The role of cognitive reserve in white matter hyperintensities: from cognitive aging to Alzheimer's spectrum

Abstract

INTRODUCTION

White matter hyperintensities (WMHs) are characteristic of Alzheimer's disease (AD), and cognitive reserve (CR) protects cognitive function. However, whether WMHs mediate the CR–cognition relationship remains unclear.

METHODS

Brain imaging, clinical features, and neuropsychological assessments were performed, and CR was measured using the Cognitive Reserve Index questionnaire. Bootstrap mediation analysis examined CR's role in specific cognitive functions, controlling for covariates.

RESULTS

Participants who were cognitively unimpaired (CU; n = 85, mean age = 68.6 ± 5.7) and who had mild cognitive impairment (MCI; n = 43, mean age = 71.8 ± 6.5) or AD (n = 61, mean age = 72.8 ± 6.2) were included. CR was positively associated with global and non‐memory cognitive functions in the CU and MCI groups. In the CU group, WMHs served as a mediator between CR and global cognitive ability.

DISCUSSION

CR may maintain the optimal cognitive function by mitigating the WMH burden independently of AD‐related brain changes.

Highlights

• Cognitive reserve (CR) positively links to non‐memory cognition.

• Cognitive reserve mitigates white matter hyperintensities to preserve cognition.

• Cognitive reserve primarily protects cognition in pre‐Alzheimer's stages.

1. BACKGROUND

White matter hyperintensities (WMHs), once considered solely an indicator of small vessel disease, have been recognized as a crucial co‐pathology in Alzheimer's disease (AD), significantly affecting cognitive decline and disease progression beyond the scope of conventional AD pathological markers. ¹ Studies consistently show that individuals with AD demonstrate elevated WMH levels across various brain regions compared to healthy individuals, and the WMH levels positively correlate with amyloid deposition in positron emission tomography (PET) scans and negatively correlate with hippocampal volume and cortical thickness. ² , ³ Previous studies showed that WMHs may interact with total tau in the cerebrospinal fluid (CSF) and accelerate the onset of cognitive decline in patients with AD, suggesting that WMHs interact with AD pathology in ways that accelerate cognitive decline. ⁴ , ⁵

WMHs are strongly correlated with declines in non‐memory cognitive functions, including executive function, psychomotor speed, and visuoconstruction abilities. ⁶ Executive function encompasses critical goal‐directed processes such as problem‐solving, planning, and multitasking. ⁷ Its decline begins during the preclinical stage of AD and can impair daily functioning, leading to difficulties in communication or cooking due to reduced multitasking capacity. ⁸ Psychomotor function, reflecting the integration of mental processes with physical movement, may manifest as slowed perceptuomotor and decision‐making processes in AD patients, increasing accident risks like delayed reactions to traffic. ⁹ , ¹⁰ Visuoconstruction, involving the recognition and organization of visual components into coherent structures, when impaired in AD, can lead to daily functional difficulties such as writing problems or frequent bumping into objects. ¹¹ , ¹²

Cognitive reserve (CR) is a concept proposed to explain individual differences in cognitive resilience in the context of aging or brain pathology. CR is believed to buffer cognitive decline in various neurodegenerative conditions, including AD and multiple sclerosis. ¹³ Specifically, individuals with higher CR can maintain optimal cognitive performance while facing brain pathologies. ¹⁴ , ¹⁵ CR is usually measured by the Cognitive Reserve Index (CRI) questionnaire, considering education, occupation, and leisure activities. ¹⁶ Regarding CR and AD‐related biomarkers, the combination of educational level, premorbid intelligence, and cognitive activities moderated the relationship between cognition and hypometabolism in AD‐signature areas, including the middle temporal lobe, inferior temporal lobe, entorhinal cortex, and fusiform gyrus. Although the protective effect of CR in individuals with intact versus impaired cognition remains unclear, studies have shown that the protective effect of CR only exists in individuals with a high level of total tau, a biomarker indicating the severity of dementia due to AD. ¹⁷ , ¹⁸ , ¹⁹ These findings may imply that CR is more prominent in the advanced stages of AD.

Prior research indicated that CR protected against the negative impact of WMH, with some studies suggesting high CR may reduce WMH volume and symptoms. ²⁰ , ²¹ Although systematic reviews on CR's direct effect on the WMH–cognition link are limited, evidence supports the notion that CR could mitigate WMHs’ adverse cognitive effects. A mediation study found that higher CR allowed for greater WMH burden at a given cognitive performance, suggesting CR lessens WMHs’ negative impact on cognitive functions. ²² Additionally, education has been shown to slow the progression from healthy cognition to mild cognitive impairment (MCI) despite WMH load. ²³

While the exact neuroprotective mechanisms of CR against WMHs and their impact on cognition are still unclear, WMHs, often caused by demyelination, primarily affect the frontal‐subcortical network, hindering neural transmission. ²⁴ The CR hypothesis proposes that individuals with higher CR compensate for WMH‐related cognitive decline by efficiently using existing neural networks or recruiting alternative frontal‐related brain networks. Aging studies further suggest that higher CR enhances frontal‐parietal functional connectivity and reduces dorsal attention network connectivity, both linked to better cognitive performance. ²⁵ Ultimately, higher CR is thought to boost the brain's resilience against WMHs and other neuropathological damage.

RESEARCH IN CONTEXT

1. Systematic review: Prior studies on WMHs, CR, and cognition in AD were reviewed. WMHs are associated with cognitive decline, while CR is known to mitigate AD‐related pathology. However, few studies have explored CR as a mediator between WMHs and cognition in AD.

2. Interpretation: Our findings show that CR is linked to non‐memory cognitive function, and deep WMHs mediate its effect on global cognition in the predementia stage of AD. This supports CR's protective role against vascular co‐pathology in AD.

3. Future directions: Longitudinal studies are required to determine whether CR impacts WMH‐related cognitive decline. Interventions aimed at enhancing CR could be a method to maintain cognition in AD.

Although the protective role of CR in AD has been widely studied, several gaps remain. First, previous research primarily focused on the role of CR in AD‐specific pathologies. However, WMHs, a critical co‐pathology in AD, are known to interact with both tau and amyloid pathologies. Therefore, the role of CR in mitigating the cognitive impact of WMHs, beyond traditional AD‐related pathology, warrants further investigation. Moreover, prior studies largely relied on global cognitive screening tools, providing limited insight into the domain‐specific effects of CR on distinct cognitive functions. This study specifically aimed to address the role of CR in mediating the relationship between WMH burden and non‐memory cognitive functions in individuals with MCI and AD. Based on the Reserve and Resilience framework, we integrated brain imaging and cognitive data from two medical centers and analyzed WMHs, AD‐signature cortical thickness, and hippocampal volume in a single model. ²⁶ We hypothesized that (1) higher CR would be associated with better global and specific cognitive performance and (2) WMHs would mediate the relationship between CR and cognitive function, particularly for non‐memory domains.

2. METHODS

2.1. Study design and participants

In this prospective investigation, we recruited 189 participants from two medical centers using brain magnetic resonance imaging (MRI) and a series of comprehensive neuropsychological assessments. Ethical approval for the study protocol (2023‐07‐012AC) was granted by the Institutional Review Board of the Taipei Veterans General Hospital, and all participants provided written informed consent.

Participants were included if they were aged ≥55 years, had at least 6 years of education, and had normal or corrected vision and hearing. Individuals with a history of neurological diseases, cerebrovascular events, psychiatric disorders, substance abuse, head injuries, or other conditions that could interfere with cognitive assessment were excluded.

2.2. Defining cognitive impairment

Cognitive impairment was defined as a score of at least one standard deviation below the mean on any single cognitive test. ²⁷ Participants were classified based on the National Institute on Aging and Alzheimer's Association (NIA‐AA) clinical staging criteria for AD and the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM‐5): (1) AD, cognitive and functional decline affecting daily activities, and (2) MCI, cognitive decline without impairment in daily living. ²⁸ , ²⁹ Cognitively unimpaired (CU) was defined as individuals who exhibit no clinically significant decline in cognitive abilities and perform within expected healthy ranges on cognitive tests for their age and education. The diagnoses were established by consensus among experienced neurologists using clinical, neuropsychological, and biomarker data.

2.3. MRI protocol

MRI was performed using a 3T MRI system (Magnetom Trio, Siemens, Erlangen, Germany) with a 32‐channel head coil in both medical centers. Scans included T1‐weighted, T2‐weighted, and fluid‐attenuated inversion recovery (FLAIR) sequences. T1‐weighted images were acquired using a magnetization‐prepared rapid gradient echo (MPRAGE) sequence with the following parameters: sagittal slicing; repetition time, 7.14 ms; echo time, 3.49 ms; field of view, 240 × 240 mm²; voxel size, 1 × 1 × 1 mm³; slice thickness, 1 mm. T2‐weighted parameters included repetition time, 3000 ms; echo time, 90 ms; field of view, 230 × 230 mm²; voxel size, 0.22 × 0.22 × 6 mm³; slice thickness, 5 mm. FLAIR images were acquired in axial/coronal slices with a repetition time of 8000 ms, echo time of 120 ms, inversion time of 2400 ms, field of view of 230 × 233 mm², voxel size of 0.45 × 0.45 × 6 mm³, and slice thickness of 5 mm.

2.4. Data processing and analysis

MRI data were processed using FreeSurfer software, averaging the entorhinal, inferior temporal, middle temporal, and fusiform cortices as AD‐signature markers (Figure 1‐1). ³⁰ The volume of the hippocampal, total WMHs, periventricular WMH, and deep WMHs was also analyzed. WMH segmentation involved several steps: skull and non‐brain tissues were removed from T2‐FLAIR images using the Brain Extraction Tool from the FMRIB Software Library (FSL, Oxford, UK), followed by Otsu thresholding and manual adjustments. The WMH regions were localized and segmented into WMH periventricular and deep components using Lesion Segmentation Tool version 3.0.0 (Figure 1‐2). Hippocampal and WMH volumes were further adjusted by the estimated total intracranial volume (eTIV) before the statistical analysis.

FIGURE 1.

Brain regions and white matter hyperintensities (WMHs) used in the mediation models. (1) Brain regions included as covariates in the mediation models: the Alzheimer's disease signature cortical areas and the hippocampus. These regions include the entorhinal cortex (EC), fusiform gyrus (FG), hippocampus (HP), inferior temporal gyrus (ITG), and middle temporal gyrus (MTG). Cortical thickness in AD signature regions and hippocampal volume are associated with memory learning and consolidation abilities. ³⁰ , ³¹ (2) Periventricular WMHs are presented in blue, and deep WMHs are presented in yellow.

2.5. Neuropsychological and CR assessments

Participants completed a series of cognitive tests and questionnaires, including the following:

1. Montreal Cognitive Assessment (MoCA): Global cognitive function scores of < 24 indicate impairment. ³² , ³³

2. 12‐Item Recall Test: Verbal memory, with delayed recall after multiple learning trials.

3. Benson Complex Figure Test: Visuoconstructive ability with a score of up to 17 based on replication accuracy. ³⁴

4. Digit span backward: Executive function and working memory, with a maximum score of 16. ³⁵

5. Boston Naming Test: Linguistic function scored for correct object naming. The short‐form version was adopted in this study, with a total score of 15. ³⁶

6. Clinical Dementia Rating (CDR): Dementia severity and functional status. ³⁷ The CDR has six aspects: memory, orientation, judgment and problem‐solving, community affairs, home and hobbies, and personal care. Each domain can be rated as 0, 0.5, 1, 2, or 3, except that personal care does not have a score of 0.5, with a higher score indicating greater severity. Global CDR is assessed by the weighted score of six domains, and the CDR‐sum of boxes (CDR‐SB) is acquired by the summation of the six domains.

7. Geriatric Depression Scale (GDS): Depressive symptoms were assessed using the 15‐item version of GDS. The maximum score is 15, with higher scores indicating greater severity. ³⁸

8. CRI: The CRI questionnaire has four indices (CRI‐education, CRI‐working activity, CRI‐leisure time, and CRI‐total), and each index can be classified into five categories: medium‐low (71 to 84), medium (85 to 114), medium‐high (115 to 130), and high (≥130). ¹⁶ Due to concerns about Type I errors and to avoid collinearity issues, CRI‐total was used as a comprehensive proxy for CR.

2.6. Statistical analysis

Group differences in continuous demographic, clinical, and neuropsychological variables were analyzed using one‐way analysis of variance (ANOVA) with post hoc least significant difference (LSD) tests for significant findings. Chi‐squared tests were used for categorical variables. Statistical significance was set at p < 0.05.

Prior to conducting mediation analyses, model assumptions were assessed. Multicollinearity was examined using the variance inflation factor. Univariate and multivariate normality were evaluated using skewness, kurtosis statistics, and Mardia's test, respectively. Given violations of the normality assumption and the relatively small sample size, the bootstrap method (5000 resamples, seed = 42) was applied for robust estimation in the mediation models.

Structural equation modeling pathway analysis was utilized to examine the direct and indirect effects of CR, with WMH volume serving as the mediator. Mediation analyses focused on specific cognitive outcomes: the MoCA, Digit Span Backward, 12‐Item Recall Test, Benson Complex Figure Copy, and the Boston Naming Test. Within each mediation model, age, sex, vascular risk factors, hippocampus volume, and cortical thickness in the AD‐signature area were included as covariates. Initial p values were reported; however, to address multiple comparisons across models, significance was set at p < 0.05 after applying false discovery rate correction (the Benjamini–Hochberg procedure).

Between‐group analyses were conducted using IBM SPSS Statistics for Mac (version 29.0; IBM Corp., Armonk, NY, USA). Mediation analyses were performed in Python version 3.12.8 (Visual Studio Code environment), utilizing the pandas (2.2.3), numpy (2.2.1), scikit‐learn (1.6.1), statsmodels (0.14.4), semopy (2.3.11), and pingouin (0.5.5) libraries.

3. RESULTS

3.1. Demographic, clinical, neuroimaging, and neuropsychological testing characteristics

Our study included 85 CU, 43 MCI, and 61 AD participants. Table 1 details their demographics, clinical characteristics, brain imaging parameters, and neuropsychological features. We observed significant age differences across groups (ηp² = 0.09), with the CU group being significantly younger (68.6 ± 5.7 years) than the MCI (71.8 ± 6.5 years; p = 0.006) and AD groups (72.8 ± 6.2 years; p < 0.001). There was no significant age difference between the MCI and AD groups (p = 0.426). Notably, sex distribution, prevalence of hypertension, diabetes, hyperlipidemia, smoking status, and vascular risk scores did not significantly differ across groups (all p > 0.05).

TABLE 1.

Demographic, clinical, brain imaging, and neuropsychological features between groups.

	CU (n = 85)	MCI (n = 43)	AD (n = 61)
	M ± SD	M ± SD	M ± SD	p
Demographic and brain structure characteristics
Age (years)	68.6 ± 5.7	71.8 ± 6.5a	72.8 ± 6.2a	<0.001 *
Sex (female/male)	49/36	23/20	26/35	0.195
Hypertension (Yes/No)	42/43	21/22	35/26	0.575
Diabetes (Yes/No)	19/66	11/32	20/41	0.366
Hyperlipidemia	53/32	27/16	39/22	0.981
Smoking (No/Quit/Current)	6/7/72	3/6/34	2/14/45	0.140
Vascular risks	1.6 ± 1.1	1.7 ± 1.1	1.8 ± 1.2	0.351
CRI‐education	108.1 ± 15.0	104.0 ± 14.9	104.0 ± 18.1	0.228
CRI‐working activity	105.6 ± 20.2	103.4 ± 18.0	103.1 ± 20.0	−0.694
CRI‐leisure time	99.2 ± 17.1	91.6 ± 20.5a	87.1 ± 20.0a	<0.001 *
CRI‐total	105.7 ± 16.3	99.5 ± 17.9	97.4 ± 18.0a	0.013 *
AD signature (mm)	2.8 ± 0.3	2.7 ± 0.2	2.5 ± 0.2a,b	<0.001 *
Hippocampal volume **	0.6 ± 0.6	0.1 ± 0.7a	−0.8 ± 0.9a,b	<0.001 *
WMH‐total **	−0.1 ± 0.2	−0.1 ± 0.3	0.2 ± 1.7	0.132
WMH‐periventricular **	−0.2 ± 0.2	−0.0 ± 0.5	0.2 ± 1.6	0.081
WMH‐deep **	−0.1 ± 0.1	−0.1 ± 0.1	0.2 ± 1.7	0.282
Neuropsychological tests
GDS	2.7 ± 2.4	2.7 ± 2.1	2.2 ± 2.4	0.349
CDR‐SB	0.3 ± 0.7	1.4 ± 0.9a	4.1 ± 2.1a,b	<0.001 *
MoCA	26.9 ± 2.8	22.7 ± 3.5a	17.8 ± 4.6a,b	<0.001 *
12‐Item Recall Test	9.0 ± 2.3	3.3 ± 2.7a	1.0 ± 1.6a,b	<0.001 *
Benson Complex Figure Copy	16.2 ± 1.1	15.9 ± 1.2	15.1 ± 2.6a,b	<0.001 *
Digit Span Backward	8.1 ± 2.9	7.1 ± 2.7	5.9 ± 2.3a,b	<0.001 *
Boston Naming Test	14.5 ± 0.8	14.0 ± 1.2a	13.3 ± 2.0a	<0.001 *

Abbreviations: AD, Alzheimer's disease; CDR‐SB, Clinical Dementia Rating–Sum of Boxes; CRI, Cognitive Reserve Index; CU, cognitively unimpaired; GDS, Geriatric Depression Scale—15 items; M, mean; MCI, mild cognitive impairment; MoCA, Montreal Cognitive Assessment; SD, standard deviation; WMH, white matter hyperintensity.

^* p < 0.05.

^**Statistical comparisons of regional morphometric variables were based on standardized residual values (z‐scores), where the effects of the estimated total intracranial volume were regressed from the volumetric data.

Regarding CR, significant group differences were found in CRI‐total (ηp² = 0.05) and CRI‐leisure time (ηp² = 0.08). Post hoc analyses showed that the CU group had significantly higher CRI‐leisure time scores than the MCI (p = 0.030) and AD groups (p < 0.001). The CU group also demonstrated higher CRI‐total scores compared to the AD group (p = 0.005). No significant group differences were observed for CRI‐working activity or CRI‐education.

In terms of brain imaging measures, significant group differences were evident in cortical thickness within AD signature regions (ηp² = 0.22) and hippocampal volume (ηp² = 0.39). Post hoc comparisons for cortical thickness indicated that both the CU (p < 0.001) and MCI groups (p < 0.001) had significantly greater thickness than the AD group, with no significant difference between CU and MCI (p = 0.075). For hippocampal volume, both CU (p < 0.001) and MCI groups (p < 0.001) showed larger volumes than the AD group, and the CU group also had a greater volume than the MCI group (p = 0.001). No significant group differences were detected in WMH volume.

Significant group differences were observed for CDR‐SB (ηp² = 0.60), MoCA (ηp² = 0.54), 12‐Item Recall (ηp² = 0.73), Benson Complex Figure Copy (ηp² = 0.07), Digit Span Backward (ηp² = 0.12), and Boston Naming Test (ηp² = 0.13). Post hoc analyses indicated that the CU group consistently performed better and had lower impairment scores of CDR‐SB than the other groups. Specifically, the CU group outperformed the MCI group on the MoCA (p < 0.001), 12‐Item Recall (p < 0.001), and Boston Naming Test (p = 0.036) and had significantly lower CDR‐SB scores (p < 0.001). The CU group also showed significantly better performance across all assessed cognitive tests (MoCA, 12‐Item Recall, Benson Complex Figure Copy, Digit Span Backward, and Boston Naming Test; all p < 0.001) and lower CDR‐SB scores (p < 0.001) compared to the AD group.

3.2. General cognitive function

A mediation model with total WMH volume as the mediator was tested for MoCA performance. In the CU group, results indicated a significant direct (β = 0.46, p < 0.001) effect, while the indirect effect was not statistically significant (p = 0.087). This suggests that higher CR was directly associated with better MoCA performance (Figure 2A). In contrast, no significant direct or indirect effects were observed in the MCI or AD groups (Figures 3A and 4A).

FIGURE 2.

Pathway analysis within cognitively unimpaired (CU) group. (A) Total white matter hyperintensity (WMH) volume, and periventricular and deep WMHs as mediators, with Montreal Cognitive Assessment as outcome. (B) Total WMH volume, and periventricular and deep WMHs as mediators, with Digit Span Backward as outcome. (C) Total WMH volume, and periventricular and deep WMHs as mediators, with Benson Copy as outcome. (D) Total WMH volume as mediator, with 12‐Item Recall as outcome. (E) Total WMH volume as mediator, with Boston Naming as outcome. Original p values are reported; *p < 0.05 after correcting for the false discovery rate.

FIGURE 3.

Pathway analysis within mild cognitive impairment (MCI) group. (A) Total white matter hyperintensity (WMH) volume as mediator, with Montreal Cognitive Assessment as outcome. (B) Total WMH volume, and periventricular and deep WMHs as mediator, with Digit Span Backward as outcome. (C) Total WMH volume as mediator, with Benson Copy as outcome. (D) Total WMH volume as mediator, with 12‐Item Recall as outcome. (E) Total WMH volume as mediator, with Boston Naming as outcome. Original p values are reported; *p < 0.05 after correcting for false discovery rate.

FIGURE 4.

Pathway analysis within Alzheimer's disease group. (A) Total white matter hyperintensity (WMH) volume as mediator, with Montreal Cognitive Assessment as outcome. (B) Total WMH volume as mediator, with Digit Span Backward as outcome. (C) Total WMH volume as mediator, with Benson Copy as outcome. (D) Total WMH volume as mediator, with 12‐Item Recall as outcome. (E) Total WMH volume as mediator, with Boston Naming as outcome. Original p values are reported; *p < 0.05 after correcting for false discovery rate.

To further investigate the regional contribution of WMHs in the CU group, two additional mediation models were conducted, using periventricular and deep WMHs as distinct mediators. Bootstrap mediation analysis revealed a significant indirect effect of CR on MoCA performance through deep WMHs (β = 0.07, p = 0.042), but not through periventricular WMHs (p = 0.102). These findings suggest that higher CR is associated with better global cognitive function, partially mediated by a reduced deep WMH burden (Figure 2A).

3.3. Executive function

For executive function (Digit Span Backward), mediation analysis with total WMH volume as the mediator yielded varied results across groups. In the CU group, a direct positive association between CR and executive function was significant (β = 0.24, p = 0.030), but no significant indirect effect was found (p = 0.379), indicating no mediation of WMHs (Figure 2B). The MCI group showed significant total (β = 0.49, p < 0.001) and direct effects (β = 0.48, p < 0.001) of CR on working memory, but no significant indirect effect via WMHs (Figure 3B). Finally, in the AD group, neither direct nor indirect effects were significant (all p > 0.05) (Figure 4B).

To further evaluate the regional contributions of WMHs to cognitive outcomes in the CU and MCI groups, additional mediation models were conducted using periventricular and deep WMHs as separate mediators. The results consistently showed no significant indirect effects across WMH regions or groups, suggesting that regional WMH burden did not mediate the relationship between CR and working memory ability in these groups (Figures 2B and 3‐2).

3.4. Visual‐construction function

For the visual‐construction function (Benson Complex Figure Copy), a mediation model using total WMH volume as the mediator revealed distinct patterns. In the CU group, a significant direct (β = 0.33, p = 0.002) effect was observed, indicating that higher CR was directly associated with better visual‐construction function (Figure 2C). Conversely, neither significant direct nor indirect effects were found in the MCI or AD groups (Figures 3C and 4C). Further, in the CU group, mediation models incorporating periventricular or deep WMH volumes as mediators did not reveal significant indirect effects on visual‐construction function (Figure 2C).

3.5. Memory function

Mediation models were conducted within each group, using total WMH volume as the mediator and 12‐Item Recall as the outcome. The results consistently revealed no significant direct or indirect effects across any of the groups (Figures 2D, 3D, and 4D). These findings suggest that total WMH burden may not play a mediating role in episodic memory, as measured by the 12‐Item Recall task.

3.6. Language function

Similarly, mediation models examining naming ability, as measured by the Boston Naming Test, revealed no significant direct or indirect effects across the CU, MCI, or AD groups when total WMH volume was used as the mediator (Figures 2E, 3‐5, and 4E), suggesting that WMH burden may not mediate the relationship between CR and naming ability.

4. DISCUSSION

This study investigated the mediating role of CR in the relationship between WMHs and cognitive function, independent of AD‐related brain structures. Our findings indicate that higher CR is positively associated with non‐memory cognitive domains, including global cognition, visuoconstruction, and executive function, particularly in CU older adults and those in early stages of AD. Furthermore, CR appears to contribute to cognitive preservation by mitigating the burden of deep WMH.

Our current findings align with previous research suggesting that CR exerts a stronger protective effect on non‐memory cognitive domains than on memory‐related functions. One study found that the effect of education was most pronounced in executive functioning, followed by attention and visuospatial abilities, with less prominent effects on memory and language in both predementia and dementia populations with AD. ³⁹

At a neural level, higher CR in CU older adults has been linked to increased functional connectivity in the frontoparietal control network and decreased connectivity in the dorsal attention network. ²⁵ Similarly, in individuals with mild AD, more years of formal education have been associated with enhanced functional connectivity between the left inferior temporal cortex and the middle frontal cortex and between the right superior frontal cortex and the right angular gyrus. ⁴⁰ Since these neural circuits are more closely tied to executive control processes than to memory functions, this may explain why CR's protective effects are more salient in preserving non‐memory cognitive abilities.

Our pathway analysis revealed no significant association between WMH burden and memory performance. Instead, episodic memory was significantly linked to both cortical thickness in AD‐signature regions (p = 0.002) and hippocampal volume (p < 0.001). These findings suggest that hippocampal atrophy, rather than WMH burden, plays a more critical role in memory function across the AD spectrum.

Our findings indicate that deep WMH volume partially mediates the relationship between CR and global cognitive capacity in the CU group. While the link between WMH and cognitive decline is established, regional contributions are debated. ⁶ Consistent with previous studies associating deep frontal, occipital, and parietal WMHs with executive, calculation, language, and visuoconstructive impairments, ⁴¹ , ⁴² our results show that deep WMH volume is negatively associated with global cognition and mediates the pathway between CR and global cognitive performance.

In our study, significant associations between CR and cognitive performance were exclusively observed in the CU and MCI groups. The existing literature, however, presents inconsistencies regarding CR's protective role across the AD spectrum. Some studies suggest CR is more effective in early disease stages, while others propose its influence becomes more prominent in moderate to advanced stages. A 5‐year longitudinal study found CR predicted executive function decline only in individuals with elevated CSF tau‐to‐amyloid ratios, implying CR's protective effects primarily occur with a certain level of AD pathology. ¹⁹ Conversely, a longitudinal study recruiting 839 amyloid beta‐positive individuals reported that higher CR was linked to a lower rate of diagnostic conversion and slower cognitive declines, particularly in memory and executive function. ⁴³ These protective effects were predominantly observed in the CU and MCI groups, leading to the conclusion that CR is critically protective during the predementia stages of AD.

The mixed findings across studies likely stem from differences in biomarker selection and covariate adjustment strategies. Our results align with the latter perspective, contributing to this body of evidence by suggesting that CR may confer cognitive protection predominantly in the predementia stage of AD, potentially through the reduction of deep WMH burden.

Our study demonstrates that higher CR supports optimal cognitive function. We defined CR by educational attainment, work activity, and leisure engagement. These findings have significant implications for dementia prevention. First, since education is a consistent protective factor against cognitive decline – from early schooling to lifelong learning – promoting educational opportunities at all ages is crucial for maintaining cognitive health. Second, engagement in leisure activities, including social interaction, mentally stimulating tasks, and physical exercise, was positively linked to better cognitive outcomes. From a public health perspective, community programs that integrate physical activity, social engagement, and cognitively enriching games offer a practical, holistic way to boost CR and delay cognitive decline in older adults.

This study has two main limitations. First, our AD classification relied on structural brain features and clinical presentation, rather than confirmation with amyloid PET or CSF biomarkers. While this method has shown satisfactory sensitivity and specificity in prior research, it carries a risk of misdiagnosis, especially in the MCI stage. ³⁰ , ⁴⁴ This could affect the validity of our findings, as individuals with initial amnestic symptoms might later be diagnosed with conditions like non‐AD dementia. Incorporating biomarker assessments would significantly improve diagnostic accuracy. ⁴⁵ Future studies should incorporate biomarker assessments to improve diagnostic accuracy and validate our findings. Second, the cross‐sectional design prevents us from drawing causal inferences regarding the relationship between CR and WMH. We cannot determine if CR‐enhancing interventions promote frontal functional connectivity. Future longitudinal studies integrating functional MRI are needed to address such causal questions.

CONFLICT OF INTEREST STATEMENT

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

CONSENT STATEMENT

Ethical approval for the study protocol (2023‐07‐012AC) was granted by the Institutional Review Board of the Taipei Veterans General Hospital, and all participants provided written informed consent before they participated in this study.

Supporting information

Supporting Information

Lin Y‐R, Lee W‐L, Fuh J‐L. The role of cognitive reserve in white matter hyperintensities: from cognitive aging to Alzheimer's spectrum. Alzheimer's Dement. 2025;17:e70167. 10.1002/dad2.70167