression analyses, compared with optimal sleep, short (model 1: beta=−0.04 [95% CI, −0.06 to −0.02]; P=0.001) and long (beta=−0.06 [95% CI, −0.1 to −0.02]; P=0.002) sleep were both associated with lower (worse) FA values. When applying multivariable models including cardiovascular risk factors and myocardial infarction, results were similar for both long and short sleep (models 2 and 3) (Table 3). Sensitivity analyses with sleep measured at the time of MRI, not baseline, and excluding patients with sleep apnea, revealed similar findings (Tables S3 through S5). When evaluating the associations of short sleep across each of the 48 white matter tracts separately, the bilateral posterior limbs of the internal capsule demonstrated associations with the biggest reduction (worse) in FA values (Figures 2 and 3). Similarly, when evaluating the associations between long sleep and FA across each white matter tract, the right superior cerebellar peduncle and the left fornix cres and stria terminalis correlated with the biggest reduction (worse) in FA values (Figures S1 and S2).

Figure 2. Brain map statistics of white matter tracts for associations between short sleep and fractional anisotropy (FA).

Betas of linear regression analysis results between short sleep and FA across 48 distinct white matter tracts using the John Hopkins University white matter atlas. The color bars represent betas from multivariable linear regression models for each white matter tract. Red signifies smaller betas, and yellow signifies larger betas.

Figure 3. Heatmap of betas and their 95% CIs for short sleep and fractional anisotropy (FA) for each white matter tract.

Heatmap showing betas and their 95% CIs (in parentheses) for the associations of short sleep and FA for each of the 48 white matter tracts. The color bar represents the betas from multivariable linear regression models. Red signifies smaller betas, and yellow signifies larger betas.

Discussion

We report the results of a large neuroimaging study that tested for an association between suboptimal sleep duration and neuroimaging markers of poor brain health in middle‐aged adults without a history of dementia or stroke. This study was motivated by the mounting evidence supporting healthy sleep as a pillar of good cardiovascular and cerebrovascular health and suboptimal sleep duration as an important risk factor for poor brain health. We found that, in this specific population of asymptomatic individuals, suboptimal sleep duration (both short and long sleep) was associated with poorer neuroimaging brain health profiles, as represented by WMHs and FA, two well‐established imaging markers of brain health. These results remained significant across several different modeling strategies and secondary analyses. Of note, stratified regional analysis indicated that the white matter tracts associated with more microstructural injury were the posterior limbs of the internal capsule, the right superior cerebellar peduncle, and the left fornix crus and stria terminalis.

Existing research on the role of sleep duration in cardiovascular disease and brain health has so far focused on acute events. A large study based on the prospective EPIC (European Prospective Investigation into Cancer) Norfolk cohort followed 9692 stroke‐free participants for a mean of 9.5 years and subsequently meta‐analyzed its results with other smaller studies, finding a 15% and 45% higher risk of stroke for short and long sleep, respectively. ¹⁶ Similarly, a study that analyzed data on 461 347 UKB participants free of cardiovascular disease found a 20% and 34% higher risk of myocardial infarction in short and long sleepers, respectively. ¹⁷

By focusing on asymptomatic middle‐aged adults, the present study adds important new evidence supporting the role of sleep duration as a risk factor for clinically silent cerebrovascular disease and poor brain health. The neuroimaging markers evaluated in this study are strong predictors of acute stroke and precede the occurrence of these events by many years. Moreover, in sensitivity analyses, findings still showed an association after removing participants with obstructive sleep apnea, another highly prevalent sleep disturbance associated with neuroimaging profiles of poor brain health, ¹⁵ , ¹⁸ suggesting that participants without this disorder may still be at risk of microstructural brain injury. Given these associations, there may be merit in considering early diagnostic interventions to identify middle‐aged individual with suboptimal sleep duration. Importantly, both the screening and initial management and assessment of abnormal sleep duration can be easily integrated into routine health visits, emphasizing its practicality and affordability. Beyond this simple initial approach to diagnosis and treatment, the identification and improvement of sleep duration will be greatly facilitated by the recent addition of sleep duration to the AHA's Life's Simple 7, which has now become the Life's Essential 8, a framework for health promotion that includes several different communication and educational materials. ⁵

The main strengths of the present study are its large sample size, atypical for neuroimaging studies based on MRI, the standardized imaging protocols used for all analyzed brain neuroimages, and availability of 2 different data points for sleep duration. A few limitations should also be mentioned. First, the ascertainment of sleep duration was based on self‐reported information instead of more accurate methods, such as actigraphy and polysomnography. ¹⁹ Second, due to the observational nature of the present study, we are precluded from making causal inferences between sleep duration and poor brain health. In addition, the directionality of such a causal relationship remains indeterminate. Third, the generalizability of our results is limited by the characteristics of the UKB study, which is being conducted in a single European country and may be prone to healthy volunteer bias, where individuals who are willing and able to participate in a study tend to be healthier than the source population. In combination, these limitations emphasize the importance of follow‐up research aimed at confirming these findings in studies that use more sophisticated methods to ascertain sleep duration and enroll more diverse populations.

Conclusions

In conclusion, the present study evaluated whether abnormal sleep duration is associated with poorer neuroimaging brain health profiles in middle‐aged adults without stroke or dementia. We found that both short and long sleep were associated with both higher risk and more severe burden of WMHs, as well as worse FA profiles. These findings extend the existing evidence focused on sleep duration as a risk factor for stroke and point to the need, and significant opportunity, for deploying early interventions aimed at routinely identifying middle‐aged individuals who have suboptimal sleep.

Sources of Funding

S.C.‐T. is funded by National Institutes of Health T32 AG019134 and P30 AG021342.

Disclosures

None.

Supporting information

Tables S1–S5

Figures S1–S2