Midlife sleep characteristics and Alzheimer's disease biomarkers 20 years later

Abstract

INTRODUCTION

Poor sleep characteristics are associated with Alzheimer's disease (AD) among older adults; however, this relationship remains underexplored earlier in life. We investigated how early midlife sleep relates to AD‐related biomarkers in late midlife.

METHODS

A total of 1325 adults (mean age = 40.3 years) self‐reported sleep quality, sleep duration, daytime sleepiness, and insomnia symptoms. Twenty years later, we assessed plasma amyloid beta (Aβ)42/40, tau phosphorylated at threonine 217 (p‐tau217), neurofilament light chain (NfL), and magnetic resonance imaging (MRI)‐derived Spatial Pattern of Abnormality for Recognition of Early AD (SPARE‐AD), characterizing AD‐like brain atrophy.

RESULTS

After multivariable adjustment, longer sleep duration was associated with lower Aβ42/40, poor sleep quality with higher NfL, daytime sleepiness with both lower Aβ42/40 and higher NfL, and insomnia symptoms with higher SPARE‐AD. No associations were observed with sleep and p‐tau217.

DISCUSSION

Poor sleep in early midlife was associated with unfavorable AD biomarker levels 20 years later, supporting the idea of sleep as a potential modifiable AD risk factor.

Highlights

• Poor sleep in early midlife was associated with adverse AD biomarkers 20 years later.

• Longer sleep duration and daytime sleepiness were linked to lower plasma Aβ42/40.

• Poor sleep quality and daytime sleepiness were linked to higher NfL.

• Insomnia symptoms were linked with a MRI‐derived biomarker of AD‐like brain atrophy.

1. INTRODUCTION

Alzheimer's disease (AD) presents a growing public health burden worldwide. Efforts to prevent or delay its onset are particularly challenging due to the disease's long preclinical phase, during which neuropathological changes often silently progress for 10 to 20 years before the first clinical symptoms emerge. ¹ , ² Identifying modifiable risk factors that influence disease processes during this window is therefore essential for effective prevention strategies.

Among the emerging candidates, sleep has gained increasing attention as a potentially modifiable target. A growing body of evidence has linked a higher risk of AD with various poor sleep characteristics, ranging from short or long sleep duration, poor sleep quality, and insomnia symptoms at night to sleepiness during the day. ³ , ⁴ Poor sleep characteristics have also been associated with key AD‐related biomarkers, including higher levels of amyloid beta (Aβ) and tau, ⁵ , ⁶ , ⁷ , ⁸ as well as non‐specific neurogenerative biomarkers such as neurofilament light chain (NfL) and neuroimaging markers like gray matter atrophy. ⁶ , ⁹ , ¹⁰ Collectively, these findings suggest that sleep may play a crucial role in the pathophysiological process of AD. However, these associations are still not fully established as other studies have reported null findings, ⁵ , ⁶ , ¹¹ , ¹² and importantly, most have focused on older adult populations. Consequently, it remains unclear whether poor sleep earlier in life – when preventive interventions might be most effective – are linked to AD‐related brain changes. Understanding these midlife relationships could highlight a critical window for intervention and identify mechanistic pathways connecting sleep to brain health.

Building on our previous work showing that poor sleep in early midlife is associated with accelerated brain aging, ¹³ we sought to extend these findings by examining how sleep in early midlife relates to plasma and neuroimaging biomarkers of AD in late midlife.

2. METHODS

2.1. Study population

We studied participants enrolled in the Coronary Artery Risk Development in Young Adults (CARDIA) study, an ongoing prospective cohort study following participants from early adulthood through early late life. ¹⁴ Detailed design and methodology have been published elsewhere. ¹⁴ Briefly, 5115 adults aged 18 to 30 years were recruited in 1985 to 1986 from population‐based samples of four US cities: Birmingham, AL; Chicago, IL; Minneapolis, MN; and Oakland, CA. A balanced sampling by race (i.e., Black or White), sex, educational level, and age was achieved within each center. Participants completed follow‐up examinations every 2 to 5 years for 35 years. At each examination, participants provided written informed consent, and study protocols were approved by Institutional Review Boards at each study site and the CARDIA Coordinating Center.

Of the 3649 participants with sleep measures assessed at Year 15 (2000 to 2001, this study's baseline), 1374 had minimal covariate data (i.e., age, sex, race, education, and study center) and plasma AD biomarker measurements at Year 35 (2020 to 2022). After excluding participants with biomarker outliers (> ± 3 standard deviations [SD]), our final analytic sample included 1325 participants. For magnetic resonance imaging (MRI) analyses, the final analytic sample consisted of 1044 participants with minimal covariate data and a MRI‐derived AD biomarker measured at Year 35 (or, if missing, at Year 30 [2015 to 2016] or Year 25 [2010 to 2011]).

2.2. Sleep variables

Participants completed a sleep questionnaire at baseline (Year 15) (Supplementary Methods). Sleep duration was self‐reported and classified as shorter (<6 h), normal (6 to <9 h), or longer (≥9 h). Sleep quality was evaluated using a 5‐point Likert scale and categorized as “poor” (fairly bad/very bad) or “good” (very good/fairly good/good). Difficulty initiating sleep, difficulty maintaining sleep, and early morning awakening were queried (yes/no) and combined into a single variable indicating the presence of at least one nocturnal insomnia symptom versus none. Participants were also asked about daytime sleepiness (yes/no). We constructed a multidimensional healthy sleep score (range 0 to 4), assigning one point each for normal sleep duration, good sleep quality, no insomnia symptoms, and no daytime sleepiness.

2.3. Plasma AD biomarkers

Blood samples collected at Year 35 were processed and stored within 90 min (stick‐to‐freezer) at −70°C until assayed between March and May 2024 at the Hospital of the University of Pennsylvania. Plasma Aβ42, Aβ40, and phosphorylated tau at threonine 217 (p‐tau217) were assayed using the Fujirebio Lumipulse G1200 analyzer. Plasma NfL was analyzed using the Quanterix Simoa HD‐X assay using the Neurology 2‐Plex B kit (Quanterix Corp) (Supplementary Methods). Plasma biomarkers were standardized and analyzed as continuous variables. P‐tau217 was log‐transformed to improve the normality of its distribution. Lower Aβ42/40 and higher p‐tau217 and NfL levels are indicative of a higher AD risk.

2.4. SPARE‐AD index

At Years 25, 30, and 35, brain MRI was conducted using 3T scanners at each of the four clinic sites. MRI acquisition parameters and processing were previously described. ¹⁵ Briefly, structural images were acquired with 1‐mm isotropic 3D T1 and T2 sequences and processed using an automated multispectral computer algorithm that classified all supratentorial brain tissue into gray matter, white matter, and cerebrospinal fluid (CSF) and identified anatomic regions of interest (ROIs). After correction of intensity inhomogeneities, ¹⁶ a multi‐atlas skull stripping algorithm was applied for the removal of extra‐cerebral tissues. ¹⁷ Subsequently, each T1‐weighted scan underwent automated segmentation to define a set of anatomical gray matter ROIs using a multi‐atlas label fusion method. ¹⁸

We examined the Spatial Pattern of Abnormality for Recognition of Early Alzheimer's disease (SPARE‐AD) index, a composite neuroimaging biomarker of brain atrophy derived from volumetric structural MRI data. ¹⁹ This index was developed to detect early stages of AD by capturing spatial patterns of brain atrophy associated with the disease. It was trained on MRI data from cognitively normal individuals and AD patients using a machine‐learning approach, which identifies a pattern of regional brain volumes that best distinguishes AD from normal aging. For a new individual, the model generates a single SPARE‐AD score. More positive SPARE‐AD values indicate a higher risk of AD (i.e., a more AD‐like pattern of brain atrophy), whereas more negative values indicate a lower risk (i.e., a more normal pattern of brain morphology). While SPARE‐AD captures AD‐like atrophy patterns, it is not a direct measurement of neuronal loss in a given person. It has been widely used in AD studies and has demonstrated excellent performance in predicting AD risk. ²⁰ , ²¹ We used measurements from Year 35 (N = 650), but, if those were missing, we used measurements from Year 30 (N = 287), and if those were also unavailable, then we used ones from Year 25 (N = 107).

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the literature using PubMed. Growing evidence links sleep characteristics to AD biomarkers; however, most studies focus on older populations and use cross‐sectional designs.

2. Interpretation: In this prospective, community‐based cohort of adults (age range: 32 to 47 years), early midlife poor sleep characteristics were associated with unfavorable levels of plasma and imaging AD‐related biomarkers 20 years later. Longer sleep duration was associated with lower plasma Aβ42/40 ratio, poor sleep quality with higher plasma NfL, and daytime sleepiness with both Aβ42/40 and NfL level. Additionally, the presence of insomnia symptoms was associated with an MRI‐derived biomarker of AD‐related brain atrophy. These findings suggest that poor sleep characteristics in early midlife are associated with brain health patterns in late midlife (age range = 53 to 69 years).

3. Future directions: Studies should evaluate whether interventions targeting sleep in midlife can modify AD biomarker and cognitive trajectories and delay disease onset.

2.5. Statistical analyses

Baseline characteristics were compared according to sleep duration categories using chi‐squared test or Fisher's exact test for categorical variables and Kruskal‐Wallis test for continuous variables. We conducted linear regression analyses to examine the associations between each sleep parameter with plasma AD biomarkers and the SPARE‐AD index. Models were first adjusted for age, sex, race, education, and study center (Supplementary Methods). Further adjustments were made for body mass index (BMI), diabetes, hypertension, smoking, alcohol, physical activity, and depressive symptoms (Supplementary Methods). Adjusted means (95% confidence interval [CI]) were reported for the SPARE‐AD outcome. We verified the hypothesis of normality and homoscedasticity with a residual analysis, and we assessed the linearity assumption for continuous variables using fractional polynomials. In addition, we examined interactions between sleep characteristics, sex, race, and apolipoprotein E (APOE) ε4 status on the association with AD biomarkers using the Wald chi‐square test. Additional analyses included (1) assessment of potential non‐linear associations between sleep duration and outcomes using quadratic terms, (2) examining sleep quality across all five possible response categories, and (3) examining the multidimensional healthy sleep score. Finally, we conducted two sensitivity analyses to ensure results robustness: associations were further adjusted for APOE ε4, and plasma AD biomarker associations were also adjusted for renal function (Supplementary Methods). Significance level was set at a two‐sided p < 0.05. All statistical analyses were carried out using R version 4.5.1.

3. RESULTS

3.1. Population characteristics

At our study baseline, the participants’ mean age was 40.3 ± 3.5 years; 57.4% were women, and 42.6% were Black. Around 82.2% of participants reported normal sleep duration, 14.5% shorter, and 3.2% longer. Compared to participants with normal sleep duration, those with shorter or longer sleep durations were more likely to be Black and less educated and to have higher BMI and higher prevalences of diabetes, hypertension, and depressive symptoms (Table 1). Moreover, those with a shorter sleep duration were more likely to consume alcohol.

TABLE 1.

Baseline demographics and characteristics of study population (N = 1325).

	Total population		Shorter sleep duration	Normal sleep duration	Longer sleep duration
	N = 1325		N = 190	N = 1074	N = 42
Variables	n (%) or median (IQR)	NA	n (%) or median (IQR)	n (%) or median (IQR)	n (%) or median (IQR)	p value *
Age, year	41 (38 to 43)	0	40 (37 to 43)	41 (38 to 43)	40 (36.3 to 42)	0.17
Female	760 (57.4)	0	107 (56.3)	613 (57.1)	28 (66.7)	0.45
Black	564 (42.6)	0	135 (71.1)	390 (36.3)	21 (50.0)	<0.0001
Education, year	16 (13 to 16)	0	15 (13 to 16)	16 (14 to 17)	14 (12 to 16)	0.0002
Current smoker	234 (17.7)	4	41 (21.8)	175 (16.3)	11 (26.2)	0.06
Alcohol consumer	713 (54.0)	4	84 (44.2)	597 (55.8)	23 (54.8)	0.01
Physical exercise, exercise units	276 (144 to 476)	1	254 (120.3 to 456.8)	279 (144 to 494)	285 (144 to 437.3)	0.15
Body mass index, kg/m2	27.0 (23.7 to 31.6)	8	29.5 (25.9 to 35.3)	26.4 (23.4 to 30.7)	28.8 (22.9 to 34.3)	<0.0001
Diabetes	39 (3.0)	17	10 (5.3)	23 (2.2)	5 (12.5)	0.0005
Hypertension	205 (15.5)	0	51 (26.8)	140 (13.0)	10 (23.8)	<0.0001
Depressive symptoms	173 (13.1)	6	44 (23.4)	117 (10.9)	7 (16.7)	<0.0001
APOE ε4 carrier	386 (31.7)	109	62 (36.7)	301 (30.3)	15 (39.5)	0.15
eGFR, mL/min/1.73 m2	101 (89 to 112)	11	99 (90–111)	102 (89 to 112)	101 (88 to 112)	0.45

Abbreviations: APOE, apolipoprotein E; eGFR, estimated glomerular filtration rate; IQR, interquartile range.

^*Chi‐square test and Fisher's exact test were used for categorical variables and Kruskal‐Wallis test for continuous variables.

Across the sample, 14.3% reported poor sleep quality, 53.6% at least one insomnia symptom, and 25.7% daytime sleepiness (Table S1). At Year 35, after a mean follow‐up of 21.2 ± 0.6 years, participants had a median plasma Aβ42/40 of 0.096 (interquartile range [IQR]: 0.088 to 0.103), p‐tau217 of 0.081 (IQR: 0.061 to 0.109) pg/mL, and NfL of 9.60 (IQR: 7.45 to 12.60) pg/mL. APOE ɛ4 carriers had more unfavorable levels of plasma Aβ42/40 and p‐tau217 than non‐carriers (Table S2). Based on published US Food and Drug Administration (FDA)‐approved amyloid‐positivity cut‐offs (p‐tau217/Aβ42 > 0.0086), ²² , ²³ 5.1% of participants were classified as amyloid positive. Moreover, participants had a mean (± SD) cognitive score of 65.7 (± 16.3) for the Digit Symbol Substitution Test, 8.6 (± 3.5) for the Rey Auditory Verbal Learning Test, 23.4 (± 11.6) for the Stroop Test, 24.5 (± 3.8) for the Montreal Cognitive Assessment, 19.8 (± 5.3) for the Category Fluency Test, and 41.2 (± 13.1) for the Letter Fluency Test (Supplementary Methods). In the MRI sample, after a mean duration of 18.5 ± 4.1 years, participants had a median SPARE‐AD index of −0.829 (IQR: −1.072 to −0.574).

3.2. Sleep and plasma AD biomarkers

Adjusting for age, sex, race, education, and study center, longer sleep duration (β = −0.40, 95% CI: [−0.70, −0.10]) and daytime sleepiness (β = −0.14, 95% CI: [−0.26, −0.02]) were associated with lower Aβ42/40 levels (Figure 1, Table S1). Analysis including a quadratic term supported a non‐linear association between sleep duration and Aβ42/40 level (p for quadratic term = 0.03; Figure S1). Poor sleep quality (β = 0.16, 95% CI: [0.01, 0.31]), especially participants reporting having very bad sleep quality (β = 0.73, 95% CI: [0.29, 1.16]; Table S3), and daytime sleepiness (β = 0.17, 95% CI: [0.05, 0.29]) were associated with higher NfL levels (Figure 1, Table s1). Results remained consistent after further adjustment for behaviors and comorbidities (e.g., BMI, smoking, alcohol, physical exercise, diabetes, hypertension, and depressive symptoms), except for the association between daytime sleepiness and Aβ42/40, which was attenuated (β = −0.08, 95% CI: [−0.21,0.04]) (Table S4). Low multidimensional healthy sleep score (0 to 1) was associated with higher NfL levels, with a similar trend for lower Aβ42/40 level (Table S3). Sensitivity analyses, including further adjustment for APOE ε4 or renal function, displayed similar results (Tables S5 and S6). No association was observed with p‐tau217. There was no evidence of effect modification by sex, race, or APOE ε4 status.

FIGURE 1.

Associations between sleep in early midlife (Year 15) and plasma Alzheimer's disease biomarkers in late midlife (Year 35) among 1325 CARDIA participants. Note that plasma biomarkers were standardized. Log transformation was applied to p‐tau217 to improve the normality of the distribution. Models were adjusted for age, sex, race, education, and study center. Aβ, amyloid beta; CARDIA, Coronary Artery Risk Development in Young Adults; NfL, neurofilament light chain; p‐tau, phosphorylated tau.

3.3. Sleep and SPARE‐AD index

In the sample of participants who underwent MRI (N = 1044), 148 (14.4%) had shorter sleep duration, 33 (3.2%) longer sleep duration, 167 (16.0%) poor sleep quality, 544 (52.4%) at least one insomnia symptom, and 262 (25.3%) daytime sleepiness. Participants with insomnia symptoms had higher SPARE‐AD atrophy pattern scores (adjusted mean difference = −0.776, 95% CI: [−0.807, −0.746]) compared to those without insomnia symptoms (−0.841, 95% CI: [−0.874, −0.809]) in a model adjusted for age, sex, race, education, and study center (Table 2). This corresponds to approximately 8% greater burden of AD‐like brain atrophy (Figure 2). When examining the number of nocturnal insomnia symptoms, we observed a dose–response relationship, with an increased number of insomnia symptoms associated with higher SPARE‐AD scores (Table S7). Participants with difficulty initiating sleep (adjusted mean difference = −0.778, 95% CI: [−0.812, −0.745] vs −0.830, 95% CI: [−0.860, −0.800]; p = 0.02), maintaining sleep (−0.736, 95% CI: [−0.790, −0.682] vs −0.822, 95% CI: [ −0.847, −0.797]; p = 0.005), and early morning awakening (−0.754, 95% CI: [ −0.805, −0.702] vs −0.821, 95% CI: [−0.846, −0.796]; p = 0.02) had each higher SPARE‐AD scores compared to those without the specific symptom. Among these, difficulty maintaining sleep had the largest AD‐like brain atrophy (Figure 2). Further adjustment for behaviors and comorbidities did not change the association between insomnia symptoms and SPARE‐AD index. However, it was attenuated after considering APOE ε4 status (Table 2). There was no evidence of effect modification by sex, race, or APOE ε4 status. Participants with low multidimensional healthy sleep score (0 to 1) also had higher SPARE‐AD scores (Table S3).

TABLE 2.

Associations between sleep in early midlife (Year 15) and SPARE‐AD index in late midlife (Year 25, 30, or 35) among 1044 CARDIA participants.

		Model 1		Model 2		Model 3
		N = 1044		N = 1018		N = 939
	N (%)	Adjusted mean (95% CI)	p	Adjusted mean (95% CI)	p	Adjusted mean (95% CI)	p
Sleep duration, h			0.78		0.81		0.99
<6	148 (14.4)	−0.799 (−0.858, ‐0.741)		−0.713 (−0.804, −0.622)		−0.709 (−0.803, 0.614)
6 to 9	849 (82.4)	−0.809 (−0.835, −0.784)		−0.713 (−0.789, −0.638)		−0.713 (−0.792, −0.635)
≥9	33 (3.2)	−0.767 (−0.890, −0.645)		−0.671 (−0.815, −0.527)		−0.714 (−0.869, −0.559)
Sleep quality			0.12		0.37		0.23
Good	877 (84.0)	−0.815 (−0.840, −0.790)		−0.730 (−0.805, −0.656)		−0.730 (−0.807, −0.654)
Poor	167 (16.0)	−0.768 (−0.822, −0.713)		−0.702 (−0.789, −0.615)		−0.690 (−0.781, −0.599)
Insomnia symptoms			0.004		0.047		0.09
No	495 (47.6)	−0.841 (−0.874, −0.809)		−0.749 (−0.827, −0.671)		−0.744 (−0.824, −0.663)
Yes	544 (52.4)	−0.776 (−0.807, −0.746)		−0.704 (−0.779, −0.629)		−0.703 (−0.781, −0.625)
Daytime sleepiness			0.55		0.89		0.93
No	773 (74.7)	−0.811 (−0.838, −0.785)		−0.730 (−0.808, −0.652)		−0.728 (−0.809, −0.648)
Yes	262 (25.3)	−0.796 (−0.840, −0.752)		−0.726 (−0.808, −0.644)		−0.726 (−0.811, −0.641)

Note: Model 1 adjusted for age, sex, race, education, and study center. Model 2: Model 1 covariates + body mass index, smoking, alcohol, physical activity, diabetes, hypertension, and depressive symptoms. Model 3: Model 2 covariates + APOE ε4 status.

Abbreviations: CARDIA, Coronary Artery Risk Development in Young Adults; CI, confidence interval; SPARE‐AD, Spatial Pattern of Abnormality for Recognition of Early Alzheimer's Disease.

FIGURE 2.

Adjusted percentage difference in SPARE‐AD index by insomnia symptoms. Note that models were adjusted for age, sex, race, education, and study center. Error bars represent 95% confidence interval. *: p < 0.05; **: p < 0.01.AD, Alzheimer's disease; SPARE‐AD, Spatial Pattern of Abnormality for Recognition of Early Alzheimer's disease.

4. DISCUSSION

In this 20‐year prospective study of community‐dwelling adults, poor sleep characteristics in early midlife were associated with multiple unfavorable AD biomarkers in late midlife, independent of key confounders. Specifically, longer sleep duration was linked to lower plasma Aβ42/40 ratio, poor sleep quality to higher plasma NfL level, and daytime sleepiness to both lower Aβ42/40 and higher NfL level. Furthermore, insomnia symptoms were associated with greater AD‐like cortical atrophy, as measured by the SPARE‐AD index. These findings suggest that poor sleep may contribute to early AD‐related neurobiological changes, even decades before clinical symptom onset.

Poor sleep is associated with increased AD risk, ³ but most evidence comes from older adults, making it difficult to establish temporality as poor sleep could be an early symptom or prodrome of AD. By assessing sleep in early midlife, our study minimizes possible reverse causality and highlights early life pathways linking sleep to AD pathology. Notably, longer sleep duration and daytime sleepiness were associated with lower plasma Aβ42/40 ratio, suggestive of increased cerebral amyloid deposition. These findings may reflect that prolonged sleep captures fragmented or non‐restorative sleep, consistent with the hypothesis that chronic sleep disruption increases synaptic activity and impairs glymphatic clearance, thereby promoting Aβ accumulation. ²⁴ , ²⁵ , ²⁶ Longer sleep may also indicate poorer overall health that may contribute to AD pathology (Table 1). Our results align with prior evidence showing that excessive daytime sleepiness in older adults is associated with increased positron emission tomography (PET) Aβ accumulation ⁸ and with a recent meta‐analysis linking sleep duration to PET Aβ, but not to plasma or CSF Aβ measures in non‐demented adults. ⁵ In midlife adults, maintaining good sleep may be crucial for preventing AD by regulating the production and clearance of Aβ.

In contrast, no associations were observed between poor sleep characteristics and plasma p‐tau217, one of the most promising plasma biomarkers of AD pathology. ²⁷ This is consistent with a meta‐analysis reporting no association between sleep duration or quality and tau measured in plasma, CSF, or PET among cognitively unimpaired adults, ⁵ as well as cohort studies showing no association between daytime sleepiness or insomnia and CSF/PET tau concentrations, ⁹ , ²⁸ , ²⁹ , ³⁰ despite experimental findings showing that acute sleep deprivation can elevate tau burden. ³¹ , ³² , ³³ The null associations in our midlife cohort may reflect low overall tau burden, relatively young age at AD biomarker assessment, or lower sensitivity of plasma measures. Together, these results suggest that sleep‐related tau changes may emerge later in the AD continuum or become more detectable following an acute sleep disruption context. ²

Several sleep characteristics were associated with markers of neurodegeneration. Poor sleep quality and daytime sleepiness were linked to higher plasma NfL levels, a marker of neuronal injury that is not specific to AD, unlike amyloid and tau plasma biomarkers. These findings are consistent with some previous studies, ⁶ , ³⁴ though not all, ¹¹ , ³⁵ examining sleep quality with circulating NfL. Similarly, one study of 260 older adults reported that greater sleepiness correlated with elevated CSF NfL concentrations, ⁹ whereas another found no association with serum NfL. ³⁶ Such discrepancies may reflect differences in sample characteristics, sleep assessment methods, or biofluid type (i.e., CSF, plasma, serum). Our study is novel in that we observed associations earlier in life, when neurodegenerative processes may be beginning. Additionally, insomnia symptoms were associated with a higher SPARE‐AD index, indicating higher AD‐related cortical atrophy. While prior neuroimaging studies have reported mixed associations between insomnia and regional brain volumes, ¹⁰ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ our multivariate MRI‐based approach detected an 8% greater burden of AD‐like brain atrophy among participants with insomnia symptoms. The SPARE‐AD index might have detected subtle, spatially distributed AD‐related structural changes that region‐specific volumetric analyses might overlook, particularly in middle‐aged individuals.

Poor sleep may contribute to AD pathology through different mechanisms, including impaired glymphatic system, altered cerebral blood flow, increased levels of Aβ, oxidative stress, and inflammatory processes. ²⁴ , ⁴¹ They may also play a role through vascular, metabolic, and psychiatric conditions or reflect an underlying sleep disorder such as sleep disorder breathing or disruptions in circadian rhythms, all of which influence cognition. ⁴² , ⁴³ Our finding that distinct sleep characteristics were associated with different AD biomarkers further suggests that sleep may be related to brain health through multiple, domain‐specific pathways rather than a single global mechanism. Future longitudinal and mechanistic research is needed to clarify these pathways and replicate these results.

To the best of our knowledge, this is the first study to examine a range of sleep characteristics in early midlife in relation to multiple AD biomarkers two decades later. The key strengths of our study include a long follow‐up period of 20 years, a diverse community‐based population, adjustment for multiple confounding factors, and assessment of both plasma and neuroimaging biomarkers. However, sleep measures were self‐reported, which could have introduced misclassification bias. The relatively low number of participants with longer sleep duration may limit the statistical power and interpretation of the results. Plasma biomarkers, while minimally invasive and scalable, may be less sensitive than CSF or PET measures for detecting early AD pathology. Correlations between biomarkers were weak, which may reflect the temporal and mechanistic differences among these measures across the AD continuum. ² Finally, because this was an observational study, causal relationships cannot be determined.

5. CONCLUSIONS

Poor sleep characteristics in early midlife were associated with adverse AD‐related plasma and neuroimaging biomarkers two decades later. These findings suggest that poor sleep is associated with AD pathology from an early stage. Future studies should determine whether early to midlife sleep interventions can modify AD‐related neurobiology. Such evidence would support promoting healthy sleep behaviors in younger populations as a strategy for AD prevention.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to declare. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

Participants provided written informed consent at each examination, and study protocols were approved by Institutional Review Boards at each study site and the CARDIA Coordinating Center.

Supporting information

Supporting information

Supporting information

DATA AVAILABILITY STATEMENT

Requests for access to the data for this study can be made at the CARDIA website: https://www.cardia.dopm.uab.edu/.