Periventricular white matter hyperintensities and functional decline

Abstract

Background/Objectives

We previously showed that global brain white matter hyperintensity volume (WMHV) was associated with accelerated long-term functional decline. We hypothesized that WMHV in particular brain regions are more predictive of functional decline.

Design

Prospective population-based study

Setting

Northern Manhattan MRI study

Participants

1195 participants free of stroke at baseline

Intervention

None

Measurements

Participants had brain MRI with axial T1, T2, and fluid attenuated inversion recovery sequences. Volumetric WMHV distribution across 14 brain regions (brainstem, cerebellum, and bilateral frontal, occipital, temporal, parietal lobes, and bilateral anterior and posterior periventricular white matter [PVWM]) was determined by combining bimodal image intensity distribution and atlas based methods. Participants had annual functional assessments with the Barthel Index (BI, range 0–100) over a mean of 7.3 years and were followed for stroke, myocardial infarction (MI), and mortality. Due to multiple collinear variables, lasso regression selected regional WMHV variables most associated with outcomes, and adjusted generalized estimating equations models estimated associations with baseline BI and change over time.

Results

Mean age was 71 (SD 9) years, 460 (39%) were male. Using lasso regularization, only right anterior PVWM met criteria for selection, and each SD greater WMHV was associated with accelerated functional decline of -0.95 additional BI points per year (95% CI −1.20, −0.70) in an unadjusted model, −0.92 points per year (95% CI −1.18, −0.67) with baseline covariate adjustment, and −0.87 points per year (95% CI −1.12, −0.62) after adjusting for incident stroke and MI.

Conclusion

In this large population-based study with long-term repeated measures of function, periventricular WMHV was particularly associated with accelerated functional decline.

Introduction

Among the elderly, functional status -- measured by performance of activities of daily living (ADL) -- is an important patient-centered outcome that estimates the population burden of different conditions. Functional status is influenced by several factors, in particular, MRI measures of subclinical cerebrovascular disease. White matter hyperintensities (WMH) in the brain are mostly caused by cerebral small vessel disease linked to traditional vascular risk factors,^{1, 2} and they have been associated with subsequent stroke^{3, 4} mortality,⁵ cognitive impairment,^{6, 7} and functional impairment⁸ -- cross-sectionally,⁹ at 3 months,¹⁰ and over 4 years of follow-up.^{8, 11} In a prior analysis in the Northern Manhattan Study (NOMAS),¹² greater WMH volume (WMHV) was associated with worse episodic memory, processing speed, and semantic memory.

Although relationships between global WMHV and functional outcomes have been demonstrated, it is conceivable that WMHV in particular brain regions may explain a greater proportion of variance in outcomes. Just as a neurologist may localize symptoms and signs to injury in particular brain regions, it is conceivable that a greater burden of WMHV involving particular structures or lobes, even if clinically asymptomatic, may cause specific deficits that affect functional status. Rather than hypothesizing relationships based on clinical knowledge of localized brain function, data-driven statistical techniques in a large population of individuals may reveal previously unknown relationships. Here, we pursue both approaches to provide data about the relationship between regional WMHV and functional outcomes, which is poorly understood to date.

There is also limited data regarding the time course of change in functional trajectories in relation to WMH. Although associations have been found with functional outcomes at a single follow-up time, the effect of WMHV on slope of change in function over time is not clear. We previously showed that overall brain WMHV was associated with accelerated long-term functional decline, independently of stroke and MI.¹³ However, it was unclear if WMHV in particular brain regions is more predictive of decline. We hypothesize that WMHV in particular brain regions, selected by data-driven analysis methods, is more predictive of functional decline compared to other regions, independently of baseline confounders and vascular events occurring during follow-up. We used both traditional and novel variable selection techniques, and analysis of repeated measures of functional status, in a large population-based imaging study.

Methods

The NOMAS MRI study is a substudy of the NOMAS prospective cohort (as previously described¹⁴) that began in 2003 and included 1290 individuals: 1) ≥age 50 years, 2) without contraindications to MRI, 3) without clinical stroke and 4) providing signed informed consent. Imaging was performed once on a 1.5T MRI system (Philips Medical Systems, Best, Netherlands), and included axial T1, axial T2, and Fluid Attenuated Inversion Recovery (FLAIR) sequences.

We developed tailored protocols using tools from the FSL software package (http://www.fmrib.ox.ac.uk/fsl) to automatically measure total WMHV and regional distributions over 14 brain regions (brainstem, cerebellum, and bilateral frontal, occipital, temporal, and parietal lobes, and bilateral anterior and posterior periventricular white matter [PVWM], defined as within 1 cm of the lateral ventricles¹⁵). The skull region was removed from images using the FSL-BET tool,¹⁶ and two-class segmentation of the whole brain was performed, into cerebrospinal fluid and brain tissue using the FSL-FAST segmentation algorithm after correction for image non-uniformity.¹⁷ The mean (μ) and standard deviation (σ) of intensity values of the brain tissue voxels were derived, and WMH voxels were defined with intensity >(μ+3.5*σ). Finally, regional WMHV were calculated using the MNI structural atlas as a reference by nonlinearly registering the subject images to the atlas template and mapping the regions delineated on the atlas to the subject images using FSL-FNIRT tool.¹⁸ Total WMHV was the sum of volumes in each of the 14 sub-regions. The reliability of the WMH volumetry method was previously established using repeated measurements of the same subjects.¹⁹ Total cranial volume (TCV) constituted the sum of whole brain volume voxels from the T1 segmentation process.

Eighty-five participants were excluded due to: lack of FLAIR data, image artifacts including motion artifact, failure of the registration method due to very large ventricles or lesions/tumors, and lesions that appeared hyperintense outside the white matter. Columbia University and University of Miami IRBs approved the study.

Baseline Evaluation

Bilingual research assistants collected interview data using standardized questions regarding the following conditions, as previously described: hypertension, diabetes mellitus, hypercholesterolemia, cigarette smoking, alcohol use, and cardiac conditions.²⁰ All participants underwent a thorough baseline examination including comprehensive medical history, physical examination, review of medical records, functional status assessed by the Barthel index (BI), and fasting blood samples.

Follow-up

Participants were followed annually by phone to assess death, new neurological or cardiac symptoms and events, interval hospitalizations, and functional status via the BI. Only two subjects were lost to follow-up after their baseline examination, and the average annual contact rate was 99%.

A positive screen for any potential cardiac or neurological event was followed by an in-person assessment to determine whether a vascular outcome occurred. In addition, all admissions and discharges of NOMAS study participants to Columbia University Medical Center (CUMC) were screened for possible outcome events that may not have been captured by telephone interview. Nearly 70% of hospitalized vascular events in the cohort are at CUMC. Hospital records were reviewed to classify outcomes as previously reported.²¹ Stroke included ischemic stroke, intracerebral hemorrhage, and subarachnoid hemorrhage. At least two stroke neurologists verified and classified all stroke cases. Myocardial infarction required at least two of the following criteria: (a) ischemic cardiac pain determined to be typical angina; (b) cardiac marker abnormalities defined as abnormal CK-MB fraction or troponin I values; and (c) ischemic EKG abnormalities. Cardiologists adjudicated MI diagnoses. Through the end of 2014, there were 53 first definite and probable MI occurring during follow-up, and 64 first strokes (59 infarcts, 3 intracerebral hemorrhages, and 2 subarachnoid hemorrhages).

Study outcome

The BI^{22, 23} measures 10 activities of daily living (ADLs) and ranges from 0–100 in 5-point increments, with 100 indicating normal functioning. Previous research has demonstrated the reliability of phone assessments of function using the BI.²⁴ Although it is an ordinal scale, the scale may be analyzed as a continuous variable for increased power to detect associations, ability to describe the course of change over time in linear form, and avoidance of potential misclassification due to crude categorization.^25–27 Only BI measurements from the time of MRI onwards were included in this analysis.

Covariates

Analytic models were adjusted for the following variables: demographic variables (age, sex, race-ethnicity), medical risk factors (body mass index [body weight in kilograms divided by the square of height in meters], hypercholesterolemia [defined by self-report, lipid lowering therapy use, or fasting total cholesterol level >240 mg/dL], diabetes mellitus [defined by self-report, fasting blood glucose level ≥126 mg/dL, or insulin/oral hypoglycemic use], hypertension [defined as a systolic blood pressure recording ≥140 mmHg or a diastolic blood pressure recording ≥90 mm Hg based on the average of two blood pressure measurements or the participant’s self-report of a history of hypertension or antihypertensive use]), smoking (defined as either nonsmoker or smoker within the last year), alcohol use (with moderate alcohol use classified as 1 drink/month to 2 drinks/day), any physical activity (versus none, as previously defined²⁸), social variables (marital status, insurance status [classified uninsured/Medicaid versus Medicare/private insurance], and number of friends [individuals whom the participant knows well enough to visit in their homes]), and cognitive/mood factors (depressed mood within the prior week and performance on mini-mental state examination [analyzed as a continuous variable]).

Statistical analysis

We sought to determine whether regional WMHV was associated with baseline BI and a steeper slope of decline over time. Since a change in the baseline or intercept of the estimated curve shifts the entire curve, such a change can be considered a change in overall or mean function, but we will refer to this kind of change as change in baseline function. We calculated distributions of WMHV, TCV, baseline covariates, and BI. We calculated Pearson correlation coefficients among all regional WMHV measurements to determine degree of collinearity.

Using linear regression, we tested for associations between each regional WMHV measurement and 5-year BI, defined as BI measured within 6 months before to 6 months after the 5-year point from baseline, in unadjusted models as well as adjusted for demographics, medical risk factors, smoking and alcohol use and physical activity, social variables, and cognitive/mood variables, as defined above.

Due to correlations among repeated measures of BI in the same individual, regression models using generalized estimating equations (GEE) with an identity link function were used to assess the association between each regional WMHV variable and repeated measurements of BI, in unadjusted models as well as adjusting sequentially for: baseline demographic variables, medical risk factors, smoking and alcohol use and physical activity, social variables, and cognitive/mood factors, as defined above. To assess whether each regional MRI variable was associated with change in BI over time (slope), we included interaction terms between time of follow-up assessment and the main predictor variable. Various model diagnostics including tests of linearity, residual plots, and goodness of fit measures were used to evaluate the final model. There was no evidence to suggest lack of linearity of BI trajectories in the final models. We chose the exchangeable (intraclass) GEE correlation structure with robust standard errors.

Next, all regional WMHV variables were included in the same model. Due to significant collinearity among regional MRI variables, Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to select regional WMHV variables for analysis among the 14 regional measurements. Lasso is a penalized regression technique that incorporates both parameter shrinkage and feature selection.²⁹ It is useful for optimizing the selection of variables among a group of collinear variables, and for identifying those variables, among multiple collinear variables, that best explain variation in the outcome. L1 norm regularization and plots of mean squared error versus log of lambda were used to select regional MRI variables. The number of variables was determined by the nadir of the mean squared error plot, and the specific variables that were selected were identified by examining L1 norm plots. Unadjusted and adjusted GEE models were run as above, testing associations between regional WMHV variables selected in the lasso procedure and repeated measurements of BI.

To assess whether interval vascular events such as clinical stroke and MI were implicated in the trajectory of functional status, we ran a second set of models in which stroke and MI were included as time-varying covariates. We tested whether the relationship between regional WMHV measurements and functional status remained even after adjusting for these events.

Results

Table 1 shows distributions of NOMAS participant baseline variables. Mean follow-up time was 7.3 years (SD 2.1). Of 1136 participants with BI of 95 or 100 at baseline and >1 follow-up, 690 (61.7%) experienced a decline in BI. Mean WMHV (as %TCV) by region are reported in Table 1, and the majority of WMHV burden involved the anterior and posterior periventricular regions. Overall mean WMHV was similar by side.

Table 1.

Baseline characteristics of the cohort

Characteristic	Frequency
Number of participants, No. (%)	1290 (36.9)
Age, mean (SD), y	64.5 (8.4)
Body mass index, mean (SD), kg/m2	28.0 (4.8)
Male, No. (%)	510 (39.5)
Race-ethnicity:
Non-Hispanic white, No. (%)	191 (14.8)
Non-Hispanic black, No. (%)	223 (17.3)
Hispanic, No. (%)	847 (65.7)
Other, No. (%)	29 (2.3)
Received at least high school education, No. (%)	592 (45.9)
Marital status, No. (%) married	543 (42.1)
Health insurance, No. (%)
Medicaid or no insurance	613 (47.5)
Medicare or private insurance	677 (52.5)
Hypertension	861 (66.7)
Physical activity:
None	564 (44.3)
Any	710 (55.7)
Diabetes mellitus	245 (19.0)
Smoking:
Never	612 (47.4)
Former	496 (38.5)
Current	182 (14.1)
Hypercholesterolemia	797 (61.8)
History of coronary heart disease	177 (13.7)
Hamilton depression scale score, mean (SD)	3.1 (3.8)
Mini mental state score, mean (SD)	26.7 (3.3)
Number of people known well enough to visit with in their homes:
None	36 (2.8)
1 or 2	124 (9.6)
3 or 4	263 (20.4)
5 or more	867 (67.2)
Regional white matter hyperintensity volume, as % of total cranial volume, mean (SD)
Brainstem	0.027 (0.021)
Cerebellum	0.019 (0.019)
Frontal lobe, left	0.051 (0.064)
Occipital lobe, left	0.005 (0.006)
Parietal lobe, left	0.018 (0.026)
Temporal lobe, left	0.017 (0.014)
Anterior periventricular, left	0.112 (0.154)
Posterior periventricular, left	0.150 (0.210)
Frontal lobe, right	0.058 (0.069)
Occipital lobe, right	0.005 (0.008)
Parietal lobe, right	0.023 (0.031)
Temporal lobe, right	0.017 (0.028)
Anterior periventricular, right	0.109 (0.150)
Posterior periventricular, right	0.154 (0.205)

Seventy of 91 pairwise correlations among regional WMHV measurements were significant, demonstrating widespread collinearity. Hence, we first report associations between each regional WMHV measure and functional outcomes, in unadjusted and adjusted models. Table 2 shows associations between each regional measure and 5-year functional status. There were significant associations between greater regional WMHV and lower 5-year BI score for all variables in unadjusted models except brainstem and cerebellum locations. In adjusted models, magnitudes of associations were lower compared to unadjusted models, but most associations remained significant. In particular, the greatest magnitudes were seen for anterior (−3.06 BI points per SD increase for left location, −3.09 for right) and posterior (−2.31 for left, −2.68 for right) periventricular white matter regions.

Table 2.

Unadjusted and adjusted associations between regional white matter hyperintensity volumes and 5-year functional status

	Unadjusted models*			Adjusted models**
Brain region	Difference in BI per SD increase in WMHV	95% CI	p-value	Difference in BI per SD increase in WMHV	95% CI	p-value
Brainstem	0.27	−0.61, 1.16	0.5	0.20	−0.70, 1.09	0.7
Cerebellum	0.06	−0.75, 0.88	0.9	0.30	−0.52, 1.12	0.5
Frontal lobe, left	−1.65	−2.46, −0.84	<.0001	−0.71	−1.55, 0.13	0.099
Occipital lobe, left	−1.93	−2.77, −1.09	<.0001	−1.51	−2.46, −0.56	0.002
Parietal lobe, left	−2.68	−3.50, −1.86	<.0001	−1.67	−2.55, −0.78	0.0002
Temporal lobe, left	−0.79	−1.64, 0.05	0.07	−0.59	−1.43, 0.25	0.17
Anterior periventricular white matter, left	−4.91	−5.71, −4.10	<.0001	−3.06	−4.06, −2.06	<.0001
Posterior periventricular white matter, left	−3.95	−4.75, −3.15	<.0001	−2.31	−3.23, −1.39	<.0001
Frontal lobe, right	−2.61	−3.43, −1.79	<.0001	−1.42	−2.31, −0.53	0.002
Occipital lobe, right	−1.38	−2.47, −0.28	0.014	−0.80	−2.00, 0.40	0.2
Parietal lobe, right	−2.96	−3.78, −2.14	<.0001	−2.02	−2.91, −1.13	<.0001
Temporal lobe, right	−1.97	−3.44, −0.50	0.009	−1.40	−2.87, 0.06	0.06
Anterior periventricular white matter, right	−4.94	−5.75, −4.13	<.0001	−3.09	−4.09, −2.09	<.0001
Posterior periventricular white matter, right	−4.23	−5.04, −3.41	<.0001	−2.68	−3.62, −1.74	<.0001

^*Each row represents output from a separate model; regional white matter hyperintensity volume measurements are not mutually adjusted for; BI=Barthel index score; SD=standard deviation; WMHV=white matter hyperintensity volume; CI=confidence interval

^**models are adjusted for: age at time of MRI, sex, race-ethnicity, diabetes, hypertension, coronary artery disease, hypercholesterolemia, physical activity, alcohol use, smoking, body mass index, marital status, insurance status, number of friends, and mini-mental state score

Next, we analyzed associations using all repeated functional measurements with adjusted GEE models (Table 3), which estimated associations with baseline BI and change in BI over time, examining each region in a separate model. Greater WMHV in the brainstem and left posterior periventricular WM were associated with higher baseline BI, and greater WMHV in the right occipital and temporal lobes was associated with lower baseline BI. Greater WMHV in several regions was associated with steeper functional decline: bilateral frontal lobes, bilateral parietal lobes, left temporal lobe, and bilateral anterior and posterior periventricular white matter.

Table 3.

Adjusted associations between regional white matter hyperintensity volumes and trajectories of functional status^*

	Associations with baseline BI			Associations with change in BI over time
Brain region	Difference in baseline BI per SD increase in WMHV	95% CI	p-value	Annual change in BI per SD increase in WMHV	95% CI	p-value
Brainstem	1.02	0.46, 1.57	0.0003	−0.17	−0.34, 0.01	0.062
Cerebellum	0.60	−0.03, 1.23	0.06	−0.05	−0.16, 0.07	0.4
Frontal lobe, left	0.40	−0.58, 1.38	0.4	−0.29	−0.56, −0.02	0.03
Occipital lobe, left	−0.90	−2.07, 0.26	0.13	−0.13	−0.41, 0.16	0.4
Parietal lobe, left	1.09	−0.02, 2.20	0.053	−0.62	−0.90, −0.33	<0.0001
Temporal lobe, left	0.51	−0.30, 1.31	0.2	−0.24	−0.46, −0.02	0.03
Anterior periventricular white matter, left	0.33	−0.85, 1.52	0.6	−0.74	−0.99, −0.49	<0.0001
Posterior periventricular white matter, left	1.09	0.13, 2.04	0.03	−0.73	−1.00, −0.45	<0.0001
Frontal lobe, right	0.08	−1.15, 1.30	0.9	−0.41	−0.69, −0.13	0.005
Occipital lobe, right	−1.77	−3.18, −0.36	0.014	−0.13	−0.43, 0.16	0.4
Parietal lobe, right	−0.43	−1.98, 1.13	0.6	−0.45	−0.73, −0.16	0.002
Temporal lobe, right	−1.44	−2.20, −0.68	0.0002	−0.18	−0.52, 0.15	0.3
Anterior periventricular white matter, right	0.63	−0.54, 1.80	0.3	−0.83	−1.08, −0.57	<0.0001
Posterior periventricular white matter, right	0.47	−0.69, 1.63	0.4	−0.69	−0.95, −0.44	<0.0001

^*Each row represents output from a separate model; regional white matter hyperintensity volume measurements are not mutually adjusted for; BI=Barthel index score; SD=standard deviation; WMHV=white matter hyperintensity volume; CI=confidence interval; models are adjusted for: age at time of MRI, sex, race-ethnicity, diabetes, hypertension, coronary artery disease, hypercholesterolemia, physical activity, alcohol use, smoking, body mass index, marital status, insurance status, number of friends, and mini-mental state score

Next, using lasso regularization, only right anterior PVWM emerged as optimal for selection, and each SD increase was associated with accelerated functional decline (Figure), of −0.95 additional BI points per year (95% CI −1.20, −0.70) in an unadjusted model, −0.92 points per year (95% CI −1.18, −0.67) with baseline covariate adjustment, and −0.87 points per year (95% CI −1.12, −0.62) after adjusting for stroke and MI. This decline was in addition to a mean decline of −1.13 (95% CI −1.29, −0.97), −1.19 (95% CI −1.36, −1.01), and −1.04 (95% CI −1.21, −0.88) BI points per year, respectively.

Figure. Unadjusted and adjusted models of associations between right anterior periventricular white matter hyperintensity volume and functional trajectories.

BI=Barthel index; CI=confidence interval; SD=standard deviation; WMHV=white matter hyperintensity volume; MI=myocardial infarction

Full baseline adjustment model is adjusted for age at time of MRI, sex, race-ethnicity, diabetes, hypertension, coronary artery disease, hypercholesterolemia, physical activity, alcohol use, smoking, body mass index at the time of MRI, marital status, insurance, number of friends, mini-mental state score.

Model adjusted for stroke and MI during follow-up is additionally adjusted for stroke and MI occurring during follow-up, as time-varying covariates

Discussion

We found that greater WMHV in multiple brain regions was associated with accelerated long-term functional decline, independently of confounders and stroke and MI occurring during follow-up. In particular, greater periventricular WMHV was associated with lower 5-year functional status and steeper annual functional decline in separate adjusted models. Due to widespread collinearity among regional WMHV measurements, we used data-driven variable selection methods to select regions that were optimally explanatory of functional trajectories. The right anterior periventricular location was selected as the only region that optimally explained the data, and each SD increase in WMHV in this region was associated with an almost twofold steeper functional decline.

Prior studies have shown a predominance of WMHV in the periventricular region, and high degrees of correlation among regional measures, as in our study.^30–32 Several studies among those with cognitive impairment have shown associations between regional WMHV and outcomes. Among 347 women, the majority with Alzheimer’s dementia, a cross-sectional analysis showed that frontal WMH were associated with greater disability, but no longitudinal analysis was performed, there was no control for multicollinearity of regional WMH measures, and WMHV was measured in only 4 brain regions.³³ In a similar cross-sectional analysis among 163 individuals with cognitive impairment or dementia, greater WMH in multiple brain regions – including periventricular -- was associated with falls, but there were similar limitations.³⁴ Periventricular WMH were associated with worse cognitive performance in another cross-sectional analysis comparing 60 individuals with cognitive impairment to 24 controls, but there was no control for multicollinearity among regional WMH measurements or variable selection used.³⁵ In a small case-control study,³⁶ WMH in multiple brain regions was associated with disability and slower gait speed, but there was no attempt to adjust for all regional WMH measures in a single model.

Several studies examined regional WMH measures among the elderly. Among 99 elderly community-dwelling individuals, greater WMH burden in the splenium of the corpus callosum was associated cross-sectionally with lower mobility and gait speed.³¹ However, variable selection techniques were not used. Among 287 individuals aged 70–90 years followed up to one year,³² deep but not periventricular WMH were associated with physical decline, but variables were selected by stepwise regression, which is not an optimal variable selection procedure for correlated covariates. Among 639 elderly individuals, periventricular and deep WMH locations were cross-sectionally associated with falls.³⁷ After acute ischemic stroke (n=187), periventricular WMH, assessed with a clinical rating scale, were associated with worse 30-day functional outcomes.³⁸ Common limitations of these prior studies included limited sample sizes, cross-sectional and not longitudinal analysis of trajectories, limited extent of regional WMHV measurements, and analysis of selected samples and not a population-based sample.

Our study is the largest longitudinal study to date with detailed regional WMH measures and analysis of trajectories of functional status. In contrast to previous studies, our data show that periventricular WMH may be most informative of future functional decline. Our data, moreover, show that this decline continues for at least 5 years after assessment of subclinical cerebrovascular disease. The current analysis expands on prior analyses in this cohort, where we found that greater total WMHV was associated with accelerated long-term functional decline, independently of confounders and stroke during follow-up.¹³

WMH in the brain are thought to be caused by ischemic processes. Small clinically silent white matter infarcts have been observed to develop into regions that appeared radiographically the same as WMH.³⁹ Many studies have shown a predominance of WMHV in periventricular regions,³⁵ areas of the brain that are fed by distal pial end arterioles that form an arterial border zone.⁴⁰ This area is particularly vulnerable to ischemia and reductions of cerebral blood flow.⁴¹ Other presumed mechanisms of WMH development include blood brain barrier disruption, inflammatory processes, ependymal loss, and venular dysfunction.⁴² Resulting WMH, predominantly periventricular, may disrupt cholinergic pathways involving the cortex, leading to dysfunction in executive, memory, and visuospatial domains that negatively impact functional status.³⁵

We performed data-driven variable selection techniques because of extensive multicollinearity among regional WMH variables, which make traditional stepwise regression, and multivariable regression, problematic. When we used analysis of each regional WMH measurement in separate models, as well as lasso regression, the periventricular locations emerged as having significant and large associations with functional decline. Although these associations do not confirm a causal relationship between specific brain region and functional trajectories, periventricular WMH may disrupt motor pathways that control gait, negatively impacting the mobility aspects of functional status.³⁴ Also, disruption of pathways involved in frontal and parietal association cortices may impact the cognitive and visuospatial aspects of functional status.

Strengths of this study include the large population-based cohort, accurate assessment of events during follow-up, minimal loss to follow-up, the use of state-of-the-art imaging and regional measurement of WMHV, and repeated measures of functional outcomes that allow trajectory analysis. A limitation is that NOMAS participants enrolled in the MRI cohort were able to return for follow up and undergo MR imaging, reflecting a healthy survivor bias, which may have reduced power to detect declines in functional status. Although we found no evidence of an effect of race-ethnicity, there is a potential for limited generalizability of the results considering the large proportion of Hispanics in this cohort. Also, we did not measure arthritis and could not adjust for this, although it is unlikely to be a confounder of relationship between regional WMHV and functional trajectories.

In conclusion, we found that greater WMHV in multiple brain regions was independently associated with accelerated long-term functional decline, particularly the periventricular regions. This has implications for prognostication, since a focus on periventricular WMH may be sufficient to be able to predict functional trajectories. Also, these findings open areas of future study focusing on periventricular WMH and the reasons that lesions in this anatomical location would preferentially affect functional status.