Structural network efficiency predicts conversion to dementia

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 as assessed by MRI. This hospital-based study consists of 503 participants with SVD without dementia aged between 50 and 85 years. On the basis of established research criteria, SVD was defined as the presence of WMH or lacunes on neuroimaging.^1,5 Symptoms of SVD include acute symptoms, such as TIAs or lacunar syndromes, or subacute manifestations, such as cognitive or motor disturbances or depressive symptoms. Baseline data collection was performed in 2006. Inclusion criteria were age between 50 and 85 years and SVD on neuroimaging (WMH or lacunes). 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. Exclusion criteria were dementia; parkinson(ism); intracranial hemorrhage; life expectancy <6 months; intracranial space-occupying lesion; (psychiatric) disease interfering with cognitive or motor testing; recent or current use of acetylcholine-esterase inhibitors, neuroleptic agents, l-dopa, or dopa-agonist/antagonists; non-SVD-related WMH mimics (e.g., multiple sclerosis); prominent visual or hearing impairment; language barrier; or MRI contraindications or known claustrophobia.⁶ Follow-up was completed in 2012. Of 503 baseline participants, 2 were lost to follow-up, 49 had died, and 54 refused in-person follow-up, but clinical endpoints were available. For this study, 436 patients were included; 67 participants were excluded because of the baseline presence of territorial infarcts (n = 55), inadequate quality of the MRI (n = 4), or failure of the DTI pipeline (n = 8).

Standard protocol approvals, registrations, and patient consents.

All participants signed an informed consent form. The Medical Review Ethics Committee region Arnhem-Nijmegen approved the study.

Dementia diagnosis.

Clinical endpoints were available for all participants (n = 436). Dementia diagnosis was made by referring the patient to the memory clinic of Radboud Alzheimer Center or by a panel consisting of a neurologist, clinical neuropsychologist, and geriatrician reviewing all available medical records and cognitive assessments. Diagnosis was based on DSM-IV⁷ criteria; probable AD was based on National Institute on Aging–Alzheimer's Association criteria⁸ and vascular dementia (VaD) was based on National Institute of Neurological Disorders and Stroke–Association Internationale pour la Recherche en l'Enseignement en Neurosciences criteria.⁹ The onset of dementia was defined as the date on which the clinical symptoms were compatible with the diagnosis. If the exact date was not known, the midpoint was used between the baseline visit and the first date the diagnosis was confirmed, or, failing this, the date the participant was admitted to a nursing home because of dementia.

MRI acquisition.

MRI was performed on a 1.5T S Magneton Sonata scanner (Siemens Medical Solutions, Erlangen, Germany), including T1 3D magnetization-prepared rapid gradient echo imaging (repetition time [TR] = 2.25 s, echo time [TE] = 3.68 ms, inversion time [TI] = 850 ms, flip angle [FA] = 15°, voxel size 1.0 × 1.0 × 1.0 mm), a fluid-attenuated inversion recovery (FLAIR) sequence (TR = 9.00 s, TE = 84 ms, TI = 2.20 s, voxel size 1.0 × 1.2 × 5.0 mm, plus an interslice gap of 1 mm), T2*-weighted gradient echo sequences (TR = 800 ms, TE = 26 ms, voxel size 1.3 × 1.0 × 6.0 mm, interslice gap 1 mm), and a DTI sequence (TR = 10.10 s, TE = 93 ms, voxel size 2.5 × 2.5 × 2.5 mm, 4 unweighted scans, 30 diffusion-weighted scans with b value of 900 s/mm²). All participants were scanned on the same scanner.¹⁰

Vascular risk factors.

Hypertension was defined as systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg or use of antihypertensive drugs. Blood pressures were measured 3 times in supine position after 5 minutes of rest. Diabetes and hypercholesterolemia were considered to be present if the participant was taking antidiabetic or lipid-lowering drugs for high cholesterol. Body mass index was calculated as weight divided by height (in meters) squared. Smoking status was obtained through standardized questionnaires, which was checked during the interview. Leisure time physical activity was assessed with a questionnaire: participants were asked to estimate the average amount of time per week during the past year spent on the physical activities.¹¹ Metabolic equivalent value (MET) was assigned to each activity. One MET is proportional to the energy expended while sitting quietly. For each activity, we estimated the energy expended in MET h/wk, by multiplying its MET value by the time spent performing it.¹¹

MRI markers.

WMH were manually segmented on FLAIR images and the total WMH volume was calculated by summing the segmented areas multiplied by slice thickness. Rating of lacunes and microbleeds was revised according to the recently published Standards for Reporting Vascular changes on neuroimaging (STRIVE) by trained raters blinded to all clinical data.¹ WMH were defined as white matter hyperintensities on FLAIR images without prominent or only faint hypointensity on the T1-weighted images, except for gliosis surrounding infarcts.¹ Lacunes were defined as hypointense areas >2 mm and ≤15 mm on FLAIR and T1, ruling out enlarged perivascular spaces (≤2 mm, except around the anterior commissure, where perivascular spaces can be large) and infraputaminal pseudolacunes.¹ There were good intrarater and interrater variability with weighted kappa of 0.87 and 0.95, respectively, for the presence of lacunes and 0.85 and 0.86 for the presence of microbleeds, calculated in 10% of the scans. Interrater variability (assessed by intraclass correlation coefficient) for total WMH volume was 0.99. Automated segmentation on T1 images was performed using Statistical Parametric Mapping (SPM5) to obtain gray matter (GM), WM, and CSF probability maps. These maps were binarized and summed to supply total volumes. These volumes and WMH volume were normalized to the total intracranial volume. The b0 images were first used to compute the coregistration parameters to the skull-stripped T1 images using functional MRI of the brain linear image registration tool (FLIRT), applied to all diffusion-weighted images. Next, we calculated the mean fractional anisotropy (FA) and mean diffusivity (MD) within the whole white matter.

Network nodes (brain regions).

Brain regions were demarcated by the automated anatomical labeling (AAL) template,¹² which resulted in 45 regions for each hemisphere. Cerebellar regions were excluded, because the tractography technique employed in this study is unsuitable for tracing cerebellar connections. Using the FSL 5.0.5 tools,¹³ the skull-stripped T1 images were nonlinearly registered to Montreal Neurological Institute 152 template using functional MRI of the brain nonlinear registration tool. Next, the transformation matrix was derived from the registration of b0 images to T1 space using FLIRT. These transformations were then used to register the AAL image to each participant's diffusion image space.

Network edges (WM connections).

The in-house developed algorithm named PATCH was employed on the raw diffusion data to correct for cardiac and head motion artifacts and eddy currents.¹⁴ Diffusion Toolkit (www.trackvis.org) was used to calculate the diffusion tensor and FA. To generate the fiber tracks of the entire brain for each participant, fiber assignment by continuous tracking was employed using Diffusion Toolkit. The tracking algorithm started at the center of all voxels with FA >0.2 and ended when the fiber tracks left the brain mask or encountered voxels with FA <0.2, or when the turning angle exceeded 60°. Two regions were considered connected if the endpoints of the reconstructed streamline lay within both regions. For each participant, a weighted edge was constructed using the mean FA for each streamline and the number of reconstructed streamlines connecting 2 regions.¹⁵ Note that this remains an indirect measure, as measuring directly the number of fibers or connection strength as a true measure is currently not possible with tractography techniques in general.¹⁶ None of the current techniques is capable of reconstructing individual nerve fibers or fiber bundles and as such, interpreting the number of reconstructed streamlines as a true measurement of the number of actual fibers is inappropriate.¹⁶ The connection strength was further normalized by the average of these volumes to correct for the different sizes of the AAL regions and different brain sizes.¹⁷ This resulted in an undirected weighted 90 × 90 matrix for each participant.

Graph theory analysis.

Graph theoretical network measures were computed using the Brain Connectivity Toolbox.¹⁸ Basic network measures include density and total network strength. Density is defined as the total number of edges in a network observed divided by the possible number of edges. Total network strength represents the sum of weighted edges of a network. Next, we calculated global and local efficiency. Efficiency between 2 regions is expressed as the inverse of the shortest path length between 2 regions. The shortest path length refers to the minimum number of weighted connection between 2 regions. Global efficiency of the network is defined as the average of efficiency for all node pairs. The local efficiency for each node is the global efficiency of all direct neighbors of that node. The global efficiency reflects the extent to which information communication is globally integrated in the network, whereas the local efficiency measures the segregation and specialization within a network.

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 assessment. Group differences in baseline characteristics were tested by independent t test, χ² test, or Mann-Whitney U test, when appropriate (table 1).

Cox proportional regression analyses were used to obtain hazard ratios (HR) for each risk factor and for each MRI parameter (GM, WM, WMH volume, lacunes, microbleeds, FA, and MD), adjusted for age, sex, education, and baseline Mini-Mental State Examination (MMSE). Next, multivariable forward stepwise Cox regression model was performed with age, sex, education, and baseline MMSE as fixed predictors when entering the significant predictors (with p < 0.05 for entry and p > 0.1 for removal). Patients who died or did not reach the endpoint were censored.

Penalized Cox regression with least absolute shrinkage and selection operator (Lasso) was chosen as a secondary analysis to confirm the findings from the multivariable approach. This method is useful for selecting predictors from a set of potential predictors, reduces overfitting, and minimizes the effects of outliers, especially in datasets with a large number of variables with respect to the number of events. Variables were standardized when appropriate before performing the Lasso model and the tuning parameter was selected through a 10-fold cross-validation. Statistical analyses were done using R (http://www.r-project.org).