White Matter Abnormalities and Cognition in a Community Sample

Abstract

White matter hyperintensities (WMH) can compromise cognition in older adults, but differences in sampling, WMH measurements, and cognitive assessments contribute to discrepant findings across studies. We examined linear and nonlinear effects of WMH volumes on cognition in 253 reasonably healthy adults. After adjusting for demographic characteristics and total brain volumes, WMH burden was not associated with cognition in those aged 20–59. In participants aged 60 and older, models accounted for ≥58% of the variance in performance on tests of working memory, processing speed, fluency, and fluid intelligence, and WMH volumes accounted for variance beyond that explained by age and other demographic characteristics. Larger increases in WMH burden over 5 years also were associated with steeper cognitive declines over the same interval. Results point to both age-related and age-independent effects of WMH on cognition in later life and suggest that the accumulation of WMH might partially explain normal age-related declines in cognition.

Introduction

Cerebrovascular disease contributes to late-life cognitive decline and is a risk factor for stroke (Kuller, Longstreth, Arnold, Bernick, Bryan, & Beauchamp, 2004) and the development of Alzheimer disease (Prins et al., 2004). White matter hyperintensities (WMH) refer to areas of hyperintense signal on T2- or proton density-weighted brain magnetic resonance imaging (MRI). They are thought to reflect ischemic brain changes (Fazekas, Schmidt, & Scheltens, 1998), but other etiologies may also contribute (Pantoni & Garcia, 1997). White matter hyperintensities appear in neurologically healthy adults (Schmidt, Hayn, Fazekas, Kapeller, & Esterbauer, 1996) and in those with stroke (Kuller et al., 2004) and dementia (Bigler, Kerr, Victoroff, Tate, & Breitner, 2002). They are strongly associated with advancing age, but also with cerebral atrophy and cerebrovascular risk factors (Gunning-Dixon & Raz, 2000; Longstreth et al., 1996; Oosterman, Sergeant, Weinstein, & Scherder, 2004; Wen, Sachdev, Chen, & Anstey, 2006). In longitudinal studies, greater WMH burden at baseline has been demonstrated to be a risk factor for poor outcomes in the elderly, including functional decline (Inzitari et al., 2007) and death (Briley, Haroon, Sergent, & Thomas, 2000; Inzitari, Cadelo, Marranci, Pracucci, & Pantoni, 1997). Additionally, many studies have shown that severity of WMH correlates inversely with cognitive test performance in older adults, but findings vary (Gunning-Dixon & Raz, 2000). For example, Hunt and colleagues (1989) found no association between WMH and cognitive functioning in otherwise healthy elderly individuals, and Boone and colleagues (1992) found an inverse correlation, but only in older adults with extensive WMH burden, suggesting a “threshold effect.” These inconsistencies probably reflect differences in the methods used to measure WMH burden. Variations in WMH lesion location, breadth of cognitive domains assessed, and subject characteristics probably also contribute to discrepant findings (Pantoni, Poggesi, & Inzitari, 2007).

The use of visual rating scales is the most common approach to quantifying WMH on MRI. Such scales yield ordinal-level data at best, and are beset by variable inter- and intra-rater reliability, ambiguous terminology, and inconsistent analyses of lesion size/quantity, location, and configuration (Bigler et al., 2002; Garrett et al., 2004; Pantoni & Garcia, 1995; Wardlaw, Ferguson, & Graham, 2004). The Fazekas scale (Fazekas, Chawluk, Alavi, Hurtig, & Zimmerman, 1987) assesses periventricular and subcortical WMH on a 4-point scale. The Scheltens scale (Scheltens et al., 1993) rates periventricular WMH on a 10-point scale and subcortical WMH on a 25-point scale. Automated computer-assisted estimates of WMH volume might overcome these limitations (Garrett et al., 2004; Wardlaw et al., 2004) and improve sensitivity to WMH-cognition associations (Gunning-Dixon & Raz, 2000). However, they have not been validated against volumes derived from manual tracing. Van Straaten and colleagues (2006) compared automated volumetric estimates with visual ratings of WMH and found that brains assigned the highest ratings of WMH severity showed much greater variability in WMH volume than those assigned lower ratings. This is troublesome because those with the greatest WMH burden are also likely to show the most severe cognitive dysfunction. Thus, visual rating scales might obscure variability that is essential to detecting WMH-related cognitive dysfunction in precisely those individuals who are most likely to exhibit it.

The location of WMH in the brain also might influence patterns of cognitive impairment by virtue of the fiber tracts affected (de Groot et al., 2000). Subcortical WMH are associated with localized grey matter reduction (Wen et al., 2006) and are thought to interrupt short association fibers linking adjacent brain regions (de Groot et al., 2000). Periventricular WMH more likely disrupt long association fibers connecting distant brain regions, and therefore might affect performance on tasks that require integration of information across more widely separated cortical areas (DeCarli et al., 1995; Scheltens et al., 1993). Whereas some evidence indicates that periventricular WMH are more strongly associated with cognition (DeCarli, Fletcher, Ramey, Harvey, & Jagust, 2005), meta-analytic reviews have not found differential relationships with cognition based on WMH location (Gunning-Dixon & Raz, 2000; Oosterman et al., 2004). Given the typical pattern of WMH accumulation over time, wherein WMH burden expands out from the periventricular region into the deep white matter (Schmidt, Enzinger, Ropele, Schmidt, & Fazekas, 2003), an alternative possibility is that the degree of periventricular WMH serves as a proxy for total aggregate WMH burden. Hence total WMH, rather than the specific location of WMH, could conceivably underlie the differential cognitive effects often associated with periventricular WMH.

In this study we examined continuous volumes of WMH in various brain regions, and related them to performance on a broad range of neuropsychological tests in a community sample of adults. We tested for “threshold effects” of WMH on cognitive test performance by evaluating whether nonlinear (quadratic) terms for WMH volumes would relate inversely to WMH volumes, but only after a threshold of ischemic burden was reached. We hypothesized that WMH-related deficits would be most prominent in the domains of executive function and mental processing speed. The independent associations of region-specific WMH and neuropsychological test performance were also deemed an important area in need of further investigation. Based on prior work in this area (de Groot et al., 2000; van den Heuvel et al., 2006), we hypothesized that periventricular WMH would show stronger associations than subcortical WMH with measures of executive function and processing speed. Finally, in order to further explore the practical and clinical significance of our findings, we explored the association between increases in WMH burden over time as they relate to changes in cognition over that same period.

Materials and Methods

Participants

A total of 301 adults entered into the Johns Hopkins Aging, Brain Imaging, and Cognition (ABC) study (Schretlen, Testa, Winicki, Pearlson, & Gordon, 2008) between 1996 and 2005. The participants were recruited via random digit dialing, written invitation to Medicare beneficiaries (over age 64), and telephone calls to pseudo-randomly selected numbers from a residential directory for the Baltimore metropolitan area. Individuals were excluded at this step if English was not their first language or if an obvious cognitive disorder precluded them from completing the telephone screening. Subjects gave written informed consent, and the study was approved by the Johns Hopkins Medicine IRB. On a single day, each participant underwent physical and neurological examinations, a psychiatric interview, laboratory blood studies, and a brain MRI scan. We excluded 6 individuals who did not undergo the brain MRI scan (e.g., due to claustrophobia, metal implantation, obesity) and 40 individuals whose scans were unusable due to technical problems (e.g., motion artifact), along with 2 individuals who did not complete cognitive testing. This left 253 participants with MRI scans who completed most or all of the cognitive tests. Individuals whose data were included for use in the present study were somewhat older (M = 57.8, SD = 18.6) than those for whom there were not usable data (M = 51.4, SD = 19.9, t (299) = 2.14, p = 0.034). Otherwise, the groups did not differ in terms of sex, race, education, body mass index, or rates of hypertension, diabetes mellitus, smoking, or history of alcohol abuse/dependence (p > 0.05). Participants ranged in age from 20 to 96 years. Sample characteristics are shown in Table 1.

Table 1.

Participant characteristics, WMH volumes, and comparison by age group

Characteristic	Age			p
	All ages (n = 253)	≥ 60 (n = 121)	< 60 (n = 132)
Age (years)	57.8 ± 18.6	74.2 ± 8.5	42.7 ± 10.9	<0.001
Education (years)	13.4 ± 3.4	13.2 ± 3.6	13.6 ± 3.2	0.372
Sex (men)	123 (48.6)	62 (51.2)	61 (46.2)	0.424
Race (Caucasian)	181 (71.5)	92 (76.0)	89 (67.4)	0.285
Hypertension	86 (34.0)	58 (47.9)	28 (21.2)	<0.001
Diabetes mellitus	31 (12.3)	21 (17.4)	10 (7.6)	0.018
Smokers	61 (24.1)	22 (18.2)	39 (29.5)	0.038
History alcohol abuse/dep.b	21 (8.3)	8 (6.6)	13 (9.8)	0.323
Body mass index	27.8 ± 5.5	26.6 ± 4.9	28.9 ± 5.8	0.001
Total WMH (cm3)a	2.03 ± 5.06	3.85 ± 6.85	0.36 ± 0.63	<0.001
Periventricular WMH (cm3)a	1.53 ± 4.07	2.92 ± 5.56	0.26 ± 0.40	<0.001
Subcortical WMH (cm3)a	0.50 ± 1.37	0.93 ± 1.86	0.10 ± 0.35	<0.001

Notes: Values are mean ± SD or n (%).

^aWMH volumes used in the statistical analyses were adjusted to reflect WMH as a percentage of total brain volume.

^bHistory of alcohol abuse or dependence.

Of the 215 individuals who were recruited during the first phase of the ABC study (1995–1998), 110 agreed to return for a follow-up assessment during which they completed an identical study protocol. Of those returning, 44 individuals were at least 55 years old at the time of initial study entry (M = 69.3, SD = 8.5) and produced usable cognitive and neuroimaging data on both occasions. The interval between initial and follow-up evaluations averaged just over 5 years (M = 5.3, SD = 0.6). The 44 returning participants were similar to the sample as a whole in terms of sex (47.7% male), race (88.6% Caucasian), and years of education (M = 13.7, SD = 3.0).

Because the aim of this investigation was to examine the neuropsychological correlates of WMH in a broadly representative community sample irrespective of WMH's etiology or participant disease status, we included all individuals who met these criteria. Based on the physical and neurological examinations, each participant was categorized as belonging to one of four categories of global health. Most participants (79.1%) were either free of medical problems (n = 66) or had relatively minor health conditions such as uncomplicated diabetes mellitus, controlled hypertension, or simple phobias (n = 134). Compound or poorly controlled illnesses, such as complicated diabetes mellitus, long-standing poorly controlled hypertension, and severe chronic obstructive pulmonary disease, were present in 36 participants, and 17 participants had severe illnesses or conditions known to affect the central nervous system, such as pancreatic cancer, dementia, or alcohol dependence.

MRI Protocol

Participants underwent a T2- and proton density-weighted MRI brain scans. Contiguous 5 mm slices through the entire brain were obtained in the oblique axial plane, with the sections angled parallel to the anterior–posterior intercommissural line. The parameters were: repetition time = 2,500 ms, echo times = 30 ms (for proton density images) and 80 ms (for T2-weighted images), field of view = 24 cm, image matrix = 256 × 256. Magnetic resonance imaging data were displayed and WMH were measured using image analysis software developed in our laboratory, ‘Measure’ version 0.7 (Barta, Dhingra, Royall, & Schwartz, 1997), which allows for outlining WMH regions of interest on each axial slice using a mouse-controlled cursor. After all WMH observed on T2 images throughout the brain were traced, their total volume was calculated via the Measure program.

In order to be considered an area of hyperintensity, the region of interest must have appeared hyperintense on both T2 and proton density-weighted images. Consistent with prior research (de Groot et al., 2000; van den Heuvel et al., 2006; Victoroff, Mack, Grafton, Schreiber, & Chui, 1994), WMH that contacted the wall of the ventricle were defined as periventricular. White matter hyperintensities were classified as subcortical if they did not contact the wall of the ventricles. In cases where an axial image either superior or inferior to the ventricles (but without a direct view of the ventricles) showed a WMH, coronally reformatted T2 and proton density-weighted images were made from the axially acquired data and examined to determine whether the WMH was contiguous with the ventricular ependymal surface.

All WMH tracings were performed by a single rater (T.D.V.) who was trained by a neuroradiologist (M.K.), and performed the tracings while blinded to participant characteristics such as age and sex. The re-tracing of 13 brains (T.D.V.) yielded an acceptable estimate of intra-rater reliability, as assessed via intra-class correlation (ICC), for both total WMH (0.97) and periventricular WMH (0.98) volumes. The inter-rater reliability of 30 cases was acceptable for both total (ICC = 0.98) and periventricular WMH (ICC = 0.94) volumes. Estimates of subcortical WMH volume were derived by subtracting periventricular WMH from total WMH volumes. Data presented for total, periventricular, and subcortical WMH are expressed as the proportion of WMH volume to total brain volume.

Neuropsychological Measures

Participants completed an extensive battery of neuropsychological tests, each of which is referenced previously (Schretlen et al., 2008). Raw scores were recorded for the Information, Similarities, Picture Completion, Block Design, Digit Span, and Digit Symbol Coding subtests of the Wechsler Adult Intelligence Scale-Revised (WAIS-R) and for Logical Memory (LM) and Visual Reproduction (VR) subtests of the revised Wechsler Memory Scale (WMS-R). Other measures included total correct responses to the Benton Facial Recognition Test (BFRT), Brief Test of Attention (BTA), Perceptual Comparison Test (PCT), and Rey-Osterrieth Complex Figure Test (RCFT; copy only). Total learning and delayed recall were recorded for the Brief Visuospatial Memory Test-Revised (BVMT-R) and Hopkins Verbal Learning Test-Revised (HVLT-R). Acceptable novel drawings were recorded for the Design Fluency Test (DFT), as were times to complete parts A and B of the Trail Making Test (TMT). We recorded total acceptable words beginning with the letters S and P (letter fluency) and animals and supermarket items (category fluency) given in four consecutive 1-min trials, and the numbers of categories completed and perseverative errors made on a modified version of the Wisconsin Card Sorting Test (mWCST). Estimated “pre-morbid” IQ was based on the revised Hopkins Adult Reading Test (HART; Schretlen et al., 2009).

Statistical Analysis

In order to reduce the number of cognitive variables under consideration, each of the 28 test scores was z-transformed, based on the sample distribution, and assigned to one of eight cognitive domains. For analyses conducted within specific age groups, z-transformations were calculated based on the means and standard deviations of data from participants within that age group. Coefficients alpha for these domains (Table 2) indicate adequate to excellent internal consistency. For all domains, higher scores indicate better performance. Spearman rho (ρ) correlation analyses were used to examine the relationships among total, periventricular, and subcortical WMH volumes, as well as their zero-order relationships with cognitive performance, age, and education. In addition, multiple regression analyses were used to examine linear and quadratic associations between total, periventricular, and subcortical WMH volumes and each of the eight cognitive domains, with cognitive scores serving as the dependent variables. Multiple regression models were adjusted for the effects of age, years of education, sex, and race because each of these correlated significantly with one or more cognitive measures. Finally, Pearson r correlations evaluated the association between interval change in WMH and interval change in performance on each of the eight cognitive domains. All change scores were calculated as time 2 values minus time 1 values. Thus, larger positive numbers indicate greater improvement in cognition from baseline and greater development of WMH over time.

Table 2.

Cognitive tests and domain characteristics

Cognitive domaina	Age
	All ages (n = 253)		≥60 (n = 121)		<60 (n = 132)
	Mean (SD)	α	Mean (SD)	α	Mean (SD)	α
Working memory
Digits forward	6.5 (1.6)	0.78	6.2 (1.3)	0.75	6.8 (1.8)	0.76
Digits backward	5.2 (1.7)		4.7 (1.3)		5.6 (2.0)
BTA, letters	7.2 (2.4)		6.3 (2.5)		8.0 (1.9)
BTA, numbers	7.3 (2.2)		6.4 (2.3)		8.0 (1.9)
Processing speed
PCT	58.7 (17.0)	0.93	49.5 (14.9)	0.93	67.1 (14.2)	0.89
TMT, Part A	40.7 (24.6)		51.9 (28.6)		30.4 (13.8)
TMT, Part B	116.8 (94.4)		153.7 (116.3)		82.8 (47.9)
Symbol digit coding	44.9 (14.6)		37.0 (13.2)		52.2 (11.7)
Fluency
Letter fluency	26.9 (9.2)	0.75	25.8 (9.7)	0.77	28.0 (8.7)	0.72
Category fluency	41.1 (10.9)		37.9 (10.5)		44.1 (10.3)
DFT	13.2 (7.0)		11.3 (6.2)		15.0 (7.3)
Crystallized intelligence
Information	19.3 (5.7)	0.87	20.5 (5.6)	0.88	18.3 (5.5)	0.86
Similarities	18.5 ( 5.3)		18.3 (5.9)		18.7 (4.8)
HART IQ	102.6 (11.1)		104.0 (10.9)		101.3 (11.1)
Fluid intelligence
RCFT	30.0 (5.3)	0.81	28.3 (6.3)	0.80	31.7 (3.6)	0.69
BFRT	21.7 (2.6)		20.5 (2.6)		22.9 (2.0)
Block design	24.6 (10.7)		19.9 (9.0)		28.9 (10.4)
Picture completion	13.9 (3.6)		12.7 (4.1)		15.1 (2.7)
Verbal memory
HVLT-R learning	23.9 (5.2)	0.88	22.1 (5.2)	0.87	25.6 (4.6)	0.86
HVLT-R recall	8.3 (2.8)		7.2 (3.0)		9.3 (2.3)
LM learning	25.5 (7.5)		23.3 (7.4)		27.5 (7.2)
LM recall	21.1 (8.5)		19.0 (8.1)		23.0 (8.5)
Visual memory
BVMT-R learning	20.7 (8.0)	0.92	16.5 (6.9)	0.91	24.5 (6.9)	0.89
BVMT-R recall	8.0 (3.0)		6.7 (2.9)		9.2 (2.5)
VR learning	31.2 (7.2)		27.5 (7.6)		34.5 (4.8)
VR recall	20.9 (11.1)		15.6 (9.6)		25.8 (10.2)
Executive functioning
mWCST categories	5.0 (1.5)	0.87	4.6 (1.7)	0.84	5.3 (1.2)	0.90
mWCST per. errors	3.4 (4.8)		4.4 (5.0)		2.6 (4.4)

Notes: BTA = Brief Test of Attention; PCT = Perceptual Comparison Test; TMT = Trail Making Test; DFT = Design Fluency Test; HART = Hopkins Adult Reading Test; RCFT = Rey-Osterrieth Complex Figure Test; BFRT = Benton Facial Recognition Test; HVLT-R = Hopkins Verbal Learning Test-Revised; LM = Logical Memory; BVMT-R = Brief Visuospatial Memory Test-Revised; VR = Visual Reproduction; mWCST = modified Wisconsin Card Sorting Test.

^aData reflect raw scores.

Results

Zero-Order Associations

Total WMH volume correlated very highly with periventricular (ρ = 0.97; p < 0.0001) and subcortical (ρ = 0.83; p < 0.0001) WMH volumes. Periventricular and subcortical WMH also shared a robust association (ρ = 0.74, p < 0.0001). Significant Spearman rho correlations between total WMH volumes and cognitive factor scores ranged from −0.29 to −0.51 (p < 0.0001) for the entire sample. Total WMH volume also correlated highly with age (ρ = 0.63 and <0.0001). Total WMH burden correlated modestly with education (ρ = −0.20 and <0.01).

Linear Multiple Regression Analyses

As is seen in Table 3, for the entire sample, greater total and periventricular WMH volumes were associated with poor performance in numerous cognitive domains. Greater subcortical WMH was associated with worse performance only on tests of working memory and fluency. However, after accounting for age, sex, race, and years of education, WMH volumes explained less than 3% of unique, incremental variance in cognitive performance.

Table 3.

Linear associations between WMH and cognitive factor scores in models including age, education, sex, and race

	All ages						Age ≥60
	Total WMH		PV WMH		SC WMH		Total WMH		PV WMH		SC WMH
	R2Δ	R2	R2Δ	R2	R2Δ	R2	R2Δ	R2	R2Δ	R2	R2Δ	R2
Working memory	0.024*	0.346	0.018*	0.341	0.028*	0.351	0.047*	0.374	0.036*	0.363	0.057*	0.384
Processing speed	0.015*	0.574	0.017*	0.577	0.003	0.563	0.019*	0.538	0.021*	0.539	0.007	0.526
Fluency	0.028*	0.356	0.027*	0.356	0.016*	0.345	0.038*	0.417	0.034*	0.413	0.028*	0.407
Crystallized intelligence	0.001	0.589	0.002	0.590	0.000	0.589	0.002	0.579	0.002	0.580	0.000	0.578
Fluid intelligence	0.011*	0.543	0.016*	0.548	0.000	0.532	0.013	0.464	0.020*	0.471	0.000	0.451
Verbal memory	0.012*	0.309	0.012*	0.309	0.006	0.303	0.021	0.230	0.021	0.230	0.010	0.219
Visual memory	0.009*	0.528	0.009*	0.528	0.006	0.525	0.015	0.408	0.014	0.406	0.011	0.403
Executive functioning	0.002	0.262	0.001	0.261	0.004	0.264	0.003	0.420	0.001	0.418	0.008	0.425

Notes: All multiple R² values significant at p < 0.0001.

*p < .05.

White matter hyperintensities appeared in most (79.5%) of young and middle-aged adults (aged 20–59), but were more common in those aged 60 and older (95%). An independent samples t-test confirmed that younger individuals had less WMH burden, with participants aged 20–59 showing smaller WMH-to-total-brain-volume ratios (M = 0.0003, SD = 0.0006) than those aged 60 and older (M = 0.004, SD = 0.007, p < 0.001). Therefore, we repeated the regression analyses separately for each age group. This revealed similar associations of WMH with working memory, processing speed, fluency, and fluid intelligence for the elderly adults. In fact, the resulting models generally accounted for more variance in cognitive test performance, and showed that WMH accounted for a greater proportion of the explained variance among elderly adults than in the sample as a whole. Conversely, WMH failed to explain significant incremental variance in any cognitive measure after accounting for the effects of age, sex, race, and years of education among the young and middle-aged participants (p > .05).

Threshold Effects: Quadratic Associations Between WMH and Cognitive Domains

For the entire sample, total and periventricular WMH volumes showed curvilinear (quadratic) associations with fluid intelligence and visual memory (Table 4). In those aged 60 and over, visual memory and processing speed both showed curvilinear associations with WMH, particularly those in the periventricular regions. However, the relationship between periventricular WMH volumes and cognition remains largely linear over the wide range of cognitive test performances with no evidence to suggest an abrupt inflection of the curve or threshold effect (data for the full sample and total WMH not presented). There were no significant quadratic relationships between subcortical WMH and cognition, and no quadratic WMH-cognition associations were present in those under age 60 (ps > 0.05).

Table 4.

Quadratic associations between cognitive factor scores, total WMH, and periventricular WMH

	All ages				Age ≥60
	Total WMH		PV WMH		Total WMH		PV WMH
	β	t	β	t	β	t	β	t
Working memory	0.21	1.83	0.23	1.85	0.27	1.62	0.31	1.71
Processing speed	0.15	1.55	0.18	1.79	0.27	1.83	0.31	2.04*
Fluency	0.14	1.17	0.19	1.57	0.25	1.51	0.32	1.88
Crystallized intelligence	0.04	0.43	0.03	0.30	0.10	0.69	0.07	0.49
Fluid intelligence	0.20	2.01*	0.25	2.42*	0.27	1.70	0.31	1.90
Verbal memory	0.05	0.40	0.08	0.61	0.07	0.36	0.11	0.55
Visual memory	0.25	2.58*	0.30	2.89*	0.41	2.55*	0.50	2.94*
Executive functioning	0.20	1.62	0.15	1.15	0.26	1.58	0.19	1.09

Notes: Significant values (*) indicate that the associations between WMH volumes and cognitive domain are best captured by a nonlinear, quadratic function, in models including age, education, sex, and race. The quadratic relationships between subcortical WMH and cognition in the full sample and older group were not significant and are not reported here. There were also no significant quadratic relationships between WMH and cognition in those aged below 60 and these data are not presented.

*p < .05.

Longitudinal Analysis

Pearson correlations revealed that greater increases in total WMH volumes over time were associated with greater declines on tests of fluency (r = −0.31, p = 0.04) and crystallized intelligence (r = −0.46, p < 0.01). An identical pattern emerged for increases in periventricular WMH volumes with fluency (r = −0.30, p < 0.05) and crystallized intelligence (r = −0.46, p < 0.01). Greater increases in subcortical WMH volumes over time were associated with steeper declines in fluency (r = −0.30, p < 0.05), crystallized intelligence (r = −0.40, p < 0.01), and visual learning and memory (r = −0.38, p = 0.01). Associations were trending in the expected direction for subcortical WMH volumes and working memory (r = −0.27, p = .08). Contrary to expectation, greater increases in subcortical WMH burden over time were associated with improvements on tests of executive functioning (r = 0.32, p = .03), likely due at least in part to the sizable practice effects associated with this task.

Discussion

Among young to middle-aged adults, we found that total WMH burden tends to be minimal and is not associated with decreased cognitive performance. In contrast, among adults aged 60 and older, WMH are common and are clearly associated with decreased cognitive test performance. Overall, the cognitive domains that related most strongly to WMH burden in this study overlap with those reported previously, as several studies identified processing speed and executive functioning as most sensitive to WMH burden (Gunning-Dixon & Raz, 2000; Oosterman et al., 2004), and one found an association between WMH volume and visual learning/memory (DeCarli et al., 1995).

Our statistical models accounted for 22% to 58% of the variance in cognitive performance in older adults. However, much of the variance accounted for by WMH burden overlaps that explained by age and other demographic variables. Nonetheless, our finding that WMH volume explained significant unique variance in cognitive test performance beyond that explained by demographic variables is noteworthy. A meta-analysis (Gunning-Dixon & Raz, 2000) found that only 4% to 9% of cognitive test score variance was attributable to WMH burden, and that removing the effects of age did not decrease the strength of the WMH-cognition associations. In fact, advancing age (of that sample) was associated with weaker relationships between WMH burden and cognition, which is difficult to explain. In contrast, our data show that WMH volumes correlate strongly with age (ρ = 0.63), likely due to the broad age range sampled. Our findings further suggest that WMH burden makes both age-dependent and -independent contributions to cognitive functioning. Notably, the strongest associations between WMH burden and cognition were observed on timed tasks (i.e., processing speed, fluency, and working memory). Given that WMH tend to occur along the length of—and expand out from—the lateral ventricles, one possibility is that WMH disrupt the association fibers of white matter tracts, and perhaps slow neural transmission. The processing speed theory of cognitive aging suggests that a reduction in simple processing speed occurs as part of the normal aging process and results in the gradual cognitive decline that is observed across many cognitive domains over the course of late adulthood (Salthouse, 1996). Thus, increasing burden of WMH might also accentuate normal age-related decline in processing speed via their hypothesized contribution to neuronal transmission.

The associations between WMH burden and cognition independently accounted for only up to 6% of the variance in cognitive test performance beyond that explained by age and other demographic factors in cross-sectional analyses. However, among older participants, greater increases in WMH burden over a 5-year period were associated with greater declines on several of the cognitive domains occurring over that same timeframe. These findings show that elderly individuals whose cerebral ischemic burden increases over time show progressively worsening cognitive functioning. When considered within the context of preventable causes of cognitive decline, these findings suggest that efforts aimed at addressing the cerebrovascular risk factors associated with WMH development and progression (e.g. hypertension) might function to prevent WMH-associated cognitive decline in later life.

With respect to the role of WMH lesion location, we observed a strong overlap between total and periventricular WMH volumes. In light of this redundancy, it is not surprising that relatively few differences were detected with regard to the associations between region-specific WMH and cognition. However, there was some evidence of differential patterns by location. While greater hyperintensity burden in both the periventricular and subcortical regions was related to poorer performance on tests reliant on the integrity of the frontal lobes (i.e., working memory, fluency), poorer performance on measures of processing speed were associated exclusively with the extent of periventricular WMH burden and not with the severity of WMHs extending into the subcortical region. This finding further supports the idea that periventricular WMH disrupt long association fibers connecting distant regions of the brain and might therefore differentially impede cognitive processes that depend on the integration of information across broad cortical areas (DeCarli et al., 1995; Scheltens et al., 1993).

Although contrary to our initial hypothesis, the finding that subcortical WMH showed the strongest association with any single cognitive domain is not particularly surprising given the topographical distribution of WMH observed in our sample. That is, lesions were most prevalent adjacent to the walls of the ventricles (i.e., periventricular WMH) and diminished in both size and frequency projecting into the deep matter (i.e., becoming subcortical WMH). In fact, all of our subjects with subcortical lesions also showed periventricular lesions. Thus, it is plausible that the more robust association between working memory and subcortical WMH reflects both the greater overall burden of WMH and their penetration further out from the wall of the lateral ventricles and into the frontal lobes. This interpretation is consistent with the findings of DeCarli and colleagues (1995) who noted similarly high correlations among total and regional WMH. These investigators, using 3-D mapping techniques, failed to find any clear distinction and/or differentiation between hyperintensities in the periventricular and subcortical regions and suggested that total WMH burden underlies the WMH–cognition associations.

Finally, little evidence of WMH–cognition threshold effects was found in this study. The significant quadratic associations that were present, on closer inspection, showed relatively linear inverse associations with cognition across a range of cognitive test scores. This finding differs from Boone and colleagues (1992), who advanced the hypothesis of a threshold effect, and DeCarli and colleagues (1995) who found partial support for it. Our failure to find a threshold effect likely reflects differences in the WMH measurement techniques, statistical approach, and especially sample characteristics. The current study examined the WMH–cognition associations in a community sample of adults between 20 and 96 years of age, using continuous volumes of WMH throughout the cerebrum. Boone and colleagues's sample included only healthy older adults, examined WMH areas rather than volumes, looked exclusively in the subcortical region, and collapsed WMH areas into ordinal categories. DeCarli and colleagues's sample consisted of high functioning (mean IQ = 125) healthy adults who were free of cardiovascular disease, and the authors used stepwise multiple regression analyses.

Several study limitations should be noted. Data for the current study were gathered over the course of several years. When data acquisition began, 5 mm slice thickness of the MR images was the standard. This convention was maintained over the course of the data acquisition period to maintain constant technique over time; however, it may have resulted in a reduced ability to detect some small WMH. Additionally, while WMH are commonly assumed to reflect cerebral ischemia, not all observed WMH is necessarily ischemic in origin (Pantoni & Garcia, 1997). Postmortem histopathological studies have demonstrated that periventricular capping is reflective of decreased myelin and edema, whereas smooth halos of periventricular WMH frequently represent ependymal lining disruption along with gliosis and myelin loss (Salthouse, 1996). However, in the absence of an autoimmune etiology, such as multiple sclerosis, irregular periventricular WMH along with punctate and confluent subcortical WMH usually denote cerebral ischemia (Fazekas et al., 1998). Finally, we conceptualized WMH as a final common pathway for a variety of disease processes and sought to explore its association with cognition irrespective of etiology. The goal of the current investigation was not to compare the patterns of WMH–cognition associations in various disease cohorts, though this would be an interesting area for future investigation.

In summary, we demonstrated that greater WMH severity is associated with decreased speed of processing, working memory, and fluency, and that greater accumulation of WMH burden over time is associated with greater cognitive decline during that same period. Our findings are compatible with the processing speed theory of cognitive aging (Salthouse, 1996), and suggest that the accumulation of WMH might partially explain normal age-related declines in processing speed. These associations primarily involve periventricular WMH, possibly owing to the concentration of long association fiber tracts this region and the putative disruption of neuronal transmission. We also found little support for the threshold hypothesis, as WMH–cognition relationships were generally linear. Finally, our results point to both age-related and age-independent effects of WMH on cognitive functioning in later life. Given that total WMH volume also represents a biomarker of small-vessel ischemic disease (Pantoni et al., 2007), addressing the causes, prevention, and treatment is an increasingly pressing public health concern (Hill & Mitchell, 2006).

Funding

Financial support for this research was provided by NIH/NIMH grant MH60504, the Therapeutic Cognitive Neuroscience Fund, and the Benjamin & Adith Miller Family Endowment on Aging, Alzheimer's and Autism Research.

Conflict of Interest

Dr. Schretlen receives royalties from the sale of the Brief Test of Attention. No other author has any potential or actual conflict of interest relevant to the subject of this manuscript.