Cardiovascular symptoms and longitudinal declines in processing speed differentially predict cerebral white matter lesions in older adults
Introduction
Cerebral white matter lesions (WML) are present in the majority of older adults and appear to play a key role in cognitive decrements, stroke, dementia, and increased mortality risk (Debette & Markus, 2010). These can be quantified as hyperintensities on Fluid Attenuated Inversion Recovery (FLAIR) magnetic resonance images (MRI). Ample evidence from several large-sample studies of community-dwelling adults implicates cardiovascular and cerebrovascular diseases as key factors in the development and progression of WML in later life, and frontal/anterior regions of the brain appear most affected at initial onset (Kennedy & Raz, 2009). With respect to changes in specific cognitive abilities, WML burden most consistently predicts declines in processing speed and executive function (Gunning-Dixon, Brickman, Cheng, & Alexopoulos, 2009; Jagust, 2013; Kloppenborg, Nederkoorn, Geerlings, & van den Berg, 2014). Thus, for example, there appears to be general consensus that cerebral small-vessel ischemia leads to WM degradation (e.g., demyelination, rarefication) within association fiber tracts and, by extension, to impaired functional connectivity across cortical networks. This process may play a key role in age-related cognitive deficits in older adults (Kennedy & Raz, 2009; Madden et al., 2012; Prins & Scheltens, 2015).
However, associations between WML burden and age-related cognitive decrements have been shown to persist even after statistical adjustment for the presence of vascular disorders and associated risk factors, such as smoking (Jagust, 2013; Kloppenborg et al., 2014). This suggests that other variables may act independently of, or interact with, vascular diseases to influence WML-cognition relations. For example, sleep duration has been shown to moderate the relation between diabetes mellitus and WML burden (Ramos et al., 2014), and depression has been implicated in both direct and indirect relations with WM degradation (Madden et al., 2012; Malloy, Correia, Stebbins, & Laidlaw, 2007). These variables may reflect specific neurochemical imbalances. Genetic predisposition may also play a role: Individuals with the E4 allele of Apolipoprotein E (APOE), a gene involved in the regulation of cholesterol transport, may be at greater risk for developing WML (Cherbuin, Leach, Christensen, & Anstey, 2007; Paternoster, Chen, & Sudlow, 2009; Raz, Yang, Dahle, & Land, 2012). Additionally, higher education and socio-economic status during youth have been linked to reduced WML prevalence in later life (Murray, McNeil, Salarirad, Whalley, & Staff, 2014), but education may also attenuate or mask the effects WM degradation on age-sensitive cognitive declines (Brickman et al., 2011; Christensen et al., 2007).
As for specific cognitive outcomes, WML-related decrements in lower-order abilities, such as processing speed, may mediate decrements in higher-order abilities, such as fluid intelligence and abstract problem-solving (Kievit et al., 2016). There is also evidence that WM degradation across brain regions is differentially associated with decrements in processing speed and in fluid abilities. For example, Marquine et al. (2010) found that total and anterior WML progression was associated with processing speed decrements, whereas posterior WML progression was linked to declines in visuo-constructional ability. More recently, MRI diffusion tensor imaging (DTI) studies have shown that WM diffusivity in select cerebral regions accounts for variability in fluid intelligence not accounted for by processing speed or age (Borghesani et al., 2013) and that multiple WM tracts are differentially associated with decrements in processing speed and in fluid intelligence (Gazes et al., 2016). However, it remains unclear whether WM deterioration mediates associations between decrements in higher order abilities, such as abstract reasoning, and declines in processing speed (Madden, Bennett, & Song, 2009).
Regional specificity in brain-cognition associations has also been observed in research using volumetric and cortical thickness measures of brain atrophy (e.g., Fjell et al., 2006; Golomb et al., 1994; Raz, Briggs, Marks, & Acker, 1999; Schretlen et al., 2000; Tisserand, Visser, van Boxtel, & Jolles, 2000). Raz et al. (2008) observed that brain volume losses across prefrontal and medio-temporal regions were differentially associated with vascular risk and fluid intelligence deficits, results they interpreted as dissociation between prefrontal and medial-temporal systems. Others have similarly suggested that different etiological profiles may underpin neurodegeneration in frontal vs. temporal regions, e.g., due to “normal” aging vs. pathological aging (Fjell, McEvoy, Holland, Dale, & Walhovd, 2014). Such differences would likely be reflected in different profiles of cognitive decline (e.g., decrements in processing speed vs. memory). Additionally, although there is some evidence to suggest that loss of brain volume may account for relations between WML progression and cognitive declines (Schmidt et al., 2005), results from other studies indicate that brain atrophy does not influence WML-cognitive associations (Borghesani et al., 2013; Marquine et al., 2010; Silbert, Nelson, Howieson, Moore, & Kaye, 2008; van den Heuvel et al., 2006).
At present, knowledge of WML-cognition relations mostly comes from cross-sectional (single time point) studies wherein WML was assessed globally and in which cognitive measures were modeled as singular outcomes (Marquine et al., 2010). Covariates, if included, have often been limited to age, sex, and educational status (Kloppenborg et al., 2014). But in a recent large-sample cohort study (with assessments at ages 73 and 76 years), Ritchie et al. (2015) examined level-change associations between multiple cognitive abilities (modeled as latent constructs) and cerebral WM volumes (both for normal appearing tissue and lesion prevalence). Changes in all cognitive abilities were negatively associated with changes in WM lesion prevalence. With respect to level-change relations, better baseline performance in all cognitive domains predicted less decline in normal appearing white matter volume, but only better baseline performance in processing speed predicted less increase in WM lesion prevalence. More recently, Ritchie et al. (2017) found that an aggregate measure of intelligence (mainly verbal ability) was modestly negatively associated with baseline level of normal appearing WM volume, but not WM lesion prevalence, when statistically adjusted for differences in eleven socio-demographic, health-related, and genetic covariates. In both of these studies, cognitive predictors were not mutually-conditioned, so the relative contribution of each cognitive domain for predicting WM health is difficult to gauge. Notwithstanding, the results suggest that specific cognitive abilities, both at baseline and longitudinally, may be differentially implicated in WM associations.
In brief, current empirical evidence most consistently frames cerebral WML burden as an intermediating condition linking vascular diseases to decrements in processing speed and executive ability. However, evidence of differential associations between distinct cognitive abilities (and levels and changes thereof) and region-specific WML burden suggests a broader scope for etiologies implicated in WML-cognition associations. These associations are likely complex, and a comprehensive analytical approach in which multiple cognitive abilities are examined concurrently with other longitudinal covariates may serve to further clarify the importance of different variables linked to WML burden.
Toward this end, in a sample of 112 older adults, we evaluated the relative strength of association of 52 variables—including socio-demographic indices, longitudinal measures of lifestyle and health, and longitudinal measures of cognitive performance (processing speed, crystallized intelligence, and fluid intelligence)—with respect to regional and total cerebral WML. We applied two different statistical methodologies for this purpose. First, we used random forest analysis (RFA), a machine learning approach, to evaluate all 52 variables as concurrently related to WML burden. RFA is an extension of regression trees, a non-parametric statistical method in which observations are recursively partitioned to identify variables most strongly associated with the outcome of interest (Aichele, Rabbitt, & Ghisletta, 2016; Breiman, 2001; Strobl, Malley, & Tutz, 2009). Unlike more common statistical approaches based on regression, RFA implicitly adjusts for all possible linear, non-linear, and higher-order interaction effects. It also provides built-in safeguards against multicollinearity and model over-fit. However, because RFA was not developed within a standard probabilistic framework, we also examined subsets of the most important predictors (as determined by RFA) using generalized linear regression (GLR), which allowed us to estimate effect sizes for the most important variables associated with WML.
In light of previous research findings, we anticipated that chronological age and symptoms of cardiovascular disease would weigh strongest in associations with WML (both within and across cerebral lobes). Additionally, we expected that processing speed decrements would be closely associated with elevated WML even after accounting for age and cardiovascular risk—and that this association would appear strongest for total WML and for frontal lobe WML. We further expected to find some evidence for an independent association between fluid intelligence and WML after accounting for differences in symptoms of cardiovascular illnesses and processing speed decrements.
Section snippets
Participants
Data for these analyses were obtained from 65 women and 47 men who were participants in the Manchester Longitudinal Study of Cognition (MLSC; see Rabbitt et al., 2004 for details). The project was approved by the Department of Psychology Internal Ethics Committee, University of Manchester; by the University of Manchester Ethics Committee, and by the Greater Manchester NIH Trust Ethics Committee. Participants’ ages ranged 50–71 years (M = 60.9) at study inception and 62–86 years (M = 73.8) at
Results
Outcomes from the random forest analysis and from the generalized linear regression analysis are presented in Table 4. Only variables included in the GLR analysis (Irel ≥ .25) are shown. Thus, regression models included between three and seven predictors. We conducted a post-hoc power analysis to approximate the probability of detecting weak, moderate, and strong effects given these different numbers of predictors as based on criteria suggested by Cohen (1992). Power to detect weak effects
Discussion
In a longitudinal study of community-dwelling middle-aged and older adults, we examined the relative strength of associations of cognitive, socio-demographic, lifestyle, and health variables with WML burden. Of these variables, we found that higher chronological age at time of MRI scan, more symptoms of cardiovascular disease, and sharper declines in processing speed were most strongly (and consistently) associated with increased WML burden—jointly accounting for approximately 49% of
Conclusions
To date, few studies have examined longitudinal measures of multiple cognitive abilities as concurrently associated with WML burden in community-dwelling older adults. And we know of no other study that has accounted for such a large range of socio-demographic, lifestyle, and health-related covariates (or that has used a sophisticated machine learning methodology) to account for these associations. In general, chronological age, symptoms of cardiovascular illness (especially hypertension), and
Funding
The authors report no conflicting financial interests pertaining to this research.
Acknowledgments
This work was supported by the Swiss National Science Foundation (grant number 100019_146535), by the Swiss National Centre of Competence in Research LIVES – Overcoming vulnerability: Life course perspectives (which is financed by the Swiss National Science Foundation, grant number: 51NF40-160590), the UK Medical Research Council, the UK Economic and Social Research Council, and the UK Wellcome Trust. The authors are grateful to these organizations for their financial assistance. We also thank
References (85)
- et al.
The association between higher order abilities, processing speed, and age are variably mediated by white matter integrity during typical aging
Neuropsychologia
(2013) - et al.
White matter hyperintensities and cognition: Testing the reserve hypothesis
Neurobiology of Aging
(2011) - et al.
What is normal aging? Effects of aging, amyloid and Alzheimer’s disease on the cerebral cortex and the hippocampus
Progress in Neurobiology
(2014) - et al.
Selective increase of cortical thickness in high-performing elderly—Structural indices of optimal cognitive aging
Neuroimage
(2006) - et al.
Mini-mental state”: A practical method for grading the cognitive state of patients for the clinician
Journal of Psychiatric Research
(1975) - et al.
White matter tract covariance patterns predict age-declining cognitive abilities
NeuroImage
(2016) Vulnerable neural systems and the borderland of brain aging and neurodegeneration
Neuron
(2013)- et al.
Pattern of normal age-related regional differences in white matter microstructure is modified by vascular risk
Brain Research
(2009) - et al.
A watershed model of individual differences in fluid intelligence
Neuropsychologia
(2016) - et al.
Diffusion tensor imaging of cerebral white matter integrity in cognitive aging
Biochimica et Biophysica Acta
(2012)