White matter hyperintensity burden in acute stroke patients differs by ischemic stroke subtype

Abstract

Objective

To examine etiologic stroke subtypes and vascular risk factor profiles and their association with white matter hyperintensity (WMH) burden in patients hospitalized for acute ischemic stroke (AIS).

Methods

For the MRI Genetics Interface Exploration (MRI-GENIE) study, we systematically assembled brain imaging and phenotypic data for 3,301 patients with AIS. All cases underwent standardized web tool–based stroke subtyping with the Causative Classification of Ischemic Stroke (CCS). WMH volume (WMHv) was measured on T2 brain MRI scans of 2,529 patients with a fully automated deep-learning trained algorithm. Univariable and multivariable linear mixed-effects modeling was carried out to investigate the relationship of vascular risk factors with WMHv and CCS subtypes.

Results

Patients with AIS with large artery atherosclerosis, major cardioembolic stroke, small artery occlusion (SAO), other, and undetermined causes of AIS differed significantly in their vascular risk factor profile (all p < 0.001). Median WMHv in all patients with AIS was 5.86 cm³ (interquartile range 2.18–14.61 cm³) and differed significantly across CCS subtypes (p < 0.0001). In multivariable analysis, age, hypertension, prior stroke, smoking (all p < 0.001), and diabetes mellitus (p = 0.041) were independent predictors of WMHv. When adjusted for confounders, patients with SAO had significantly higher WMHv compared to those with all other stroke subtypes (p < 0.001).

Conclusion

In this international multicenter, hospital-based cohort of patients with AIS, we demonstrate that vascular risk factor profiles and extent of WMH burden differ by CCS subtype, with the highest lesion burden detected in patients with SAO. These findings further support the small vessel hypothesis of WMH lesions detected on brain MRI of patients with ischemic stroke.

White matter hyperintensity (WMH) is a common radiographic marker seen in the deep and periventricular white matter on fluid-attenuated inversion recovery (FLAIR) MRI.¹ In patients with acute ischemic stroke (AIS), WMH is associated with susceptibility to infarct growth² and poor poststroke outcomes.³ As a radiographic manifestation of chronic cerebrovascular disease, WMH burden is thought to result from the impact of multiple vascular risk factors on the small vasculature and is known to be associated with ongoing injury, including higher rates of WMH accumulation and other cerebrovascular manifestations.⁴ In stroke-free adults, known vascular risk factors such as hypertension,⁵ carotid atherosclerosis,⁶ diabetes mellitus,⁷ and cigarette smoking⁸ are strongly associated with increased WMH volume (WMHv). However, limited data are available on risk factors for WMH in AIS, with age and elevated homocysteine levels having been previously identified.⁹

In this study, we aimed to assess common vascular risk factors contributing to WMH severity in patients with AIS. However, reliable WMH assessment in AIS is complicated by the varying methodology, ranging from a semiquantitative rating scale¹⁰ to semiautomated/fully automated WMH measurements.^11,12 Furthermore, many of these approaches are challenging to apply to large clinical datasets of WMH. To address this gap, we used a designated artificial intelligence–driven deep-learning pipeline to derive robust WMHv and to systematically investigate determinants of WMHv in AIS and its subtypes in a retrospective hospital-based cohort of 3,301 patients with AIS. We hypothesize that WMH burden is highest in patients with small artery occlusion (SAO) and that classic vascular risk factors contribute to a higher WMH burden.

Methods

Study design and participants

The MRI-Genetics Interface Exploration (MRI-GENIE) study is a large, international collaboration of 12 sites contributing 3,301 patients with AIS with phenotypic, radiographic, and genotypic data. A detailed description of the study design and concept has been published previously.¹³ In summary, each site received approval of its respective institutional review board. Patients were recruited through the Stroke Genetics Network (SiGN), with recruitment dates ranging from 1999 to 2012.¹⁴ MRI data were assembled in a central imaging repository for assessment of neuroimaging phenotypes. Here, we assessed 2,781 patients for whom FLAIR imaging was available for automated assessment. MRI sequences were obtained in the acute phase (median time to scan <1 day [upper quartile range 1–4 days] from symptom onset). After quality control (QC), 2,529 patients with AIS with automatedly extracted WMHv from clinical axial FLAIR images remained available for analysis (figure 1).¹⁵

Figure 1. Flowchart of case selection for analysis.

FLAIR = fluid-attenuated inversion recovery; MRI-GENIE = MRI-Genetics Interface Exploration; QC = quality control.

Standard protocol approvals, registrations, and patient consents

All patients with AIS were recruited in a hospital-based setting. Ethics or institutional review board approval was obtained as appropriate for each individual participating study. Informed consent, including sharing of deidentified demographic, imaging, and genotyping data, was obtained from all patients or their legally authorized representative. Demographic and genotyping data were collected by the Data Management and Genotyping Core of SiGN.

Automated WMHv extraction

WMHv was successfully extracted from clinical axial FLAIR images of 2,529 patients (76.6%) in the MRI-GENIE cohort. All FLAIR images were acquired as per the local AIS clinical imaging protocol. On average, the images had a mean resolution of 0.7 mm in-plane (minimum 0.4 mm, maximum 1.9 mm) and 6.3 mm through-plane (minimum 1.0 mm, maximum 65.0), whole-brain acquisition. A fully automated pipeline using deep learning was developed specifically for quantification of WMHv on clinical-grade MRI scans and applied to all patients.¹⁵ A detailed description, including assessment of the WMH pipeline performance, has been previously published.¹⁵ In brief, the pipeline has 3 main processing steps with 2 QC checks. First, an initial QC assessment based on in-plane and through-plane resolution, as well as number of slices, is performed to identify scans with insufficient information for WMH extraction. Brains were extracted from the clinical axial FLAIR images using a dedicated deep-learning architecture (library found at github.com/adalca/neuron). After brain extraction, patients with unexpectedly low or high age-stratified brain volume were identified and manually assessed for incomplete brain extraction. To account for differences in image acquisition, this was followed by an intensity normalization to harmonize image intensities across sites and scanners. Then, automated WMH segmentation was performed, again by using a deep-learning architecture with atlas-based, spatial priors to include information on the distribution of WMH and differentiation from AIS artifacts such as edema, movement artifact, or prior stroke, with the algorithm compensating for large infarctions to avoid underestimation of WMH.^15–17 Reasons for excluding scans during the first-pass QC (n = 254, 10% of all participants with FLAIR) were as follows: low number of slices, mislabeled MRI sequence, or significant motion artifact (n = 97, 3.8%). In the second-pass QC, images with different slice direction (coronal [n = 72, 1.1%] or sagittal [n = 3, 0.1%]), persistent motion artifact (n = 8, 0.3%), incomplete brain extraction (n = 62, 2.5%), or other issues (n = 8, 0.3%) were excluded. Finally, we excluded 4 patients with WMHv measurement of 0 cm³ (n = 4, 0.2%).

AIS subtyping

All patients underwent systematic stroke subtyping with the Causative Classification of Ischemic Stroke (CCS).¹⁸ A web-based standardized algorithm incorporates results of patient history, physical examination findings from the clinical stroke assessment, and diagnostic testing to systematically assign the CCS subtype. Two major CCS classifications have been established: phenotypic CCS, reflecting the abnormal test results at the time of stroke that does not rely on judgment regarding the likely etiology, and causative CCS, which requires integration of all diagnostic tests and history to provide the most likely mechanism for the stroke. Throughout this article, we refer to causative CCS as CCS. If presented with multiple competing etiologies, the web-based CCS algorithm assigns the most probable cause. CCS subtypes include large artery atherosclerosis (LAA), major cardioembolic stroke (CE major), SAO, Other, and Undetermined cause of stroke (i.e., 5-item CCS subtypes).¹⁸ LAA, CE major, SAO, and Other categories include those for whom the subtype was considered possible, probable, or evident. The Undetermined category included the following: cryptogenic embolism, other cryptogenic stroke, and minor cardioembolic stroke, as well as those with incomplete information or unclassified stroke. Each reader underwent formal training and certification before adjudicating CCS subtypes.¹⁸

Statistical analysis

Demographic data and vascular risk factors, including age, sex, race, atrial fibrillation, coronary artery disease (CAD), diabetes mellitus, hypertension, prior stroke, and smoking status, were abstracted from patient records by each site. Numeric variables are expressed as mean ± SD or median and interquartile range (IQR), depending on normal or nonparametric distribution. Categorical variables are expressed as counts and frequencies. Statistical comparison was performed across the 5-item CCS subtypes with the χ² test to compare categorical data, analysis of variance for age, and Kruskal-Wallis test for WMHv. Given the skewed distribution of WMHv, we used natural log-transformed WMHv for all regression analyses. Included patients were compared to those who failed the imaging QC using mixed logistic regression model of included/excluded status with the demographic variables as fixed-effects variables and study site as a random-effects variable for each demographic parameter. After adjustment for site as a random variable, the participants passing or failing QC did not differ significantly (data not shown). Furthermore, univariable linear regression was used to identify predictors of WMHv. Variables passing p < 0.1 were included in the multivariable model. Multiple linear mixed-effect modeling was used to identify independent determinants of WMHv. The multiple linear mixed-effects model was adjusted for site as a random variable because key variables such as age, race, and distribution of vascular risk factors (atrial fibrillation, CAD, diabetes mellitus, and hypertension) differed by site (data not shown). Cases with missing information for vascular risk factors were excluded from the final multivariable model (table 1).

Table 1.

Basic demographics of the MRI-GENIE cohort and comparison by stroke subtype

Data availability

Requests for individual-level data from MRI-GENIE study should be addressed to the MRI-GENIE principal investigator (N.S.R.) and are subject to approval by the study group and clearance by the local regulatory board.

Results

Imaging data for this retrospective analysis were assembled between 2012 and 2017. WMH analysis for 2,529 patients with AIS (figure 1) was conducted using deidentified clinical FLAIR images in June 2018. Mean age was 63.4 (SD = 14.6) years, and 39.3% (n = 993) of all patients were female. The majority of patients were white (n = 2,141, 84.7%), had a medical history of hypertension (n = 1,668, 66.4%), and were either current or former tobacco smokers (n = 1,323, 54.1%). Table 1 includes other risk factor distributions, including atrial fibrillation, CAD, diabetes mellitus, and history of stroke. MRI-GENIE patients excluded from this analysis due to lack of or poor-quality images (n = 772) did not differ significantly in age, sex, or other vascular risk factors from those included (n = 2,529).

Vascular risk factor profiles in CCS subtypes

Distribution of age, sex, and vascular risk factors differed significantly across the 5-item CCS subtypes except for history of prior stroke (table 1). Patients with LAA were less likely to be female (n = 176, 32.3%) or to have atrial fibrillation (n = 30, 5.6%). Patients classified as having a major CE stroke were more likely to be female (n = 186, 47.2%), were on average older (mean 71.8 [SD 11.9] years), were more likely to have atrial fibrillation (n = 267, 68.8%) or CAD (n = 111, 28.6%), and were slightly more likely to have hypertension (n = 288, 73.3%) compared with patients with the other stroke subtypes, although the differences were not significant. Patients classified as having SAO had the lowest frequency of atrial fibrillation (n = 12, 3.1%) and the highest frequency of diabetes mellitus (n = 107, 28.1%). Notably, patients classified as Other stroke subtype were the youngest (mean 49.0 [SD 13.6] years) and had the lowest frequency of history of CAD (n = 18, 10.2%) and diabetes mellitus (n = 27, 15.1%).

WMH in CCS subtypes

The median WMHv (example outlines in figure 2) in the entire AIS cohort was 5.86 cm³ (IQR 2.18–14.61 cm³). The unadjusted median WMHv was highest in patients classified as having CE major (8.13 cm³ [IQR 3.65–17.12 cm³]) and second highest in those with SAO (7.53 cm³ [IQR 2.84–18.45 cm³]). The lowest WMHv with a median volume of 2.16 cm³ (IQR 0.93–5.29 cm³) was observed in patients classified as having Other cause of stroke (table 1).

Figure 2. Automated WMH outline in ischemic stroke patients with (A) mild, (B) moderate, and (C) severe WMH burden.

Results were extracted by the automated MRI-Genetics Interface Exploration (MRI-GENIE) pipeline. WMH = white matter hyperintensity.

Univariable associations with WMHv in AIS

In univariable analysis, age was the strongest predictor of WMHv (β = 0.05, 95% confidence interval [CI] 0.05–0.05, p < 0.001). Furthermore, atrial fibrillation (β = 0.46, 95% CI 0.31–0.61), CAD (β = 0.41, 95% CI 0.27–0.55), diabetes mellitus (β = 0.35, 95% CI 0.23–0.48), hypertension (β = 0.82, 95% CI 0.72–0.93), and prior stroke (β = 0.55, 95% CI 0.37–0.72) were significant univariable predictors of WMHv (all lower than p < 0.001). Likewise, a history of current or former smoking contributed to a higher WMHv in univariable analysis (β = 0.13, 95% CI 0.03–0.24, p = 0.015) (table 2).

Table 2.

Univariable and multivariable predictors of WMH

Multivariable associations with WMHv in AIS

The multivariable analysis included all variables passing the threshold of p < 0.1 and was adjusted for site as a random variable. Age (β = 0.05, 95% CI 0.04–0.05), hypertension (β = 0.35, 95% CI 0.27–0.47), history of stroke (β = 0.45, 95% CI 0.32–0.63), current or former smoking status (β = 0.19, 95% CI 0.10–0.28) (all p < 0.001), and diabetes mellitus (β = 0.11, 95% CI 0.03–0.24, p = 0.041) remained independent predictors of WMHv (table 2).

Adjusted WMHv in CCS subtypes

WMHv for the 5-item CCS subtypes was reassessed when adjusted for the identified multivariable predictors of WMHv, as well as for site as a random variable, by comparing the residuals of the multivariable linear mixed-effects model by CCS subtype. Cases classified as SAO demonstrated the highest adjusted residual WMHv (1.47 cm³ [IQR −1.52 to 3.15 cm³]) in the Bonferroni-adjusted group-wise comparison of CCS subtypes (figure 3).

Figure 3. WMH by CCS subtype.

Residuals of white matter hyperintensity (WMH) volume (WMHv) adjusted for age, vascular risk factors, and site. Cases with small artery occlusion (SAO) have the highest WMH burden compared to other Causative Classification of Ischemic Stroke (CCS) subtypes (*p < 0.05, **p < 0.005). CE major = major cardioembolic stroke; IQR = interquartile range; LAA = large artery atherosclerosis.

CCS-subtype specific associations with WMHv

Contributors to WMH burden were also assessed by CCS subtype (table 3). In LAA, age (β = 0.05, 95% CI 0.04–0.05, p < 0.001) and history of stroke (β = 0.68, 95% CI 0.34–1.01, p < 0.001) were independently associated with higher WMHv. In CE major, only age (β = 0.05, 95% CI 0.04–0.05, p < 0.001) emerged as an independent predictor of WMHv in the multivariable model. In contrast, age (β = 0.04, 95% CI 0.03–0.05, p < 0.001), hypertension (β = 0.43, 95% CI 0.16–0.70, p = 0.002), and prior stroke (β = 0.52, 95% CI 0.11–0.93) emerged as independent predictors of WMHv in SAO. In cases with Other stroke subtype, age (β = 0.04, 95% CI 0.03–0.06, p < 0.001) and prior stroke (β = 0.90, 95% CI 0.13–1.66, p = 0.021) were associated with higher WMHv, and female sex (β = −0.37, 95% CI −0.72 to −0.02, p = 0.040) was associated with lower WMHv. Lastly, in patients with Undetermined cause of stroke, age (β = 0.05, 95% CI 0.04–0.05, p < 0.001), hypertension (β = 0.58, 95% CI 0.42–0.74, p < 0.001), prior stroke (β = 0.38, 95% CI 0.15–0.61, p = 0.001), and smoking (β = 0.23, 95% CI 0.09–0.37, p = 0.001) were independently associated with higher WMHv, whereas CAD was associated with lower WMHv (β = −0.26, 95% CI −0.45 to −0.05, p = 0.011).

Table 3.

Predictors of WMHv by CCS subtype

Discussion

In this large multicenter, hospital-based cohort of 2,529 patients with AIS, we demonstrate that vascular risk factor profiles differ across CCS subtypes. Furthermore, we show that, as in stroke-free populations, age, hypertension, smoking (all p < 0.001), and diabetes mellitus (p = 0.041) are independent predictors of high WMHv. In addition, when the analysis is adjusted for confounding variables, patients with SAO exhibit the highest amount of WMH burden, whereas patients with CE major have the highest unadjusted WMHv.

These findings independently validate prior studies examining patients with thoroughly ascertained stroke subtypes. In a study of 891 patients with AIS with the Trial of Org 10172 in Acute Stroke Treatment (TOAST) stroke subtype classification,¹⁹ patients with SAO had the highest amount of WMH burden across all AIS subtypes.⁹ Recently, a large multicenter study semiautomatedly assessed WMHv in 5,035 patients with AIS of Korean descent.³ Vascular risk factor profiles were compared across WMH quintiles. Overall, hypertension, diabetes mellitus, and atrial fibrillation emerged as independently associated with WMH quintile when adjusted for age and sex. While the findings were overall similar, our study highlights the importance of assessing vascular risk factors by stroke subtype, particularly when a standardized subtyping-tool like CCS is used. We show that in patients with SAO, a medical history of hypertension and prior stroke are independently associated with larger WMH burden, highlighting potentially addressable risk factors in SAO. In addition, prior stroke is a predictor of WMH burden in LAA, SAO, and other and undetermined cases, possibly hinting at a vicious cycle in which existing cerebrovascular burden increases the risk for further cerebral tissue injury. In cases of undetermined stroke, CAD, hypertension, and smoking also are significantly associated with WMH burden. However, this may reflect the potentially competing underlying stroke etiologies. Furthermore, in multivariable modeling, we show that in LAA, only age and prior stroke were independently associated with WMHv, while in CE major, only age independently contributed to WMH burden. Such specific CCS subtype findings support the concept of a different underlying etiologic disease processes.

Consistent with prior studies,^20–22 age remained the most significant independent determinant of WMHv. However, the effect of age on WMH burden may differ across the lifespan, and it may interact with other vascular risk factors. For example, in a cohort of 560 patients with AIS at the extremes of age (young <55 years and old >75 years), different vascular risk profiles emerged.²³

Among other vascular risk factors, hypertension has a well-established role in WMH accumulation in population-based cohorts⁵; furthermore, given its robust association with WMHv in this large cohort of patients with AIS, prior studies^9,24 have most likely been underpowered. Likewise, diabetes mellitus has been implicated in the development and lesion size of WMH in stroke-free adults,^7,25 but the current analysis is among the first to demonstrate an association between the WMH severity and diabetes mellitus in patients with AIS.

Furthermore, we observed a relationship of AF with higher WMH burden in univariable analysis, which no longer persisted after adjusting for age and other potential confounders. Given the higher incidence of AF among elderly patients and potential collinearity between these 2 risk factors, the effect of AF on WMH burden in patients with AIS could not be definitively assessed and requires further study.

Lastly, we confirmed smoking as an independent predictor of WMH burden. The importance of smoking exposure on WMH risk is further highlighted by a study from 2015 demonstrating a dose-dependent effect of smoking in stroke-free adults.²⁶

Our study provides a key piece of evidence on the association between WMH burden and small vessel stroke. This is of growing importance because multiple mechanisms are considered for the pathophysiology of chronic cerebral ischemic changes that appear as WMH lesions on T2/FLAIR MRI. Among these competing factors, arteriolar sclerosis, capillary endothelial activation, and immunoreactivity for hypoxia-inducible factors 1 and 2 as a manifestation of ongoing hypoxia have been described.²⁷ More recently, the focus for elucidating WMH pathology shifted toward the investigation of microstructural changes in normal-appearing white matter, which possibly precede formation of new WMH lesions and WMH lesion progression.²⁸ Such microstructural changes have been associated with hypertension, smoking, and diabetes mellitus,²⁹ raising the possibility of addressing modifiable vascular risk factors before the formation of new or the expansion of existing WMH. The association of microstructural white matter changes with worsened functional recovery after ischemic stroke further highlights the importance of white matter integrity.³⁰

Recent studies also suggest that WMH has the potential to regress over time. In a study investigating 190 patients with minor stroke, a repeat MRI at 1 year demonstrated WMH regression in 71 patients.³¹ Patients with WMH regression also had a greater reduction in blood pressure, although this requires further validation in prospective cohorts. Patients with increasing WMH had a higher likelihood of experiencing a recurrent ischemic event.³¹ Similarly, patients with higher levels of WMH had a higher likelihood of 90-day stroke recurrence.³²

This study has important limitations. First, imaging data were collected retrospectively, and their availability varied by site. Because brain MRIs were collected in the acute phase of AIS, variability in the quality of acquisition and clinical indication for neuroimaging is a significant factor in this study. Great care has been exercised in ascertaining the quality and operational utility of the MRI-GENIE neuroimaging database.¹³ Furthermore, QC for the automated WMHv segmentation was the key feature of this innovative MRI analysis pipeline and is described in detail elsewhere.¹⁵ An additional limitation of this analysis is related to a large proportion of AIS cases with undetermined stroke subtype (n = 1,022). Because 5-item CCS-subtyping was performed with a standardized web-based protocol by trained adjudicators, this may be due to either ≥2 equally likely competing stroke etiologies resulting in the classification of undetermined or a lack of sufficient diagnostic data for a final CCS classification. The proportion of stroke cases classified as undetermined in the presented MRI-GENIE cohort is slightly lower than in the original SiGN study (39% vs 43%).³³ In SiGN, this category was driven mainly by cases with incomplete clinical evaluations (55%), minor cardioembolic sources (18%), other cryptogenic sources (15%), and multiple competing etiologies (9%). Overall, the proportion of completely unclassified patients was small (4%). Despite the number of undetermined cases, the 5-item CCS has the advantage of a standardized, reproducible web-based assessment with excellent interrater agreement (κ = 0.86).³⁴

One advantage of examining stroke subtypes in the hospital-based cohort of patients with AIS is the relative uniformity of the cohort, including timing of assessment, risk factor ascertainment at the time of stroke diagnosis, and the ability to verify the stroke event with the acute MRI imaging for those cases that may otherwise have atypical semiology or symptom duration. Central strengths of our study include the following: (1) MRI-GENIE is a large imaging and genetic database specifically developed to enable future studies of genetic architecture of acute and chronic neuroimaging traits in patients with AIS; (2) all cases underwent systematic stroke subtyping using the standardized web-based CCS tool¹⁸; and (3) WMH was segmented with the use of an artificial intelligence–enabled automated segmentation pipeline¹⁵ specifically designed to analyze the multicenter, clinical brain MRI of patients with AIS. The automated WMH pipeline has the advantage of yielding robust and reproducible WMH segmentations across multiple sites and has applicability to other acute stroke cohorts. The pipeline has demonstrated excellent agreement with manual assessments of WMHv in patients with AIS.¹⁵ Overall, the applied WMH pipeline contributes to the generalizability of our results to other AIS cohorts. Systematic, large-scale WMH assessment in AIS will allow the study of the underlying genetic architecture and assessment of the role of WMH in AIS severity and outcome.

We demonstrated that patients with SAO exhibit the highest amount of WMH after adjustment for confounders compared to other AIS subtypes, supporting the hypothesis that WMH lesions seen in patients with AIS are the result of small vessel disease. Furthermore, we have shown that the vascular risk profile differs by CCS, with modifiable risk factors being important contributors to the overall WMH burden in patients with AIS. Our findings in part reconcile the previously described differences in risk factor profiles for WMH in stroke-free and AIS populations. Effectively addressing these vascular risk factors could provide an important avenue for modifying WMH disease burden and thus potentially preventing the detrimental downstream effects of high WMH burden in patients with AIS.

Glossary

AIS

acute ischemic stroke

CAD

coronary artery disease

CCS

Causative Classification of Ischemic Stroke

CE major

major cardioembolic stroke

CI

confidence interval

FLAIR

fluid-attenuated inversion recovery

IQR

interquartile range

LAA

large artery atherosclerosis

MRI-GENIE

MRI-Genetics Interface Exploration

QC

quality control

SAO

small artery occlusion

SiGN

Stroke Genetics Network

TOAST

Trial of Org 10172 in Acute Stroke Treatment

WMH

white matter hyperintensity

WMHv

WMH volume

Appendix 1. Authors

Appendix 2. Coinvestigators

Study funding

Funding provided by NIH–National Institutes of Neurological Disorders and Stroke (MRI-GENIE: R01NS086905—principal investigator N. Rost; K23NS064052, R01NS082285—N. Rost; SiGN: U01 NS069208—J. Rosand, S. Kittner; R01NS059775, R01NS063925, R01NS082285, P50NS051343, R01NS086905, U01 NS069208—O. Wu), NIH National Institute of Biomedical Imaging and Bioengineering (P41EB015902—P. Golland, U01NS030678—B. Kissela, D. Kleindorfer; EB015325—O. Wu), ISGS: R01NS423733—principal investigator J. Meschia, Swedish Heart and Lung Foundation, Lund University, Region Skåne, the Freemasons Lodge of Instruction Eos Lund, Skåne University Hospital, the Foundation of Färs&Frosta—one of Sparbanken Skåne's ownership foundations, and the Swedish Stroke Association—A. Lindgren, Swedish Research Council and the Swedish Heart and Lung Foundation, the Swedish State under the ALF agreement—C. Jern, Spanish Ministry of Science and Innovation, Instituto de Salud Carlos III (Funding for Research in Health [PI051737], [PI10/02064], [PI12/01238], and [PI15/00451—J. Jiménez-Conde]), Fondos FEDER/EDRF Red de Investigación Cardiovascular (RD12/0042/0020—J. Jimenez-Conde), Fundació la Marató TV3 (76/C/2011—J. Jiménez-Conde) and Recercaixa’13 (JJ086116—J. Jiménez-Conde), and Wistron Corp (P. Golland). This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement 753896 (M.D. Schirmer). R.L. is a senior clinical investigator of FWO Flanders.

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.