Association of White Matter Hyperintensity Markers on MRI and Long-term Risk of Mortality and Ischemic Stroke

Standard Protocol Approvals, Registrations, and Patient Consents

The Second Manifestations of Arterial Disease–Magnetic Resonance (SMART-MR) study was approved by the medical ethics committee of the University Medical Center Utrecht according to the guidelines of the Declaration of Helsinki of 1975. Written informed consent was obtained from all patients participating in the SMART-MR study.

Study Population

We used data from the SMART-MR study.¹⁷ The SMART-MR study is a prospective cohort study at the University Medical Center Utrecht with the aim of investigating risk factors and consequences of brain changes on MRI in patients with manifest arterial disease.¹⁷ A total of 1,309 middle-aged and older adult patients referred to our medical center for treatment of manifest arterial disease (cerebrovascular disease, manifest coronary artery disease, abdominal aortic aneurysm, or peripheral arterial disease) were included for baseline measurements between 2001 and 2005.¹⁷ During a 1-day visit to the University Medical Center Utrecht, ultrasonography of the carotid arteries to measure the intima–media thickness (mm), blood and urine samplings, a physical examination, neuropsychological assessment, and a 1.5T brain MRI scan were performed.¹⁷ We used questionnaires for the assessment of demographics, medical history, risk factors, medication use, and cognitive and physical functioning.¹⁷

Vascular Risk Factors

We assessed age, sex, smoking habits, and alcohol intake at baseline using questionnaires. The body mass index (BMI) was calculated (kg/m²) by measuring weight and height. We measured systolic blood pressure (mm Hg) and diastolic blood pressure (mm Hg) 3 times with a sphygmomanometer and the average of these measurements was calculated. Hypertension was defined as a mean systolic blood pressure of >160 mm Hg, a mean diastolic blood pressure of >95 mm Hg, or the self-reported use of antihypertensive drugs.¹⁷ To determine glucose and lipid levels, an overnight fasting venous blood sample was taken. We defined diabetes mellitus as a fasting serum glucose level of ≥7.0 mmol/L or use of glucose-lowering medication or a known history of diabetes.¹⁷ The degree of carotid artery stenosis at both sides was assessed with color Doppler-assisted duplex scanning using a 10 MHz linear-array transducer (ATL Ultramark 9).¹⁸ Blood flow velocity patterns were used to evaluate the severity of carotid artery stenosis and the greatest stenosis observed on the left or right side of the common or internal carotid artery was taken to determine the severity of carotid artery disease.¹⁸ We defined a carotid artery stenosis ≥70% as a peak systolic velocity >210 cm/s.¹⁸

Brain MRI

MRI of the brain was performed on a 1.5T whole-body system (Gyroscan ACS-NT; Philips Medical Systems, Best, the Netherlands) using a standardized scan protocol.^17,19 Transversal fluid-attenuated inversion recovery (FLAIR) (repetition time [TR] 6,000 ms; echo time [TE] 100 ms; inversion time [TI] 2,000 ms), T1-weighted (TR 235 ms; TE 2 ms), T1-weighted inversion recovery (TR 2,900 ms; TE 22 ms; TI 410 ms), and T2-weighted images (TR 2,200 ms; TE 11 ms) were acquired with a voxel size of 1.0 × 1.0 × 4.0 mm³ and contiguous slices.^14,19 A neuroradiologist blinded to patient characteristics visually rated brain infarcts on the T1-weighted, T2-weighted, and FLAIR images of the MRI scans. We defined lacunes as focal lesions between 3 and 15 mm according to the STRIVE criteria.⁴ Nonlacunar lesions were categorized into large infarcts (i.e., cortical infarcts and subcortical infarcts not involving the cerebral cortex) and those located in the brainstem or cerebellum.¹⁴

WMH Volumes

WMH and brain volumes (intracranial volume and total brain volume) were obtained using the k-nearest neighbor (kNN) automated segmentation program on the T1-weighted, FLAIR, and T1-weighted inversion recovery sequences of the MRI scans.^19,20 WMH segmentations were visually assessed by an investigator (R.G.) using an image processing framework (MeVisLab 2.7.1.; MeVis Medical Solutions AG, Bremen, Germany) to ensure that cerebral infarcts were correctly removed from the WMH segmentations.¹⁴ Next, we performed ventricle segmentation using the fully automated lateral ventricle delineation (ALVIN) algorithm in Statistical Parametric Mapping 8 (SPM8, Wellcome Trust Centre for Neuroimaging, University College London, UK) for MATLAB (The MathWorks, Inc., Natick, MA).¹⁴ The procedure is described in detail elsewhere.^14,21 We labeled WMH according to their continuity with the margins of the lateral ventricle and their extension from the lateral ventricle into the white matter.¹⁴ Periventricular WMH were defined as lesions contiguous with the margins of the lateral ventricles and extending up to 10 mm from the lateral ventricle into the white matter.¹⁴ We defined confluent WMH as lesions contiguous with the margins of the lateral ventricles and extending more than 10 mm from the lateral ventricles into the white matter.¹⁴ We defined deep WMH as lesions that were separated from the margins of the lateral ventricles.¹⁴ Examples of periventricular, confluent, and deep WMH visualized in our algorithm are shown in figure 1. Total WMH volume was calculated as the sum of deep WMH and periventricular or confluent WMH.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 White Matter Hyperintensities (WMH) on Fluid-Attenuated Inversion Recovery (FLAIR) Images With Corresponding Visualizations in the Automated Algorithm

Examples of confluent (A), periventricular (B), and deep (C) WMH on FLAIR images with the corresponding visualizations in our algorithm shown below. The deep WMH lesion (arrow) is reconstructed in the coronal view, while the periventricular and confluent WMH are viewed from a transverse perspective. Note that the coronal reconstruction of the deep WMH lesion (C) may be influenced by the slice thickness and the lesion may be more punctiform. The confluent WMH lesion in (A) showed a volume of 11.57 mL with an accompanying deep WMH volume of 0.25 mL. The periventricular WMH lesion in (B) showed a volume of 4.98 mL without any accompanying deep WMH lesions. The deep WMH lesion in (C) showed a volume of 0.02 mL with an accompanying periventricular and deep WMH volume of 2.12 and 0.49 mL, respectively.

WMH Types

We categorized patients into the following 3 WMH types: periventricular WMH type without deep WMH, periventricular WMH type with deep WMH, and a confluent WMH type. We did not categorize the latter type according to presence or absence of deep WMH as the number of patients with a confluent WMH without deep WMH (n = 5) was insufficient to perform statistical analyses.¹⁴

WMH Shape Markers

We hypothesized that a more irregular shape of WMH may indicate more severe cerebral parenchymal damage based on previous histopathologic studies.^{6,-,8,13,22,23} The degree to which deep WMH are punctiform or ellipsoidal may also provide information on CSVD severity.¹⁵

Irregularity of periventricular or confluent WMH was quantified using the concavity index and fractal dimension. In previous work, we established that the concavity index was a robust shape marker that showed a normal distribution in the study sample and provided information on WMH shape irregularity based on volume and surface area.^14,24 The concavity index was calculated by reconstructing convex hulls and calculating volume and surface area ratios of lesions, in which a higher concavity index value corresponds to a more irregular shape of periventricular or confluent WMH.¹⁴ Fractal dimension was calculated using the box counting method and was used to quantify irregularity of periventricular or confluent WMH and of deep WMH.^14,25,26 A higher fractal dimension value indicated a more irregular WMH shape. As patients frequently show multiple deep WMH, a mean value for the fractal dimension was calculated across all deep WMH per patient.

The degree to which deep WMH are punctiform or ellipsoidal was assessed using the eccentricity, which was calculated by dividing the minor axis of a deep WMH lesion by its major axis.^14,15 A high eccentricity value corresponded to a punctiform deep WMH lesion, whereas a low value corresponded to an ellipsoidal lesion.^27,28 A mean value for the eccentricity was calculated across all deep WMH per patient.

Clinical Outcomes

Patients received a questionnaire every 6 months to provide information on outpatient clinic visits and hospitalization.¹⁸ If a fatal or nonfatal event was reported, original source documents were obtained and reviewed to determine the cause of the event. All possible events were audited independently by 3 physicians of the end point committee.¹⁸ Follow-up of patients was performed until death, refusal of further participation, or loss to follow-up. Vascular-related death was defined as death caused by myocardial infarction, stroke, sudden death (unexpected cardiac death occurring within 1 hour after onset of symptoms, or within 24 hours given convincing circumstantial evidence), congestive heart failure, or rupture of an abdominal aortic aneurysm.¹⁸ We defined nonvascular-related death as death caused by cancer, infection, unnatural death, or death from another nonvascular cause.¹⁸ Ischemic stroke was defined as the occurrence of relevant clinical features that caused an increase in impairment of at least 1 grade on the modified Rankin Scale, with or without a new relevant ischemic lesion on brain imaging.¹⁸

Study Sample

Of the 1,309 patients included, MRI data were irretrievable for 19 patients and 239 patients had missing data of one or more MRI sequences due to logistic reasons or motion artifacts. Forty-four of the remaining 1,051 patients had unreliable brain volume data due to motion artifacts in all 3 MRI sequences. Four patients were excluded due to undersegmentation of WMH by the automated segmentation algorithm and another 4 patients were excluded because they did not have any WMH greater than 5 voxels. As a result, 999 patients were included in the current study.

Statistical Analysis

Baseline characteristics of the study sample were reported as means or percentages where applicable.

Patients were followed from the date of the MRI until ischemic stroke, death, loss to follow-up, or end of follow-up (March 2017), whichever came first. Cox proportional hazard analysis was used to estimate hazard ratios (HRs) for the occurrence of all-cause, vascular-related, and nonvascular-related death and ischemic stroke associated with WMH volumes, type, and shape markers. The proportional hazards assumption was checked by inspection of Schoenfeld residuals. We concluded that the proportional hazards assumption was met for all covariates.

To reduce the risk of bias due to complete case analysis, we performed chained equations imputation on missing covariates to generate 10 imputed datasets using SPSS 25.0 (Chicago, IL).²⁹ The Cox regression analyses were performed on the imputed datasets and the pooled results were presented. We used SAS 9.4 (SAS Institute, Cary, NC) and SPSS 25.0 (Chicago, IL) to perform the statistical analyses.

Data Availability

For use of anonymized data, a reasonable request has to be made in writing to the study group and the third party has to sign a confidentiality agreement.