Abstract
Background:
The relationship between cerebral small vessel disease (SVD) and dementia has been studied without considering white matter (WM) volume, the microstructural integrity of the WM surrounding the SVD, and grey matter (GM).
Objective:
We prospectively investigated the relationship between these structures and the risk of dementia, and formed a prediction model to investigate which characteristics (macro- or microstructural) explained most of the variance.
Methods:
The RUN DMC study is a prospective cohort study among 503 non-demented participants with an age between 50 and 85 years at baseline, with baseline assessment in 2006 and follow-up assessment in 2012. Two were lost to follow-up (yielding a 99.6% response-rate). Cox regression analysis was used, to calculate hazard ratios for dementia, of baseline MRI characteristics. Tract-Based Spatial Statistics (TBSS) analysis was used to assess the added value of microstructural integrity of the WM.
Results:
Mean age at baseline was 65.6 years (SD 8.8) and 56.8% was male. 43 participants developed dementia (8.6% ), resulting in a 5.5-year cumulative risk of 11.1% (95% CI 7.7–14.6). Low WM and hippocampal volume are significant predictors for dementia. WM, WM hyperintensities, and hippocampal volume explained most of the variance. TBSS analyses showed no additional value of diffusion parameters.
Conclusions:
WM and hippocampal volume were the main predictors for the development of incident dementia at 5-year follow-up in elderly with SVD. There was no additional diagnostic value of the diffusion tensor imaging parameters on top of the macrostructural characteristics.
Keywords
INTRODUCTION
White matter hyperintensities (WMH) and lacunes occur in over 90% of all individuals aged 60 years and over. Expressions of cerebral small vessel disease (SVD) [1], including WMH [2], lacunes [3], and microbleeds [4] have always been studied in isolation, rather than in conjunction with respect to incident dementia, often without taking the surrounding (volume of) normal appearing white matter (NAWM) into account. The approach is usually similar for grey matter (GM) structures that have found to be related with dementia, such as neocortical or hippocampal volume (HV) [5]. Obviously, the white matter (WM) connects and thereby influences the structure and function of these GM areas [6]. Given this, it seems plausible to also take these GM structures into consideration when assessing the relation between SVD and cognitive performance or dementia and vice versa.
Apart from the with conventional magnetic resonance imaging (MRI) visible SVD, the microstructural integrity of the WMH itself and that of the surrounding NAWM, which can be assessed with diffusion tensor imaging (DTI) [7, 8] has suggested to play a role in dementia in cross sectional studies[9]. However, prospective studies are lacking.
We therefore investigated the relationship between the total spectrum of SVD, the GM (neocortex, basal ganglia, thalamus), the hippocampus, and the microstructural integrity of the WM altogether on the risk of dementia. With these macro- and microstructural parameters, we constructed a prediction model to indicate which MRI-parameter accounts for the highest variance. We did so within the RUN DMC study, a prospective cohort study among 503 individuals with SVD after a mean follow-up of over 5 years.
MATERIALS AND METHODS
Study population
The Radboud University Nijmegen Diffusion Tensor and Magnetic resonance Cohort (RUN DMC) study prospectively investigates risk factors and clinical consequences of brain changes during aging as assessed with MRI among 503 50–85 year old non-demented elderly with cerebral SVD. On the basis of established research criteria, SVD was defined as the presence of lacunes and/or WMH on neuroimaging [10]. Patients were referred either because of acute symptoms, such as transient ischemic attack (TIA) or lacunar syndromes, or subacute complaints such as cognitive, motor disturbances, and/or depressive symptoms [10]. The baseline data collection was performed in 2006. Participants who underwent routine diagnostic brain imaging, (for among others: vascular causes (TIA, stroke), headache, mild traumatic brain injury, and cognitive complaints) were eligible for participation. Inclusion criteria were: age between 50- and 85 and cerebral SVD on neuroimaging (WMH and/or lacunes). Main exclusion criteria were dementia; (psychiatric) disease interfering with cognitive testing or follow-up; WMH or SVD mimics (e.g., multiple sclerosis), and MRI contraindications [11].
Follow-up was completed in 2012. Of 503 baseline participants, two were lost to follow-up (but not deceased according to the Dutch Municipal Personal Records database), 49 had died, and 54 refused in person follow-up, but clinical endpoints were available (Fig. 1). One was additionally excluded because of baseline MRI artifacts, yielding a final sample of 500 participants. This study was approved by the Medical Review Ethics Committee region Arnhem-Nijmegen and all participants gave written informed consent prior to inclusion.
Dementia case finding
At baseline all participants were free of dementia. Dementia case finding was performed for those who participated in the face-to-face follow-up and for those who died or refused follow-up participation (they gave permission to perform follow-up based on available clinical data). The Mini–Mental State Examination (MMSE) [12] was used as a first screen among face-to-face participants. A score below 26 or a decline of three points or more from baseline was considered screen positive (n = 34), of whom 20 were subsequentlyexamined for dementia at the Radboud Alzheimer Center (7 were diagnosed with dementia, and 13 were not). For the remaining 14, who refused additional analysis, a consensus diagnosis of dementia was made by a panel, consisting of a neurologist, clinical neuropsychologist and a geriatrician with expertise in dementia. They reviewed all available neuropsychological [11] and imaging information, which included: (I) the difference in neuropsychological performance between baseline and follow-up; (II) outcome of the Mini International Neuropsychiatric Interview MINI [13]; (III) the follow-up MRI scan, or, if not available (in 7 cases), the baseline MRI-scan; and (IV) for the interpretation of these tests, age and level of education were taken into account [14], next to interference with daily living, confirmed by family or caregivers. Of these 14 participants, seven were diagnosed with dementia.
Medical records were reviewed from the participants who were not available for follow-up assessment (49 deceased, 54 follow-up data available, no center visit); in addition their general practitioners and medical specialists were contacted for information on their cognitive status. Dementia was mentioned in 37 participants. After review by members from the panel, 29 were diagnosed with dementia, and eight were not. In total, this resulted in 43 incident cases of dementia during a mean follow-up period of 5.2 (SD 0.7) years.
The diagnosis of dementia was based on the Diagnostic and Statistical Manual of Mental Disorders (IV) [15] criteria; probable Alzheimer’s disease was based on the NIA-AA criteria [16], and vascular dementia was based on NINDS-AIREN criteria [17]. Individuals not fulfilling these criteria were classified as possible Alzheimer’s Dementia with etiologically mixed presentation [16] and frontotemporal dementia. The onset of dementia was defined as the date on which the clinical symptoms allowed for the diagnosis [3]. When the date of diagnosis was not exactly known, we used the mid-point between the baseline visit and the first date the diagnosis was confirmed [18], or the date someone was placed in a nursing home because of dementia.
MRI resonance imaging protocol and analysis
MRI scans of all participants were acquired on a single 1.5-Tesla MRI (Magnetom Sonata, SiemensMedical Solutions, Erlangen, Germany). The protocol included the following whole brain scans: a T1-weighted 3D magnetization-prepared rapid gradient-echo (MPRAGE) imaging (voxel size 1.0 × 1.0 × 1.0 mm); Fluid-attenuated inversion recovery (FLAIR); voxel size 1.0 × 1.2 × 5.0 mm, with an interslice gap of 1 mm); a transversal T2** weighted gradient echo s (voxel size 1.3 × 1.0 × 6.0 mm, with an interslice gap of 1 mm) and a DTI sequence (voxel size 2.5 × 2.5 × 2.5 mm; 4 unweighted scans, 30 diffusion weighted scans with b-value = 900 s·mm–²). WMH were manually segmented on FLAIR images and the total WMH volume was calculated by summing the segmented areas multiplied by slice thickness. The rating of baseline lacunes and microbleeds were rated according to the recently published Standards for Reporting Vascular changes on neuro-imaging (STRIVE), by trained raters blinded to all clinical data [1] with good intra and inter-rater variability (weighted kappas of 0.87 and 0.95 for lacunes and 0.85 and 0.86 for microbleeds, calculated in 10% of the scans). To obtain GM (volume of the neocortex, basal ganglia, and thalamus), WM and cerebrospinal spinal fluid (CSF) volume, the T1 MPRAGE images were segmented using Statistical Parametric Mapping 5 unified segmentation routines [11, 19]. Total GM, WM, and CSF volumes were subsequently calculated by summing all voxel volumes that had a p > 0.5 for belonging to that tissue class. The intracranial volume (ICV) was a summation of total GM, total WM, and CSF volume. HV were manually segmented on the MPRAGE image using the interactive software program “ITK-SNAP” as described previously [20, 21]. (http://www.itksnap.org). All volumes were normalized to total ICV [22].
DTI analysis
Diffusion data were analyzed in the total WM, according to an earlier detailed procedure [23]. In short, diffusion data were pre-processed using an in-house developed algorithm for patching artifacts from cardiac and head motion [24]. Corrections of Eddy current and motion artifacts from affine misalignment were performed simultaneously, which was based on minimization of the residual diffusion tensor error [25]. Fractional anisotropy (FA) and mean diffusivity (MD) images were calculated using DTIFit within the FSL toolbox, which were fed into the TBSS pipeline [26]. A FA skeleton was created by conducting a thinning procedure on the mean FA image. This skeleton was thresholded at 0.3 and the skeleton projection vectors were then applied to the MD.
Other measurements
Education was classified using 7 categories (1 being less than primary school, 7 reflecting academic degree) [14]. We then dichotomized in primary or less (level 1 and 2), or more than primary education (level 3 to 7). Depressive symptoms were assessed with the Centre of Epidemiologic Studies Depression Scale (CES-D); they were considered present with CES-D ≥16 and/or current use of anti-depressive medication, taken for depression [11, 27].
Statistical analysis
Person-years at risk were calculated from date of the baseline assessment, until onset of dementia, death, or date of the follow-up whichever came first. Patients who died or did not reach the endpoint were censored. WMH volume was log transformed because of the skewed distribution of the data. As frontotemporal dementia is basically genetic, we excluded this participant from the analysis.
The cumulative risk of incident dementia was estimated with Kaplan-Meier analysis, stratified by severity (in quartiles) of SVD characteristics (WM and WMH volume, lacunes, and microbleeds), GM characteristics, and MD within WMH and NAWM. Subsequently, Kaplan-Meier curves were compared between the subgroups using a log-rank test.
Cox regression analysis was used, to calculate hazard ratios for dementia. Three models were constructed to predict dementia. First, age, gender, education, and baseline MMSE were forced into the model and a backward stepwise selection procedure was used to enter imaging characteristics into the model until all variables in the model had p values smaller than 0.15. Among the imaging characteristics, only SVD characteristics were considered in model 1, GM volumes were added in model 2, and DTI characteristics were added in model 3. C-statistic was used to assess the discriminatory performance of the three models. Two-sided p-values of less than 0.05 were considered to indicate statistical significance. For the TBSS analyses, we assessed voxel-wise correlations between the skeletal DTI parameters (FA and MD) and dementia, adjusting for age, gender, education, brain volume, MMSE, and SVD characteristics. Statistical analysis were performed using IBM SPSS Statistics version 20 and R version 2.15 (http://www.R-project.org) software packages.
RESULTS
Baseline characteristics are shown in Table 1. 43 participants developed dementia during a mean follow-up of 5.2 years (SD 0.7) resulting in a 5.5-year cumulative risk of dementia of 11.1% (95% CI 7.7–14.6).
The risk of dementia was highest in participants with the highest WMH volume compared with the lowest volume (14.4% versus 0.8% , Log-Rank p < 0.001), whereas this risk did not differ by absence or presence of lacunes or microbleeds at baseline. Lower baseline WM, GM, and HV were associated with a higher 5-year risk of dementia (risks for lower versus upper quartile of the volume 13.9% versus 0.8% , log-rank p < 0.001 for WM, 24.5% versus 4.0% Log-rank p < 0.001 for GM, and 21.2% versus 5.8% , log-rank p = 0.003 for HV). Participants with the lowest quartile of microstructural integrity (both within WMH and NAWM) had a higher risk for dementia than participants in the highest quartile of structural integrity (Fig. 2).
Low WM, CSF, and HV at baseline were significant predictors of incident dementia, with adjustment for age, gender, education, baseline MMSE and territorial infarcts (Table 2).
After additional adjustment for hippocampal and GM volume, WM volume remained a significant predictor of incident dementia (HR 0.65 95% CI 0.44–0.96, p = 0.034). After subsequent adjustment for all SVD characteristics separately, a lower HV also remained a risk factor of dementia (HR 0.68, 95% CI 0.47–0.99, p = 0.045), but CSF volume did not. After additional adjustment for baseline depressivesymptoms the strength of the associations did not markedly change.
Finally, in the prediction models of incident dementia, model 1 showed that WM volume, WMH volume, and microbleeds retained in the model. After adding the volumes of GM structures (GM and HV) in model 2, HV was also retained in the model. Adding DTI characteristics in model 3 did not alter the predictive characteristics (Table 3). There were no indications that the proportional hazards assumption was violated. The performance of the three models was comparable for the three models (model 1: C-statistic 0.84; CI 0.74–0.93, model 2: C-statistic 0.85; CI 0.76–0.95) and for model 3: C-statistic 0.86; CI 0.76–0.95). TBSS analysis revealed no significant differences between participants with and without incident dementia, for both FA and MD parameters (data not shown).
DISCUSSION
Lower WM and HV at baseline significantly increased the 5-year risk of dementia in individuals with cerebral SVD, but WMH, lacunes, and microbleeds did not. The degree of microstructural integrity of the WM did not affect this risk. To the best of our knowledge, this is the first longitudinal study that assessed the relationship between the whole spectrum of SVD, GM structures, and the microstructural integrity altogether and the risk of dementia in elderly with SVD.
Other studies on dementia or cognitive performance were cross-sectional, did not include both GM and WM volume, did not include the whole SVD spectrum, did not use TBSS analysis, or did not take dementia as a clinical endpoint [28–31].
Strengths of our study include its longitudinal assessment of a population that covers the whole spectrum of cerebral SVD and the large sample size, and the ascertainment of endpoints in 99.6% of our participants. Collection of our data in a single center allowed us to assemble baseline and follow-up data according to identical procedures, reducing the risk of procedural bias. Furthermore, we manually segmented the WMH and HV without prior knowledge of the clinical data. The relationship between brain imaging characteristics and dementia was investigated with extensive adjustments for other brain imaging characteristics, reducing confounding. Finally, brain volumes were normalized to the total intracranial volume.
Several methodological issues need to be addressed. First, the nosological dementia diagnosis in our study was a clinical diagnosis, supported by MR imaging at the moment of diagnosis, and if not available, baselineMR imaging. In some cases, a distinction between Alzheimer’s dementia and vascular dementia based on clinical data is hard to make, because neurodegeneration and vascular diseases often co-occur [32, 33]. For this reason, we investigated ‘overall dementia’ as outcome measure [34]. Second, it is possible that we missed some patients with incident dementia, because the cut-off point of 26 in the MMSE, although widely used, might not be sensitive enough, especially for early cases, vascular dementia, or dementia in higher educated participants. We think that if misclassification has occurred, it may have led to an underestimation of the effect. Third, we were not informed on the genetic APOE status, CSF biomarkers, or PET scan at baseline of our participants, which prevented us from further increasing the predictive value of the model. Finally, we feel that our study has a high generalizability to patients between 50 and 85 years presenting with SVD on neuroimaging in a general neurology clinic due to comparable baseline complaints andbaseline characteristics across studies. All our participants were functionally independent at baseline, with a mean MMSE of 28.1 (SD 1.6), reflecting estimates in the general population [35]. The median of WMH in our study is higher than found in population-based studies. For this reason, it is not surprising that our overall incidence rate of 16.4 per 1000 person years is higher than the overall incidence rate of 10.7 per 1000 person years found in a large population basedstudy [18].
We investigated both SVD and GM structures on the risk of dementia, because the WM represents the structural connectivity between different brain structures, including the neocortex and hippocampus, which are altogether crucial for cognitive function. It might be that the loss of neocortical volume is, at least in part, secondary to SVD, although the exact mechanismsremain to be elucidated. It could be that severe WMH represent areas of axonal loss that result in, reduced structural connectivity of the cortical areas once connected by these axons (with attendant loss of volume) for example by anterograde or Wallerian degeneration) [36, 37]. However, it might also be possible that neocortical atrophy leads to axonal damage as visualized by WMH on conventional FLAIR imaging. Irrespective of the actual mechanism, this may result in disconnection of the areas crucial in cognitive function, with attendant cognitive decline or even dementia.
In contrast to another longitudinal study [3], the risk of dementia was not significantly increased per increasing volumes of WMH, although the strength of the association was strikingly similar to that observed in the Rotterdam Study (HR 1.78, p = 0.167, compared with an HR of 1.6, p < 0.05). However, similar to our study after additional adjustments for SVD characteristics, this association was no longer significant. Furthermore we have adjusted for additional confounders, including HV, which the other study did not [34]. In accordance with our findings, it may be that the relation between WMH and dementia in that study was confounded by HV given the strong relation between WMH volume and HV [38].
Conflicting reports exist on impaired WM integrity in dementia compared to healthy controls [9]. Two recent longitudinal studies, [39, 40] showed that microstructural changes in the WM predict a faster decline in several cognitive domains, but dementia has never been included as an outcome measure. We found no additional effect of microstructural integrity of the WM on the risk of dementia. Possibly the microstructural changes have a more subtle clinical correlate, such as lower scores on separate cognitive domains, but are not severe enough to result in a dementia syndrome; especially not in the presence of presumably more severe expressions of SVD already visible on conventional imaging such as WMH and lacunes. Furthermore, at baseline, most participants had a relatively intact WM integrity, with only a mild to moderate WMH load. This might have limited the statistical power to detect additional value of the DTIparameters.
In our prognostic models, age is the foremost important predictor of dementia. The addition of WM, WMH, and HV did only marginally increase the variance explained. We did not find any additional diagnostic value of the DTI parameters. We did not cross-validate the prediction model in our dataset, because of the relatively low number of outcome events, therefore validation in an independent dataset should be performed as a gold standard.
Our results imply a pre-clinical period, with lower WM and HV in those who might develop dementia in the future, where intervention might be possible before onset of clinical symptoms.
We demonstrated that in elderly with cerebral SVD, low WM volume, next to HV is significantly related to the development of incident dementia after 5 years. Baseline diffusion parameters (both the global parameters in the prediction model as well as assessed voxel-wise by TBSS analyses) were not related to incident dementia, and had no diagnostic value on top of the macrostructural characteristics. Possibly the microstructural changes have a more subtle clinical correlate, and are therefore not severe enough to result in a dementia syndrome in 5 years time. Possibly these novel markers of subtle damage need to be assessed years or even decades before the first clinical symptoms emerge. Future research is needed to investigate additional value of change in DTI parameters over time in relation to incident dementia.
