Browse the corpus

In acute ischemic stroke, the relationship between regional cerebral blood flow (CBF) and tissue injury has led to the widespread acceptance of perfusion-weighted imaging to define tissue at risk of infarction.1,2 Perfusion-weighted imaging, using a variety of measures, has shown promise in predicting tissue outcome3,4 and has led to the selection of patients for clinical trials based on presenting perfusion characteristics.1

ed to the widespread acceptance of perfusion-weighted imaging to define tissue at risk of infarction.1,2 Perfusion-weighted imaging, using a variety of measures, has shown promise in predicting tissue outcome3,4 and has led to the selection of patients for clinical trials based on presenting perfusion characteristics.1 The majority of studies using perfusion imaging have relied on metrics derived from contrast-based magnetic resonance imaging (MRI) or computed tomography, such as time-to-peak or Tmax, which correlate with CBF.1,5 These biomarkers are used in clinical trials and clinical decision making,4,6 but their relationship to pathophysiological processes is not always clear.7 Arterial spin labeling (ASL) is increasingly used to measure perfusion in the context of acute ischemic stroke with the advantages that it is noninvasive, can be performed serially without repeated contrast administration without reference to renal function, and allows absolute CBF quantification.8,9 ASL is validated for measuring perfusion in acute stroke and compares favorably to conventional techniques.9,10 Multiple postlabeling delay (PLD) ASL has been less widely used in stroke-imaging studies but improves absolute quantification of CBF, even in the presence of delayed blood arrival times.10,11 The addition of vessel encoding attributes signal to the different feeding arteries and has the potential to further improve accuracy of CBF quantification in regions supplied by multiple arteries without compromising signal-to-noise ratio.11

ntification of CBF, even in the presence of delayed blood arrival times.10,11 The addition of vessel encoding attributes signal to the different feeding arteries and has the potential to further improve accuracy of CBF quantification in regions supplied by multiple arteries without compromising signal-to-noise ratio.11 After the advent of thrombectomy, there is a need to improve our understanding of the relationships between vessel status, reperfusion, and tissue outcome.12 This study aimed to investigate the relationship between changes in absolute CBF and tissue outcome in patients with acute ischemic stroke. Methods Patients and Volunteers Six healthy volunteers were recruited and imaged under an agreed technical development protocol approved by the institution’s Research Governance Office. Patients with ischemic stroke were recruited into a prospective observational cohort study regardless of age or stroke severity under research protocols agreed by the UK National Research Ethics Service committees (refs: 12/SC/0292 and 13/SC/0362). Inclusion criteria for this analysis were the following: DWI lesion on presenting scan, presenting scan within 18 hours of symptom onset, patient or representative able to give a clear medical history and participate in the consent process, and age >18 years. Patients with a contraindication to MRI, lacunar stroke defined on DWI, or severely impaired conscious level (score >1 on question 1a of the National Institute for Health Stroke Scale) were excluded.

atient or representative able to give a clear medical history and participate in the consent process, and age >18 years. Patients with a contraindication to MRI, lacunar stroke defined on DWI, or severely impaired conscious level (score >1 on question 1a of the National Institute for Health Stroke Scale) were excluded. Imaging All scans were acquired using a 3.0T Siemens Verio scanner (Siemens Healthcare, Erlangen, Germany). CBF was measured using a multiple PLD vessel-encoded pseudocontinuous ASL (VEPCASL) sequence.11 The labeling plane was positioned ≈8 cm below the level of the circle of Willis, through the proximal V3 segment of the vertebral arteries. A single-shot echo-planar imaging readout was used, acquiring 24 slices sequentially from inferior to superior to give whole brain coverage (repetition time [TR]=4080 ms, echo time [TE]=14 ms, voxel size=3.4×3.4×4.5 mm). Volumes were acquired after a range of 6 PLDs (0.25, 0.5, 0.75, 1, 1.25, and 1.5 seconds). Vessel encoding of the 4 arteries in the labeling plane was achieved by means of 8 paired encoding cycles in a range of orientations.11,13 Calibration scans were acquired using identical parameters to the VEPCASL sequence but without ASL or background suppression applied, to allow absolute CBF quantification and to correct for uneven spatial sensitivity. Total acquisition time for the ASL data was 5 minutes and 55 seconds.

in a range of orientations.11,13 Calibration scans were acquired using identical parameters to the VEPCASL sequence but without ASL or background suppression applied, to allow absolute CBF quantification and to correct for uneven spatial sensitivity. Total acquisition time for the ASL data was 5 minutes and 55 seconds. Retrospective motion correction was applied using the MCFLIRT tool found in the Oxford Centre for Function MRI of the Brain (FMRIB)'s Software Library (FSL).14,15 Calculation of the perfusion signals arriving from each artery at each PLD was achieved using a maximum a posteriori approach to the general Bayesian framework for vessel-encoded data.13,16 A kinetic curve was fitted to the data for each feeding artery separately to estimate CBF and bolus arrival time with a corresponding variance using a variational Bayes approach.8,17 The resulting vessel-specific CBF maps were summed to generate a map of the total CBF from all arteries. Other scanning protocols included diffusion-weighted imaging (3 directions, 1.8×1.8×2.0 mm, field of view [FoV]=240 mm, 4 averages, b=0 and 1000 s/mm2, TR=9000 ms, TE=98 ms, 50 slices, and acquisition time=2 minutes and 53 seconds) with apparent diffusion coefficient calculation; T1-weighted structural imaging (magnetization prepared rapid acquisition gradient echo, 1.8×1.8×1.0 mm, FoV=228 mm, TR=2040 ms, TE=4.55 ms, and acquisition time=3 minutes and 58 seconds); and T2-weighted turbo spin echo fluid attenuated inversion recovery (1.9×1.9×2.0 mm, FoV=240 mm, TR=9000 ms, and TE=96 ms).

ulation; T1-weighted structural imaging (magnetization prepared rapid acquisition gradient echo, 1.8×1.8×1.0 mm, FoV=228 mm, TR=2040 ms, TE=4.55 ms, and acquisition time=3 minutes and 58 seconds); and T2-weighted turbo spin echo fluid attenuated inversion recovery (1.9×1.9×2.0 mm, FoV=240 mm, TR=9000 ms, and TE=96 ms). Healthy volunteers were scanned 3 times—at time 0, at 24 hours, and at 1 week—and underwent 4 repetitions of the VEPCASL sequence at each scan time. Patients were imaged at presentation, 2 hours, 24 hours, 1 week (3–9 days), and 1 month (14–42 days), whenever possible. Acute scans were defined where the presenting MRI scans were acquired within 6 hours of symptom onset. When intravenous thrombolysis was administered, the first MRI scan occurred during the infusion of alteplase. Image Registration Rigid body registration using FMRIB’s Linear Image Registration Tool (FLIRT) was used for within time point registration.14 Nonlinear registration of structural scans was used between time points to limit potential error introduced by subacute edema.15 Contralateral nonischemic masks were created after registration of the perfusion deficit masks to standard (MNI152) space, before reflection and registration back to native image space. Definitions and Regions of Interest In healthy volunteers, 6 regions of interest (ROIs), which were evenly distributed throughout the cerebral cortex, were derived from the Harvard-Oxford Atlas and registered into ASL image space. The regions chosen were insula, lateral occipital, middle temporal, paracingulate, postcentral, and precentral.18

of Interest In healthy volunteers, 6 regions of interest (ROIs), which were evenly distributed throughout the cerebral cortex, were derived from the Harvard-Oxford Atlas and registered into ASL image space. The regions chosen were insula, lateral occipital, middle temporal, paracingulate, postcentral, and precentral.18 In stroke patients, infarction at presentation was defined using semiautomated delineation of apparent diffusion coefficient below an externally validated threshold of 620×10−6 mm2/s.19 At 24 hours, infarction was manually delineated using trace DWI (b=1000 s/mm2) and at 1 week using T2-weighted fluid attenuated inversion recovery imaging. ROIs were defined as follows: Ischemic core: within both presenting and final infarct definitions. Infarct growth: within the final infarct but not within the presenting infarct, which was further divided into: i. Early infarct growth (infarct growth within the trace DWI infarct at 24 hours but not in the ischemic core); and ii. Late infarct growth (infarct growth not within the 24-hour lesion, but within the final infarct). Peri-infarct: tissue that survived adjacent to the final infarct (within a dilated infarct mask [using a 3×3×3 voxel kernel], but not within the final infarct itself).

i. Early infarct growth (infarct growth within the trace DWI infarct at 24 hours but not in the ischemic core); and ii. Late infarct growth (infarct growth not within the 24-hour lesion, but within the final infarct). Peri-infarct: tissue that survived adjacent to the final infarct (within a dilated infarct mask [using a 3×3×3 voxel kernel], but not within the final infarct itself). Within the ischemic core, diffusion lesion pseudonormalization was defined as regions within this ROI that had renormalized apparent diffusion coefficient values by 24 hours (>620×10−6 mm2/s). Patients were divided by reperfusion status using the Modified Treatment in Cerebral Ischemia Scale, with Modified Treatment in Cerebral Ischemia grades 2b and 3 representing reperfusion.4,20 Tissue segmentation of the presenting structural T1-weighted image using FMRIB’s Automated Segmentation Tool (FAST) defined gray matter partial volume estimates,21 which were registered into perfusion image space. Analyses were performed within gray matter masks with a partial volume estimate of ≥50%, where ASL CBF estimates are more reliable. Correlation With Clinical Outcomes To investigate the relationship between perfusion dynamics and clinical recovery, changes in CBF from presentation to the 24-hour and 1-week time points within the presenting perfusion deficit were compared with changes in National Institute for Health Stroke Scale scores. Perfusion deficits were defined using a threshold of 20 mL/100 g/min to guide manual delineation of the region.

inical recovery, changes in CBF from presentation to the 24-hour and 1-week time points within the presenting perfusion deficit were compared with changes in National Institute for Health Stroke Scale scores. Perfusion deficits were defined using a threshold of 20 mL/100 g/min to guide manual delineation of the region. Statistics Means and SD of CBF within the gray matter voxels of each ROI were extracted, and patient and voxelwise means and 95% confidence intervals were calculated. Repeatability within healthy volunteers and the contralateral ROIs of patients was quantified using the coefficient of variation (SD/mean) and ANOVA. Receiver-operating characteristic curve analyses were performed to determine the utility of using VEPCASL-derived CBF at acute presentation (within 6 hours of symptom onset) in predicting infarction. Youden indices were used to estimate optimum CBF thresholds.22 Optimum CBF values to predict final infarction were used to estimate thresholds of ischemic core (infarct growth in reperfusers) and tissue at risk (infarct growth in nonreperfusers). To explore the effects of time of imaging on ability of CBF to predict infarction, patients were further divided into those imaged 0 to 3 and 3 to 6 hours. Results Six healthy volunteers and 40 consecutive patients were prospectively enrolled and underwent serial VEPCASL imaging. Patient demographics are presented in Table. Table. Demographic Data

Statistics Means and SD of CBF within the gray matter voxels of each ROI were extracted, and patient and voxelwise means and 95% confidence intervals were calculated. Repeatability within healthy volunteers and the contralateral ROIs of patients was quantified using the coefficient of variation (SD/mean) and ANOVA. Receiver-operating characteristic curve analyses were performed to determine the utility of using VEPCASL-derived CBF at acute presentation (within 6 hours of symptom onset) in predicting infarction. Youden indices were used to estimate optimum CBF thresholds.22 Optimum CBF values to predict final infarction were used to estimate thresholds of ischemic core (infarct growth in reperfusers) and tissue at risk (infarct growth in nonreperfusers). To explore the effects of time of imaging on ability of CBF to predict infarction, patients were further divided into those imaged 0 to 3 and 3 to 6 hours. Results Six healthy volunteers and 40 consecutive patients were prospectively enrolled and underwent serial VEPCASL imaging. Patient demographics are presented in Table. Table. Demographic Data Healthy Volunteers The average gray matter perfusion of healthy volunteers was 62±15 mL/100 g/min (mean±SD), range 40 to 78 mL/100 g/min. The coefficient of variation between the 4 repeated measures of CBF at the same scan time point was 8.5%. Between scan time points, the coefficient of variation was 9.7% (ANOVA, P<0.001), was 16% between individuals (P<0.001), and was 15% between atlas regions (P<0.001). Regional CBF ranged from 50±4 mL/100 g/min in the post central gyrus to 73±8 mL/100 g/min in the insula. Data are presented in Figure I in the online-only Data Supplement.

points, the coefficient of variation was 9.7% (ANOVA, P<0.001), was 16% between individuals (P<0.001), and was 15% between atlas regions (P<0.001). Regional CBF ranged from 50±4 mL/100 g/min in the post central gyrus to 73±8 mL/100 g/min in the insula. Data are presented in Figure I in the online-only Data Supplement. CBF Variability in Patients Within the contralateral ROIs of patients, the weighted mean CBF was 52±42 mL/100 g/min (Figure 1). For patients with both presenting and 24-hour scans, 2-way ANOVA demonstrated a significant effect of the individual patient on the contralateral ROI values (ANOVA, 77% of variance, P=0.01) but a nonsignificant effect of the day of the scan (ANOVA, 8% of variance, P=0.06). Dynamic CBF values from the contralateral ROIs of individual patients are presented in Figure II in the online-only Data Supplement. There was no effect of tissue-type plasminogen activator infusion or blood pressure on contralateral ROI values (P=0.95 and P=0.61). Figure 1. Upper: Voxelwise mean cerebral blood flow (CBF) in each region of interest within 6 h of symptom onset (error bars, 95% confidence interval). Lower: Patient-level mean CBF in each region of interest (whisker plot). ****P<0.0001; *P<0.05.

CBF Variability in Patients Within the contralateral ROIs of patients, the weighted mean CBF was 52±42 mL/100 g/min (Figure 1). For patients with both presenting and 24-hour scans, 2-way ANOVA demonstrated a significant effect of the individual patient on the contralateral ROI values (ANOVA, 77% of variance, P=0.01) but a nonsignificant effect of the day of the scan (ANOVA, 8% of variance, P=0.06). Dynamic CBF values from the contralateral ROIs of individual patients are presented in Figure II in the online-only Data Supplement. There was no effect of tissue-type plasminogen activator infusion or blood pressure on contralateral ROI values (P=0.95 and P=0.61). Figure 1. Upper: Voxelwise mean cerebral blood flow (CBF) in each region of interest within 6 h of symptom onset (error bars, 95% confidence interval). Lower: Patient-level mean CBF in each region of interest (whisker plot). ****P<0.0001; *P<0.05. CBF in the Ischemic Hemisphere of Patients At a voxel level, analysis demonstrated a graduated severity of hypoperfusion within 6 hours of stroke onset (Figure 1, top). Ischemic core voxels had a lower mean perfusion at presentation (mean±SD, 17±23 mL/100 g/min) than that in the regions of early and late infarct growth (21±26 and 25±35 mL/100 g/min, respectively; ANOVA, P<0.0001). Mean CBF in the peri-infarct ROI (29±30 mL/100 g/min) was less than that in the contralateral hemisphere (52±42 mL/100 g/min; t test, P<0.0001) but less severely hypoperfused than ROIs that infarcted (t test, P=0.002). Within the ischemic core, CBF values within regions of diffusion lesion pseudonormalization had a lower CBF than the ischemic core as a whole (8±9 mL/100 g/min; t test, P<0.0001).

e contralateral hemisphere (52±42 mL/100 g/min; t test, P<0.0001) but less severely hypoperfused than ROIs that infarcted (t test, P=0.002). Within the ischemic core, CBF values within regions of diffusion lesion pseudonormalization had a lower CBF than the ischemic core as a whole (8±9 mL/100 g/min; t test, P<0.0001). At a patient level, mean CBF values showed a similar pattern to mean voxelwise ROI means (Figure 1, bottom; ANOVA, P=0.02). The variation in CBF values within and between a given type of ROI resulted in considerable overlap of values measured. Receiver-operating characteristic curve analysis of CBF generated an area under the curve (AUC) of 0.71 for predicting final infarct using all patients scanned within 6 hours. Using only those patients with demonstrated reperfusion to predict the ischemic core generated a CBF threshold, defined by the Youden analysis, of 22 mL/100 g/min and AUC of 0.75. The optimum threshold for final infarction in nonreperfusers was 25 mL/100 g/min with an AUC of 0.72 (Figure III in the online-only Data Supplement). Subgroups of patients who were imaged at 0 to 3 and 3 to 6 hours had similar optimum thresholds and AUCs to the larger 0 to 6 hours cohort, except those who reperfused and were imaged before 3 hours, when the optimum threshold was 14 mL/100 g/min with an AUC of 0.76 (Table I in the online-only Data Supplement).

ement). Subgroups of patients who were imaged at 0 to 3 and 3 to 6 hours had similar optimum thresholds and AUCs to the larger 0 to 6 hours cohort, except those who reperfused and were imaged before 3 hours, when the optimum threshold was 14 mL/100 g/min with an AUC of 0.76 (Table I in the online-only Data Supplement). Serial CBF Measures There was marked heterogeneity in the patterns of perfusion within identically defined ROIs between patients (Figure 2). Both sustained ischemia and reperfusion to varying degrees were seen in the ischemic core and early and late infarct growth ROIs. The only uniform pattern was seen in peri-infarct ROIs, which by definition survived, where all patients demonstrated a CBF value of >20 mL/100 g/min by 24 hours. Examples of the dynamic changes in CBF within individual patients with differing degrees of reperfusion, together with example CBF data, can be seen in Figures 3 through 5. Where hyperemia was seen post reperfusion at 24 hours and 1 week, the CBF had returned to low or normal levels in all patient ROIs at 1 month (Figures 2 and 4). Figure 2. Serial cerebral blood flow (CBF) values from different regions of interest at a patient level (presentation within 6 h of symptom onset). Figure 3. Upper: Example cerebral blood flow (CBF) maps from a patient without reperfusion at presentation and 24 h, with superimposed regions of interest below (red, ischemic core and blue, infarct growth). Lower: Absolute CBF quantification in the 4 regions of interest (red, ischemic core; blue, infarct growth; green, peri-infarct; and black, contralateral).

ow (CBF) maps from a patient without reperfusion at presentation and 24 h, with superimposed regions of interest below (red, ischemic core and blue, infarct growth). Lower: Absolute CBF quantification in the 4 regions of interest (red, ischemic core; blue, infarct growth; green, peri-infarct; and black, contralateral). Figure 4. Upper: Example cerebral blood flow (CBF) maps from a patient with delayed reperfusion and localized hyperemia at presentation, 24 h, and 1 wk, with superimposed regions of interest below (red, ischemic core and blue, infarct growth). Lower: Absolute CBF quantification in the 4 regions of interest (red, ischemic core; blue, infarct growth; green, peri-infarct; and black, contralateral). Figure 5. Upper: Example cerebral blood flow (CBF) maps from a patient with partial reperfusion at presentation, 24 h, and 1 wk, with superimposed regions of interest below (red, ischemic core and blue, infarct growth). Lower: Absolute CBF quantification in the 4 regions of interest (red, ischemic core; blue, infarct growth; green, peri-infarct; black, contralateral). Correlation With Clinical Outcomes Increases in CBF within the presenting perfusion deficit by 1 week predicted a reduction in National Institute for Health Stroke Scale score (r2=54%; P=0.03). No correlation was seen between presentation and 24 hours (P=0.9).

Figure 5. Upper: Example cerebral blood flow (CBF) maps from a patient with partial reperfusion at presentation, 24 h, and 1 wk, with superimposed regions of interest below (red, ischemic core and blue, infarct growth). Lower: Absolute CBF quantification in the 4 regions of interest (red, ischemic core; blue, infarct growth; green, peri-infarct; black, contralateral). Correlation With Clinical Outcomes Increases in CBF within the presenting perfusion deficit by 1 week predicted a reduction in National Institute for Health Stroke Scale score (r2=54%; P=0.03). No correlation was seen between presentation and 24 hours (P=0.9). Discussion Knowledge of perfusion dynamics partially explained tissue outcome in this cohort of patients with acute ischemic stroke. At presentation, mean voxelwise CBF measured using multiple PLD VEPCASL was consistent with tissue outcome in the ROIs of patients. The absolute CBF values were comparable to those derived using contrast and radiolabeled techniques in both healthy individuals and patients with stroke.23–26 A biologically plausible pattern was observed: ROIs that underwent infarction earliest had a lower presenting CBF than those of later infarct growth. For example, regions of ischemic core had the lowest blood flow at presentation, but within this ROI, the voxels that underwent diffusion pseudonormalization had lower still presenting CBF values. This phenomenon is thought to represent facilitated diffusion from early vasogenic edema and a more severe injury,27,28 and the CBF in this ROI was similar to cited thresholds of membrane failure (8 mL/100 g/min).29

this ROI, the voxels that underwent diffusion pseudonormalization had lower still presenting CBF values. This phenomenon is thought to represent facilitated diffusion from early vasogenic edema and a more severe injury,27,28 and the CBF in this ROI was similar to cited thresholds of membrane failure (8 mL/100 g/min).29 Receiver-operating characteristic curve analyses at presentation demonstrated that the ability of CBF to predict final infarction was only fair. CBF thresholds were similar to those estimated using contrast techniques and with equivalent AUCs.30,31 There was only small improvement in the AUCs when those with and without reperfusion were considered separately, implying that knowledge of presenting CBF and a final reperfusion status is not sufficient to predict tissue outcome. Subdividing patients into those imaged before and after 3 hours did not improve CBF prediction of infarction but did identify a lower threshold for defining ischemic core in patients imaged before 3 hours. This is consistent with preclinical data describing increasing CBF thresholds for infarction over time.32 Factors that confound using a single CBF measurement to predict tissue fate include duration and degree of previous hypoperfusion, dynamics of reperfusion, tissue type, previous exposure to ischemia, ischemia–reperfusion injury, and individual susceptibility to hypoperfusion,33–36 along with measurement errors such as residual motion artifacts and insensitivity to delayed blood arrival.

e fate include duration and degree of previous hypoperfusion, dynamics of reperfusion, tissue type, previous exposure to ischemia, ischemia–reperfusion injury, and individual susceptibility to hypoperfusion,33–36 along with measurement errors such as residual motion artifacts and insensitivity to delayed blood arrival. At a patient level, this finding was borne out by the marked overlap of presenting CBF values in different ROIs, demonstrated in Figure 1. A similar pattern of mean CBF values to that of the voxelwise analysis was seen across the different tissue outcomes. Again, a single CBF measurement to predict tissue outcome at presentation in an individual is not sufficient, even with knowledge of final perfusion status.

in different ROIs, demonstrated in Figure 1. A similar pattern of mean CBF values to that of the voxelwise analysis was seen across the different tissue outcomes. Again, a single CBF measurement to predict tissue outcome at presentation in an individual is not sufficient, even with knowledge of final perfusion status. The predominant source of variability in the measurement of CBF is the variation seen between individuals, as opposed to temporal variation within individuals and other factors including noise. This individual variation may reflect differences in age,26 hypertension,37 and burden of preexisting cerebrovascular disease.26 This has implications when quantifying CBF values relative to the contralateral hemisphere. Alternatively, systematic measurement bias that remained consistent across time points, such as anatomic factors affecting labeling efficiency, tendency to move in the scanner, and variations in vascular anatomy, may have explained some of the observed interindividual variation. Although there was some variation over time in the contralateral ROIs of patients, CBF did not vary systematically between the presenting scan and at 24 hours, which might have been expected if diaschisis had a significant effect.38,39

vascular anatomy, may have explained some of the observed interindividual variation. Although there was some variation over time in the contralateral ROIs of patients, CBF did not vary systematically between the presenting scan and at 24 hours, which might have been expected if diaschisis had a significant effect.38,39 Analysis in healthy volunteers showed small, but statistically significant, variations in CBF between individuals, day of scan, and region of the brain, comparable to other work.40 These variations in healthy CBF values may explain some of the variations seen in patients. However, the magnitude of the variation in healthy volunteers was low and well within the recommended limits for variation in ASL studies (20%),41 and less than that seen in contrast-based perfusion MRI.42

rable to other work.40 These variations in healthy CBF values may explain some of the variations seen in patients. However, the magnitude of the variation in healthy volunteers was low and well within the recommended limits for variation in ASL studies (20%),41 and less than that seen in contrast-based perfusion MRI.42 Serial perfusion data within individuals emphasized that perfusion dynamics are consistent with tissue outcome in some, but not all, patients. Data in Figure 2 from identical ROIs across patients show that reperfusion characteristics are heterogeneous. In the peri-infarct ROIs, which represent tissue that survives, CBF consistently recovers by 24 hours to a greater level than the thresholds for infarction. However, reperfusion is not sufficient for survival of ischemic tissue outside the ischemic core, demonstrated by the reperfusion observed in regions of infarct growth. It is likely that only early reperfusion will allow tissue survival, but even this may not be sufficient for recovery. Delayed tissue injury despite recanalization is well described,43 and measuring serial absolute tissue perfusion may help to differentiate no-reflow phenomena from ischemia–reperfusion injury.44 Increases in CBF within the presenting perfusion deficit correlated with clinical recovery at 1 week but not at 24 hours. The lack of correlation at 24 hours may be because the study is too small for these patient-level analyses. Additionally, the correlation at 1 week may have been emphasized by hyperemia in reperfusers at 1 week (Figure 4).

the presenting perfusion deficit correlated with clinical recovery at 1 week but not at 24 hours. The lack of correlation at 24 hours may be because the study is too small for these patient-level analyses. Additionally, the correlation at 1 week may have been emphasized by hyperemia in reperfusers at 1 week (Figure 4). All ASL studies are limited by the challenges of measuring late arriving blood and white matter perfusion. Even after using multiple PLD ASL techniques, it is not possible to distinguish late arriving blood from voxels with minimal CBF. Combined with effects of partial volume contamination from white matter and CSF, this may lead to systematical underestimation of CBF. Other limitations include data loss in this study, a combination of motion artifact and loss to follow-up, which is more likely to occur in severe stroke syndromes. Advances using prospective motion correction, improved acquisition techniques, and shorter imaging times may mitigate some of these sources of error in the future. Conclusions This study explores the relationship between tissue-level perfusion and highly characterized tissue outcome in patients with acute ischemic stroke. Without using exogenous contrast, CBF values were derived that were consistent with other more invasive techniques. The ability to acquire serial data highlighted the heterogeneity of perfusion characteristics between individuals and the need for complementary information, including tissue susceptibility and metabolism, to fully understand tissue fate in acute stroke.

derived that were consistent with other more invasive techniques. The ability to acquire serial data highlighted the heterogeneity of perfusion characteristics between individuals and the need for complementary information, including tissue susceptibility and metabolism, to fully understand tissue fate in acute stroke. Acknowledgments We wish to acknowledge the facilities provided by the Oxford Acute Vascular Imaging Centre and the staff of the Oxford Acute Stroke Programme. Sources of Funding This study was supported by the National Institute for Health Research Oxford Biomedical Research Centre Programme, the National Institute for Health Research Clinical Research Network, the Dunhill Medical Trust (grant number: OSRP1/1006), the Royal Academy of Engineering, and the Centre of Excellence for Personalized Healthcare funded by the Wellcome Trust and Engineering and Physical Sciences Research Council (grant number WT088877/Z/09/Z). Disclosures Dr Ford has received grant support from the National Institute of Health Research. Dr Ford has also received personal remuneration for advisory work from Astra Zeneca, Daiichi Sankyo, and Pfizer. Dr Ford has received lecture fees from Medtronic. Dr Chappell has received royalties for commercial licenses from the FMRIB software library. Dr Chappell and Dr Okell have received royalties from commercial licenses from Siemens from the vessel-encoding image-processing software. The other authors report no conflicts. Supplementary Material * G.W.J. Harston and T.W. Okell contributed equally.

Disclosures Dr Ford has received grant support from the National Institute of Health Research. Dr Ford has also received personal remuneration for advisory work from Astra Zeneca, Daiichi Sankyo, and Pfizer. Dr Ford has received lecture fees from Medtronic. Dr Chappell has received royalties for commercial licenses from the FMRIB software library. Dr Chappell and Dr Okell have received royalties from commercial licenses from Siemens from the vessel-encoding image-processing software. The other authors report no conflicts. Supplementary Material * G.W.J. Harston and T.W. Okell contributed equally. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.116.014707/-/DC1.

Patients with diabetes mellitus are at increased risk of cardiovascular events and cardiovascular mortality.1 The risk of stroke in patients with diabetes mellitus is increased 2-fold compared with individuals without diabetes mellitus1; the risk of recurrent stroke is also increased.2 Trials of intensive glucose-lowering3 or of specific glucose-lowering agents,4–7 with the exception of pioglitazone8 and semaglutide,9 have not been shown to significantly reduce the risk of stroke in patients with type 2 diabetes mellitus even after prolonged follow-up.

the risk of recurrent stroke is also increased.2 Trials of intensive glucose-lowering3 or of specific glucose-lowering agents,4–7 with the exception of pioglitazone8 and semaglutide,9 have not been shown to significantly reduce the risk of stroke in patients with type 2 diabetes mellitus even after prolonged follow-up. Empagliflozin is a potent and selective inhibitor of SGLT2 (sodium glucose cotransporter 2) used in the treatment of type 2 diabetes mellitus. In the EMPA-REG OUTCOME trial (Empagliflozin Cardiovascular Outcome Event Trial in Type 2 Diabetes Mellitus Patients) in patients with type 2 diabetes mellitus and high cardiovascular risk, empagliflozin added to standard of care significantly reduced the risk of the primary outcome 3-point major adverse cardiovascular events (the composite of cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke; hazard ratio [HR], 0.86; 95.02% confidence interval [CI], 0.74–0.99; P=0.04).10 This was driven primarily by a reduction in the risk of cardiovascular death (HR, 0.62; 95% CI, 0.49–0.77; P<0.001). There were no significant differences between empagliflozin and placebo in the risk of myocardial infarction (HR, 0.87; 95% CI, 0.70–1.09; P=0.23) or stroke (HR, 1.18; 95% CI, 0.89–1.56; P=0.26).10 Given the importance of stroke prevention in patients with type 2 diabetes mellitus and the numeric difference in the proportion of patients with stroke events between the empagliflozin and placebo groups in the EMPA-REG OUTCOME trial, we performed a comprehensive analysis of cerebrovascular events in EMPA-REG OUTCOME, including sensitivity and subgroup analyses.

patients with type 2 diabetes mellitus and the numeric difference in the proportion of patients with stroke events between the empagliflozin and placebo groups in the EMPA-REG OUTCOME trial, we performed a comprehensive analysis of cerebrovascular events in EMPA-REG OUTCOME, including sensitivity and subgroup analyses. Methods Study Design The design of EMPA-REG OUTCOME has been described.10,11 Briefly, the study population comprised patients with type 2 diabetes mellitus, established cardiovascular disease, and estimated glomerular filtration rate (MDRD [Modification of Diet in Renal Disease] equation) >30 mL min−1 1.73 m−2. Patients were randomized 1:1:1 to receive empagliflozin 10 mg, empagliflozin 25 mg, or placebo in addition to standard of care. Throughout the trial (or after week 12 for glucose-lowering medication), investigators were encouraged to treat cardiovascular risk factors to achieve optimal standard of care according to local guidelines. Patients were asked to attend the clinic at prespecified times, including a follow-up visit 30 days after the end of treatment. The trial was to continue until ≥691 patients had experienced an adjudicated event included in the primary outcome. Patients who prematurely discontinued study medication continued to be followed for ascertainment of cardiovascular outcomes, adverse events, and vital status.

up visit 30 days after the end of treatment. The trial was to continue until ≥691 patients had experienced an adjudicated event included in the primary outcome. Patients who prematurely discontinued study medication continued to be followed for ascertainment of cardiovascular outcomes, adverse events, and vital status. The trial was conducted in accordance with the principles of the Declaration of Helsinki and the International Conference on Harmonization Good Clinical Practice guidelines and was approved by local authorities. An independent ethics committee or institutional review board approved the clinical protocol at every participating center. All patients provided written informed consent before study entry.

lsinki and the International Conference on Harmonization Good Clinical Practice guidelines and was approved by local authorities. An independent ethics committee or institutional review board approved the clinical protocol at every participating center. All patients provided written informed consent before study entry. Outcomes Definitions of the major clinical outcomes in EMPA-REG OUTCOME have been described.10 The definitions of transient ischemic attack (TIA) and stroke are provided in the online-only Data Supplement. Cardiovascular outcome events and deaths were prospectively adjudicated by 2 Clinical Events Committees (for cardiac and neurological events). We assessed time to first stroke (fatal or nonfatal), time to fatal stroke, time to first nonfatal stroke, time to first nonfatal disabling stroke (defined as adjudicated nonfatal stroke with investigator-reported seriousness criterion of persistent or significant disability/incapacity; stroke disability scores were not used), recurrent stroke during the trial, time to first nonfatal disabling stroke or fatal stroke, time to first cardiovascular death or nonfatal stroke, time to first TIA, and time to first nonfatal or fatal stroke or TIA. Ischemic stroke was classified post hoc according to TOAST criteria (Trial of Org 10 172 in Acute Stroke Treatment)12 by the neurological Clinical Events Committee. The Clinical Events Committee charter for classification of ischemic stroke is provided in the online-only Data Supplement.

onfatal or fatal stroke or TIA. Ischemic stroke was classified post hoc according to TOAST criteria (Trial of Org 10 172 in Acute Stroke Treatment)12 by the neurological Clinical Events Committee. The Clinical Events Committee charter for classification of ischemic stroke is provided in the online-only Data Supplement. Analyses It was prespecified that analyses would compare the pooled empagliflozin dose groups versus placebo. Outcomes were analyzed using a modified intent-to-treat approach in the treated set (patients treated with ≥1 dose of study drug), using the time to first stroke event irrespective of whether another outcome event had occurred. Data for patients who did not have an event were censored on the last day they were known to be free of the outcome. Sensitivity analyses of fatal or nonfatal stroke were performed based on events that occurred during treatment or ≤90, ≤30, or ≤7 days after a patient’s last intake of study drug (treated set plus 90 days, treated set plus 30 days, and treated set plus 7 days) and based on events that occurred during treatment or ≤30 days after a patient’s last intake of study drug in patients who received ≥30 days of study medication (cumulative; on-treatment set). A sensitivity analysis of TIA was performed on the treated set plus 90 days. Analyses were based on a Cox proportional hazards model, with treatment, age, sex, baseline body mass index, baseline HbA1c, baseline estimated glomerular filtration rate, and region as factors. Subgroup analyses included a subgroup factor and a treatment-by-subgroup factor interaction as additional effects. All analyses were performed at a nominal level of α=0.05 2-sided without adjustment for multiplicity. Cumulative incidence function estimates were corrected for death as a competing risk. Because of the declining numbers of patients at risk, cumulative incidence plots have been truncated at 48 months.

ditional effects. All analyses were performed at a nominal level of α=0.05 2-sided without adjustment for multiplicity. Cumulative incidence function estimates were corrected for death as a competing risk. Because of the declining numbers of patients at risk, cumulative incidence plots have been truncated at 48 months. The percentages of patients with recurrent stroke were analyzed descriptively in the treated set. In addition, analyses were conducted of the percentages of patients with stroke in patients with maximum decreases from baseline in systolic blood pressure ≥30 and <30 mm Hg, with maximum increases from baseline in hematocrit ≥90th and <90th percentiles, in patients who had an event consistent with volume depletion (based on 8 preferred terms in the Medical Dictionary for Regulatory Activities) and in patients who had an atrial fibrillation event (based on the Medical Dictionary for Regulatory Activities preferred term). Changes from baseline in systolic blood pressure and hematocrit at the last value on treatment and at follow-up were analyzed descriptively. Prespecified analyses were the modified intent-to-treat analyses and analyses in the on-treatment set for time to first fatal or nonfatal stroke, nonfatal stroke and TIA, and the assessment of recurrent strokes. Other analyses were post hoc.

The percentages of patients with recurrent stroke were analyzed descriptively in the treated set. In addition, analyses were conducted of the percentages of patients with stroke in patients with maximum decreases from baseline in systolic blood pressure ≥30 and <30 mm Hg, with maximum increases from baseline in hematocrit ≥90th and <90th percentiles, in patients who had an event consistent with volume depletion (based on 8 preferred terms in the Medical Dictionary for Regulatory Activities) and in patients who had an atrial fibrillation event (based on the Medical Dictionary for Regulatory Activities preferred term). Changes from baseline in systolic blood pressure and hematocrit at the last value on treatment and at follow-up were analyzed descriptively. Prespecified analyses were the modified intent-to-treat analyses and analyses in the on-treatment set for time to first fatal or nonfatal stroke, nonfatal stroke and TIA, and the assessment of recurrent strokes. Other analyses were post hoc. Results Study Population A total of 7020 patients at 590 sites in 42 countries received ≥1 dose of study drug. Baseline characteristics of the study population have been described.10 Briefly, mean (SD) age was 63.1 (8.6) years, mean (SD) body mass index was 30.6 (5.3) kg/m2, 71.5% were male, 25.9% had estimated glomerular filtration rate <60 mL min−1 1.73 m−2, 39.6% had microalbuminuria or macroalbuminuria, 38.7% had systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg, 23.3% had a history of stroke, 5.5% of patients had atrial fibrillation, 89.1% were taking anticoagulant or antiplatelet therapies, and 82.7% were taking acetylsalicylic acid.10,13 In total, 97% of patients completed the study, and 25% prematurely discontinued study drug. The median duration of treatment was 2.6 years, and the median observation time was 3.1 years. Vital status was available for 99% of patients.

coagulant or antiplatelet therapies, and 82.7% were taking acetylsalicylic acid.10,13 In total, 97% of patients completed the study, and 25% prematurely discontinued study drug. The median duration of treatment was 2.6 years, and the median observation time was 3.1 years. Vital status was available for 99% of patients. Stroke and TIA During the trial, 3.0% (69/2333) of patients in the placebo group and 3.5% (164/4687) of patients in the empagliflozin group had ≥1 adjudicated fatal or nonfatal stroke. Ischemic stroke was reported in 2.7% and 3.2% of patients in the placebo and empagliflozin groups and hemorrhagic stroke in 0.3% and 0.2% of patients in these groups, respectively. A further 0.1% of patients in each group had a stroke for which the type was not assessable. There was no marked imbalance between the placebo and empagliflozin groups in any specific type of ischemic stroke. Cardioembolism was the most common type of ischemic stroke that could be determined (Table I in the online-only Data Supplement).

patients in each group had a stroke for which the type was not assessable. There was no marked imbalance between the placebo and empagliflozin groups in any specific type of ischemic stroke. Cardioembolism was the most common type of ischemic stroke that could be determined (Table I in the online-only Data Supplement). In the prespecified modified intent-to-treat analysis of time to first stroke, there was no significant difference between empagliflozin and placebo in the occurrence of stroke (HR, 1.18; 95% CI, 0.89–1.56; P=0.26).10 The cumulative incidence of time to first stroke is shown in Figure 1A. In sensitivity analyses based on events that occurred during treatment or ≤90, ≤30, or ≤7 days after the last dose of study drug, there was no significant difference in the occurrence of stroke between empagliflozin and placebo, and the HR moved toward unity compared with the modified intent-to-treat analysis (Figure I in the online-only Data Supplement). The numeric difference in the proportion of patients with stroke between the empagliflozin and placebo groups was largely driven by events that occurred >90 days after a patient’s last intake of trial medication (Figure 1B; Figure I in the online-only Data Supplement). Three patients treated with placebo and 18 patients treated with empagliflozin experienced their first stroke >90 days after the last intake of trial medication (of whom 1 patient in the placebo group and 11 patients in the empagliflozin group experienced their first stroke >1 year after the last intake of trial medication).

patients treated with placebo and 18 patients treated with empagliflozin experienced their first stroke >90 days after the last intake of trial medication (of whom 1 patient in the placebo group and 11 patients in the empagliflozin group experienced their first stroke >1 year after the last intake of trial medication). Figure 1. Time to first fatal or nonfatal stroke. A, Modified intent-to-treat analyses in the treated set; events observed from randomization to the end of the study in treated set (patients treated with ≥1 dose of study drug). B, Sensitivity analysis in treated set plus 90 days; events observed during treatment or ≤90 days after a patient’s last intake of trial medication in treated set (patients treated with ≥1 dose of study drug). Cumulative incidence function. Hazard ratios (HR) are based on Cox regression analyses. CI indicates confidence interval.

ity analysis in treated set plus 90 days; events observed during treatment or ≤90 days after a patient’s last intake of trial medication in treated set (patients treated with ≥1 dose of study drug). Cumulative incidence function. Hazard ratios (HR) are based on Cox regression analyses. CI indicates confidence interval. The proportion of patients with recurrent stroke during the trial was similar between the empagliflozin and placebo groups (13 [0.3%] and 8 [0.3%], respectively; Table II in the online-only Data Supplement). Nonfatal disabling stroke (based on investigator-reported seriousness criterion [not stroke disability scores]) was reported in 10 patients (0.2%) on empagliflozin and 6 patients (0.3%) on placebo (HR, 0.82; 95% CI, 0.30–2.26; P=0.70). Fatal stroke was reported in similar proportions of patients in the empagliflozin and placebo groups (0.3% and 0.5%, respectively; HR, 0.72; 95% CI, 0.33–1.55; P=0.40) as was the composite of nonfatal disabling stroke or fatal stroke (0.6% and 0.7%, respectively; HR, 0.81; 95% CI, 0.43–1.50; P=0.50; Figure 2). Figure 2. Time to first stroke, transient ischemic attack, and composite outcomes in modified intent-to-treat analyses. Cox regression analyses. Events from randomization to the end of the study in treated set (patients treated with ≥1 dose of study drug). Analyses were prespecified for time to first fatal or nonfatal stroke, time to first nonfatal stroke, and time to first transient ischemic attack. CI indicates confidence interval; and HR, hazard ratio.

yses. Events from randomization to the end of the study in treated set (patients treated with ≥1 dose of study drug). Analyses were prespecified for time to first fatal or nonfatal stroke, time to first nonfatal stroke, and time to first transient ischemic attack. CI indicates confidence interval; and HR, hazard ratio. As empagliflozin reduced the risk of cardiovascular death by 38%,10 the composite outcome of cardiovascular death or nonfatal stroke was analyzed to account for cardiovascular death as a competing risk. Empagliflozin significantly reduced the risk of this composite outcome (HR, 0.79; 95% CI, 0.66–0.94; P=0.009; Figure 2). There was no significant difference in the risk of TIA (HR, 0.85; 95% CI, 0.51–1.42; P=0.54) or the composite of stroke or TIA (HR, 1.05; 95% CI, 0.82–1.35; P=0.87) with empagliflozin versus placebo (Figure 2; Figure II in the online-only Data Supplement).

As empagliflozin reduced the risk of cardiovascular death by 38%,10 the composite outcome of cardiovascular death or nonfatal stroke was analyzed to account for cardiovascular death as a competing risk. Empagliflozin significantly reduced the risk of this composite outcome (HR, 0.79; 95% CI, 0.66–0.94; P=0.009; Figure 2). There was no significant difference in the risk of TIA (HR, 0.85; 95% CI, 0.51–1.42; P=0.54) or the composite of stroke or TIA (HR, 1.05; 95% CI, 0.82–1.35; P=0.87) with empagliflozin versus placebo (Figure 2; Figure II in the online-only Data Supplement). Subgroup Analyses In exploratory analyses of time to first stroke in >30 prespecified subgroups by baseline characteristics, analyses by region and HbA1c showed nominal heterogeneity at P<0.05 (with no adjustment for multiple tests; Figure 3; Table III in the online-only Data Supplement). Compared with the total population, the HR for stroke in patients in Europe was higher (2.04; 95% CI, 1.26–3.29; P value for interaction of treatment and region: 0.01; Table III in the online-only Data Supplement). Baseline characteristics, including background medications, were similar between treatment groups within a given region (Table IV in the online-only Data Supplement). However, there were small differences in baseline characteristics between patients in Europe and North America (Table IV in the online-only Data Supplement), including a greater proportion of patients in Europe with a history of stroke (Europe: 29.9% placebo, 25.7% empagliflozin; North America: 14.9% placebo, 18.1% empagliflozin). Despite this, patients treated with placebo had a markedly lower stroke rate in Europe than North America (7.8/1000 versus 15.2/1000 patient-years). This pattern was not observed in patients treated with empagliflozin (stroke event rates were 15.7/1000 and 12.3/1000 patient-years in Europe and North America, respectively). In contrast to the HRs for stroke, the HR for TIA in European patients was <1 (0.61; 95% CI, 0.27–1.35) and in North American patients was 1.22 (95% CI, 0.53–2.78; P value for interaction of treatment and region: 0.24).

e event rates were 15.7/1000 and 12.3/1000 patient-years in Europe and North America, respectively). In contrast to the HRs for stroke, the HR for TIA in European patients was <1 (0.61; 95% CI, 0.27–1.35) and in North American patients was 1.22 (95% CI, 0.53–2.78; P value for interaction of treatment and region: 0.24). Figure 3. Time to first stroke in subgroups defined by baseline characteristics. Post hoc Cox regression analyses. Events of fatal or nonfatal stroke observed from randomization to end of study in treated set (patients treated with ≥1 dose of study drug). Race: Black and Other not included in Cox regression as <14 patients with an event in these subgroups. Region: Africa not included in Cox regression as <14 patients with an event in this subgroup. Cardiovascular risk: no high cardiovascular risk not included in Cox regression as <14 patients with an event in this subgroup. Heart failure at baseline was based on narrow standardized Medical Dictionary for Regulatory Activities query cardiac failure. P value is for homogeneity of the treatment group difference among subgroups (test for group by covariate interaction) with no adjustment for multiple tests. P=0.054 for age. The size of the oval is proportional to the number of patients in the subgroup. ACE indicates angiotensin-converting enzyme; ARB, angiotensin receptor blocker; CI, confidence interval; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate (according to Modification of Diet in Renal Disease formula); HbA1c, glycated hemoglobin; HR, hazard ratio; and SBP, systolic blood pressure. *Plus Australia and New Zealand.

ing enzyme; ARB, angiotensin receptor blocker; CI, confidence interval; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate (according to Modification of Diet in Renal Disease formula); HbA1c, glycated hemoglobin; HR, hazard ratio; and SBP, systolic blood pressure. *Plus Australia and New Zealand. Compared with the total population, the HR for stroke in patients with baseline HbA1c ≥8.5% was higher (P value for interaction: 0.01; Table III in the online-only Data Supplement). An analysis by baseline HbA1c deciles showed no significant treatment-by-subgroup interaction (Figure III in the online-only Data Supplement). Subgroup analysis by risk factors for stroke such as previous stroke, atrial fibrillation, smoking, and hypertension at baseline showed no statistically significant interaction with treatment for risk of stroke (Figure 3; Table III in the online-only Data Supplement). Changes in Systolic Blood Pressure, Changes in Hematocrit, and Events Consistent With Volume Depletion or Atrial Fibrillation in Relation to Stroke Because treatment with empagliflozin is associated with reductions in systolic blood pressure and small increases in hematocrit, we assessed changes in systolic blood pressure and hematocrit in relation to stroke. We assessed the occurrence of stroke in patients who did and did not have events consistent with volume depletion or atrial fibrillation.

lozin is associated with reductions in systolic blood pressure and small increases in hematocrit, we assessed changes in systolic blood pressure and hematocrit in relation to stroke. We assessed the occurrence of stroke in patients who did and did not have events consistent with volume depletion or atrial fibrillation. Systolic blood pressure decreased in patients treated with empagliflozin (mean change from baseline to last value on treatment: −3.3 [SE, 0.3] mm Hg) but had returned to its baseline level at follow-up (30 days after end of treatment; Table). Patients with the largest decreases from baseline in systolic blood pressure (≥30 mm Hg) did not have an increased risk of stroke compared with other patients (Table V in the online-only Data Supplement). Table. Systolic Blood Pressure and Hematocrit at Baseline, Last Value on Treatment, and at Follow-Up In the empagliflozin group, hematocrit increased during treatment (mean change from baseline to last value on treatment: 3.61% [SE, 0.06]), but decreased toward baseline at follow-up (30 days after end of treatment; Table). Patients with the largest increases from baseline in hematocrit (increases ≥90th percentile, which corresponded to a change in hematocrit of 9 percentage points) did not have an increased risk of stroke compared with patients not meeting this threshold (Table V in the online-only Data Supplement).

of treatment; Table). Patients with the largest increases from baseline in hematocrit (increases ≥90th percentile, which corresponded to a change in hematocrit of 9 percentage points) did not have an increased risk of stroke compared with patients not meeting this threshold (Table V in the online-only Data Supplement). The proportion of patients with a stroke event was not higher in those who did versus did not have an event consistent with volume depletion in placebo or empagliflozin groups (Table VI in the online-only Data Supplement). The risk of stroke was comparable between patients with and without atrial fibrillation at baseline or during the study in the placebo and empagliflozin groups (Figure 3; Tables III and VII in the online-only Data Supplement).

The proportion of patients with a stroke event was not higher in those who did versus did not have an event consistent with volume depletion in placebo or empagliflozin groups (Table VI in the online-only Data Supplement). The risk of stroke was comparable between patients with and without atrial fibrillation at baseline or during the study in the placebo and empagliflozin groups (Figure 3; Tables III and VII in the online-only Data Supplement). Discussion In the EMPA-REG OUTCOME trial in patients with type 2 diabetes mellitus and high cardiovascular risk, empagliflozin added to standard of care significantly reduced the risk of the primary outcome of 3-point major adverse cardiovascular events by reducing the risk of cardiovascular death. There was no significant difference in the occurrence of stroke between empagliflozin and placebo in the prespecified modified intent-to-treat analysis, but there was a numeric difference between treatment groups. In these new analyses, we show that this numeric difference was driven by nonfatal ischemic stroke, with no isolated increase in any subtype of ischemic stroke, and that there was no significant difference between empagliflozin and placebo in the risk of stroke in on-treatment sensitivity analyses or in the risk of recurrent, fatal, or nonfatal disabling strokes, or TIA, which has similar pathophysiological mechanisms as stroke. Further sensitivity analyses demonstrated that the numeric difference in the proportion of patients with stroke between empagliflozin and placebo in the modified intent-to-treat analysis was primarily because of events that occurred >90 days after the last intake of study drug. In this context, it is important to note that measures of the hemodynamic effects of empagliflozin, specifically systolic blood pressure and hematocrit, returned to near baseline levels within 30 days after the last intake, making a causal association with empagliflozin unlikely and making it unlikely that there is an increased risk of stroke after empagliflozin is stopped.

f the hemodynamic effects of empagliflozin, specifically systolic blood pressure and hematocrit, returned to near baseline levels within 30 days after the last intake, making a causal association with empagliflozin unlikely and making it unlikely that there is an increased risk of stroke after empagliflozin is stopped. In subgroup analyses of time to first stroke, analyses by region and HbA1c showed nominal heterogeneity. The increased HR for stroke with empagliflozin compared with placebo in patients in Europe compared with the total population could not be explained by differences in baseline characteristics between regions. Given the large number of subgroup factors and tests conducted, the differences in HR between Europe and other regions, and between patients with HbA1c ≥8.5% and <8.5% at baseline, are within the realm of chance variation. Analyses of time to first stroke in subgroups by other baseline characteristics, including factors associated with risk for stroke such as atrial fibrillation, smoking, previous stroke, and hypertension,14–17 showed no statistically significant interaction with treatment for the risk of stroke. The risk of experiencing a stroke was comparable between patients with and without atrial fibrillation at baseline or during the study. A slightly lower proportion of patients treated with empagliflozin than placebo had anticoagulants introduced postbaseline,10 and it cannot be excluded that this could have contributed to the numeric difference in stroke.

comparable between patients with and without atrial fibrillation at baseline or during the study. A slightly lower proportion of patients treated with empagliflozin than placebo had anticoagulants introduced postbaseline,10 and it cannot be excluded that this could have contributed to the numeric difference in stroke. Treatment with empagliflozin is associated with hemoconcentration, as shown by increases in hematocrit, and with reductions in systolic blood pressure.10 Concerns have been raised that elevated hematocrit and hypotension may be associated with an increased risk of stroke caused by sludging and hypoperfusion, respectively. In a meta-analysis of observational studies of patients with and without diabetes mellitus, orthostatic hypotension was associated with an increased risk of cardiovascular events, including stroke.18 In EMPA-REG OUTCOME, mean baseline hematocrit in the empagliflozin group was 41.4%, mean baseline systolic blood pressure was 135 mm Hg, and patients with the largest increases in hematocrit and the largest decreases in systolic blood pressure did not have an increased risk of stroke. The proportion of patients with a stroke event was consistent between patients who did and did not have an event consistent with volume depletion in both treatment groups.

and patients with the largest increases in hematocrit and the largest decreases in systolic blood pressure did not have an increased risk of stroke. The proportion of patients with a stroke event was consistent between patients who did and did not have an event consistent with volume depletion in both treatment groups. A reduction in the risk of stroke was observed with intensive blood pressure lowering versus standard therapy in the ACCORD study (Action to Control Cardiovascular Risk in Diabetes), but this occurred after a mean follow-up of 4.7 years despite a large difference in systolic blood pressure (14.2 mm Hg) after 1 year.19 Thus, the lack of a risk reduction for stroke with empagliflozin in EMPA-REG OUTCOME may have been expected given the modest reduction in systolic blood pressure provided by empagliflozin over a median treatment time of 2.6 years and from a baseline of 135 mm Hg with 95% of patients taking antihypertensive therapy at baseline. The risk of stroke in the EMPA-REG OUTCOME trial was similar between patients with controlled (systolic blood pressure <140 mm Hg and diastolic blood pressure <90 mm Hg) versus uncontrolled blood pressure at baseline. Limitations of these analyses include that the results cannot be extrapolated beyond the treatment duration or observation time of the trial or to patient populations with other clinical characteristics.

A reduction in the risk of stroke was observed with intensive blood pressure lowering versus standard therapy in the ACCORD study (Action to Control Cardiovascular Risk in Diabetes), but this occurred after a mean follow-up of 4.7 years despite a large difference in systolic blood pressure (14.2 mm Hg) after 1 year.19 Thus, the lack of a risk reduction for stroke with empagliflozin in EMPA-REG OUTCOME may have been expected given the modest reduction in systolic blood pressure provided by empagliflozin over a median treatment time of 2.6 years and from a baseline of 135 mm Hg with 95% of patients taking antihypertensive therapy at baseline. The risk of stroke in the EMPA-REG OUTCOME trial was similar between patients with controlled (systolic blood pressure <140 mm Hg and diastolic blood pressure <90 mm Hg) versus uncontrolled blood pressure at baseline. Limitations of these analyses include that the results cannot be extrapolated beyond the treatment duration or observation time of the trial or to patient populations with other clinical characteristics. In conclusion, in patients with type 2 diabetes mellitus and high cardiovascular risk in the EMPA-REG OUTCOME trial, empagliflozin, when compared with placebo, was not associated with either a reduction or an increase in the risk of cerebrovascular events.

Limitations of these analyses include that the results cannot be extrapolated beyond the treatment duration or observation time of the trial or to patient populations with other clinical characteristics. In conclusion, in patients with type 2 diabetes mellitus and high cardiovascular risk in the EMPA-REG OUTCOME trial, empagliflozin, when compared with placebo, was not associated with either a reduction or an increase in the risk of cerebrovascular events. Acknowledgments Medical writing assistance, supported financially by Boehringer Ingelheim, was provided by Elizabeth Ng and Wendy Morris of FleishmanHillard Fishburn, London, United Kingdom, during the preparation of this article. The authors are fully responsible for all content and editorial decisions, were involved at all stages of article development, and have approved the final version. Sources of Funding This trial was funded by the Boehringer Ingelheim and Eli Lilly and Company Diabetes Alliance.

Acknowledgments Medical writing assistance, supported financially by Boehringer Ingelheim, was provided by Elizabeth Ng and Wendy Morris of FleishmanHillard Fishburn, London, United Kingdom, during the preparation of this article. The authors are fully responsible for all content and editorial decisions, were involved at all stages of article development, and have approved the final version. Sources of Funding This trial was funded by the Boehringer Ingelheim and Eli Lilly and Company Diabetes Alliance. Disclosures Dr Zinman has received personal fees from AstraZeneca, Boehringer Ingelheim, Eli Lilly and Company, Janssen, Merck & Co, Novo Nordisk, Sanofi, and Takeda and has received grants from Boehringer Ingelheim, Novo Nordisk, and Merck & Co. Dr Inzucchi has received personal fees from Alere, AstraZeneca, Daiichi-Sankyo, Janssen, Intarcia, Merck & Co, Novo Nordisk, Poxel, Sanofi, Regeron, Lexicon, vTv Pharmaceuticals, and Eli Lilly and Company; received personal fees and nonfinancial support from Boehringer Ingelheim; received nonfinancial support from Takeda; and has received grants from the National Institute of Diabetes and Digestive and Kidney Diseases and the National Institute of Neurological Disorders and Stroke. Dr Lachin has received personal fees from Boehringer Ingelheim, Merck & Co, Gilead Sciences, Janssen, Novartis, and AstraZeneca. Dr Wanner has received personal fees from Boehringer Ingelheim, Janssen, and Novo Nordisk; and has received grants from Boehringer Ingelheim and the European Foundation for the Study of Diabetes. Dr Fitchett has received personal fees from Boehringer Ingelheim, Novo Nordisk, AstraZeneca, Sanofi, and Merck & Co. Dr Albers reports personal fees from AstraZeneca, Codman, Covidien, iSchemaView, Genentech, Johnson & Johnson, and Lundbeck; has received grants from Lundbeck; and has a patent on Automated arterial input function detection issued. Dr Diener has received honoraria for participation in clinical trials, contribution to advisory boards or presentations from Abbott, Allergan, AstraZeneca, Bayer Vital, Bristol-Myers Squibb, Boehringer Ingelheim, CoAxia, Corimmun, Covidien, Daiichi-Sankyo, D-Pharm, Fresenius, GlaxoSmithKline, Janssen-Cilag, Johnson & Johnson, Knoll, Lilly, Merck Sharp & Dohme, Medtronic, MindFrame, Neurobiological Technologies, Novartis, Novo Nordisk, Paion, Parke-Davis, Pfizer, Sanofi-Aventis, Schering-Plough, Servier, Solvay, St. Jude, Sygnis, Talecris, Thrombogenics, WebMD Global, Wyeth, and Yamanouchi, and has received financial support for research projects from AstraZeneca, GlaxoSmithKline, Boehringer Ingelheim, Lundbeck, Novartis, Janssen-Cilag, Sanofi-Aventis, Sygnis, and Talecris.

er, Sanofi-Aventis, Schering-Plough, Servier, Solvay, St. Jude, Sygnis, Talecris, Thrombogenics, WebMD Global, Wyeth, and Yamanouchi, and has received financial support for research projects from AstraZeneca, GlaxoSmithKline, Boehringer Ingelheim, Lundbeck, Novartis, Janssen-Cilag, Sanofi-Aventis, Sygnis, and Talecris. The Department of Neurology at the University Duisburg-Essen has received research grants from the German Research Council, German Ministry of Education and Research, European Union, the National Institutes of Health, Bertelsmann Foundation, and Heinz-Nixdorf Foundation. Dr Diener served as editor of Aktuelle Neurologie, Arzneimitteltherapie, Kopfschmerznews, Stroke News, as coeditor of Cephalalgia, and is on the editorial board of Lancet Neurology, Stroke, European Neurology, and Cerebrovascular Disorders. Dr Diener chairs the treatment guidelines committee of the German Society of Neurology and contributed to the European Heart Rhythm Association and the European Society of Cardiology guidelines for the treatment of atrial fibrillation. Dr Kohler, M. Mattheus, Dr Johansen, Dr Woerle, and Dr Broedl are employees of Boehringer Ingelheim. Supplementary Material Guest Editor for this article was Natalia S. Rost, MD, MPH. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.116.015756/-/DC1.

Perivascular spaces (PVSs) are tiny cavities of cerebrospinal fluid that surround arterioles that penetrate the brain parenchyma.1 They are most frequently found in the inferior basal ganglia (BG), centrum semiovale (CS), and midbrain.2 Although it is normal to have a few visible PVSs on neuroimaging,3 an increased burden of BG and CS-PVSs has been associated with increasing age,4–7 hypertension,4–6,8 renal impairment,9 white matter hyperintensity (WMH),4–6,8,10 and lacunes.4,6,8,10 BG-PVSs in addition have also been associated with male sex,6 mean systolic and diastolic blood pressure,8,11 deep or infratentorial cerebral microbleeds,7,8 and also stroke caused by small vessel occlusion.10 BG-PVSs have been considered a marker of hypertensive arteriopathy secondary to endothelial dysfunction7,8 and have recently been proposed as 1 of the 4 components of the Total Small Vessel Disease (SVD) Score.12 In contrast, CS-PVSs have been associated with lobar microbleeds in healthy adults and in those with cognitive impairment.5,8 A high burden of CS-PVSs has also been noted in patients with cerebral amyloid angiopathy (CAA).7,13 It has, therefore, been hypothesized that in contrast to BG-PVSs, CS-PVSs may be a neuroimaging marker of CAA by representing fluid and metabolic waste clearance dysfunction because of vascular amyloid deposition.7,8,14

A high burden of CS-PVSs has also been noted in patients with cerebral amyloid angiopathy (CAA).7,13 It has, therefore, been hypothesized that in contrast to BG-PVSs, CS-PVSs may be a neuroimaging marker of CAA by representing fluid and metabolic waste clearance dysfunction because of vascular amyloid deposition.7,8,14 Although PVSs have shown potential as an imaging biomarker of hypertensive angiopathy and CAA, the long-term prognostic implications of PVSs among patients with transient ischemic attack (TIA) and ischemic stroke have yet to be determined. Ethnic differences in PVS are also uncertain. To address these unanswered questions, we performed 2 large prospective studies, consisting of >2000 whites and Chinese with TIA/ischemic stroke from 2 independent cohorts to determine the associations of PVSs with ethnicity, vascular risk factors, other neuroimaging markers of SVD, and long-term risks of stroke and death.

ddress these unanswered questions, we performed 2 large prospective studies, consisting of >2000 whites and Chinese with TIA/ischemic stroke from 2 independent cohorts to determine the associations of PVSs with ethnicity, vascular risk factors, other neuroimaging markers of SVD, and long-term risks of stroke and death. Methods We prospectively studied patients with TIA/ischemic stroke from 2 cohorts: the OXVASC (Oxford Vascular) Study and the University of Hong Kong (HKU). In brief, OXVASC is an ongoing population-based study of all acute vascular events occurring within a population of all 92 728 individuals, irrespective of age, who are registered with 100 general practitioners in 9 general practices of Oxfordshire, United Kingdom.15 The analysis herein includes 1080 consecutive cases of TIA/ischemic stroke recruited from November 1, 2004, to September 30, 2014, who had a cerebral magnetic resonance imaging (MRI). The imaging protocol of OXVASC has been described in detail elsewhere.16 Briefly, from April 1, 2002, to March 31, 2010 (phase 1), MRI and magnetic resonance angiography were done in selected patients when clinically indicated. From April 1, 2010, onward (phase 2), brain MRI and magnetic resonance angiography of intra- and extracranial vessels became the first-line imaging methods.16 A further 1076 consecutive patients who were predominantly Chinese with a diagnosis of acute ischemic stroke who received an MRI scan and magnetic resonance angiography of the intra- and extracranial blood vessels at the HKU MRI Unit were recruited during March 1, 2008, to September 30, 2014.

st-line imaging methods.16 A further 1076 consecutive patients who were predominantly Chinese with a diagnosis of acute ischemic stroke who received an MRI scan and magnetic resonance angiography of the intra- and extracranial blood vessels at the HKU MRI Unit were recruited during March 1, 2008, to September 30, 2014. We collected demographic data, atherosclerotic risk factors, and details of hospitalization of index event during face-to-face interview and cross-referenced these with primary care records and hospital records in both cohorts. Cause of TIA/ischemic stroke was classified according to the modified Trial of ORG 10172 in Acute Stroke Treatment (TOAST) criteria.17

otic risk factors, and details of hospitalization of index event during face-to-face interview and cross-referenced these with primary care records and hospital records in both cohorts. Cause of TIA/ischemic stroke was classified according to the modified Trial of ORG 10172 in Acute Stroke Treatment (TOAST) criteria.17 Details of scan parameters are documented in Table I in the online-only Data Supplement. Two neuroradiologists (H.K.F.M. and W.K.) supervised the interpretation of the MRI images. PVSs were defined as small (<3 mm) punctate (if perpendicular to the plane of scan) or linear (if longitudinal to the plane of scan) hyperintensities on T2 images in the BG or CS based on a previously validated scale.18 In patients with asymmetrical number of PVSs, the side with the higher number of PVSs was counted.18 Burden of PVSs was then stratified into 3 groups: <11, 11 to 20, and >20 (frequent–severe).18 Definitions of subcortical and periventricular WMH, microbleeds, and lacunes are provided in the online-only Data Supplement. The intrarater κ for burden of PVS (<11, 11–20, and >20) was 0.86 (BG) and 0.84 (CS) in OXVASC and 0.86 (BG) and 0.72 (CS) in HKU (50 scans in each center). Seventy-five MRI scans from HKU were cross-interpreted by investigators in OXVASC with an interrater κ of 0.64 for both BG and CS-PVSs.

nline-only Data Supplement. The intrarater κ for burden of PVS (<11, 11–20, and >20) was 0.86 (BG) and 0.84 (CS) in OXVASC and 0.86 (BG) and 0.72 (CS) in HKU (50 scans in each center). Seventy-five MRI scans from HKU were cross-interpreted by investigators in OXVASC with an interrater κ of 0.64 for both BG and CS-PVSs. All patients in OXVASC were followed up regularly by a research nurse or physician after 1, 3, 6, 12, 24, 60, and 120 months after the index event. Patients recruited from HKU were followed up by a clinician every 3 to 6 months, or more frequently if clinically indicated. All patients were assessed for recurrent stroke (ischemic and hemorrhagic) and death (vascular and nonvascular; see definitions in the online-only Data Supplement). Where needed, details of clinical outcomes were supplemented by electronic or paper medical records from individual primary care practices, hospitals, and the Deaths General Register Office. Patients gave written informed consent after an event or assent was obtained from relatives for patients who were unable to provide consent. Both cohorts were approved by the local research ethics committee.

All patients in OXVASC were followed up regularly by a research nurse or physician after 1, 3, 6, 12, 24, 60, and 120 months after the index event. Patients recruited from HKU were followed up by a clinician every 3 to 6 months, or more frequently if clinically indicated. All patients were assessed for recurrent stroke (ischemic and hemorrhagic) and death (vascular and nonvascular; see definitions in the online-only Data Supplement). Where needed, details of clinical outcomes were supplemented by electronic or paper medical records from individual primary care practices, hospitals, and the Deaths General Register Office. Patients gave written informed consent after an event or assent was obtained from relatives for patients who were unable to provide consent. Both cohorts were approved by the local research ethics committee. Statistical Analysis We compared differences in baseline and imaging characteristics in the OXVASC and HKU cohorts using Student t test for continuous variables and χ2 test for categorical variables. The predictors of >20 BG and CS-PVSs were determined using a logistic regression model adjusted for center and MRI scanner strength. Variables including age, male sex, vascular risk factors (hypertension, hyperlipidaemia, diabetes mellitus, smoking, and atrial fibrillation), renal impairment (defined as glomerular filtration rate <60 mL/min/1.73 m2, as measured by the Modification of Diet in Renal Disease Study equation19), periventricular and subcortical WMH, deep and lobar microbleed number and lacunes were entered into a univariate analysis model. All variables were subsequently entered into a multivariate analysis model to determine the independent predictors of >20 BG and CS-PVSs. The multivariate model to determine the independent predictors of >20 BG-PVSs was also adjusted for CS-PVSs and vice versa.

and lacunes were entered into a univariate analysis model. All variables were subsequently entered into a multivariate analysis model to determine the independent predictors of >20 BG and CS-PVSs. The multivariate model to determine the independent predictors of >20 BG-PVSs was also adjusted for CS-PVSs and vice versa. In a logistic regression model, we determined the odds of a TIA/ischemic stroke because of SVD with increasing burden of BG and CS-PVSs, adjusted for age, sex, vascular risk factors, center, and MRI scanner strength. We used Kaplan–Meier survival analysis to calculate the 5-year risk of a recurrent stroke (ischemic and hemorrhagic) and all-cause mortality, censored at death or March 31, 2015, according to the burden of PVSs. We also determined, by Cox regression analysis, the unadjusted and adjusted (for age, sex, vascular risk factors, center, and MRI scanner strength) risks of recurrent stroke (ischemic and hemorrhagic), mortality (vascular and nonvascular) in patients with 11 to 20 and >20 BG and CS-PVSs compared with <11 PVSs as reference. Finally, we performed a stratified analysis to determine whether the prognosis of PVSs differed in patients with no or mild versus moderate or severe periventricular and subcortical WMH. All analyses were done with SPSS version 20. Role of the Funding Source The funding source had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding authors had full access to all the data in the study and had the final responsibility for the decision to submit for publication.

All analyses were done with SPSS version 20. Role of the Funding Source The funding source had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding authors had full access to all the data in the study and had the final responsibility for the decision to submit for publication. Results The 2 study populations contributed a total of 2156 patients. After excluding 154 patients with incomplete clinical or imaging data, 2002 (OXVASC n=1028, 542 TIA, 486 ischemic stroke; HKU n=974, all ischemic stroke) were included in the final analysis. Baseline clinical and imaging characteristics of patients are shown in Table 1. Proportion of patients according to TOAST classification is shown in Table II in the online-only Data Supplement. HKU patients had a higher proportion of men (P=0.001) and were more likely to have hypertension and diabetes mellitus (P<0.0001), whereas OXVASC patients were more likely to have hyperlipidaemia or a history of smoking (P<0.0001; Table 1).

to TOAST classification is shown in Table II in the online-only Data Supplement. HKU patients had a higher proportion of men (P=0.001) and were more likely to have hypertension and diabetes mellitus (P<0.0001), whereas OXVASC patients were more likely to have hyperlipidaemia or a history of smoking (P<0.0001; Table 1). Patients from OXVASC had a higher burden of >20 BG (22.4% versus 7.1%; P<0.0001) and CS-PVSs (45.8% versus 10.4%; P<0.0001) compared with those from HKU (Table 1). These differences in PVS burden remained, despite stratification of individuals into stroke subtypes (Table III in the online-only Data Supplement). OXVASC patients also had more severe periventricular WMH (P<0.0001; Table 1). In contrast, those from HKU had a greater burden of subcortical WMH (P<0.0001) and microbleeds (P<0.0001; Table 1). These differences remained in analyses confined to patients who received an MRI with a 3T scanner (Table IV in the online-only Data Supplement). However, within OXVASC, patients who received a 3T MRI (n=446) had a greater burden of >20 BG-PVSs (25.8% versus 19.8%; P=0.022) and CS-PVSs (55.6% versus 38.3%; P<0.0001) compared with patients who received a 1.5T MRI (n=582). Table 1. Clinical and Imaging Characteristics of the Study Population

Patients from OXVASC had a higher burden of >20 BG (22.4% versus 7.1%; P<0.0001) and CS-PVSs (45.8% versus 10.4%; P<0.0001) compared with those from HKU (Table 1). These differences in PVS burden remained, despite stratification of individuals into stroke subtypes (Table III in the online-only Data Supplement). OXVASC patients also had more severe periventricular WMH (P<0.0001; Table 1). In contrast, those from HKU had a greater burden of subcortical WMH (P<0.0001) and microbleeds (P<0.0001; Table 1). These differences remained in analyses confined to patients who received an MRI with a 3T scanner (Table IV in the online-only Data Supplement). However, within OXVASC, patients who received a 3T MRI (n=446) had a greater burden of >20 BG-PVSs (25.8% versus 19.8%; P=0.022) and CS-PVSs (55.6% versus 38.3%; P<0.0001) compared with patients who received a 1.5T MRI (n=582). Table 1. Clinical and Imaging Characteristics of the Study Population Burden of BG and CS-PVSs increased with age, baseline history of hypertension, atrial fibrillation, and renal impairment (P<0.05; Table V in the online-only Data Supplement). Burden of BG and CS-PVSs was also greater in patients with lacunes and severe WMH (P<0.05; Table V in the online-only Data Supplement). In a multivariate analysis, >20 BG or CS-PVSs were associated with increasing age (multivariate adjusted odds ratio [OR], BG: 1.05; 95% confidence interval [CI], 1.03–1.07; P<0.0001; CS: OR, 1.01; 95% CI, 1.00–1.03; P=0.020) and subcortical WMH (BG: OR, 1.44; 95% CI, 1.20–1.72; P<0.0001; CS: OR, 1.28; 95% CI, 1.09–1.50; P=0.003). More than 20 BG-PVSs were also associated with atrial fibrillation (OR, 1.58; 95% CI, 1.10–2.29; P=0.014) and periventricular WMH (OR, 2.01; 95% CI, 1.66–2.44; P<0.0001; Table 2; Table VI in the online-only Data Supplement). Although underlying significant (>50%) large artery atherosclerosis was not related to >20 BG-PVSs (multivariate adjusted OR, 1.10; 95% CI, 0.77–1.56; P=0.61), an independent association between >20 CS-PVSs with significant large artery disease was noted (OR, 1.44; 95% CI 1.07–1.93; P=0.015). Whites, as compared with Chinese, were at increased odds of >20 BG (multivariate adjusted OR, 2.09; 95% CI, 1.35–3.22; P=0.001) and CS-PVSs (OR, 8.82; 95% CI, 6.25–12.46; P<0.0001). These ORs remained similar after additional adjustment of MRI magnet strength (BG: OR, 2.50; 95% CI, 1.56–4.02; P=0.0002; CS: OR, 11.93; 95% CI, 8.15–17.47; P<0.0001).

ese, were at increased odds of >20 BG (multivariate adjusted OR, 2.09; 95% CI, 1.35–3.22; P=0.001) and CS-PVSs (OR, 8.82; 95% CI, 6.25–12.46; P<0.0001). These ORs remained similar after additional adjustment of MRI magnet strength (BG: OR, 2.50; 95% CI, 1.56–4.02; P=0.0002; CS: OR, 11.93; 95% CI, 8.15–17.47; P<0.0001). Table 2. Clinical Correlates of >20 Perivascular Spaces A 26.8% of the study population was classified to have TIA/ischemic stroke because of SVD (Table II in the online-only Data Supplement). These patients were associated with a higher BG and CS-PVS burden (multivariate adjusted OR compared with <11 PVSs, 11–20 BG-PVSs: OR, 2.44; 95% CI, 1.45–4.10; >20 BG-PVSs: OR, 2.82; 95% CI, 1.60–4.97; P=0.0002; 11–20 CS-PVSs: OR, 2.54; 95% CI, 1.32–4.88; >20 CS-PVSs: OR, 4.20; 95% CI, 2.19–8.06; P<0.0001). After a mean follow-up of 42±23 months (OXVASC 45±26 months, HKU 37±19 months, 6924 patient-years of follow-up), 199 recurrent strokes occurred (85.4% ischemic; Table 1). Two hundred sixty-six patients died, 34.6% of which were vascular deaths. The 5-year risk of recurrent ischemic stroke and intracerebral hemorrhage (ICH) in patients with <11, 11 to 20, and >20 BG-PVSs was 8.5%, 11.5%, and 19.3% (log-rank test: P<0.0001) and 1.6%, 2.3%, and 3.7%, respectively (P=0.038; Figure 1). An increasing burden of BG-PVSs was also associated with a higher all-cause mortality (P<0.0001; Figure 1). In contrast, burden of CS-PVSs was not associated with recurrent ischemic stroke (P=0.76), intracerebral hemorrhage (ICH; P=0.96), or all-cause mortality (P=0.33; Figure 2).

.7%, respectively (P=0.038; Figure 1). An increasing burden of BG-PVSs was also associated with a higher all-cause mortality (P<0.0001; Figure 1). In contrast, burden of CS-PVSs was not associated with recurrent ischemic stroke (P=0.76), intracerebral hemorrhage (ICH; P=0.96), or all-cause mortality (P=0.33; Figure 2). Figure 1. Risk of (A) recurrent stroke, (B) recurrent ischemic stroke, (C) intracerebral hemorrhage, and (D) all-cause mortality among transient ischemic attack/ischemic stroke patients with increasing basal ganglia perivascular space burden. BG-PVS indicates basal ganglia perivascular spaces. Figure 2. Risk of (A) recurrent stroke, (B) recurrent ischemic stroke, (C) intracerebral hemorrhage, and (D) all-cause mortality among transient ischemic attack/ischemic stroke patients with increasing centrum semiovale perivascular space burden. CS-PVS indicates centrum semiovale perivascular spaces.

Figure 1. Risk of (A) recurrent stroke, (B) recurrent ischemic stroke, (C) intracerebral hemorrhage, and (D) all-cause mortality among transient ischemic attack/ischemic stroke patients with increasing basal ganglia perivascular space burden. BG-PVS indicates basal ganglia perivascular spaces. Figure 2. Risk of (A) recurrent stroke, (B) recurrent ischemic stroke, (C) intracerebral hemorrhage, and (D) all-cause mortality among transient ischemic attack/ischemic stroke patients with increasing centrum semiovale perivascular space burden. CS-PVS indicates centrum semiovale perivascular spaces. On Cox regression analysis, strong univariate association between increasing burden of BG-PVSs with all-cause mortality was noted (P<0.0001), but this association disappeared after adjustment for age and sex (P=0.058) and on multivariate adjustment (P=0.16; Table VII in the online-only Data Supplement). In contrast, the strong univariate associations between increasing burden of BG-PVSs with recurrent ischemic stroke persisted with multivariate adjustment (hazard ratio compared with <11 BG-PVSs, 11–20: HR, 1.15; 95% CI, 0.78–1.68; >20: HR, 1.82; 95% CI, 1.18–2.80; P=0.011; Table 3; Table VIII in the online-only Data Supplement). BG-PVS burden was not independently associated with ICH (P=0.10; Table 3). An increasing burden of CS-PVSs was not related to ischemic stroke (P=0.42), ICH (P=0.69), or mortality (P=0.072; Table 3; Table VII in the online-only Data Supplement). When patients were stratified by MRI scanner in OXVASC, the prognostic value of BG and CS-PVSs remained similar for prediction of recurrent stroke (BG-PVS: P=0.15; CS-PVS: P=0.45; Table IX in the online-only Data Supplement). Similarly, no heterogeneity was observed when analyses for risk of recurrent stroke were stratified by patients with no or mild versus moderate–severe WMH (BG-PVS: P=0.92; CS-PVS: P=0.076; Table X in the online-only Data Supplement). In an unadjusted model, burden of BG-PVSs, microbleeds, periventricular WMH, subcortical WMH, and presence of lacunes were all associated with recurrent ischemic stroke (P<0.05; Table XI in the online-only Data Supplement). Forward stepwise multivariate Cox regression model adjusting for all neuroimaging markers of SVD revealed that burden of BG-PVSs (P=0.001) and microbleeds (P=0.001) as independent predictors of recurrent ischemic stroke (Table XI in the online-only Data Supplement).

P<0.05; Table XI in the online-only Data Supplement). Forward stepwise multivariate Cox regression model adjusting for all neuroimaging markers of SVD revealed that burden of BG-PVSs (P=0.001) and microbleeds (P=0.001) as independent predictors of recurrent ischemic stroke (Table XI in the online-only Data Supplement). Table 3. Cox Regression Analyses of Recurrent Stroke With Increasing Burden of Perivascular Spaces Versus <11 Perivascular Spaces Discussion Our study has combined the 2 largest current cohorts from the west and the east of the clinical implications of BG and CS-PVSs in patients with TIA/ischemic stroke and is the first to determine the ethnic differences in prevalence. In our study, PVSs were coded according to a validated rating scale,18 with excellent intrarater variability and good interrater variability when scans were cross-interpreted between the 2 centers. Our study is also the first to determine the long-term prognostic implications of PVSs in patients with TIA/ischemic stroke.

our study, PVSs were coded according to a validated rating scale,18 with excellent intrarater variability and good interrater variability when scans were cross-interpreted between the 2 centers. Our study is also the first to determine the long-term prognostic implications of PVSs in patients with TIA/ischemic stroke. Our results support those from previous studies that BG and CS-PVSs are both markers of hypertensive angiopathy.6,8,14 We too found that PVSs were associated with age4–6 and WMH.5,6 Concordant with previous studies,4–8,10 we also found that compared with CS-PVSs, BG-PVSs were a stronger marker of hypertensive angiopathy, with greater associations with periventricular and subcortical WMH and that BG-PVSs were more strongly associated with TIA/ischemic stroke because of SVD. Although previous studies have also noted significant associations of BG-PVS with deep microbleeds,7,8 this finding did not reach statistical significance in our cohorts (P=0.063). The stronger association of BG-PVSs with hypertensive angiopathy was also reflected in our long-term follow-up data. Compared with <11 PVSs, TIA/ischemic stroke patients with >20 BG-PVSs were at 1.8-fold increased risk of recurrent ischemic stroke on multivariate analysis. There was a trend toward patients with increasing burden of BG-PVSs being similarly at increased risk of subsequent ICH and mortality. Furthermore, we were able to demonstrate that the prognostic implications of BG-PVSs were independent of other neuroimaging markers of SVD.

ent ischemic stroke on multivariate analysis. There was a trend toward patients with increasing burden of BG-PVSs being similarly at increased risk of subsequent ICH and mortality. Furthermore, we were able to demonstrate that the prognostic implications of BG-PVSs were independent of other neuroimaging markers of SVD. In contrast, although previous studies have revealed an association of CS-PVSs with lobar microbleeds5,8 and CAA,7 suggesting that CS-PVSs may be an imaging biomarker of CAA,5,7,8 CS-PVSs were not associated with lobar microbleeds nor adverse clinical events including ICH in our cohorts. It should be noted, however, that studies that have ascertained the relationship of CS-PVSs with lobar microbleeds were based on either healthy adults or subjects recruited from a memory clinic,5,8 with an expected lower prevalence of vascular risk factors and hence less severe imaging markers of SVD compared with patients in our study. It is widely accepted that PVSs may be difficult to identify in patients with extensive WMH.18 This is particularly the case for CS-PVSs that are often masked by subcortical WMHs. Indeed, in our high-risk cohort, where 36% of individuals had moderate–severe subcortical WMH, the true prevalence of CS-PVSs would without doubt be underestimated.

at PVSs may be difficult to identify in patients with extensive WMH.18 This is particularly the case for CS-PVSs that are often masked by subcortical WMHs. Indeed, in our high-risk cohort, where 36% of individuals had moderate–severe subcortical WMH, the true prevalence of CS-PVSs would without doubt be underestimated. Our results also demonstrate that significant ethnic differences in PVS prevalence exist. We showed a similar prevalence of >20 BG and CS-PVSs in OXVASC to a previous study of whites with TIA/ischemic stroke.4 In contrast, however, our study showed that Chinese with TIA/ischemic stroke had a much lower prevalence of PVSs. The prevalence of BG and CS-PVSs among Chinese with ischemic stroke has previously been reported.9,20 One study showed that 10.7% of subjects had >40 BG-PVSs,20 and in another, ≈40% of subjects had >10 BG or CS-PVSs.9 These 2 studies, however, were purely based on patients with lacunar stroke subtype.9,20 In a large cohort of neurologically healthy Japanese individuals, a low prevalence of >20 BG-PVS and CS-PVS of 2.5% and 22.6% was similarly noted.8 Such low prevalence of >20 PVSs in the HKU cohort would have limited the statistical power when determining the clinical correlates of PVSs and attributed to some of the differences observed when compared with OXVASC. Atrial fibrillation was also noted to be significantly associated with >20 BG-PVSs, whereas underlying large artery disease was significantly associated with >20 CS-PVSs in our cohorts. Further studies to confirm and to delineate the underlying mechanisms of our findings would be required. Finally, our findings are also limited by patients in OXVASC being scanned on 4 different scanners during the 10-year study period. However, although this could have been a potential source of heterogeneity, the prognostic values for prediction of recurrent stroke with increasing burden of PVSs were similar across the 4 scanners, suggesting that the prognostic value of PVSs is robust to variations in scanner type and sequences. PVS size, symmetry, or ventricular size was not studied in our cohorts. Hence, we were only able to study clinical and imaging correlates and prognostic implications according to PVS number18 but not its size or symmetry.

uggesting that the prognostic value of PVSs is robust to variations in scanner type and sequences. PVS size, symmetry, or ventricular size was not studied in our cohorts. Hence, we were only able to study clinical and imaging correlates and prognostic implications according to PVS number18 but not its size or symmetry. Our study has several clinical implications. First, in 2 large cohorts, our results confirm BG-PVSs as a marker of SVD, independent of WMH. These results, therefore, justify the inclusion of BG-PVSs into the recently derived Total SVD Score.12 In the current version,12 patients with >11 BG-PVSs are given 1 point, as are patients with severe periventricular WMH or moderate–severe subcortical WMH. Whether alternative cutoffs (eg, >20 BG-PVSs) should be used instead in view of the relatively low prognostic value of patients with 11 to 20 BG-PVSs noted in our study would require further research. Second, although the burden of CS-PVSs may possibly have prognostic implications in healthy individuals or those seen in the Memory Clinic, the role of CS-PVSs as a prognostic imaging marker in the TIA/ischemic stroke population seems to be limited. In conclusion, in addition to identifying ethnic differences in frequency of PVSs, we found that BG-PVSs are markers of hypertensive angiopathy and predict risk of recurrent ischemic stroke in patients with TIA/ischemic stroke, independent of WMH. In contrast, the prognostic value of CS-PVSs in TIA/ischemic stroke is limited.

lusion, in addition to identifying ethnic differences in frequency of PVSs, we found that BG-PVSs are markers of hypertensive angiopathy and predict risk of recurrent ischemic stroke in patients with TIA/ischemic stroke, independent of WMH. In contrast, the prognostic value of CS-PVSs in TIA/ischemic stroke is limited. Acknowledgments We acknowledge the use of the facilities of the Acute Vascular Imaging Centre, Oxford, United Kingdom, Cardiovascular Clinical Research Facility, Oxford, United Kingdom, and Magnetic Resonance Imaging Unit, Department of Diagnostic Radiology, University of Hong Kong. Dr K.-K. Lau obtained funding, collected data, did the statistical analysis and interpretation, and wrote and revised the article. Drs L. Li, C.E. Lovelock, G. Zamboni, T.-T. Chan, M.-F. Chiang, and K.-T. Lo collected data. Dr Küker provided study supervision and acquired data. Dr Mak provided study supervision and funding, acquired and interpreted imaging data, and revised the manuscript. Dr Rothwell conceived and designed the overall study, provided study supervision and funding, acquired, analysed, and interpreted the data, and wrote and revised the manuscript.

pervision and acquired data. Dr Mak provided study supervision and funding, acquired and interpreted imaging data, and revised the manuscript. Dr Rothwell conceived and designed the overall study, provided study supervision and funding, acquired, analysed, and interpreted the data, and wrote and revised the manuscript. Sources of Funding Oxford Vascular Study has been funded by the Wellcome Trust, Wolfson Foundation, UK Stroke Association, British Heart Foundation, Dunhill Medical Trust, National Institute for Health Research (NIHR), Medical Research Council, and the NIHR Oxford Biomedical Research Centre. Magnetic Resonance Imaging studies from University of Hong Kong (HKU) have been funded by the SK Yee Medical Foundation and HKU Strategic Research Theme in Neurosciences. Dr Rothwell is in receipt of an NIHR Senior Investigator Award and a Wellcome Trust Senior Investigator Award. Dr K.-K. Lau is funded by a University of Oxford Croucher Scholarship. Disclosures None Supplementary Material Guest Editor for this article was Gregory W. Albers, MD. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.016694/-/DC1.

A significant challenge in the treatment of stroke survivors is the rehabilitation of chronic motor disabilities. Although behavioral therapies such as constraint-induced movement therapy1 or robot-aided sensorimotor stimulation2 can improve upper-limb motor function, they require some level of peripheral motor function to engage with the therapy. This residual function is variable across patients and absent in the setting of complete hemiplegia. An alternative to behavioral therapies is to engage with the patient’s central nervous system directly. Specifically, a brain–computer interface (BCI) system can measure movement-related signals from the central nervous system and provide meaningful feedback to the central nervous system to direct plasticity. BCIs have recently emerged as novel and potentially powerful tools to restore function in chronic stroke survivors.3 Early results present promising demonstrations that BCI-controlled orthoses or functional electric stimulators can lead to improvements in motor function in chronic stroke survivors.3–8 These stroke-specific BCI systems for rehabilitation have focused on signals stemming from perilesional cortex, contralateral to the affected hand for BCI control. Because the ability to modulate perilesional cortical activity decreases with increasing cortical damage,9 it may be particularly important for neurorehabilitation systems to focus on the ipsilateral, contralesional cortex in those patients who are most severely affected.

ralateral to the affected hand for BCI control. Because the ability to modulate perilesional cortical activity decreases with increasing cortical damage,9 it may be particularly important for neurorehabilitation systems to focus on the ipsilateral, contralesional cortex in those patients who are most severely affected. Although movement-related neural activity occurs in the ipsilateral and the contralateral cortices,10,11 the role of the unaffected hemisphere in stroke recovery is uncertain. Specifically, decreases in contralesional activity are associated with optimal recovery in some studies.12,13 Other studies show that increases in contralesional activity may be related to motor recovery,14,15 particularly in patients with incomplete recovery.16 As motor recovery is inversely correlated with the extent of corticospinal tract transection,17 we hypothesized that using contralesional hemisphere activity to drive a BCI-controlled exoskeleton may lead to functional improvements. Previously, we demonstrated that chronic stroke survivors can control BCIs using electroencephalographic (EEG) signals from the contralesional hemisphere associated with the intention to move the affected limb.18 However, it was uncertain whether emphasizing the relationship between activation of ipsilateral cortex and resultant sensory feedback would be beneficial.

survivors can control BCIs using electroencephalographic (EEG) signals from the contralesional hemisphere associated with the intention to move the affected limb.18 However, it was uncertain whether emphasizing the relationship between activation of ipsilateral cortex and resultant sensory feedback would be beneficial. This feasibility study tested an EEG-BCI system that used signals related to affected hand motor imagery, recorded from the unaffected hemisphere, to control the affected hand via a powered exoskeleton. This study is the first to specifically focus on the unaffected hemisphere with a BCI rehabilitation system and the first to provide BCI-driven therapy in the patients’ homes. This setting is important because it increases the likelihood that this approach can be scaled more widely across the stroke-affected population. Methods To determine whether a BCI-controlled exoskeleton using EEG signals from the unaffected hemisphere can lead to functional rehabilitation, we created a novel home-based system called the IpsiHand. We then examined whether a 12-week training period led to functional improvements in chronic, hemiparetic stroke survivors.

To determine whether a BCI-controlled exoskeleton using EEG signals from the unaffected hemisphere can lead to functional rehabilitation, we created a novel home-based system called the IpsiHand. We then examined whether a 12-week training period led to functional improvements in chronic, hemiparetic stroke survivors. Patient Characteristics Ten chronic hemiparetic stroke survivors with moderate-to-severe upper-limb hemiparesis, enrolled at least 6 months after first-time hemispheric stroke, completed the study. Because motor recovery plateaus after 3 months,19 the study was designed as a self-controlled study comparing motor function before and after the intervention to establish the feasibility of the BCI-driven therapy studied. The Table contains patient demographics and baseline motor function. The online-only Data Supplement contains detailed inclusion and exclusion criteria. Moderate-to-severely impaired patients were specifically targeted because they are less likely to recover through other methods and therefore require an alternative rehabilitation strategy, such as a BCI. The Washington University School of Medicine Institutional Review Board approved the study protocol, and all patients provided written informed consent. Table. Patient Characteristics and ARAT Scores

Patient Characteristics Ten chronic hemiparetic stroke survivors with moderate-to-severe upper-limb hemiparesis, enrolled at least 6 months after first-time hemispheric stroke, completed the study. Because motor recovery plateaus after 3 months,19 the study was designed as a self-controlled study comparing motor function before and after the intervention to establish the feasibility of the BCI-driven therapy studied. The Table contains patient demographics and baseline motor function. The online-only Data Supplement contains detailed inclusion and exclusion criteria. Moderate-to-severely impaired patients were specifically targeted because they are less likely to recover through other methods and therefore require an alternative rehabilitation strategy, such as a BCI. The Washington University School of Medicine Institutional Review Board approved the study protocol, and all patients provided written informed consent. Table. Patient Characteristics and ARAT Scores BCI System Design The BCI system (Figure 1A) combined a novel powered exoskeleton with a commercial EEG amplifier and active electrodes. The exoskeleton opened and closed the patient’s hand in a 3-finger pinch grip (1 degree of freedom). A detailed description of the system is contained in the Methods in the online-only Data Supplement. Consistent with our previous work,18 the system used spectral power changes to control hand position. Because stroke patients typically have difficulty extending their extremities, BCI control associated motor imagery with opening the affected hand. Each trial began with the hand fully closed, and spectral power at the control feature was used to update the hand position, providing visual and proprioceptive feedback. During rest trials, patients were instructed to try to keep the exoskeleton closed by imagining that they were resting. During movement trials, patients were instructed to try to open their hand via motor imagery.

at the control feature was used to update the hand position, providing visual and proprioceptive feedback. During rest trials, patients were instructed to try to keep the exoskeleton closed by imagining that they were resting. During movement trials, patients were instructed to try to open their hand via motor imagery. Figure 1. Study methodology. A, The exoskeleton used attached to a patient’s affected hand via straps on the forearm, palm of the hand, and intermediate phalanges of the index and middle finger, whereas the thumb was held stationary. The exoskeleton was controlled by a microprocessor in the forearm assembly that processed electroencephalographic (EEG) signals. A linear actuator drove hand movements in a 3-finger pinch grip based on the decoded EEG. B, The study tested whether training with the brain–computer interface (BCI)–controlled exoskeleton would lead to functional improvements. Patients that met the inclusion criteria completed 3 EEG screenings. Patients with consistent movement-related EEG activations then completed baseline motor evaluations and BCI system training. Finally, patients completed a 12-wk home-based BCI protocol with follow-up motor evaluations at 2-wk intervals.

improvements. Patients that met the inclusion criteria completed 3 EEG screenings. Patients with consistent movement-related EEG activations then completed baseline motor evaluations and BCI system training. Finally, patients completed a 12-wk home-based BCI protocol with follow-up motor evaluations at 2-wk intervals. EEG Screening After meeting the inclusion criteria, patients underwent an EEG screening protocol to ensure that a consistent control signal was present for device control. Each patient completed 3 separate screenings to assess the stability of potential BCI control signals. EEG electrodes were applied by a trained biomedical engineer, and EEG signals were collected while patients performed a visually cued motor screening task consisting of trials of (1) rest, (2) unaffected hand movements, (3) affected hand motor imagery, and (4) bilateral motor imagery. Spectral power, or the power in the EEG signal as a function of frequency, was calculated using an autoregressive spectral estimation method. The coefficient of determination (r2), the percent of variance in spectral power that was accounted for by the difference between affected hand motor imagery and rest trials, was calculated for each channel and frequency. After completing 3 EEG screenings, the EEG data were examined for the presence of consistent spectral power changes during affected hand motor imagery. BCI control features were required to be associated with imagined movements of the affected hand and located in unaffected hemisphere motor regions. These sessions were not designed to achieve BCI mastery but to identify patients with consistent cortical activations (ie, μ [8–12 Hz] or β (12–30 Hz) power decreases) in at least 2 of 3 sessions. The feature in the unaffected hemisphere with the strongest r2 value was chosen as the patient-specific BCI control feature. Patients without consistent spectral power changes were unable to continue in the study.

istent cortical activations (ie, μ [8–12 Hz] or β (12–30 Hz) power decreases) in at least 2 of 3 sessions. The feature in the unaffected hemisphere with the strongest r2 value was chosen as the patient-specific BCI control feature. Patients without consistent spectral power changes were unable to continue in the study. Outcome Measures The primary outcome measure was the Action Research Arm Test (ARAT).20 Secondary outcome measures included: (1) the Canadian Occupational Performance Measure,21 (2) the Motricity Index, (3) the modified Ashworth Scale at the elbow joint, (4) grip strength, (5) pinch strength, and (6) the active range of motion (AROM) at the metacarpophalangeal joint of digits 2 to 5. As this study was the first to use a BCI system for stroke rehabilitation in the home setting, we measured the BCI control quality by comparing the topographies of spectral power changes in the laboratory and home-based sessions. We assessed compliance by recording the total number of days and time that each patient used the system.

was the first to use a BCI system for stroke rehabilitation in the home setting, we measured the BCI control quality by comparing the topographies of spectral power changes in the laboratory and home-based sessions. We assessed compliance by recording the total number of days and time that each patient used the system. Study Protocol The study timeline is shown in Figure 1B. After completing the EEG screenings, patients completed 2 pretherapy motor evaluations in which all primary and secondary outcome measures were measured by an occupational therapist. On these days, the exoskeleton was also fit to the patient’s hand. In addition, patients and their caregivers were trained to use the system. This included (1) donning the exoskeleton and EEG cap, (2) examining the EEG readouts to verify that physiological signals were collected, (3) software operation, and (4) system maintenance. After the baseline motor evaluations and training, each patient was sent home with a BCI system to complete 12 weeks of training. Patients were instructed to use the BCI system on a minimum of 5 days per week. Patients completed 1 to 12 10-minute runs of the BCI task per day depending on their stamina and time constraints. At 2-week intervals, patients came to the laboratory for follow-up motor evaluations consisting of the ARAT and Canadian Occupational Performance Measure. At these follow-up sessions and as needed, an occupational therapist or a biomedical engineer communicated with the patients to ensure compliance with the study, answer questions about the device, fix any malfunctions, and discuss EEG signal quality, which was assessed regularly by a biomedical engineer. After 12 weeks, patients were again tested on all primary and secondary outcome measures. Different occupational therapists collected baseline and completion outcome measures, and all occupational therapists were blinded to observed EEG changes.

EEG signal quality, which was assessed regularly by a biomedical engineer. After 12 weeks, patients were again tested on all primary and secondary outcome measures. Different occupational therapists collected baseline and completion outcome measures, and all occupational therapists were blinded to observed EEG changes. Analysis of Outcome Measures A paired-sample t test was used to evaluate the statistical significance of ARAT changes and continuous secondary outcome measures (grip strength, pinch strength, and AROM). Signed-rank tests were used for all other outcome measures because their measurement scales were ordinal. Because the exoskeleton drove extension of the second and third digits, AROM values for the second and third digits and fourth and fifth digits were averaged separately. Changes were examined for both the overall and subcomponents of the ARAT and Motricity Index.

measures because their measurement scales were ordinal. Because the exoskeleton drove extension of the second and third digits, AROM values for the second and third digits and fourth and fifth digits were averaged separately. Changes were examined for both the overall and subcomponents of the ARAT and Motricity Index. Neurophysiological Correlates To examine potential mechanisms of action, we calculated the correlation between the change in ARAT and changes in BCI control accuracy, total usage time, and EEG modulation changes. To quantify BCI performance, we calculated the average hand position in the second half of each trial. The BCI accuracy for each run of the BCI task was calculated by taking the difference in this average position between movement and rest trials. EEG modulation was determined by calculating the coefficient of determination (r2 value) quantifying the difference in EEG spectral power between motor imagery and rest trials. The change in BCI accuracy and EEG modulation was defined as the slope of a robust multilinear regression representing the change per run of the BCI task. The relationship between the ARAT change and change in both BCI control accuracy and EEG modulation was measured with Spearman r. To control for the location and frequency used for BCI control, we performed 3 control analyses: (1) change in EEG modulation at the same frequency but at the location contralateral to the control site (ipsilesional motor cortex), (2) change in EEG modulation at the same frequency used for control but at a nonmotor electrode site (F3), and (3) change in EEG modulation at the location used for BCI control (contralesional motor cortex) but at a different frequency (50 Hz). Because patients performed the BCI task at home, poor-quality EEG activity was observed on some days. Thus, we included only those runs in which BCI control signals significantly (P<0.01) differed between movement and rest trials.

used for BCI control (contralesional motor cortex) but at a different frequency (50 Hz). Because patients performed the BCI task at home, poor-quality EEG activity was observed on some days. Thus, we included only those runs in which BCI control signals significantly (P<0.01) differed between movement and rest trials. Results Ten patients completed the study. Patient characteristics are summarized in the Table, and the online-only Data Supplement contains a detailed description of patient recruitment. In short, of the 22 patients who completed EEG screenings, 18 (81%) were suitable for further BCI therapy, 13 (59%) began the therapy, and 10 (45%) completed the study. The drop off was because of a variety of causes, including unrelated medical diagnoses, inability to comply with the time commitment, and poor orthosis fit. BCI Control After initial training, patients and their caregivers were able to apply EEG electrodes in the home setting to record physiological EEG signals. Figure 2 shows exemplary movement-related EEG activity observed in the laboratory and while at home. The patient demonstrated bilateral μ- and β-band power decreases in both settings. Furthermore, the patient had very similar spatial and spectral patterns of movement-related EEG activity during both sessions. The significant decrease in power during motor imagery in the BCI control task led to a high level of accuracy with discriminable patterns of exoskeleton movement during rest and motor imagery.

Furthermore, the patient had very similar spatial and spectral patterns of movement-related EEG activity during both sessions. The significant decrease in power during motor imagery in the BCI control task led to a high level of accuracy with discriminable patterns of exoskeleton movement during rest and motor imagery. Figure 2. Exemplar electroencephalographic (EEG) activity and brain–computer interface (BCI) control. A, During an exemplar laboratory-based screening session, the patient (patient 10, left affected) demonstrated significant decreases in μ- and β-band spectral power bilaterally. The color scale shows signed r2 values indicating increases (positive values) and decreases (negative values) in spectral power during motor imagery. A BCI control feature (red box) ipsilateral to the affected hand was chosen (contact C3). B, During a home-based BCI control session, a similar spatiospectral pattern of movement-related EEG activity was observed. C, The mean (±SE) of the hand position in movement and rest trials shows that the patient achieved a high level of BCI control (0% fully closed, 100% fully open).

ted hand was chosen (contact C3). B, During a home-based BCI control session, a similar spatiospectral pattern of movement-related EEG activity was observed. C, The mean (±SE) of the hand position in movement and rest trials shows that the patient achieved a high level of BCI control (0% fully closed, 100% fully open). Because our hypothesis focused on the contralesional hemisphere, the features used to drive the BCI system were from electrodes over the contralesional motor cortex. Movement-related EEG activations were also observed from the ipsilesional hemisphere in 8 of the 10 patients. Although the frequency used for BCI control varied across patients, all BCI control features were μ- and β-band power suppressions, also referred to as event-related desynchronization.22 Patients used the device on 37 to 72 days. Patients performed 74 to 465 10-minute runs of the BCI task for a total of 740 to 4650 minutes of online BCI control in addition to the daily screening task. Details of the patient-specific BCI control are included in the online-only Data Supplement.

lated desynchronization.22 Patients used the device on 37 to 72 days. Patients performed 74 to 465 10-minute runs of the BCI task for a total of 740 to 4650 minutes of online BCI control in addition to the daily screening task. Details of the patient-specific BCI control are included in the online-only Data Supplement. Functional Outcomes The 2 baseline motor assessments were averaged to determine each patient’s baseline motor function. ARAT changes throughout the study protocol are shown in Figure 3A. Patients had a statistically significant mean ARAT increase of 6.2 points. Importantly, 5.7 points has been estimated to represent the minimal clinically important difference in chronic stroke survivors.23 Specifically, 6 of the 10 patients had ARAT improvements above this level. In addition to this per-protocol analysis, a significant increase in ARAT score was also found using an intention-to-treat analysis as described in the online-only Data Supplement. Grasp strength, Motricity Index, the grip and grasp ARAT subscores, and Canadian Occupational Performance Measure performance and satisfaction ratings were also significantly increased after therapy, whereas pinch strength, AROM, and the pinch and gross ARAT subscores were not changed. Figure 4 and Table II in the online-only Data Supplement summarize changes across outcomes. Other than minor fatigue, no negative effects were observed.

rmance and satisfaction ratings were also significantly increased after therapy, whereas pinch strength, AROM, and the pinch and gross ARAT subscores were not changed. Figure 4 and Table II in the online-only Data Supplement summarize changes across outcomes. Other than minor fatigue, no negative effects were observed. Figure 3. Improvement in motor function. A, Each line shows the change in Action Research Arm Test (ARAT) during the study. At completion, 6 of 10 patients had ARAT increases surpassing the minimal clinically important difference (MCID; 5.7 points). B, ARAT increases were related to the rate of change in brain–computer interface (BCI) accuracy (Spearman r=0.75, P=0.013). C, ARAT increases were not related to the time of device use (Spearman r=0.47, P=0.17). Figure 4. Summary of outcome measures. Each box shows the distribution of each outcome measurement at baseline and study completion. Boxes show the 25th percentile, median, and 75th percentile; bars indicate the range of values; and outliers >2.7 SDs from the mean are marked with a +. Measures with statistically significant (P<0.05) changes are indicated with an *. ARAT indicates Action Research Arm Test; and COPM, Canadian Occupational Performance Measure.

ow the 25th percentile, median, and 75th percentile; bars indicate the range of values; and outliers >2.7 SDs from the mean are marked with a +. Measures with statistically significant (P<0.05) changes are indicated with an *. ARAT indicates Action Research Arm Test; and COPM, Canadian Occupational Performance Measure. Neurophysiological Correlates Across patients, there was a significant correlation between the change in ARAT score and the change in BCI accuracy (defined as the difference between the hand position in the movement and rest trials) per BCI task run (Figure 3B; Spearman r=0.75, P=0.013). There was not a significant relationship between the change in ARAT score and the total device usage time (Figure 3C; Spearman r=0.47, P=0.17). Finally, we sought to determine whether there was a relationship between ARAT and EEG changes (Figure 5). There was a trend toward a positive relationship between ARAT score changes and the change in the EEG modulation per run of the BCI task at the location and frequency used for BCI control and in a site in the contralateral motor cortex (BCI control feature: Spearman r=0.48, P=0.16, contralateral motor cortex: Spearman r=0.62, P=0.06).

d toward a positive relationship between ARAT score changes and the change in the EEG modulation per run of the BCI task at the location and frequency used for BCI control and in a site in the contralateral motor cortex (BCI control feature: Spearman r=0.48, P=0.16, contralateral motor cortex: Spearman r=0.62, P=0.06). Figure 5. Relationship between changes in electroencephalographic (EEG) activity and Action Research Arm Test (ARAT) improvements. Ranked changes in motor function (ARAT) and changes in EEG activations (r2 value) per brain–computer interface (BCI) run are shown. A, Analyses were performed using EEG activity at the site and frequency used for BCI control, at the frequency used for BCI control but an electrode in the contralateral hemisphere, at the frequency used for BCI control but an electrode in the frontal lobe (F3; serving as a spatial control), and at the site used for BCI control but at 50 Hz (serving as a spectral control). B, There was a positive relationship that trended toward significance at both the BCI control feature (top left) and in the contralateral motor cortex (top right) but not at a location outside the motor cortex (bottom left) or a task-irrelevant frequency (bottom right).

rol but at 50 Hz (serving as a spectral control). B, There was a positive relationship that trended toward significance at both the BCI control feature (top left) and in the contralateral motor cortex (top right) but not at a location outside the motor cortex (bottom left) or a task-irrelevant frequency (bottom right). Discussion This study provides evidence for the potential role of the unaffected hemisphere in rehabilitation via a BCI-controlled exoskeleton. Specifically, patients had an average ARAT improvement surpassing the minimal clinically important difference.23 In addition, improvements were observed in some, but not all, objective secondary measures of function. Although pinch strength, AROM, and the ARAT pinch subcomponent did not change, these measures are less sensitive in more severely impaired patients and were likely affected by a qualitative increase in spasticity observed, particularly in patients who had received botox 90 to 120 days before study onset. Furthermore, the grasp and grip ARAT subcomponents and grip strength, which all involve distal hand function, significantly improved. It is uncertain whether observed improvements in general distal hand function that did not localize to pinch were because of the poor spatial specificity of EEG or the sensitivity of pinch-specific subcomponents. Finally, we also observed statistically significant increases in a self-scored subjective measure of each patient’s use of their affected arm in functional tasks (Canadian Occupational Performance Measure). These findings build on previous evidence that BCI-controlled rehabilitation systems can facilitate motor recovery.4–8 There are several features that distinguish this work from previous studies. First, this study was the first to focus exclusively on using the unaffected hemisphere in a BCI rehabilitation system. Second, the BCI drove the velocity of the exoskeleton, providing a closer temporal pairing between brain activity and proprioceptive feedback than previous systems.4,6

this work from previous studies. First, this study was the first to focus exclusively on using the unaffected hemisphere in a BCI rehabilitation system. Second, the BCI drove the velocity of the exoskeleton, providing a closer temporal pairing between brain activity and proprioceptive feedback than previous systems.4,6 The choice of a BCI control signal for poststroke motor rehabilitation requires careful consideration, particularly given the conflicting evidence on the unaffected hemisphere after stroke.12–15,24–28 By pairing cortical activations with peripheral feedback, we hypothesized that we would induce plasticity in the remaining (ipsilateral) central nervous system pathways. As noted, there was a significant relationship between the change in ARAT scores and the rate of change in BCI control accuracy that could not be explained by the volume of device use. Further, there was a trend toward a significant relationship between the rate of change in EEG activity and ARAT score specific to the bilateral motor system, but not in the frontal lobe or at task-irrelevant frequencies. Therefore, although what can be asserted from a mechanistic standpoint is somewhat limited, the results indicate that the choice of a BCI control feature in the unaffected hemisphere may have played an important role in the benefits of the intervention.

em, but not in the frontal lobe or at task-irrelevant frequencies. Therefore, although what can be asserted from a mechanistic standpoint is somewhat limited, the results indicate that the choice of a BCI control feature in the unaffected hemisphere may have played an important role in the benefits of the intervention. There are many potential explanations that could account for the functional improvements observed. Specifically, although postrecovery increases in activity have been found in both the affected and unaffected hemispheres,16,24,26,29 the reorganization of interhemispheric connectivity between the contralesional and ipsilesional motor cortices may also play a role in functional recovery.17,28 Further studies designed to better define the mechanism of action will be beneficial to better understand the characteristics of patients who will benefit optimally from BCIs controlled from the unaffected hemisphere. Because the integrity of the ipsilesional corticospinal tract is strongly correlated with motor recovery,17 we would hypothesize that the corticospinal tract integrity is essential in determining what role the contralesional hemisphere will play in recovery. Specifically, in patients with the greatest corticospinal tract damage, we would expect recovery to require an alternative pathway, such as fibers descending ipsilateral to the contralesional motor cortex.

ospinal tract integrity is essential in determining what role the contralesional hemisphere will play in recovery. Specifically, in patients with the greatest corticospinal tract damage, we would expect recovery to require an alternative pathway, such as fibers descending ipsilateral to the contralesional motor cortex. This study was also unique in that the system was used in the home setting without daily oversight. Traditional BCI systems for rehabilitation have been used in a laboratory setting with trained experts operating them.4–8 The ability to provide therapy in a patient’s home without constant supervision would likely reduce the cost of therapy, increase the time of therapy, and give patients flexibility in scheduling therapy. For this approach to achieve large-scale implementation, several practical aspects will need to be addressed, including building the system in a cost-effective fashion, optimizing the orthosis and EEG headset design for enhanced user experience and compliance, and integrating the hardware and software to enable seamless remote maintenance and minimize the need for EEG quality checks.

ractical aspects will need to be addressed, including building the system in a cost-effective fashion, optimizing the orthosis and EEG headset design for enhanced user experience and compliance, and integrating the hardware and software to enable seamless remote maintenance and minimize the need for EEG quality checks. There are also several limitations to note. Because of the home-based setting, it was impossible to ensure that data were free from artifacts. Although the majority of patients had good-quality EEG recordings in the majority of sessions, a few patients met this standard in <50% of sessions. In addition, because the study sample is small in size and was restricted to those with enough motivation to complete the study protocol, the scope and generalizability of the results is uncertain. Also, pinch strength, all Motricity Index subcomponents, ARAT pinch and gross subcomponents, and AROM did not improve. Whether this was because of the poorer sensitivity of these subcomponents combined with the small sample size, the poor spatial resolution of the EEG signals used, or a limitation of the therapy is uncertain. Finally, the study was uncontrolled. Previous work has shown ARAT improvements can be achieved in chronic stroke patients after interventions such as constraint-induced movement therapy or standard physical therapy,30 but patients in these studies began with a much higher baseline ARAT score than the current cohort. Also of note, while shorter in duration (2 weeks), a randomized controlled trial of a BCI-controlled hand orthosis in patients with a similar baseline motor function showed no improvement in a control group receiving a sham therapy.6 Taken together, there remains an open question of whether more severely affected chronic stroke patients benefit from a BCI intervention exclusively versus prolonged physical therapy; a question that will ultimately be answered with a randomized clinical trial. However, this work provides important early evidence that training with a BCI-driven orthosis can be implemented in the home environment and is associated with a meaningful functional improvement.

xclusively versus prolonged physical therapy; a question that will ultimately be answered with a randomized clinical trial. However, this work provides important early evidence that training with a BCI-driven orthosis can be implemented in the home environment and is associated with a meaningful functional improvement. Conclusions This feasibility study shows a statistically significant and clinically meaningful improvement in the motor function of chronic stroke survivors after using a home-based BCI-controlled exoskeleton. The use of control features in the contralesional hemisphere shows evidence of the potential relevance of the unaffected hemisphere for functional rehabilitation. Collectively, although this study represents an important step toward developing and translating BCI-driven rehabilitation protocols for chronic stroke survivors, the effectiveness of BCI-driven therapies must be proven in large randomized controlled trials before full acceptance. Sources of Funding This study was funded by Neurolutions, Inc. Disclosures Dr Bundy, Dr Schalk, R. Coker, Dr Moran, and Dr Leuthardt own stock in Neurolutions, Inc. Study data were reviewed by an unaffiliated neurologist before submission as part of a comprehensive conflict of interest management plan. The other authors report no conflicts. Supplementary Material * Drs Huskey and Leuthardt contributed equally. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.116.016304/-/DC1.

Stroke is the leading cause of serious long-term disability in the United States and the fifth leading cause of death.1,2 The stroke illness trajectory presents a unique challenge that is distinct from most other serious illnesses;3 the presentation is sudden and unexpected, and whereas a small proportion of patients will receive potentially curative acute treatment, patients with severe stroke typically face treatment decisions that leave little time for deliberation and can lead either to an early death in the setting of withdrawal or withholding of life-sustaining interventions or enable survival with a wide range of disability.3 In some cases, survival may be worse than death.4 Given their neurological impairment, conversations about goals of care usually occur between providers and surrogate decision makers, rather than with the patient themselves.

thholding of life-sustaining interventions or enable survival with a wide range of disability.3 In some cases, survival may be worse than death.4 Given their neurological impairment, conversations about goals of care usually occur between providers and surrogate decision makers, rather than with the patient themselves. These observations highlight the importance of integrating palliative care into the acute stroke care setting.5 Palliative care is a multidisciplinary approach to medical care that focuses on improving communication, decision making, and quality of life for patients with serious illness and their families. Early integration of palliative care into acute stroke care has been recently endorsed by the American Heart6/American Stroke Association7 and Neurocritical Care Society,8 but data remain limited about how to best implement these recommendations and how to measure their benefit.9–11 Palliative care services are available in an increasing number of US hospitals, but substantial variations remain in access and use across regional, socioeconomic, racial, and ethnic groups.12,13

re Society,8 but data remain limited about how to best implement these recommendations and how to measure their benefit.9–11 Palliative care services are available in an increasing number of US hospitals, but substantial variations remain in access and use across regional, socioeconomic, racial, and ethnic groups.12,13 In an effort to evaluate hospital quality of care, the Center for Medicare and Medicaid Services (CMS) currently uses 30-day mortality after ischemic stroke as a quality of care surrogate14 and provides access to adjusted rates on the CMS hospital compare website (https://www.medicare.gov/hospitalcompare/search.html). Inpatient mortality after stroke varies widely depending on several patient and hospital characteristics, including the hospital’s use of do not resuscitate (DNR) orders.7,15,16 Use of palliative care after stroke is likely also associated with higher in-hospital and 30-day mortality rates and may not indicate lower hospital quality of care: this is not accounted for in the CMS stroke mortality quality measure. The overall goal of this study was to characterize current practices around the use of palliative care in a nationally representative sample of patients with stroke by (1) identifying patient and hospital characteristics associated with palliative care utilization, and (2) assessing how the use of palliative care influences inpatient mortality.

tudy was to characterize current practices around the use of palliative care in a nationally representative sample of patients with stroke by (1) identifying patient and hospital characteristics associated with palliative care utilization, and (2) assessing how the use of palliative care influences inpatient mortality. Methods Database We performed a retrospective observational study in patients with stroke admitted to US acute care hospitals using discharge data from the publicly available national inpatient sample (NIS), healthcare cost and utilization project, and agency for healthcare research and quality.17–19 The NIS is a cross-sectional, all-payer, inpatient care data set in the United States, consolidated on an annual basis. It is the largest inpatient health data set in the United States. Unweighted, it contains data from >7 million hospital stays from >1000 hospitals each year, which represent a stratified sample of 20% of all nonfederal hospitals. Weighted, it estimates >35 million hospitalizations nationally. Discharge data include demographics, socioeconomics, primary and secondary diagnoses, procedures, and length of stay (LOS). The NIS database contains deidentified information and is exempt from institution review board approval at our institution.

spitals. Weighted, it estimates >35 million hospitalizations nationally. Discharge data include demographics, socioeconomics, primary and secondary diagnoses, procedures, and length of stay (LOS). The NIS database contains deidentified information and is exempt from institution review board approval at our institution. Stroke Data Selection We identified adult (age, >18 years) stroke admissions from 2010 to 2012 using International Classification of Diseases-Ninth Revision (ICD-9) diagnosis codes. Codes 433.X1–occlusion and stenosis of cerebral artery with infarction, 434.X1–occlusion of cerebral artery with infarction, and 436–acute but ill-defined cerebrovascular disease, irrespective of their diagnosis position, were used to identify ischemic strokes. Code 430 (first diagnosis only) was used to identify subarachnoid hemorrhage (SAH) and 431 (first diagnosis only) for intracerebral hemorrhage (ICH). Cases were excluded if there was a concomitant ICD-9 code for traumatic brain injury or rehabilitation stay.20

diagnosis position, were used to identify ischemic strokes. Code 430 (first diagnosis only) was used to identify subarachnoid hemorrhage (SAH) and 431 (first diagnosis only) for intracerebral hemorrhage (ICH). Cases were excluded if there was a concomitant ICD-9 code for traumatic brain injury or rehabilitation stay.20 Demographic and socioeconomic factors were identified from the primary data set. Race/ethnicity had a high degree of missing data compared with other variables because of state suppression or partial reporting by hospitals. We identified individuals with intubation and PEG (percutaneous endoscopic gastrostomy) tube placement separately as proxy for life-prolonging care in these patients. In addition, cancer, heart disease, and dementia were identified because of implications for end-of-life care, and atrial fibrillation was identified because of its increased risk of large cardioembolic strokes.

neous endoscopic gastrostomy) tube placement separately as proxy for life-prolonging care in these patients. In addition, cancer, heart disease, and dementia were identified because of implications for end-of-life care, and atrial fibrillation was identified because of its increased risk of large cardioembolic strokes. Palliative Care Palliative care was identified using the ICD-9-CM procedure, code V66.7 (palliative care encounter [PCE]), in the hospital discharge data. This code is added by billing staff when components of palliative care, such as comfort care, end-of-life care, and hospice care, are mentioned in the treatment record of the patient and is independent of whether or not a palliative care specialist was consulted or not.21 The PCE code is not used for pain and symptom management. This article uses the term PCE to indicate the presence of V66.7 code in the patient’s medical record. Several scenarios about the use of the V66.7 code in end-of-life and hospice care admissions and its interpretation by multiple national databases, such as CMS and US News and World Report, have been described.22 Recently, this code was examined in patients with ICH using NIS data from the previous decade.23

al record. Several scenarios about the use of the V66.7 code in end-of-life and hospice care admissions and its interpretation by multiple national databases, such as CMS and US News and World Report, have been described.22 Recently, this code was examined in patients with ICH using NIS data from the previous decade.23 Death We used the healthcare cost and utilization project database uniform discharge disposition to track death during hospitalization. We compared the timing of death in PCE versus non-PCE patients. We defined early death as death occurring with hospital LOS ≤2 days. We explored implications of early death for stroke mortality as a CMS measure of high-quality care in the setting of PCE. As the healthcare cost and utilization project database format changed in 2012, this combined analysis was limited to 2010 and 2011 data. Statistical Analysis Pearson χ2 test was used to compare proportions between categories of PCE versus no PCE. Logistic regression was used to evaluate independent associations with PCE use. Covariates for logistic regression included age, race, sex, hospital characteristics, all-patient refined diagnosis-related group severity, and year. Statistical significance was defined as a P value of <0.05. Statistical analysis was performed using STATA data analysis and statistical software.

ociations with PCE use. Covariates for logistic regression included age, race, sex, hospital characteristics, all-patient refined diagnosis-related group severity, and year. Statistical significance was defined as a P value of <0.05. Statistical analysis was performed using STATA data analysis and statistical software. The all-patient refined diagnosis-related group, which assesses risk of mortality using an algorithm developed by 3 mol/L health information systems, was used to determine disease severity and its correlation with PCE. All-patient refined diagnosis-related group is a proprietary 4-point ordinal scale (minor, moderate, major, and extreme risk of mortality) derived from age, primary and secondary diagnoses, and procedures.24 Results We identified 395 411 adult patients with stroke. The majority of patients had ischemic strokes (86%) followed by ICH (10%) and SAH (4%). The mean age was 70.1 years (SD, 16), 52% were women and 69% were white. Among all patients with stroke, 24 641 (6.2%) received PCE, and this proportion increased with each study year from 5.4% in 2010 to 6.9% in 2012 (Table I in the online-only Data Supplement).

chemic strokes (86%) followed by ICH (10%) and SAH (4%). The mean age was 70.1 years (SD, 16), 52% were women and 69% were white. Among all patients with stroke, 24 641 (6.2%) received PCE, and this proportion increased with each study year from 5.4% in 2010 to 6.9% in 2012 (Table I in the online-only Data Supplement). Palliative Care and Patient Characteristics Bivariate analysis of pertinent variables is presented in Table 1. Although specific stroke severity scales (National Institutes of Health Stroke Scale, ICH score, Hunt/Hess) were not available in this cohort, proxies of overall illness severity, including the all-patient refined diagnosis-related group severity subclass and codes for intubation and coma, were associated with an increased rate of PCE use, whereas PEG placement was less common among patients with PCE (Table 1). Table 1. Patient and Hospital Characteristics in Relation to Palliative Care Encounter (Bivariate Analysis) Using multivariate analysis, we found a variety of patient characteristics that were independently associated with the use of PCE (Table 2), including older age and female sex. Compared with whites, the rate of PCE use was significantly lower in blacks (odds ratio [OR], 0.62), Hispanics (OR, 0.67), and Asians (OR, 0.73). ICH, while representing only 10% of overall strokes, was associated with a higher rate of PCE use than ischemic stroke (OR, 3.40). Table 2. Logistic Regression: Predictors of Palliative Care Encounter

Using multivariate analysis, we found a variety of patient characteristics that were independently associated with the use of PCE (Table 2), including older age and female sex. Compared with whites, the rate of PCE use was significantly lower in blacks (odds ratio [OR], 0.62), Hispanics (OR, 0.67), and Asians (OR, 0.73). ICH, while representing only 10% of overall strokes, was associated with a higher rate of PCE use than ischemic stroke (OR, 3.40). Table 2. Logistic Regression: Predictors of Palliative Care Encounter The mean LOS for all patients with stroke receiving PCE was 6.8 days (95% confidence interval, 6.66–6.87), which was significantly longer than in patients who did not receive PCE (5.7 days; 95% confidence interval, 5.64–5.69). When looking at each stroke subtype separately, this association was evident for patients with ischemic stroke (7.4 versus 6.2 days). Conversely, PCE was associated with shorter LOS in patients with ICH (5.0 versus 8.3 days) or SAH (6 versus 12 days; Table 3). Table 3. LOS With and Without PCE Palliative Care and Hospital Characteristics Hospitals with higher PCE use included large hospitals (OR, 1.24 compared with small hospitals), urban teaching (OR, 1.1 compared with rural), nonprofit hospitals (OR, 1.22 compared with government hospitals), and western states (OR, 1.5 compared with northeast hospitals). In general, hospitals with higher mortality also had higher use of PCE. However, this trend comes with a wide variability showing some hospitals with low PCE use and high mortality, as well hospitals with low mortality and high PCE use (Figure 1).

pitals), and western states (OR, 1.5 compared with northeast hospitals). In general, hospitals with higher mortality also had higher use of PCE. However, this trend comes with a wide variability showing some hospitals with low PCE use and high mortality, as well hospitals with low mortality and high PCE use (Figure 1). Figure 1. Rates of death and palliative care encounter (PCE) among patients with stroke by age. Palliative Care and Inpatient Mortality Among all patients with stroke, 36 397 (9.2%) died in hospital, and the rate of death declined from 2010 (10.9%) to 2012 (9.8%; P<0.001; Table I in the online-only Data Supplement). Patient characteristics that were independently associated with higher mortality after stroke included older age (≥80; OR, 2.64 compared with <60), female sex (OR, 1.04), white race (OR, 0.77 for black versus white), ICH (OR, 4.76 compared with ischemic stroke), and non-Medicare insurance (OR self-pay 1.79 and private insurance 1.27 compared with Medicare). Hospital characteristics independently associated with higher mortality after stroke included small hospitals, hospitals in the northeast region, hospitals in rural areas, and public hospitals.

mpared with ischemic stroke), and non-Medicare insurance (OR self-pay 1.79 and private insurance 1.27 compared with Medicare). Hospital characteristics independently associated with higher mortality after stroke included small hospitals, hospitals in the northeast region, hospitals in rural areas, and public hospitals. Among the patients who died, more than one third (38%) had received PCE (Table II in the online-only Data Supplement). The proportion of PCE was highest among patients dying with ICH (42%), followed by ischemic stroke (36%) and SAH (33%). Nonwhite races were less likely to die in hospital, and among those who died, nonwhites were significantly less likely to receive PCE. The rates of death and PCE both increased with age, whereas the difference between the 2 decreased: among young patients (<60 years), more than twice as many patients died than received PCE. In the oldest patients (>90), more patients received PCE than died (Figure 2). Figure 2. Proportion of deaths per hospital length of stay among patients who received palliative care encounter (A). All patients (B through D) by stroke type. ICH indicates intracerebral hemorrhage; and SAH, subarachnoid hemorrhage.

Among the patients who died, more than one third (38%) had received PCE (Table II in the online-only Data Supplement). The proportion of PCE was highest among patients dying with ICH (42%), followed by ischemic stroke (36%) and SAH (33%). Nonwhite races were less likely to die in hospital, and among those who died, nonwhites were significantly less likely to receive PCE. The rates of death and PCE both increased with age, whereas the difference between the 2 decreased: among young patients (<60 years), more than twice as many patients died than received PCE. In the oldest patients (>90), more patients received PCE than died (Figure 2). Figure 2. Proportion of deaths per hospital length of stay among patients who received palliative care encounter (A). All patients (B through D) by stroke type. ICH indicates intracerebral hemorrhage; and SAH, subarachnoid hemorrhage. In the group of patients who died in hospital, the mean LOS was 7.0 days (95% confidence interval, 6.9–7.1). PCE use was associated with longer LOS in patients with ischemic stroke and with shorter LOS in patients with ICH and SAH. In decedents, PCE use was associated with a shorter LOS overall (6.2 versus 7.5 days) but with a longer LOS in patients with SAH (Table 3). In other words, PCE was associated with early death. The percentage of all PCE-related deaths was the highest in the earliest days of hospitalization both overall and for each stroke type (Figures 1 and 3).

associated with a shorter LOS overall (6.2 versus 7.5 days) but with a longer LOS in patients with SAH (Table 3). In other words, PCE was associated with early death. The percentage of all PCE-related deaths was the highest in the earliest days of hospitalization both overall and for each stroke type (Figures 1 and 3). Discussion Using a well-established database of inpatient admissions in the United States, we found an overall rate of coding of PCEs among patients with stroke of 6.2% and an inpatient mortality rate of 9.2%. We observed substantial variation across patient and hospital characteristics and a strong correlation between palliative care use and death. Our findings have important implications for the use of hospital mortality rates as a CMS quality measure.

g patients with stroke of 6.2% and an inpatient mortality rate of 9.2%. We observed substantial variation across patient and hospital characteristics and a strong correlation between palliative care use and death. Our findings have important implications for the use of hospital mortality rates as a CMS quality measure. Palliative Care, Mortality, and Quality of Care Across Age, Race, and Sex Consistent with the National Vital Statistics Report,1 we observed that stroke admissions and in-hospital mortality increased with age. We found that the rate of PCE increased with age, and this difference was especially pronounced among decedents, where less than half of those younger than 60 years of age received PCE but all of those older than 90 years(Figure 2). In other words, older patients who die do so in the setting of PCE, whereas younger patients are more likely to die without PCE. This finding suggests a lower emphasis on life-prolonging care among older patients with stroke but may not indicate poor quality care. On the contrary, a recent study showed that the presence of DNR orders in patients with acute ischemic stroke, highly associated with older age and mortality, did not predict a lower incidence of stroke quality of care indicators.25

rolonging care among older patients with stroke but may not indicate poor quality care. On the contrary, a recent study showed that the presence of DNR orders in patients with acute ischemic stroke, highly associated with older age and mortality, did not predict a lower incidence of stroke quality of care indicators.25 Use of PCE and hospital mortality in the setting of PCE were also all more common in white people compared with blacks and Hispanics. This finding is consistent with a well-described racial variation in end-of-life care showing consistently lower rates of advance care planning,26 DNR orders,27 palliative care use28 and end-of-life discussions,29 and a higher rate of life-prolonging treatment, including PEG tube placement30 among black patients with serious illness. Quality care indicators, however, are observed less frequently in black patients with stroke compared with white patients,31 and hospital deaths occur alongside high adherence to high-quality, evidence-based stroke care.16 Although not as clear, evidence for the association of female sex with palliative care use, as suggested in our study, has some support in the literature. Previous studies in patients with stroke have indicated higher rates of DNR orders and withdrawal of life-sustaining treatments in women and whites32–35 but slightly lower quality of care.36,37

nce for the association of female sex with palliative care use, as suggested in our study, has some support in the literature. Previous studies in patients with stroke have indicated higher rates of DNR orders and withdrawal of life-sustaining treatments in women and whites32–35 but slightly lower quality of care.36,37 Finally, our results suggest substantial practice variation of PCE use, consistent with the variation previously shown in end-of-life care after stroke, in particular in regards to the use of DNR orders,16,38 and prognostication.39 The variation seen in our study may affect mortality in this group of patients, casting a shadow over the meaning and validity of mortality-based hospital comparisons that fail to account for PCE.

usly shown in end-of-life care after stroke, in particular in regards to the use of DNR orders,16,38 and prognostication.39 The variation seen in our study may affect mortality in this group of patients, casting a shadow over the meaning and validity of mortality-based hospital comparisons that fail to account for PCE. PCE (V66.7) Versus Palliative Care Services The report card published by the center to advance palliative care showed that access to palliative care specialist services in US hospitals has increased in the past decade, but that a variability persists in regards to hospital size, location, and tax status.12 In 2015, one third of US hospitals with 50 or more beds reported no palliative care services.12 The reports’ findings of a reduced rate of palliative care specialist services in smaller and nonacademic centers parallels the lower PCE rates in smaller, for-profit hospitals seen in our study, suggesting similar practice variations for specialist palliative care and palliative end-of-life care. Similarly, the geographic variation in palliative care specialist availability12 corresponds with our finding of a higher PCE rate in the western states of the United States without substantial variation in mortality rates.

ting similar practice variations for specialist palliative care and palliative end-of-life care. Similarly, the geographic variation in palliative care specialist availability12 corresponds with our finding of a higher PCE rate in the western states of the United States without substantial variation in mortality rates. PCE and LOS Among decedents, PCE was associated with a shorter LOS suggesting an earlier death through PCE and less days of aggressive life-sustaining treatment. LOS with PCE was longer in patients with ischemic stroke and shorter for ICH, which may be explained by a larger proportion of less-severe strokes on the one hand and later palliative care engagement for severe but nondeadly ischemic strokes on the other. This hypothesis was supported when we restricted the analysis to patients with ischemic stroke who died in the hospital: the trend reversed to shorter LOS with PCE. When looking only at patients with ICH, the association of PCE with shorter LOS was both seen in all patients with ICH and in decedents, possibly because of the high mortality and prognostic pessimism40 in this stroke type. For the small group of patients with SAH, PCE was associated with shorter LOS, but here, the trend reversed when we looked only at patients who died. One possible explanation may be a difference in the culture of the medical services, given that SAH is typically managed by different medical teams than ischemic stroke and ICH.

small group of patients with SAH, PCE was associated with shorter LOS, but here, the trend reversed when we looked only at patients who died. One possible explanation may be a difference in the culture of the medical services, given that SAH is typically managed by different medical teams than ischemic stroke and ICH. Limitations This study has several limitations, including those related to the retrospective analysis of the NIS database and the nature of an analysis based on ICD-9 coding. Large numbers in this data set lead to statistical significance even with small clinical changes. Owing to the nature of a preexisting database, important patient characteristics, such as stroke severity scales, are unavailable (eg, National Institutes of Health Stroke Scale). Second, ICD-9 coding is typically performed by the billing departments of hospitals based on language used by providers in their documentation. Provider documentation and billing guidelines may vary across individual departments, hospital types, geographic regions, and by individual administrative personnel. In addition, because the V66.7 code is not linked to reimbursement, the documentation may be less reliable. It is possible that our observations indicate an increase in the coding of palliative care rather than an increase in the actual use of palliative care over time. However, the patterns observed in this study correlate with other studies, suggesting a proportionate use of the PCE code. For example, the variability of PCEs across sampling year, age, region and hospital size, and ownership correlate with the availability of palliative care specialist services shown in the center to advance palliative care report card.13 Finally, documentation of PCE does not reflect the entirety of palliative care that a patient receives through primary or specialty palliative care. It also does not act as a surrogate for the degree to which goals of care, early comfort care measures, or surrogate decision making were addressed. Such services may be provided by the primary treating team without specific coding. More research is needed to build palliative and patient-centered care as a measurable healthcare quality metric.

surrogate for the degree to which goals of care, early comfort care measures, or surrogate decision making were addressed. Such services may be provided by the primary treating team without specific coding. More research is needed to build palliative and patient-centered care as a measurable healthcare quality metric. Conclusions Palliative care is increasing among patients with stroke, especially in larger hospitals. Disparities and variability in PCE and mortality across age, sex, race, region, and hospital characteristics are apparent. When evaluating 30-day mortality as a marker of quality of care, the presence or absence of PCE needs to be taken into account. Figure 3. Hospital mortality and hospital palliative care encounter (PCE) use. Each dot represents one hospital. Hospitals with <10 stroke admissions or <10 palliative care encounters per year were excluded. Disclosures Dr Creutzfeldt receives support from the Cambia Health Foundation. The other authors report no conflicts. Supplementary Material The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.016893/-/DC1.

See related article, p 2650 Coronary artery bypass graft (CABG) surgery is the most commonly performed major cardiovascular operation. Carotid artery stenosis is present in ≈6% to 8% of all patients undergoing CABG and is associated with an increased risk of stroke during and after CABG.1,2 Prophylactic treatment of asymptomatic concomitant carotid artery stenosis is managed in different ways, for example, by carotid artery angioplasty and stenting (CAS) or carotid endarterectomy (CEA), either simultaneously with CABG, before CABG, or delayed after CABG (staged or reverse staged). For many years, staged or synchronous CEA has been advocated by many cardiovascular surgeons in the attempt to reduce the perioperative and long-term risk of stroke associated with carotid artery stenosis but only very few patients with this disease entity have been included in controlled clinical trials.3,4 Only data from uncontrolled studies with variable inclusion criteria and end point assessment are available in patients with coexisting cardiac and carotid atherosclerotic disease undergoing CABG without carotid revascularization (by CAS or CEA).5 Some studies have even found that asymptomatic carotid stenosis did not increase the risk of post-CABG stroke.6,7 Therefore, in the absence of any randomized controlled trial, no systematic high-level evidence exists that staged or synchronous CEA and CABG confer any short-term benefit over CABG without CEA. Moreover, improvements in medical therapy have considerably reduced the average long-term risk of ipsilateral stroke in patients with asymptomatic carotid stenosis.8 Thus, any potential long-term benefit conferred by prophylactic CEA may be offset by the relatively high procedural risk reported in systematic reviews.9,10 Therefore, the CABACS trial (Coronary Artery Bypass Graft Surgery in Patients With Asymptomatic Carotid Stenosis) aimed to compare the perioperative safety and long-term efficacy of synchronous CEA and CABG versus isolated CABG in patients with asymptomatic high-grade carotid artery stenosis.

in systematic reviews.9,10 Therefore, the CABACS trial (Coronary Artery Bypass Graft Surgery in Patients With Asymptomatic Carotid Stenosis) aimed to compare the perioperative safety and long-term efficacy of synchronous CEA and CABG versus isolated CABG in patients with asymptomatic high-grade carotid artery stenosis. Methods Study Design This investigator-initiated trial was designed as a multicenter, randomized (one-to-one), open, group sequential trial with 2 parallel arms and blinded end point adjudication. After initial commitment of 35 major German cardiac surgery centers, 25 German and 1 Czech center could be initiated and 17 centers finally recruited. Ethics approval was obtained from each center. The trial was conducted according to Good Clinical Practice guidelines and the principles stated in the latest revision of the Declaration of Helsinki. The trial protocol has been described previously.11 Two amendments to the protocol were implemented in 2012 and 2014 and approved by the local ethics committees of each participating center (online-only Data Supplement). Enrollment was terminated early because of withdrawal of funding following insufficient recruitment.

he trial protocol has been described previously.11 Two amendments to the protocol were implemented in 2012 and 2014 and approved by the local ethics committees of each participating center (online-only Data Supplement). Enrollment was terminated early because of withdrawal of funding following insufficient recruitment. Participants The inclusion criteria were as follows: asymptomatic (past 180 days) internal carotid artery stenosis ≥80% (following criteria of the ECST [European Carotid Surgery Trial],12 main criterion: in-stenosis peak systolic velocity ≥300 cm/s, corresponding to ≥70% NASCET [North American Symptomatic Carotid Endarterectomy Trial]13,14), carotid artery stenosis treatable with CEA, negative pregnancy test in premenopausal women, written informed consent and full legal capacity, ability of the patient to participate in follow-up examinations.

velocity ≥300 cm/s, corresponding to ≥70% NASCET [North American Symptomatic Carotid Endarterectomy Trial]13,14), carotid artery stenosis treatable with CEA, negative pregnancy test in premenopausal women, written informed consent and full legal capacity, ability of the patient to participate in follow-up examinations. Exclusion criteria were as follows: nonatherosclerotic stenosis (eg, dissection, floating thrombus, fibromuscular dysplasia, tumor, and postradiation), complete occlusion or previous stenting of the carotid artery to be treated, additional higher grade intracranial or intrathoracic stenosis (tandem stenosis), recent (past 180 days) ischemic symptoms ipsilateral to carotid stenosis or occlusion, contralateral carotid occlusion or other known indication for carotid revascularization (apart from scheduled CABG), myocardial infarction (non–ST-segment–elevation myocardial infarction or ST-segment–elevation myocardial infarction) within the past 7 days (reduced to 48 hours for non–ST-segment–elevation myocardial infarction after the first amendment) or hemodynamically unstable patients, known high risk for cardiogenic embolism requiring anticoagulation (mechanical heart valve, chronic atrial fibrillation [omitted after the first amendment], left ventricular thrombus, left ventricular aneurysm), evidence for intracranial bleeding within the past 90 days, modified Rankin Scale score of >3 or severe aphasia, patients unlikely to survive >1 year because of concomitant diseases, planned combined cardiac valve replacement or any other cardiac surgery beyond CABG (±CEA) during the procedure, major surgery (apart from study procedures) planned within 8 weeks from randomization, and participation in another clinical trial. Written informed consent was obtained from each patient before trial participation.

combined cardiac valve replacement or any other cardiac surgery beyond CABG (±CEA) during the procedure, major surgery (apart from study procedures) planned within 8 weeks from randomization, and participation in another clinical trial. Written informed consent was obtained from each patient before trial participation. Randomization Eligible patients were randomized to either isolated CABG or synchronous CEA + CABG. To achieve comparable groups, patients were allocated in a concealed way by central preoperative randomization ≥1 day before surgery. To avoid unbalanced prognostic factor distributions, we used a web-based stratified block randomization (strata: center, age [(<60 or ≥60 years], sex [male or female], modified Rankin Scale [score of 0–1 or 2–3]) with randomly varying the block size.

aled way by central preoperative randomization ≥1 day before surgery. To avoid unbalanced prognostic factor distributions, we used a web-based stratified block randomization (strata: center, age [(<60 or ≥60 years], sex [male or female], modified Rankin Scale [score of 0–1 or 2–3]) with randomly varying the block size. Interventions The eligibility of a patient was determined by both a certified study surgeon and a certified study neurologist with experience in cerebrovascular ultrasound examination. Additional imaging of the brain or cerebral circulation was not required or documented as part of the study. CABG with or without CEA under treatment with aspirin was to be performed as soon as possible (maximum within 7 days) after randomization. Standards for surgical treatment were formulated by a surgical quality subcommittee, which also had to approve every surgeon for the study. Each (cardio-) vascular surgeon had to meet the following standards to be certified: anonymous confirmation of the last 30 consecutively performed CEA surgeries, affirmed by the head of department, alone or in combination with anonymous confirmation of the last 150 consecutively performed CABG surgeries, affirmed by the head of department. With regard to carotid revascularization, no policies were prescribed for the use of a shunt, the type of patches, and the use of neurological monitoring or the method of carotid reconstruction (eversion endarterectomy or thromboendarterectomy). With regard to coronary revascularization, the use of extracorporeal circulation (on- or off-pump CABG) was left to the surgeon`s decision. Standards for medical treatment were formulated by a Best Medical treatment subcommittee.

r the method of carotid reconstruction (eversion endarterectomy or thromboendarterectomy). With regard to coronary revascularization, the use of extracorporeal circulation (on- or off-pump CABG) was left to the surgeon`s decision. Standards for medical treatment were formulated by a Best Medical treatment subcommittee. Patients were followed up after 7 days (±1 day), 30 days (±3 days), and 1 year (±30 days) with a neurological examination including evaluation on the National Institutes Health Stroke Scale and the modified Rankin Scale at each visit as well as cerebrovascular ultrasound and DemTect at 30 days and 1 year. The DemTect is a generic dementia screening test, which consists of 5 subtests: a word list, a number transcoding task, a word fluency task, digit span reverse, and delayed recall of the word list. The transformed total score (maximum 18) is corrected for age and does not show any ceiling effect.15 A parallel test version was used at 30 days to avoid retest effects.16 In addition, a yearly telephone follow-up for 5 years after the operation was also prespecified and is still ongoing.

ayed recall of the word list. The transformed total score (maximum 18) is corrected for age and does not show any ceiling effect.15 A parallel test version was used at 30 days to avoid retest effects.16 In addition, a yearly telephone follow-up for 5 years after the operation was also prespecified and is still ongoing. Outcomes The primary composite outcome was the rate of any stroke or death of any cause up to 30 days after the operation (synchronous CEA and CABG versus isolated CABG) or 30 days after randomization for patients not receiving surgery (protocol violations). Stroke was clinically defined as any persistent focal or global neurological deficit lasting longer than 24 hours and presumed to be of no other than vascular origin. All perioperative stroke events were assessed by a study neurologist. Secondary end points are listed in the Table I in the online-only Data Supplement. For the sake of comparability with previous research, we also included a post hoc secondary composite outcome event of any stroke, myocardial infarction, or death. All suspected strokes, myocardial infarctions, and all deaths were adjudicated by an independent end point committee, which was blinded to treatment allocation. Data Management Data on paper case report forms were collected at the ZKSE in Essen. Good Clinical Practice compliant remote data capture (Oracle Clinical) was used for entering, managing, and validating the data from all centers.

All suspected strokes, myocardial infarctions, and all deaths were adjudicated by an independent end point committee, which was blinded to treatment allocation. Data Management Data on paper case report forms were collected at the ZKSE in Essen. Good Clinical Practice compliant remote data capture (Oracle Clinical) was used for entering, managing, and validating the data from all centers. Statistical Methods We initially based the target sample size of 2 groups with 580 participants (total planned sample size 1160) on an estimate of 8.5% frequency of the primary outcome of stroke or death in the synchronous CEA and CABG arm, a clinically relevant 4.5% absolute reduction of this risk to 4.0% in the isolated arm, resulting in 84% power at a 2-sided level of significance of 0.05. The sample size of the first planned interim analysis was 550 (for details see8). In accordance with the second amendment, the first interim analysis was performed after the first 100 patients, as recommended by the Data Safety Monitoring Board, and the final analysis was planned after inclusion of 300 patients. A generalized linear mixed-effects model that includes the fixed factors treatment group and the randomization factors age (<60 or ≥60 years), sex (male or female), modified Rankin Scale (score of 0–1 or 2–3), and a random (intercept) factor center was used to test and estimate the effect on the primary end point. This preplanned, confirmatory analysis of the primary end point was based on the Wald test statistic with corresponding 2-sided P value for the treatment effect and was applied to the intention-to-treat population. Because of the reduced sample size, we did not apply the originally planned group-sequential design with 1 interim analysis but instead performed 1 final confirmatory analysis at a significance level α of 5%. For robustness, we also ran sensitivity analyses using exact Monte Carlo estimation for χ2 tests and analyses excluding major protocol violations (defined in the statistical analysis plan as not meeting any inclusion or exclusion criteria or not finishing the allocated therapy [including change of treatment group, an operation done by a noncertified surgeon, an end point event between randomization and surgical treatment]). For all other exploratory analyses, we also applied an exploratory significance level α of 5% for χ2, log-rank, and Wilcoxon–Mann–Whitney tests but focused more on point (mean, relative risk, and hazard ratio) and interval estimators of effects (95% confidence intervals).

between randomization and surgical treatment]). For all other exploratory analyses, we also applied an exploratory significance level α of 5% for χ2, log-rank, and Wilcoxon–Mann–Whitney tests but focused more on point (mean, relative risk, and hazard ratio) and interval estimators of effects (95% confidence intervals). Details on the biometric analyses—especially those pertaining to the secondary end points—were defined in the statistical analysis plan before data bank closure. Analyses were performed using SAS version 9.4. Role of Funding Source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all data from the trial and had final responsibility for the results on submission for publication.

Details on the biometric analyses—especially those pertaining to the secondary end points—were defined in the statistical analysis plan before data bank closure. Analyses were performed using SAS version 9.4. Role of Funding Source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all data from the trial and had final responsibility for the results on submission for publication. Results Between December 2010 and December 2014, 129 patients were randomized in 17 centers with an average recruitment rate of 2.6 per year. Two patients withdrew consent between randomization and treatment and were excluded from the analysis. The intention-to-treat population consisted of 127 patients who were allocated to the 2 arms (Figure 1). Baseline characteristics of both treatment groups are shown in Table 1. Two patients in the synchronous CEA and CABG group did receive neither CABG nor CEA and had no outcome events up to 1 year. One patient in the isolated CABG arm received aortic valve replacement instead of CABG and likewise had no outcome events up to 1 year. Two patients with perioperative ipsilateral ischemic events (1 stroke and 1 TIA) received CAS at 27 and 92 days after isolated CABG. Two patients in the isolated CABG group received CAS of asymptomatic carotid stenosis >30 days after CABG, and 3 patients received CEA of asymptomatic carotid stenosis >30 days after CABG with no subsequent outcome events.

ilateral ischemic events (1 stroke and 1 TIA) received CAS at 27 and 92 days after isolated CABG. Two patients in the isolated CABG group received CAS of asymptomatic carotid stenosis >30 days after CABG, and 3 patients received CEA of asymptomatic carotid stenosis >30 days after CABG with no subsequent outcome events. Table 1. Baseline Characteristics of the Patients Included in the Intention-to-Treat Population Figure 1. CONSORT flow diagram (Consolidated Standards of Reporting Trials) of patients in the CABACS (Coronary Artery Bypass Graft Surgery in Patients With Asymptomatic Carotid Stenosis) trial. CABG indicates coronary artery bypass grafting; and CEA, carotid endarterectomy.

Table 1. Baseline Characteristics of the Patients Included in the Intention-to-Treat Population Figure 1. CONSORT flow diagram (Consolidated Standards of Reporting Trials) of patients in the CABACS (Coronary Artery Bypass Graft Surgery in Patients With Asymptomatic Carotid Stenosis) trial. CABG indicates coronary artery bypass grafting; and CEA, carotid endarterectomy. The rates of the primary composite end point of any stroke or death from any cause within 30 days after operation in the intention-to-treat population were 12/65 (18.5%) in the patients who underwent synchronous CEA and CABG and 6/62 (9.7%) in the patients who underwent isolated CABG (absolute risk reduction, 8.8%; 95% confidence interval, −3.2% to 20.8%; PWALD=0.12). The following major protocol violations resulted in exclusion from the per-protocol analysis: operation by a noncertified surgeon (n=11), myocardial infarction before CABG (n=1), informed consent not provided (n=1), no CABG performed (n=3), prior stenting of carotid stenosis to be treated according to the study protocol (n=1), and no high-grade carotid artery stenosis (n=2). In the per-protocol analysis, primary composite end point event rates at 30 days were 11 of 56 patients (19.6%) who received synchronous CEA and CABG and 6 of 53 patients (11.3%) who received isolated CABG (absolute risk reduction, 8.3%; 95% confidence interval, −5.1% to 21.8%; PWALD=0.21).

rotid artery stenosis (n=2). In the per-protocol analysis, primary composite end point event rates at 30 days were 11 of 56 patients (19.6%) who received synchronous CEA and CABG and 6 of 53 patients (11.3%) who received isolated CABG (absolute risk reduction, 8.3%; 95% confidence interval, −5.1% to 21.8%; PWALD=0.21). Similarly, there was no evidence for a significant treatment-group effect for all secondary end points at 30 days and 1 year (Figure 2) although absolute event rates for isolated CABG were lower for most secondary end points. Sensitivity analyses, showing the same trend but no significant differences, in the per-protocol population are provided in the Table I in the online-only Data Supplement.

all secondary end points at 30 days and 1 year (Figure 2) although absolute event rates for isolated CABG were lower for most secondary end points. Sensitivity analyses, showing the same trend but no significant differences, in the per-protocol population are provided in the Table I in the online-only Data Supplement. Figure 2. Secondary end points at 30 days and 1 year (forest plot of risk ratios and hazard ratios plotted on a logarithmic scale). 1For day 30 and year 1, absolute and relative frequencies; for time-to-event analysis, 1-year Kaplan–Meier estimates; for length of hospital and ICU stay, mean and SD. 2For day 30 and year 1, relative risk; for time-to-event analysis, unadjusted hazard ratios for treatment variable from Cox proportional hazards regression; missing effect sizes either not available or not calculated; 3Confirmatory analysis of the primary endpoint was based on the Wald test statistic; for day 30 and year 1, exact Monte Carlo estimation for χ2 test P values; for time-to-event analysis, log-rank test P values; for DemTect scale difference, length of hospital stay and ICU stay exact Wilcoxon–Mann–Whitney test P values. 4Technical failure of intervention can only be measured for the synchronous carotid endarterectomy (CEA) and coronary artery bypass grafting (CABG) arm. CI indicates confidence interval.

log-rank test P values; for DemTect scale difference, length of hospital stay and ICU stay exact Wilcoxon–Mann–Whitney test P values. 4Technical failure of intervention can only be measured for the synchronous carotid endarterectomy (CEA) and coronary artery bypass grafting (CABG) arm. CI indicates confidence interval. A complete clinical follow-up after 1 year could be obtained in 54 (83.1%) patients in the synchronous CEA and CABG arm and in 56 (90.3%) patients in the isolated CABG arm. Reasons for premature study termination by treatment group are displayed in Figure 1. All patients were on antithrombotic treatment, mostly aspirin, after 1 year and concomitant treatment was balanced among treatment groups (Table 2). Kaplan-Meier estimates of any stroke or vascular death-free survival after 1 year are shown in Figure 3. Pre-specified subgroup analyses as well as two post hoc analyses on the degree of ipsilateral and presence of contralateral carotid artery stenosis were in line with the main analysis (Table II online-only Data Supplement). Table 2. Antithrombotic Treatment During Follow-Up (Patient-Based for Drug Classes, Multiple Responses) Figure 3. Kaplan–Meier estimates of survival free from stroke or vascular death up to 1 year in the intention-to-treat population (Plog-rank=0.30). CABG indicates coronary artery bypass grafting; and CEA, carotid endarterectomy.

Table 2. Antithrombotic Treatment During Follow-Up (Patient-Based for Drug Classes, Multiple Responses) Figure 3. Kaplan–Meier estimates of survival free from stroke or vascular death up to 1 year in the intention-to-treat population (Plog-rank=0.30). CABG indicates coronary artery bypass grafting; and CEA, carotid endarterectomy. Discussion CABACS was conducted as a randomized trial to compare synchronous CEA of asymptomatic high-grade carotid artery stenosis versus no carotid operation in patients undergoing CABG surgery. Because of the low power of the trial following early termination, there was no evidence for a treatment-group effect although patients in the synchronous CEA and CABG arm had double the rate of stroke or death within 30 days or within 1 year compared with CABG without CEA. This observation was also found in predefined subgroups. All secondary end points were also more in favor of the isolated CABG arm but likewise failed to demonstrate a significant difference. Of note, we observed a >2-fold higher overall event rate compared with previously published data, which may result from the relatively high age of our study population, treatment quality, the systematic follow-up by study neurologists, or chance because of the relatively small study sample.5,17,18 In contrast, only few patients showed postoperative worsening of cognitive functions as described previously,19 but cognitive testing was not possible in all patients, particularly in those with stroke or severe physical impairment.

p by study neurologists, or chance because of the relatively small study sample.5,17,18 In contrast, only few patients showed postoperative worsening of cognitive functions as described previously,19 but cognitive testing was not possible in all patients, particularly in those with stroke or severe physical impairment. Our trial is the first multicenter randomized controlled trial with a rigorous design to investigate synchronous CEA versus no carotid operation in patients undergoing CABG. In addition, our trial provides data for perioperative events within 30 days and a 1-year follow-up including cognitive testing. The long-term 5-year follow-up is still ongoing. Two previous (bi- respectively monocentric) RCTs comparing synchronous CEA and CABG with delayed CEA after CABG suggested a lower perioperative risk of stroke in patients undergoing synchronous CEA and CABG compared with delayed CEA after CABG alone.3,4 In these studies, however, the very low 30-day risk of stroke or death in the synchronous CEA and CABG arm (1% and 2.8%, respectively) is contradictory to a systematic review and large observational studies and therefore unlikely to represent routine clinical practice.9,18,20

pared with delayed CEA after CABG alone.3,4 In these studies, however, the very low 30-day risk of stroke or death in the synchronous CEA and CABG arm (1% and 2.8%, respectively) is contradictory to a systematic review and large observational studies and therefore unlikely to represent routine clinical practice.9,18,20 Conclusions on the safety of isolated CABG should be made with caution. Although a systematic review and meta-analysis suggested low complication rates in patients with unilateral asymptomatic 70% to 99% carotid artery stenosis undergoing CABG without CEA,5 we found a considerably higher rate of perioperative stroke or death in the isolated CABG arm, which may raise concerns also about the isolated CABG approach. However, patient populations (and surgeons) may not have been comparable to other studies and therefore the only way to directly compare different operative strategies versus isolated CABG remains a head-to-head randomized controlled trial. Compared with prior (staged) CEA, a synchronous CEA and CABG approach requires only 1 anesthesia and does not expose patients to the risk of myocardial infarction while waiting for CABG. On the contrary, overall stroke risk seems to be lower with prior staged than with simultaneous CEA and there seems to be no substantial difference in total vascular morbidity and mortality between the 2 approaches.9,20 CAS has emerged as a possible alternative to CEA, but most clinical trials in patients with high-grade carotid artery stenosis have shown higher perioperative complication rates for CAS than for CEA, particularly in older men, which constitute the majority of patients undergoing CABG.21–23 In a systematic review of cohort studies of predominantly asymptomatic patients with unilateral carotid disease undergoing staged CAS and CABG, the 30-day risk of any stroke or death was 9.1%.24 More recent observational studies have even favored CAS before CABG.20,25,26 However, dual-antiplatelet therapy is mandated for at least 4 weeks after CAS, which for many surgeons constitutes a reason to postpone CABG because of the risk of bleeding, exposing a patient to an additional risk of myocardial infarction.

24 More recent observational studies have even favored CAS before CABG.20,25,26 However, dual-antiplatelet therapy is mandated for at least 4 weeks after CAS, which for many surgeons constitutes a reason to postpone CABG because of the risk of bleeding, exposing a patient to an additional risk of myocardial infarction. Whether asymptomatic high-grade carotid artery stenosis unrelated to CABG requires revascularization is the subject of ongoing studies (SPACE-2 [Stent-Protected Angioplasty in Asymptomatic Carotid Artery Stenosis vs Endarterectomy27], ECST-2 [European Carotid Surgery Trial 2; ISRCTN97744893], CREST-2 [Carotid Revascularization and Medical Management for Asymptomatic Carotid Stenosis Trial; NCT02089217])

ascularization is the subject of ongoing studies (SPACE-2 [Stent-Protected Angioplasty in Asymptomatic Carotid Artery Stenosis vs Endarterectomy27], ECST-2 [European Carotid Surgery Trial 2; ISRCTN97744893], CREST-2 [Carotid Revascularization and Medical Management for Asymptomatic Carotid Stenosis Trial; NCT02089217]) Our trial has several limitations. The prespecified sample size was not reached because of withdrawal of funding after insufficient recruitment and therefore the study was underpowered to demonstrate statistically significant effects for the minimal, clinically relevant effect expected during the planning phase.8 Moreover, investigators were not blinded to treatment allocation. However, main outcome events were adjudicated by blinded observers. Finally, few centers enrolled the large majority of patients, thus limiting the generalizability of our findings. Although all major German cardiovascular centers were invited and the majority participated in this trial, ≈90% of simultaneous CEA and CABG operations in Germany were performed outside of the trial.17 A similar problem was encountered in the SPACE-2 trial, which was also stopped early because of slow enrolment.27 Although all CABACS centers were required to keep screening logs, only a minority assessed all CABG patients for eligibility and thus bias by selective inclusion remains unknown. Reported reasons for the low study inclusion were surgery preferences of the referring physicians, lack of timely screening for carotid artery stenosis in patients scheduled for CABG, insufficient time to obtain informed consent for study participation, and symptomatic stenosis. Furthermore, an unknown number of patients with asymptomatic carotid artery stenosis scheduled for CABG received staged CEA or CAS before coronary bypass surgery but no registry data or complication rates are available for these patients.

ime to obtain informed consent for study participation, and symptomatic stenosis. Furthermore, an unknown number of patients with asymptomatic carotid artery stenosis scheduled for CABG received staged CEA or CAS before coronary bypass surgery but no registry data or complication rates are available for these patients. In conclusion, although we cannot rule out a treatment effect, the very high rate of perioperative strokes does not seem to justify simultaneous CEA in patients with high-grade asymptomatic carotid artery stenosis undergoing CABG. Whether any carotid revascularization (CAS or CEA) is warranted in patients with unilateral asymptomatic high-grade carotid artery stenosis requiring CABG remains to be proven. Follow-up studies of the CABACS trial should test staged CAS or CEA followed by CABG versus isolated CABG and be performed in countries with a younger patient population and with reimbursement of staged or synchronous carotid revascularization only if a patient is treated within the trial.

ng CABG remains to be proven. Follow-up studies of the CABACS trial should test staged CAS or CEA followed by CABG versus isolated CABG and be performed in countries with a younger patient population and with reimbursement of staged or synchronous carotid revascularization only if a patient is treated within the trial. Acknowledgments We thank the members of the Data Safety and Monitoring Board for their esteemed support: Peter Ringleb (Universitätsklinikum Heidelberg, Neurologische Klinik, Sektion Vaskuläre Neurologie, Heidelberg, Germany), Hans-Henning Eckstein (Technische Universität München, Klinik und Poliklinik für Vaskuläre und Endovaskuläre Chirurgie, München, Germany), Ross Naylor (University of Leicester, Department of Cardiovascular Sciences, Leicester, United Kingdom), Siegfried Hagl (Universitätsklinikum Heidelberg, Abteilung Herzchirurgie [Emeritus], Heidelberg, Germany), and Andreas Ziegler (Universität zu Lübeck und Universitätsklinikum Schleswig-Holstein, Institut für Biometrie und Statistik, Lübeck, Germany). Sources of Funding The trial was financially supported by the German Research Council (DFG, Bonn, Germany, WE2585-3). Disclosures None. Supplementary Material Guest Editor for this article was Seemant Chaturvedi, MD. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.017570/-/DC1.

Many studies have reported greater stroke severity and worse outcome in women albeit with some conflicting results.1,2 Potential explanations for these inconsistencies include different source populations, lack of statistical power, and methodological limitations. First, age difference at the time of stroke, if not properly accounted for, may lead to finding a spurious worse outcome in women.3,4 Second, another potential confounding factor often overlooked is functional status before the index stroke, which may influence stroke severity and prognosis even after adjustment for age. Disentangling the respective effect of sex, age, and prior handicap is important because it would shift the focus from sex differences to age-related differences in outcomes. In this respect, the properties of the modified Rankin Scale (mRS), the commonly used measure of handicap in stroke research, may also impact the observed sex differences.

ve effect of sex, age, and prior handicap is important because it would shift the focus from sex differences to age-related differences in outcomes. In this respect, the properties of the modified Rankin Scale (mRS), the commonly used measure of handicap in stroke research, may also impact the observed sex differences. Finally, most previous studies did not include the full spectrum of symptomatic ischemic cerebrovascular events, particularly transient ischemic attack (TIA) and minor stroke and also pre-hospital deaths and severe strokes in community institutions. Only by including the full range of event severity can analyses reliably address sex differences in severity outcomes. To our knowledge, no published study has met these 3 conditions. Our objective was, therefore, to assess potential sex differences in the severity and prognosis of ischemic stroke/TIA between men and women in a large prospective population-based study of incidence and outcome of TIA and stroke (OXVASC [Oxford Vascular Study]) while carefully taking age and pre-morbid functional status into account.

s, therefore, to assess potential sex differences in the severity and prognosis of ischemic stroke/TIA between men and women in a large prospective population-based study of incidence and outcome of TIA and stroke (OXVASC [Oxford Vascular Study]) while carefully taking age and pre-morbid functional status into account. Methods Study Design and Source Population The OXVASC is an ongoing population-based study of the incidence and outcome of all acute vascular events.5 The study population comprises 92 728 individuals, irrespective of age, who are registered with ≈100 general practitioners (GPs) in 9 general practices in Oxfordshire, United Kingdom. In the United Kingdom, the vast majority of individuals register with a GP who is the gatekeeper of the health system. The GP provides their primary healthcare and holds a lifelong record of their medical history. All participating practices on OXVASC held accurate age sex patient registers and allowed regular searches of their computerized diagnostic coding systems. All practices refer patients to only 1 secondary care centre. The OXVASC was approved by the local research ethics committee (OREC A: 05/Q1604/70). All eligible patients gave informed consent, or study assent was obtained from next of kin if the patient was unable to consent.

Methods Study Design and Source Population The OXVASC is an ongoing population-based study of the incidence and outcome of all acute vascular events.5 The study population comprises 92 728 individuals, irrespective of age, who are registered with ≈100 general practitioners (GPs) in 9 general practices in Oxfordshire, United Kingdom. In the United Kingdom, the vast majority of individuals register with a GP who is the gatekeeper of the health system. The GP provides their primary healthcare and holds a lifelong record of their medical history. All participating practices on OXVASC held accurate age sex patient registers and allowed regular searches of their computerized diagnostic coding systems. All practices refer patients to only 1 secondary care centre. The OXVASC was approved by the local research ethics committee (OREC A: 05/Q1604/70). All eligible patients gave informed consent, or study assent was obtained from next of kin if the patient was unable to consent. From the OXVASC source population, we studied all consecutive patients aged ≥18 years with a first TIA or ischemic stroke between April 1, 2002, and March 31, 2014. Cohort entry was taken as the date of the first recorded TIA or stroke occurring during the study period. Patients were followed until death or the end of the study period (September 30, 2014), whichever occurred first.

tients aged ≥18 years with a first TIA or ischemic stroke between April 1, 2002, and March 31, 2014. Cohort entry was taken as the date of the first recorded TIA or stroke occurring during the study period. Patients were followed until death or the end of the study period (September 30, 2014), whichever occurred first. Procedures and Data Collection The OXVASC study has been described in detail elsewhere.5 Briefly, capture and near-complete ascertainment of all incident or recurrent vascular events were achieved by means of several overlapping methods of hot and cold pursuit. These included (1) a daily, rapid-access TIA and stroke clinic to which participating GPs and the local emergency department team referred individuals with suspected TIA or minor stroke; (2) daily searches of admissions to medical, stroke, neurology, and other relevant wards; (3) daily searches of the local emergency department attendance register; (4) daily searches of in-hospital death records via the bereavement office; (5) monthly searches of all death certificates and coroner’s reports for out-of-hospital deaths; (6) monthly searches of GP diagnostic coding and hospital discharge codes; and (7) monthly searches of all brain and vascular imaging referrals.

r; (4) daily searches of in-hospital death records via the bereavement office; (5) monthly searches of all death certificates and coroner’s reports for out-of-hospital deaths; (6) monthly searches of GP diagnostic coding and hospital discharge codes; and (7) monthly searches of all brain and vascular imaging referrals. Patients were assessed by a study physician as soon as possible after a cerebrovascular event. A standardized detailed past medical history was recorded, as well as clinical history of the current medical event, findings of the physical examination, current medications, and investigations. All information was recorded from the patients, relatives, their hospital records, and their general practice records. All interventions occurring subsequent to the event were also recorded. If a patient died before assessment, an eyewitness account was obtained whenever possible and any relevant records reviewed. If death occurred outside the hospital or before the investigation, any autopsy result was reviewed. Clinical details were sought from primary care physicians or other clinicians on all deaths of possible vascular pathogenesis.

ent, an eyewitness account was obtained whenever possible and any relevant records reviewed. If death occurred outside the hospital or before the investigation, any autopsy result was reviewed. Clinical details were sought from primary care physicians or other clinicians on all deaths of possible vascular pathogenesis. Handicap was assessed with the mRS,6 including pre-morbid mRS before the vascular event of interest. Raters were all trained in the use of the mRS using an instructional DVD with accompanying written materials produced by the University of Glasgow that has been previously used in large-scale clinical trials.7 Neurological impairment was assessed as soon as possible after the event with the National Institutes of Health Stroke Scale (NIHSS).8 All cases were reviewed by a senior neurologist (P.M.R.), and the presumed pathogenesis of stroke was classified according to the modified Trial of ORG 10172 in Acute Stroke Treatment criteria.9 All patients were followed up by a research nurse or study physician at 1, 3, 6, 12, 60, and 120 months from the time of the first TIA or stroke. If a recurrent vascular event was suspected, the patient was assessed by a study neurologist, with recurrent events also identified by the ongoing daily study surveillance.

Handicap was assessed with the mRS,6 including pre-morbid mRS before the vascular event of interest. Raters were all trained in the use of the mRS using an instructional DVD with accompanying written materials produced by the University of Glasgow that has been previously used in large-scale clinical trials.7 Neurological impairment was assessed as soon as possible after the event with the National Institutes of Health Stroke Scale (NIHSS).8 All cases were reviewed by a senior neurologist (P.M.R.), and the presumed pathogenesis of stroke was classified according to the modified Trial of ORG 10172 in Acute Stroke Treatment criteria.9 All patients were followed up by a research nurse or study physician at 1, 3, 6, 12, 60, and 120 months from the time of the first TIA or stroke. If a recurrent vascular event was suspected, the patient was assessed by a study neurologist, with recurrent events also identified by the ongoing daily study surveillance. Data Analysis Descriptive statistics were used to summarize the baseline characteristics of the cohort stratified by sex. Ordinal regression analysis was used to compare the handicap (pre-morbid mRS) before the occurrence of the cerebrovascular event, between women and men adjusting for age, history of stroke or dementia, and other variables listed in the Table. Similarly, we compared stroke severity (using NIHSS in 7 categories) between men and women while adjusting for the same variables, as well as for pre-morbid mRS. Handicap after stroke was evaluated in several ways. We first compared mRS scores 1 month, 6 months, 1 year, and 5 years after ischemic stroke between women and men using ordinal regression. Then, to account for pre-morbid handicap in the assessment of poststroke mRS, we assessed change in mRS score between pre-morbid score and 1 month, 6 months, 1 year, and 5 years after stroke using ordinal regression analysis adjusting for age. Change in mRS score was classified in the following 3 categories: no increase, 1 point increase, ≥2 points increase. In a subsequent analysis, we also assessed changes in mRS as a binary outcome (increase, yes/no) using logistic regression. Finally, among patients with a pre-morbid mRS score of ≤2, we compared the percentage of women and men reaching an mRS score of ≥3 at 1 month. Crude and age-adjusted rates of recurrent stroke and all-cause mortality along with 95% confidence intervals (CIs) were estimated, stratified by sex, based on the Poisson distribution. Cox proportional hazards models were used to estimate separately the hazard ratio (HR) of recurrent stroke and all-cause mortality in women compared with men, adjusting for baseline characteristics listed in the Table, as well as for stroke severity. When studying recurrent stroke, the analysis was repeated using Cox proportional hazards models modified for competing risks of death.10 All potential confounders were measured at cohort entry.

tality in women compared with men, adjusting for baseline characteristics listed in the Table, as well as for stroke severity. When studying recurrent stroke, the analysis was repeated using Cox proportional hazards models modified for competing risks of death.10 All potential confounders were measured at cohort entry. To assess the robustness of our results, all analyses were repeated among patients with an incident (first in life-time rather than just first in study period) cerebrovascular event only and with censoring at the time of any recurrent stroke. Table. Characteristics of the Cohort of 2553 Patients With a First Ischemic Stroke or Transient Ischemic Attack in the Study Period CIs were calculated using a significance level of 5%. All statistical procedures were performed using SAS version 9.4 (SAS Institute Inc, Cary, NC).

To assess the robustness of our results, all analyses were repeated among patients with an incident (first in life-time rather than just first in study period) cerebrovascular event only and with censoring at the time of any recurrent stroke. Table. Characteristics of the Cohort of 2553 Patients With a First Ischemic Stroke or Transient Ischemic Attack in the Study Period CIs were calculated using a significance level of 5%. All statistical procedures were performed using SAS version 9.4 (SAS Institute Inc, Cary, NC). Results The cohort comprised 2553 patients, 1293 (50.6%) women and 1260 (49.4%) men with a first TIA or ischemic stroke during the study period, with a mean follow-up of 4.2 years (SD, 3.4). The baseline characteristics of women and men at the time of the event are shown in the Table. Women were older and had a lower prevalence of coronary artery disease, peripheral vascular disease, and hyperlipidemia compared with men after taking age into account. The same pattern was found in analysis confined to the elderly (≥85 year old, who were predominantly women; Table I in the online-only Data Supplement). However, despite a lesser burden of vascular comorbidities overall, women had a higher pre-morbid mRS score compared with men (ordinal regression odds ratio [OR], 2.01; 95% CI, 1.74–2.33; Figure 1). This difference in pre-morbid mRS persisted after adjusting for age (OR, 1.58; 95% CI, 1.36–1.84), for age, prior stroke, and dementia (OR, 1.62; 95% CI, 1.37–1.92) or for all variables listed in the Table (OR, 1.77; 95% CI, 1.48–2.13). Among 2194 patients with a known marital status, 44.2% of women and 73.2% of men were married or leaving with a partner (age-adjusted OR, 0.34; 95% CI, 0.28–0.41).

5% CI, 1.36–1.84), for age, prior stroke, and dementia (OR, 1.62; 95% CI, 1.37–1.92) or for all variables listed in the Table (OR, 1.77; 95% CI, 1.48–2.13). Among 2194 patients with a known marital status, 44.2% of women and 73.2% of men were married or leaving with a partner (age-adjusted OR, 0.34; 95% CI, 0.28–0.41). Figure 1. Distribution of modified Rankin Scale (mRS) 1 month after stroke among men and women (A) and mRS before stroke (B). CI indicates confidence interval; and OR, odds ratio. A total of 949 (37.2%) patients had a TIA and 1604 (62.8%) had an ischemic stroke with no difference in proportion between women and men (age-adjusted P=0.13; Figure I in the online-only Data Supplement). The event was incident in 86.4% of patients and recurrent in 13.6%, with similar figures in both sexes (age-adjusted P=0.97). Women had slightly more events originating from the carotid territory than men (74% and 69.7%, respectively) but not after accounting for age (age-adjusted P=0.17; Figure II in the online-only Data Supplement). The pathogenesis was most frequently undetermined in both women and men (31.6% and 31.9%, respectively, age-adjusted P=0.04) followed by cardioembolic (Figure III in the online-only Data Supplement). The higher frequency of cardioembolic strokes in women (28.1%) compared with men (24.3%) was no longer apparent when taking age into account (age-adjusted P=0.97).

rmined in both women and men (31.6% and 31.9%, respectively, age-adjusted P=0.04) followed by cardioembolic (Figure III in the online-only Data Supplement). The higher frequency of cardioembolic strokes in women (28.1%) compared with men (24.3%) was no longer apparent when taking age into account (age-adjusted P=0.97). Regarding the distribution of cerebrovascular event severity measured by NIHSS, women had more severe events than men (OR, 1.49; 95% CI, 1.23–1.80). However, this difference was halved when adjusted for age (OR, 1.23; 95% CI, 1.01–1.50) and was no longer apparent when adjusting for age and pre-morbid mRS (OR, 1.10; 95% CI, 0.90–1.35), or age, pre-morbid mRS, and other variables listed in the Table (OR, 1.19; 95% CI, 0.94–1.52). The apparent effect of female sex on severity did not vary according to age (P=0.85 for interaction). Interestingly, adjusting for all baseline comorbidities listed in the Table but omitting pre-morbid mRS in the model still resulted in an apparently higher severity in women (OR, 1.27; 95% CI, 1.00–1.62). In a subgroup analysis restricted to patients with ischemic stroke only (ie, excluding TIA), women had more severe stroke than men however (age- and mRS-adjusted OR, 1.26; 95% CI, 1.02–1.56; Figure 2). This increased severity was not statistically significant when also adjusting for ischemic stroke subtype (OR, 1.22; 95% CI, 0.98–1.53). Only 21 patients were treated with tissue-type plasminogen activator at the time of stroke, 8 (1%) of 802 women and 13 (1.6%) of 798 men with an ischemic stroke.

1.02–1.56; Figure 2). This increased severity was not statistically significant when also adjusting for ischemic stroke subtype (OR, 1.22; 95% CI, 0.98–1.53). Only 21 patients were treated with tissue-type plasminogen activator at the time of stroke, 8 (1%) of 802 women and 13 (1.6%) of 798 men with an ischemic stroke. Figure 2. Distribution of National Institutes of Health Stroke Scale score among men and women with ischemic stroke only. mRS indicates modified Rankin Scale; OR, odds ratio; TIA, transient ischemic attack; and TOAST, Trial of ORG 10172 in Acute Stroke Treatment.

1.02–1.56; Figure 2). This increased severity was not statistically significant when also adjusting for ischemic stroke subtype (OR, 1.22; 95% CI, 0.98–1.53). Only 21 patients were treated with tissue-type plasminogen activator at the time of stroke, 8 (1%) of 802 women and 13 (1.6%) of 798 men with an ischemic stroke. Figure 2. Distribution of National Institutes of Health Stroke Scale score among men and women with ischemic stroke only. mRS indicates modified Rankin Scale; OR, odds ratio; TIA, transient ischemic attack; and TOAST, Trial of ORG 10172 in Acute Stroke Treatment. Women had a worse mRS score compared with men 1 month after the ischemic stroke, excluding patients with a TIA (crude OR, 1.87; 95% CI, 1.56–2.24 and age-adjusted OR, 1.35; 95% CI, 1.12–1.63; Figure 1). Changes in mRS score from pre-morbid state to 1 month after the ischemic stroke, stratified by age and sex, are shown in Figure 3. Female sex was not associated with a higher risk of increased mRS score either in the whole cohort (age-adjusted OR, 1.00; 95% CI, 0.82–1.21) or in the subgroup of patients aged ≥65 (age-adjusted OR, 0.95; 95% CI, 0.76–1.18). Results were unchanged when assessing mRS change as a binary outcome (age-adjusted OR, 0.91 [95% CI, 0.73–1.14] and 0.86 [95% CI, 0.67–1.10], respectively). Similarly, among patients with a low pre-morbid mRS score (0–2), women were not more likely than men to have a score of ≥3 1 month after the stroke (age-adjusted OR, 1.15; 95% CI, 0.90–1.48). Results were similar for mRS changes at 6 months, except a slightly higher mRS change in women with low pre-morbid mRS (Results I in the online-only Data Supplement). Results were also virtually the same at 1 and 5 years, showing no higher risk of increased mRS score in women compared with men (data not shown). Of note, post-stroke inpatient rehabilitation admission was similar for women and men (43.9% and 41.2%, respectively); this was confirmed after adjusting for age and NIHSS (adjusted OR, 0.91; 95% CI, 0.73–1.13). Poststroke depression was also as frequent in women and men (23.8% and 21.4%, respectively).

ata not shown). Of note, post-stroke inpatient rehabilitation admission was similar for women and men (43.9% and 41.2%, respectively); this was confirmed after adjusting for age and NIHSS (adjusted OR, 0.91; 95% CI, 0.73–1.13). Poststroke depression was also as frequent in women and men (23.8% and 21.4%, respectively). Figure 3. Distribution of the change in modified Rankin Scale score before and 1 month after the ischemic stroke by sex and age category. During 10 670 person-years of follow-up, 219 (16.9%) women and 199 (15.8%) men had a recurrent ischemic stroke with similar age-adjusted rates, 4.60 (4.01–5.27) per 100 persons per year in women and 4.64 (4.03–5.33) per 100 persons per year in men, respectively. In Cox multivariate analysis, female sex was not associated with a higher risk of recurrent ischemic stroke compared with male (HR, 0.97; 95% CI, 0.79–1.20). Results were unchanged when controlling for competing risks (data not shown).

er year in women and 4.64 (4.03–5.33) per 100 persons per year in men, respectively. In Cox multivariate analysis, female sex was not associated with a higher risk of recurrent ischemic stroke compared with male (HR, 0.97; 95% CI, 0.79–1.20). Results were unchanged when controlling for competing risks (data not shown). Death occurred in 620 (48%) women and 482 (38%) men during follow-up. Mortality rates increased with age and were similar in women and men when stratified by age whether during the first year or during the entire study period (Figure 4). In Cox univariate analysis, women had a 32% higher risk of death compared with men (HR, 1.32; 95% CI, 1.17–1.48). However, this apparent higher mortality in women was no longer seen when controlling for age (HR, 0.95; 95% CI, 0.84–1.07) and for baseline comorbidities and severity of stroke (HR, 0.82; 95% CI, 0.72–0.94). All results were virtually the same in the sensitivity analyses restricting the cohort to incident cerebrovascular events only (data not shown). Figure 4. Mortality rates per 1000 person-years during the first year or during the entire study period.

Death occurred in 620 (48%) women and 482 (38%) men during follow-up. Mortality rates increased with age and were similar in women and men when stratified by age whether during the first year or during the entire study period (Figure 4). In Cox univariate analysis, women had a 32% higher risk of death compared with men (HR, 1.32; 95% CI, 1.17–1.48). However, this apparent higher mortality in women was no longer seen when controlling for age (HR, 0.95; 95% CI, 0.84–1.07) and for baseline comorbidities and severity of stroke (HR, 0.82; 95% CI, 0.72–0.94). All results were virtually the same in the sensitivity analyses restricting the cohort to incident cerebrovascular events only (data not shown). Figure 4. Mortality rates per 1000 person-years during the first year or during the entire study period. Discussion In this population-based cohort study, we showed the impact of potential confounders on the apparent sex differences in severity and outcome of stroke/TIA. Overall, about one third of women and men presented with a TIA, they had a similar proportion of events originating from the anterior circulation, and the cardioembolic pathogenesis was equally represented in men and women when taking age into account. However, women had a higher pre-morbid mRS score compared with men, even when adjusting for age and comorbidities so that the apparently higher severity of cerebrovascular events in women did not remain after adjustment for age and pre-morbid mRS. When ignoring pre-morbid mRS score, women had a worse functional status than men after ischemic stroke. However, change in mRS score 1 month after ischemic stroke was similar between women and men, including among those with low pre-morbid mRS. Finally, women were not at higher risk of recurrent ischemic stroke and had, in fact, a significantly lower mortality rate than men after accounting for age and comorbidities.

ke. However, change in mRS score 1 month after ischemic stroke was similar between women and men, including among those with low pre-morbid mRS. Finally, women were not at higher risk of recurrent ischemic stroke and had, in fact, a significantly lower mortality rate than men after accounting for age and comorbidities. Our study included TIAs along with ischemic strokes to cover the whole spectrum of ischemic cerebrovascular events as opposed to most previous studies.2 If there was a sex difference in cerebral susceptibility to ischemia, as some have argued, then the proportion of TIA versus stroke might differ. However, the proportion of strokes was similar in both sexes even though women were on average older than men. Moreover, the apparent higher severity of stroke in women was mostly explained by an older age and a worse handicap before the event rather than by sex. Interestingly, the pre-morbid mRS score was still worse in women compared with men after adjusting for age and several comorbidities at baseline. Frailty may still play a role in the sex difference in pre-morbid mRS because of unmeasured confounders. Other factors related to the properties of the mRS scale might also contribute to this difference. Prior studies on stroke severity yielded inconsistent results,2,11–20 but they did not account for both age and some measure of handicap before stroke or specifically assess ischemic stroke/TIA while taking both factors into account. Therefore, future studies on sex differences, including clinical trials planning sex-stratified analyses, should properly account for age and collect information on handicap before the event of interest. Moreover, in many instances, it may be more relevant to examine effectiveness based on prior functional status rather than sex because we showed no clinically substantial differences between sexes. In the subgroup of patients with ischemic stroke only, however, women had slightly higher NIHSS scores than men even after taking age and pre-morbid mRS into account because of the greater proportion of cardioembolic stroke in women due to their older age. Of note, as previously reported, the proportion of cardioembolic stroke was not more frequent in women after adjusting for age21; previous studies using Trial of ORG 10172 in Acute Stroke Treatment classification and reporting a higher proportion of this subtype in women did not take age into account.17,22

r older age. Of note, as previously reported, the proportion of cardioembolic stroke was not more frequent in women after adjusting for age21; previous studies using Trial of ORG 10172 in Acute Stroke Treatment classification and reporting a higher proportion of this subtype in women did not take age into account.17,22 Interestingly, change in mRS score between pre-morbid estimate and 1 and 6 months after stroke was similar in women and men. Moreover, rates of recurrence of stroke/TIA were similar, and mortality rate was lower in women in accordance with other studies.11,23,24 Conversely, using various scales, some studies found that women had a worse functional status than men after stroke but did not adjust for prior functional status.2,25 This finding can be expected if women have already a worse functional status before stroke, as previously reported.26–28 One study did not find any difference in poststroke functional status using Barthel Index at 6 months after adjusting for pre-stroke disability and comorbidities,28 whereas another still found a worse functional status at 3 months in women in multivariate analysis adjusting for pre-morbid mRS but methodological differences may explain this result.26 A recent study on functional outcome after intracerebral hemorrhage also did not show any sex difference in adjusted analyses, including age and pre-morbid mRS among other factors.29 Finally, it remains possible that inconsistencies between studies are partly related to the lack of precision of the mRS scale, and our conclusions may not apply to other disability or handicap scales.

age also did not show any sex difference in adjusted analyses, including age and pre-morbid mRS among other factors.29 Finally, it remains possible that inconsistencies between studies are partly related to the lack of precision of the mRS scale, and our conclusions may not apply to other disability or handicap scales. Our study has several strengths. OXVASC is a prospective population-based cohort study with nearly complete ascertainment of vascular events in a well-defined geographic region. Therefore, selection bias toward the exclusion of old women with severe stroke living in an institution for instance is unlikely to have occurred. Patients with minor events were also included because the study population was not limited to patients seen in emergency department or hospitalized contrary to many previous studies. Finally, the prospective and standardized data collection by stroke specialists and access to lifelong primary care records ensure completeness and accuracy of data and limit the potential for recall bias. Some potential limitations of our study must, however, be acknowledged. First, the OXVASC population is mostly composed of white patients, and our results may not be entirely generalizable to non-white populations. Second, some residual confounding by unmeasured factors cannot be excluded, but accounting for age and pre-morbid mRS score already corrected fully for apparent sex differences. Moreover, there was no apparent difference in acute care as measured by tissue-type plasminogen activator use and admission to inpatient rehabilitation. Third, although pre-morbid and post-event mRS were measured using standardized forms, scoring was obviously not done while blinded for sex. However, data in OXVASC were not collected for this particular study to test the hypothesis of sex differences in stroke outcomes so that systematic differences in rating are unlikely to entirely explain our results. Similarly, in none of the previous studies evaluating stroke outcomes in men and women was handicap assessed while blinded for sex. Finally, the limitations of using ordinal regression to model mRS scores after stroke are well recognized. However, this approach is currently the most commonly used and allowed us to compare our findings with previous studies.

ies evaluating stroke outcomes in men and women was handicap assessed while blinded for sex. Finally, the limitations of using ordinal regression to model mRS scores after stroke are well recognized. However, this approach is currently the most commonly used and allowed us to compare our findings with previous studies. In conclusion, we found no substantial sex differences in severity, outcome, or recurrence of stroke/TIA. Future studies addressing sex differences in cerebrovascular diseases should take into account functional status before the event of interest as most apparent differences between women and men may be because of age and pre-morbid functional status rather than sex per se. Acknowledgments We are grateful to all the staff in the general practices that collaborated in the Oxford Vascular Study: Abingdon Surgery, Stert St, Abingdon; Malthouse Surgery, Abingdon; Marcham Road Family Health Centre, Abingdon; The Health Centre, Berinsfield; Key Medical Practice; Kidlington; 19 Beaumont St, Oxford; East Oxford Health Centre, Oxford; Church Street Practice, Wantage. We also acknowledge the use of the facilities of the Acute Vascular Imaging Centre, Oxford. Sources of Funding The Oxford Vascular Study is funded by the Wellcome Trust, Wolfson Foundation, and the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre. Dr Renoux is the recipient of a Chercheur-Boursier Award from the Fonds de la recherche du Québec – santé (FRQ-S). Dr Rothwell is in receipt of an NIHR Senior Investigator Award and a Wellcome Trust Senior Investigator Award.

tion, and the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre. Dr Renoux is the recipient of a Chercheur-Boursier Award from the Fonds de la recherche du Québec – santé (FRQ-S). Dr Rothwell is in receipt of an NIHR Senior Investigator Award and a Wellcome Trust Senior Investigator Award. Disclosures None. Supplementary Material Guest Editor for this article was Tatjana Rundek, MD, PhD. The views expressed are those of the author(s) and not necessarily those of the National Health Service, the National Institute for Health Research, or the Department of Health. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.018187/-/DC1.

Brain white matter hyperintensity (WMH) growth and cortical thinning are commonly seen on magnetic resonance imaging (MRI) in community-dwelling older people.1–5 The incidence of these features is highly variable between individuals but those with the largest WMH volumes and/or thinnest cortices are at increased risk of stroke, dementia, and cognitive and physical impairment.6–8 Effective interventions are dependent on understanding the mechanisms of WMH growth and cortical thinning and whether one feature may be an underlying cause of the other. Cross-sectional studies have found associations between larger whole brain WMH volume and reduced gray matter (GM) volume, density, and thickness.3,5,9–12 Those with larger WMH volumes generally had relatively reduced cortical thickness3 and density5,11 in frontotemporal and inferior parietal regions. Others have found similar cross-sectional patterns of negative associations between whole brain WMH volume and regional GM volume in the default mode network (which includes medial temporal lobe structures, the inferior parietal lobe, and cuneus) using a region of interest analysis.10 Larger WMH volume in small vessel disease patients has also been associated with reduced structural connectivity in frontotemporal and inferior parietal regions.7

e in the default mode network (which includes medial temporal lobe structures, the inferior parietal lobe, and cuneus) using a region of interest analysis.10 Larger WMH volume in small vessel disease patients has also been associated with reduced structural connectivity in frontotemporal and inferior parietal regions.7 These studies were cross-sectional and so cannot ascertain a direction of causation or effect. Additionally, the regions of cortical thinning that were associated with WMH volume did not overlie regions with the greatest incidence of WMH,4 for example, centrifugally around the ventricles and superiorly towards the cranial vertex. Longitudinal studies are required to determine whether larger WMH volumes at baseline and larger increases in WMH volume are associated with subsequent regional cortical thinning. A longitudinal study of the association between cortical morphology and WMH volume growth has previously been conducted in Cerebral Autosomal Dominant Arteriopathy With Subcortical Infarcts and Leucoencephalopathy (CADASIL) patients.13 This study found that although lacunar lesions were strongly related to worsening cortical morphology, WMH volume was not strongly related to worsening cortical morphology in CADASIL.

ously been conducted in Cerebral Autosomal Dominant Arteriopathy With Subcortical Infarcts and Leucoencephalopathy (CADASIL) patients.13 This study found that although lacunar lesions were strongly related to worsening cortical morphology, WMH volume was not strongly related to worsening cortical morphology in CADASIL. In the present study, we assessed longitudinal associations between change in WMH volume and change in cortical thickness to determine if the relationship between WMH and regional cortical thinning could be causal in community-dwelling subjects from ages 73 to 76.3,11 If the relationship between WMH and the regions of cortex that were thinned was potentially causal, then we hypothesized that there would be an association between (1) baseline whole WMH volume and change in cortical thickness over the next few years; (2) baseline cortical thickness and change in WMH volume; and (3) change in WMH volume and change in cortical thickness. To test these hypotheses, we measured progression of WMH volume and cortical thinning from ≈73 to ≈76 years of age in community-dwelling subjects from the Lothian Birth Cohort 1936 (LBC1936) study. Methods Study Approval and Subject Consent Approval for the LBC1936 study protocol was obtained from the Multicentre Research Ethics Committee for Scotland (MREC/01/0/56) and Lothian Research Ethics Committee (LREC/2003/2/29). All subjects gave written, informed consent.

In the present study, we assessed longitudinal associations between change in WMH volume and change in cortical thickness to determine if the relationship between WMH and regional cortical thinning could be causal in community-dwelling subjects from ages 73 to 76.3,11 If the relationship between WMH and the regions of cortex that were thinned was potentially causal, then we hypothesized that there would be an association between (1) baseline whole WMH volume and change in cortical thickness over the next few years; (2) baseline cortical thickness and change in WMH volume; and (3) change in WMH volume and change in cortical thickness. To test these hypotheses, we measured progression of WMH volume and cortical thinning from ≈73 to ≈76 years of age in community-dwelling subjects from the Lothian Birth Cohort 1936 (LBC1936) study. Methods Study Approval and Subject Consent Approval for the LBC1936 study protocol was obtained from the Multicentre Research Ethics Committee for Scotland (MREC/01/0/56) and Lothian Research Ethics Committee (LREC/2003/2/29). All subjects gave written, informed consent. Subjects In the present study, we assessed 351 (Nmale=202) community-dwelling subjects from the LBC1936 study14,15 that had full brain MRI measures, clinical and cognitive assessments at imaging baseline (mean age 72.71±0.72 years), and follow-up (mean age 76.40±0.64 years). These subjects were not deliberately selected, rather they were simply those who agreed to participate in brain scanning and had complete data sets at baseline and follow-up (Figure 1).

asures, clinical and cognitive assessments at imaging baseline (mean age 72.71±0.72 years), and follow-up (mean age 76.40±0.64 years). These subjects were not deliberately selected, rather they were simply those who agreed to participate in brain scanning and had complete data sets at baseline and follow-up (Figure 1). Figure 1. Subject recruitment flow chart. WMH indicates white matter hyperintensity. We recorded Mini Mental State Examination (MMSE) scores at baseline and follow-up to screen for possible dementia.14,15 We did not exclude subjects based on MMSE but included MMSE as an adjustment variable. We recorded the following vascular risk factors (VRF) during a clinical research facility visit: history of hypertension, hypercholesterolemia, diabetes mellitus, smoking, history of cardiovascular disease, and body mass index. History of cardiovascular disease includes self-reported incidences of coronary heart disease, stroke, peripheral arterial disease, and aortic disease. We also took blood samples and measured blood pressure but did not include these variables (eg, systolic blood pressure and glycated hemoglobin) in the present analysis to maximize the number of subjects with complete data sets and to be consistent with previous work.3 Further, we have previously shown that historical variables, for example, history of hypertension, have greater associations with WMH than measured variables, for example, systolic blood pressure.4

the present analysis to maximize the number of subjects with complete data sets and to be consistent with previous work.3 Further, we have previously shown that historical variables, for example, history of hypertension, have greater associations with WMH than measured variables, for example, systolic blood pressure.4 Brain MRI Acquisition Brain MRI acquisition parameters were described in detail previously.16 Briefly, all subjects had brain MRI on the same 1.5 tesla GE Signa Horizon HDx clinical scanner (General Electric, Milwaukee, WI), maintained on a careful quality assurance programme, at baseline and follow-up. The scanning protocol was the same at baseline and follow-up and acquired T1-, T2-, T2*-, and fluid-attenuated inversion recovery–weighted images.16 WMH Volume and Cortical Thickness Measurement We measured intracranial volume and whole WMH volume in milliliters using a validated multispectral image processing method that combines T1-, T2-, T2*-, and fluid-attenuated inversion recovery–weighted MRI sequences for segmentation.17–19 We measured cortical thickness using the fully automated Civet image processing pipeline developed at the Montreal Neurological Institute.20,21 Civet measures cortical thickness at 81 924 vertices (the perpendicular distance between GM and WM surfaces) across the cortex.20–23 For clarity, we refer to vertex as the perpendicular distance between GM and WM surfaces, not the cranial vertex.

rocessing pipeline developed at the Montreal Neurological Institute.20,21 Civet measures cortical thickness at 81 924 vertices (the perpendicular distance between GM and WM surfaces) across the cortex.20–23 For clarity, we refer to vertex as the perpendicular distance between GM and WM surfaces, not the cranial vertex. We manually verified WMH volume masks and cortical thickness maps as per procedures described previously.17–19,22,23 Specifically, we followed STRIVE (Standards for Reporting Vascular Changes on Neuroimaging) guidelines for segmenting WMH, manually removing cortical and subcortical infarcts from WMH masks.19 Finally, the reliability of cortical thickness20–22 and WMH17–19 measurements were tested and reported previously. Statistical Analysis All statistical analyses were performed in Matrix Laboratory (MATLAB) R2014a (© 1994–2014 The MathWorks, Inc). We assessed changes in overall mean cortical thickness (mean thickness of the whole cortical mantle), whole brain WMH volume, and continuous variables used in adjustment, for example, body mass index, from 73 years to 76 years using paired t-tests. Log-transforming the positively skewed WMH distributions had little effect on our results, and therefore, we maintained their original scale (proportion of intracranial volume) to simplify interpretation. We assessed changes in binary variables used in adjustment, for example, history of hypertension, using z-tests of proportion.

rming the positively skewed WMH distributions had little effect on our results, and therefore, we maintained their original scale (proportion of intracranial volume) to simplify interpretation. We assessed changes in binary variables used in adjustment, for example, history of hypertension, using z-tests of proportion. Cortical vertex-wise regression analyses were performed using the SurfStat MATLAB toolbox (http://www.math.mcgill.ca/keith/surfstat). We tested 5 vertex-wise regression models where (1) cortical thickness at 73 years at each vertex was the dependent variable and WMH volume at 73 years was the independent variable; (2) cortical thickness at 76 years at each vertex was the dependent variable and WMH volume at 76 years was the independent variable; (3) change in cortical thickness at each vertex was the dependent variable and WMH volume at 73 years was the independent variable; (4) cortical thickness at 73 years at each vertex was the dependant variable and change in WMH volume was the independent variable; and (5) change in cortical thickness at each vertex was the dependent variable and change in WMH volume was the independent variable. We defined change in WMH and cortical thickness as individual measurements at 76 years minus measurements at 73 years.

ant variable and change in WMH volume was the independent variable; and (5) change in cortical thickness at each vertex was the dependent variable and change in WMH volume was the independent variable. We defined change in WMH and cortical thickness as individual measurements at 76 years minus measurements at 73 years. We used false discovery rate to correct for multiple comparisons and calculated Q values, that is, false discovery rate–corrected P values,24 for all vertex-wise regressions thresholded at 0.05. As reported by others,3 all models were controlled for sex, MMSE, age in days, years of education, body mass index, and VRF. Finally, we also included childhood (age 11) intelligence quotient as a controlling variable to test whether any associations between WMH and cortical thickness were because of the influence of premorbid levels of cognitive ability.

re controlled for sex, MMSE, age in days, years of education, body mass index, and VRF. Finally, we also included childhood (age 11) intelligence quotient as a controlling variable to test whether any associations between WMH and cortical thickness were because of the influence of premorbid levels of cognitive ability. Results Baseline Only and Follow-Up Subject Comparisons There were no significant differences at baseline (73 years) between subjects who did return for follow-up cortical thickness measurement and subjects who did not return for follow-up (at 76 years) in overall mean cortical thickness (3.11 mm versus 3.10 mm, t=0.75; P=0.46); WMH volume (0.78% intracranial volume versus 0.84% intracranial volume, t=−0.78; P=0.43); history of cardiovascular disease (27.3% versus 26.2%, z=0.29; P=0.39); current smoking (6.5% versus 8.2%, z=−0.76; P=0.22); hypercholesterolemia (40.0% versus 43.1%, z=−0.71; P=0.24); hypertension (46.5% versus 51.8%, z=−1.2; P=0.11); diabetes mellitus (10.1% versus 11.3%, z=−0.43; P=0.33); body mass index (27.8 versus 27.8, t=−0.21; P=0.84); years of education (10.85 years versus 10.89 years, t=−0.448; P= 0.65); nor age 11 intelligence quotient (101.9 versus 100.2, t=1.24; P=0.22). However, subjects who did return had higher MMSE scores at baseline than those who did not return (28.9 versus 28.6, t=2.49; P=0.01). The remaining results are only for the 351 longitudinal subjects who had full clinical, cognitive, and brain MRI data at baseline and follow-up.

Beat-to-beat BPV for 5 minutes is only weakly correlated with day-to-day BPV on HBPM and premorbid visit-to-visit BPV but shares the same physiological associations, suggestive of a similar pathophysiology.7 Increased beat-to-beat BPV8 and diminished baroreceptor sensitivity (derived from beat-to-beat BP monitoring)9 are potentially associated with a worse outcome after a major acute stroke and may be associated with an increased risk of recurrent events.8 However, previous studies were small with significant methodological problems. Therefore, we determined the predictive value of beat-to-beat BPV in a prospective cohort of patients with recent transient ischemic attack or minor stroke. Materials and Methods Requests for access to the data and analysis tools in this article will be openly considered. Please contact P.M.R. for further information. Study Population Consecutive patients were recruited between September 2010 and 2015 from the OXVASC (Oxford Vascular Study)10 transient ischemic attack and minor stroke clinic. The OXVASC population consists of 92 728 individuals registered with 100 primary-care physicians in Oxfordshire, United Kingdom.10 All consenting patients underwent a standardized medical history and examination, ECG, blood tests, and a stroke protocol magnetic resonance imaging brain and contrast-enhanced magnetic resonance angiography (or CT brain and carotid Doppler ultrasound or CT angiogram), an echocardiogram, and 5-day ambulatory cardiac monitoring. All patients were reviewed by a study physician, the diagnosis verified by the senior study neurologist (P.M.R.), etiology determined by a panel of stroke neurologists, and were followed-up face-to-face at 1, 3, 6, and 12 months, and ≤2, 5, or 10 years. Recurrent events were determined at face-to-face follow-up and by multiple overlapping methods of ascertainment, including daily review of hospital admissions, review of death certificates and coroner’s records, manual review of general practitioner records, and linkage to hospital event statistics and death registries.

Results Baseline Only and Follow-Up Subject Comparisons There were no significant differences at baseline (73 years) between subjects who did return for follow-up cortical thickness measurement and subjects who did not return for follow-up (at 76 years) in overall mean cortical thickness (3.11 mm versus 3.10 mm, t=0.75; P=0.46); WMH volume (0.78% intracranial volume versus 0.84% intracranial volume, t=−0.78; P=0.43); history of cardiovascular disease (27.3% versus 26.2%, z=0.29; P=0.39); current smoking (6.5% versus 8.2%, z=−0.76; P=0.22); hypercholesterolemia (40.0% versus 43.1%, z=−0.71; P=0.24); hypertension (46.5% versus 51.8%, z=−1.2; P=0.11); diabetes mellitus (10.1% versus 11.3%, z=−0.43; P=0.33); body mass index (27.8 versus 27.8, t=−0.21; P=0.84); years of education (10.85 years versus 10.89 years, t=−0.448; P= 0.65); nor age 11 intelligence quotient (101.9 versus 100.2, t=1.24; P=0.22). However, subjects who did return had higher MMSE scores at baseline than those who did not return (28.9 versus 28.6, t=2.49; P=0.01). The remaining results are only for the 351 longitudinal subjects who had full clinical, cognitive, and brain MRI data at baseline and follow-up. Baseline, Follow-Up, and Changes in Cognitive, VRF, Cortical Thickness, and WMH Measurements Baseline, follow-up, and changes in cognitive, VRF, cortical thickness, and WMH measurements are shown in Table 1. There were significant increases in the proportions of subjects with hypertension, hypercholesterolemia, and cardiovascular disease and a decrease in MMSE from baseline (73 years) to follow-up (76 years). Overall, mean cortical thickness generally decreased (Cohen’s d of change=−0.45) with age, and WMH volume generally increased (Cohen’s d of change=0.93) with age (Table 1). The Spearman correlation matrix between all brain changes and independent variables (Table 2) shows that cortex thinning was generally more pronounced in older subjects (ρ=−0.13; P=0.02). All other independent variables had limited partial effects on WMH and cortical thinning (beyond the effect of time point; Table 2).

The Spearman correlation matrix between all brain changes and independent variables (Table 2) shows that cortex thinning was generally more pronounced in older subjects (ρ=−0.13; P=0.02). All other independent variables had limited partial effects on WMH and cortical thinning (beyond the effect of time point; Table 2). Table 1. Baseline, Follow-Up, and Changes in Cognitive, VRF, Cortical Thickness, and WMH Measurements Table 2. Spearman Correlation Matrix of Overall Mean Cortical Thickness and WMH Changes and Independent Variables Cross-Sectional and Longitudinal Global Correlations Between Overall Mean Cortical Thickness and WMH Volume Cross-sectional global correlations between overall mean cortical thickness and WMH volume at 73 years (r=−0.06; P=0.27) and 76 years (r=−0.08; P=0.12) were not significant. Pairwise longitudinal global correlations between (1) WMH volume at 73 years and change in overall mean cortical thickness (r=−0.07; P=0.19); (2) overall mean cortical thickness at 73 years and change in WMH volume (r=0.01; P= 0.82); and (3) change in WMH volume and change in overall mean cortical thickness (r=−0.02; P=0.67) were also not significant.

elations between (1) WMH volume at 73 years and change in overall mean cortical thickness (r=−0.07; P=0.19); (2) overall mean cortical thickness at 73 years and change in WMH volume (r=0.01; P= 0.82); and (3) change in WMH volume and change in overall mean cortical thickness (r=−0.02; P=0.67) were also not significant. Cross-Sectional Vertex-Wise Regression Models of Regional Cortical Thickness and WMH Volume Cross-sectional vertex-wise regression models of cortical thickness and WMH volume at 76 years are shown in Figure 2. Cross-sectional data from 73 years are not shown because the pattern of associations between cortical thickness and WMH volume was almost identical at 73 and 76 years. All models are corrected for VRF, MMSE, education level, and sex. Inclusion of age 11 intelligence quotient as a controlling variable made little difference to the cortical t-maps (data not shown). Figure 2. Cross-sectional (76y panel) and longitudinal (C-C, gC-wB, and wC-gB panels) t-maps of vertex-wise associations between cortical thickness and whole white matter hyperintensity (WMH) volume. Warm colors show where greater WMH volume is associated with reduced cortical thickness. The significance of these associations is shown in Figure 3. C-C indicates cortical thickness change and WMH volume change from 73 to 76 years; gC-wB, cortical thickness change and WMH volume at 73 years; wC-gB, WMH volume change and cortical thickness at 73 years.

volume is associated with reduced cortical thickness. The significance of these associations is shown in Figure 3. C-C indicates cortical thickness change and WMH volume change from 73 to 76 years; gC-wB, cortical thickness change and WMH volume at 73 years; wC-gB, WMH volume change and cortical thickness at 73 years. Warm colors in Figure 2 show regions where greater WMH volume was associated with reduced cortical thickness. The significance of cross-sectional associations is shown on the left panel of Figure 3. There were consistent patterns of negative cross-sectional associations at 73 and 76 years within and surrounding the Sylvian fissure extending superiorly to the parietal lobe, posteriorly to the occipital lobe, and anteriorly to the frontal lobe. Therefore, having greater WMH volume was cross-sectionally associated with reduced cortical thickness in specific regions only, that is, within and surrounding the Sylvian fissure. Associations between greater WMH volume and greater cortical thickness in superior regions (cold colors in Figure 2) were all nonsignificant.

erefore, having greater WMH volume was cross-sectionally associated with reduced cortical thickness in specific regions only, that is, within and surrounding the Sylvian fissure. Associations between greater WMH volume and greater cortical thickness in superior regions (cold colors in Figure 2) were all nonsignificant. Figure 3. Significance of cross-sectional (X-S) and longitudinal (L) vertex-wise associations between cortical thickness and white matter hyperintensity (WMH) volume. There were consistent patterns of negative cross-sectional associations at 73 (data not shown) and 76 years within and surrounding the Sylvian fissure extending superiorly to the parietal lobe, posteriorly to the occipital lobe, and anteriorly to the frontal lobe (left). The clear grey Q-map (right) shows that no longitudinal associations between cortical thickness and WMH volume were significant. A scatter plot of the peak cross-sectional association in the Sylvian fissure and surrounding area at 76 years (ρ=−0.276; Q=0.004) is shown in Figure 3.

ecurrent events were determined at face-to-face follow-up and by multiple overlapping methods of ascertainment, including daily review of hospital admissions, review of death certificates and coroner’s records, manual review of general practitioner records, and linkage to hospital event statistics and death registries. Participants were excluded if they were <18 years of age, cognitively impaired (Mini-Mental State Examination<23), pregnant; had a recent myocardial infarction, unstable angina, heart failure (New York Heart Association, 3–4 or ejection fraction, <40%), or untreated bilateral carotid stenosis (>70%); and if they had atrial fibrillation during testing. The study was approved by the Oxfordshire Research Ethics Committee.

Figure 3. Significance of cross-sectional (X-S) and longitudinal (L) vertex-wise associations between cortical thickness and white matter hyperintensity (WMH) volume. There were consistent patterns of negative cross-sectional associations at 73 (data not shown) and 76 years within and surrounding the Sylvian fissure extending superiorly to the parietal lobe, posteriorly to the occipital lobe, and anteriorly to the frontal lobe (left). The clear grey Q-map (right) shows that no longitudinal associations between cortical thickness and WMH volume were significant. A scatter plot of the peak cross-sectional association in the Sylvian fissure and surrounding area at 76 years (ρ=−0.276; Q=0.004) is shown in Figure 3. Longitudinal Vertex-Wise Regression Models of Regional Cortical Thickness and WMH Volume Longitudinal vertex-wise associations between (1) baseline WMH volume and change in cortical thickness (gC-wB in Figure 2); (2) baseline cortical thickness and change in WMH volume (wC-gB in Figure 2); and (3) change in WMH volume and change in cortical thickness (C-C in Figure 2) were all nonsignificant across the cortex (Q>0.05; Figure 3). Therefore, having a larger WMH volume at 73 years (or larger change in WMH volume between 73 and 76 years) did not predict greater cortical thinning between 73 and 76 years at any part of the cortex. Neither did a thinner cortex at 73 years predict greater WMH growth between 73 and 76 years.

cortex (Q>0.05; Figure 3). Therefore, having a larger WMH volume at 73 years (or larger change in WMH volume between 73 and 76 years) did not predict greater cortical thinning between 73 and 76 years at any part of the cortex. Neither did a thinner cortex at 73 years predict greater WMH growth between 73 and 76 years. The longitudinal association between WMH change and overall mean cortical thickness change was descriptively much stronger in subjects with MMSE≤26 (r=−0.220; P=0.41 versus r=−0.003; P=0.96) but this was not statistically significant potentially because of the small number of subjects with MMSE≤26 (N=16). Discussion We have replicated cross-sectional associations between greater WMH volume and regional cortical thinning around the Sylvian fissure3,5,9–12; however, we found no longitudinal associations between (1) baseline WMH volume and change in cortical thickness; (2) baseline cortical thickness and change in WMH volume; or (3) change in WMH volume and change in cortical thickness at any part of the cortex in community-dwelling subjects from 73 to 76 years. The cross-sectional associations found here between greater WMH volume and reduced cortical thickness in the region of the Sylvian fissure are consistent with previous GM volume,10 voxel-based morphometry,5,11 and cortical thickness3 studies. As with previous studies, the regions of cortical thinning–WMH associations that we found are not consistent with the most frequent WMH locations and areas of expansion,4 for example, centrifugally around the ventricles and superiorly towards the cranial vertex.

l-based morphometry,5,11 and cortical thickness3 studies. As with previous studies, the regions of cortical thinning–WMH associations that we found are not consistent with the most frequent WMH locations and areas of expansion,4 for example, centrifugally around the ventricles and superiorly towards the cranial vertex. Our results suggest that WMH volume and cortical atrophy both worsen with age and that their individual differences share some causes—thus the cross-sectional associations. However, their changes from 73 to 76 years do not appear to be associated, and such correlated change would have been one indicator of a possible causal association. This conclusion is consistent with a longitudinal study in CADASIL patients that, although finding strong associations between lacunar lesions and cortical morphological changes, found a limited association between cortical morphological changes and WMH volume.13,25

been one indicator of a possible causal association. This conclusion is consistent with a longitudinal study in CADASIL patients that, although finding strong associations between lacunar lesions and cortical morphological changes, found a limited association between cortical morphological changes and WMH volume.13,25 Strengths of our study include the ability to test longitudinal and cross-sectional associations between WMH volume and regional cortical thickness in a large sample of community-dwelling subjects. Other strengths include the age-homogeneous subjects with childhood intelligence quotient assessments and who are now in the eighth decade of life where the risk of dementia increases substantially.26 As well as the narrow age range, other novel features of the LBC1936 study (eg, all subjects are white Caucasian) may have minimized any potentially strong confounding effects that factors such as age, mixed ethnicity, and geography might have had in a less homogeneous sample. We measured WMH volume and cortical thickness using well-validated quantitative techniques that we manually checked and quality controlled post-pipeline for each subject at both time points.17,20,22 The raw brain MRI from which we measured WMH volume and cortical thickness were obtained using the same protocol on the same carefully maintained scanner at both time points.16

l-validated quantitative techniques that we manually checked and quality controlled post-pipeline for each subject at both time points.17,20,22 The raw brain MRI from which we measured WMH volume and cortical thickness were obtained using the same protocol on the same carefully maintained scanner at both time points.16 Despite these strengths and our replication of previous cross-sectional findings,3,5,10,11 our study has limitations. The follow-up time of 3 years is a major limitation because it may not have been long enough to detect correlated changes between WMH and cortical thinning. We chose 3-year follow-up (rather than a longer time) to maximize subject retention and to be consistent with previous studies, for example, Austrian Stroke Prevention Study.1,27 Further, we have previously detected cross-sectional differences in WMH because of age within the narrow age band (<3 years) in the LBC1936 study.28 We are studying these subjects again at 6 years follow-up, and this may provide better evidence for any potentially causal relationships not identified here. We will attempt to ascertain the reasons for subjects lost to follow-up and will use full information maximum likelihood analyses, checked against analyses of completers, to minimize the effect of loss to follow-up. We defined change as individual measurements at 76 years minus measurements at 73 years. We are aware that there are other ways of assessing change, for example, those often applied to cognitive variable change.29 However, the approach we used here is often applied to measure changes in brain morphology.30 Although the homogeneous nature of the LBC1936 cohort may provide increased power from having less need to control for confounding variables, for example, age and ethnicity, it limits the generalizability of our results. The longitudinal subjects we assessed here generally had higher MMSE than subjects who did not return for follow-up, and this may also limit the generalizability of our results, for example, longitudinal associations between WMH, and cortical thinning may be stronger in subjects with lower cognitive scores. We could not adequately test this here because of the small number of subjects with MMSE≤26 (N=16), and future work is required to determine whether associations are stronger in cognitively impaired subjects.

udinal associations between WMH, and cortical thinning may be stronger in subjects with lower cognitive scores. We could not adequately test this here because of the small number of subjects with MMSE≤26 (N=16), and future work is required to determine whether associations are stronger in cognitively impaired subjects. Although the locations of cross-sectional associations that we (and others3,5,9–12) found between WMH and cortical thinning do not directly reflect common areas for WMH expansion, areas of associations were proximate to the tapetum of the corpus callosum fiber tracts which extend inferiorly and anteriorly into the temporal lobes.31 Further work is required to determine whether the locations of cross-sectional WMH and cortical thinning associations are because of an indirect connection through the perisylvian cortex and tapetum of the corpus callosum. Finally, it is difficult to prove or disprove a causal relationship between WMH and cortical thinning in observational studies. However, we adjusted for a number of variables known to influence WMH and cortical thinning, and although correlation is not necessarily causation, correlation is fundamental to causation.32,33 Therefore, the lack of longitudinal correlations implies the lack of a causal relationship from 73 to 76 years.

ional studies. However, we adjusted for a number of variables known to influence WMH and cortical thinning, and although correlation is not necessarily causation, correlation is fundamental to causation.32,33 Therefore, the lack of longitudinal correlations implies the lack of a causal relationship from 73 to 76 years. Notwithstanding these limitations, we have shown that although they both worsen with age, WMH volume progression and regional cortical thinning do not seem to have a correlative/causal longitudinal relationship from 73 to 76 years. Further longitudinal studies with longer follow-up times and with more time points at different ages are required to determine whether causal relationships become apparent over longer periods of time and/or at different stages of life. The underlying cause(s) of WMH growth and cortical thinning have yet to be fully determined. Acknowledgments We thank the funders, participants, research centers, clinical, and administrative staff who contributed to the LBC1936 study (detailed fully at http://www.lothianbirthcohort.ed.ac.uk/).

Notwithstanding these limitations, we have shown that although they both worsen with age, WMH volume progression and regional cortical thinning do not seem to have a correlative/causal longitudinal relationship from 73 to 76 years. Further longitudinal studies with longer follow-up times and with more time points at different ages are required to determine whether causal relationships become apparent over longer periods of time and/or at different stages of life. The underlying cause(s) of WMH growth and cortical thinning have yet to be fully determined. Acknowledgments We thank the funders, participants, research centers, clinical, and administrative staff who contributed to the LBC1936 study (detailed fully at http://www.lothianbirthcohort.ed.ac.uk/). Sources of Funding This work was funded by a Scottish Funding Council Early Career Researcher grant to the Scottish Imaging Network—A Platform for Scientific Excellence (http://www.sinapse.ac.uk; DAD); Research into Ageing program grant (Drs Deary and Starr) and the Age UK-funded Disconnected Mind project (Drs Deary, Starr, and Wardlaw), with additional funding from the UK Medical Research Council (Drs Deary, Starr, and Wardlaw, and M.E. Bastin); and Scottish Funding Council through the Scottish Imaging Network—A Platform for Scientific Excellence (Dr Wardlaw).

and Starr) and the Age UK-funded Disconnected Mind project (Drs Deary, Starr, and Wardlaw), with additional funding from the UK Medical Research Council (Drs Deary, Starr, and Wardlaw, and M.E. Bastin); and Scottish Funding Council through the Scottish Imaging Network—A Platform for Scientific Excellence (Dr Wardlaw). Disclosures Dr Wardlaw reports money (grants) paid to The University of Edinburgh from Medical Research Council, Age UK, Row Fogo Charitable Trust, and Scottish Funding Council for her efforts on the LBC1936 study and various imaging projects. Dr Deary reports money (grants) paid to The University of Edinburgh from Medical Research Council and Age UK for his efforts on the LBC1936 study. Dr Deary reports money paid to him for board membership on Medical Research Council. All other authors have no disclosures. * Drs Deary and Wardlaw contributed equally.

Patients with episodic hypertension in clinic after a previous transient ischemic attack or stroke have a high risk of recurrent stroke,1,2 residual visit-to-visit variability in blood pressure (BP) on antihypertensive treatment has a poor prognosis, despite good control of mean BP,3 and benefits of some antihypertensive drugs in the prevention of stroke may partly result from reduced variability in systolic BP (SBP).3,4 Home day-to-day BP variability (home BP monitoring [HBPM] BPV) is similarly associated with an increased stroke risk,5,6 particularly for variability in morning BP6 and is reduced by similar medications. In contrast, short-term BPV on awake ambulatory BP monitoring (ABPM) is only weakly predictive of cardiovascular events,2 as is within-visit variability in office BP, with short-term BPV also correlating poorly with visit-to-visit BPV.1,2 However, the predictive value of beat-to-beat BPV for 5 minutes has not been determined.

rast, short-term BPV on awake ambulatory BP monitoring (ABPM) is only weakly predictive of cardiovascular events,2 as is within-visit variability in office BP, with short-term BPV also correlating poorly with visit-to-visit BPV.1,2 However, the predictive value of beat-to-beat BPV for 5 minutes has not been determined. Beat-to-beat BPV for 5 minutes is only weakly correlated with day-to-day BPV on HBPM and premorbid visit-to-visit BPV but shares the same physiological associations, suggestive of a similar pathophysiology.7 Increased beat-to-beat BPV8 and diminished baroreceptor sensitivity (derived from beat-to-beat BP monitoring)9 are potentially associated with a worse outcome after a major acute stroke and may be associated with an increased risk of recurrent events.8 However, previous studies were small with significant methodological problems. Therefore, we determined the predictive value of beat-to-beat BPV in a prospective cohort of patients with recent transient ischemic attack or minor stroke. Materials and Methods Requests for access to the data and analysis tools in this article will be openly considered. Please contact P.M.R. for further information.

ination<23), pregnant; had a recent myocardial infarction, unstable angina, heart failure (New York Heart Association, 3–4 or ejection fraction, <40%), or untreated bilateral carotid stenosis (>70%); and if they had atrial fibrillation during testing. The study was approved by the Oxfordshire Research Ethics Committee. BP Measurement Two sitting clinic BPs, 5 minutes apart, were measured at ascertainment and 1 month in the nondominant arm, by trained personnel after 5 minutes of rest. From ascertainment, patients agreeing to perform HBPM performed 3 home readings for 10 minutes, 3× daily (after waking, midmorning, and evening) with a Bluetooth-enabled, regularly calibrated telemetric IEM Stabil-o-Graph or A&D UA-767 BT. Patients were instructed to relax in a chair for 5 minutes before measuring BP in the nondominant arm or the higher-reading arm when the mean SBP differed by >20 mm Hg. Anonymized measures were securely transmitted via Bluetooth radio and a mobile phone to a password-protected website (t+ Medical, Abingdon, United Kingdom) and medication prescribed according to guidelines,11 most frequently with perindopril, indapamide, or amlodipine, to a target of <130/80. The day before the 1-month follow-up, ABPM was performed with an A&D TM-2430 monitor in the nondominant arm. BP was measured every 30 minutes during the day and 60 minutes at night.

and medication prescribed according to guidelines,11 most frequently with perindopril, indapamide, or amlodipine, to a target of <130/80. The day before the 1-month follow-up, ABPM was performed with an A&D TM-2430 monitor in the nondominant arm. BP was measured every 30 minutes during the day and 60 minutes at night. Beat-to-beat BPV was measured for 5 minutes at the ascertainment visit or 1-month clinic in a quiet, dimly lit, temperature-controlled room (21–23°C). Continuous 3-lead ECG and finger arterial BP were acquired at 200 Hz (Finometer MIDI) via a Powerlab 8/35 (ADInstruments), from the nondominant arm when possible. Automated calibration was performed until the recording was stable, but turned off during each test, and readings calibrated offline to the mean of 2 supine, oscillometric brachial readings.

arterial BP were acquired at 200 Hz (Finometer MIDI) via a Powerlab 8/35 (ADInstruments), from the nondominant arm when possible. Automated calibration was performed until the recording was stable, but turned off during each test, and readings calibrated offline to the mean of 2 supine, oscillometric brachial readings. Analysis BPV on beat-to-beat monitoring was calculated for 5 minutes. Ectopic beats and artefacts were automatically detected, visually reviewed, and removed by linear interpolation. Patients in atrial fibrillation during the recording were excluded. Variability was calculated as the coefficient of variation (CV) about a linear regression across 5 minutes to remove drift in the waveform (residual CV). HBPM variability was derived from the last 7 days of recording before the 1-month follow-up visit, from the average SBP or diastolic BP (DBP) calculated from the last 2 readings of each cluster of 3. Awake BPV on ABPM was derived after automated and manual exclusion of artefacts according to standard criteria.12 BPV was derived as the CV (CV=SD/mean) and the residual CV about a moving average on HBPM. Reproducibility of BPV on HBPM was determined in 100 patients between the second and third weeks of monitoring as Pearson r and intraclass correlation coefficient. In 50 patients, beat-to-beat BPV was measured at baseline and the 1-month visit according to the same protocol to determine reproducibility of measurement by Pearson r and the intraclass correlation coefficient.

tients between the second and third weeks of monitoring as Pearson r and intraclass correlation coefficient. In 50 patients, beat-to-beat BPV was measured at baseline and the 1-month visit according to the same protocol to determine reproducibility of measurement by Pearson r and the intraclass correlation coefficient. Risk of recurrent cardiovascular events was determined per unit increase in mean and variability in SBP or DBP and per SD for the population for each method of measurement by Cox proportional hazards regression, with and without adjustment for age, sex, and major cardiovascular risk factors (hypertension, diabetes mellitus, family history, smoking, atrial fibrillation, and dyslipidemia), and in combined models adjusting for other measures of BPV. The effect of adjustment of beat-to-beat and day-to-day BPV for regression to the mean was estimated by scaling the difference between the mean BPV for each quartile of BPV and the population mean by the intraclass correlation coefficient.13 Literature Review Pubmed and EMBASE were searched from inception until March 1, 2017, with the terms (“blood pressure” OR “BP” OR “hypertension” OR “BPV” OR “baroreflex” OR “BRS” OR “baroreflex sensitivity”) AND (“stroke” OR “cerebr*” OR “prognosis” OR “death” OR “mortality” OR “cerebrovascular accident” OR “cerebrovascular event” OR “cerebrovascular” OR “leukoaraiosis” OR “white matter hyperintensities” OR “white matter disease” OR “small vessel disease”). All articles reporting recurrent cardiovascular events per unit of beat-to-beat BPV were identified.

gnosis” OR “death” OR “mortality” OR “cerebrovascular accident” OR “cerebrovascular event” OR “cerebrovascular” OR “leukoaraiosis” OR “white matter hyperintensities” OR “white matter disease” OR “small vessel disease”). All articles reporting recurrent cardiovascular events per unit of beat-to-beat BPV were identified. Results Of 520 patients, 26 had poor-quality recordings because of excessive ectopics or poor-quality finometer recordings because of poor peripheral circulation, whereas 22 were excluded from beat-to-beat analyses because of atrial fibrillation during the recording, which limits the accuracy of BPV measurement, leaving 472 patients with valid beat-to-beat recordings. Four hundred sixty-six of 520 patients had adequate HBPM (2.9 readings per cluster for median 29 days) and 461 of 520 had adequate ABPM (Table 1), with 405 patients with adequate monitoring undergoing all forms of recording. There were weak-positive associations between BPV measured with different methods (beat-to-beat CV versus HBPM residual CV: r=0.119, P=0.017; beat-to-beat CV versus awake SBP CV: r=0.04, P=0.37; HBPM residual CV versus awake SBP CV: r=0.20, P<0.001) but limited associations with demographic variables (Table 1). Table 1. Demographics of 472 Patients With Adequate Beat-to-Beat Recording in Sinus Rhythm During the Recording

Results Of 520 patients, 26 had poor-quality recordings because of excessive ectopics or poor-quality finometer recordings because of poor peripheral circulation, whereas 22 were excluded from beat-to-beat analyses because of atrial fibrillation during the recording, which limits the accuracy of BPV measurement, leaving 472 patients with valid beat-to-beat recordings. Four hundred sixty-six of 520 patients had adequate HBPM (2.9 readings per cluster for median 29 days) and 461 of 520 had adequate ABPM (Table 1), with 405 patients with adequate monitoring undergoing all forms of recording. There were weak-positive associations between BPV measured with different methods (beat-to-beat CV versus HBPM residual CV: r=0.119, P=0.017; beat-to-beat CV versus awake SBP CV: r=0.04, P=0.37; HBPM residual CV versus awake SBP CV: r=0.20, P<0.001) but limited associations with demographic variables (Table 1). Table 1. Demographics of 472 Patients With Adequate Beat-to-Beat Recording in Sinus Rhythm During the Recording BPV on beat-to-beat monitoring in the 405 patients undergoing all forms of monitoring was associated with an increased risk of ischemic stroke, any stroke, and all cardiovascular events, independently of mean SBP (Table 2), before and after adjustment for age and sex, with a significant association with the risk of recurrent ischemic stroke remaining after adjustment for other cardiovascular risk factors (hazard ratio per SD, 1.40 [1.00–1.94]; P=0.047). Relationships were similar for all patients undergoing each form of monitoring and largely unchanged by adjustment for mean SBP (Table I in the online-only Data Supplement). The hazard ratio per 1% increase in beat-to-beat CV for stroke was 1.24 (1.07–1.43; P=0.004; Table II in the online-only Data Supplement). BPV on home monitoring was not as strongly associated with stroke risk but was associated with all-cause mortality and a composite of death and cardiovascular events (Table 2). Beat-to-beat DBP variability was not predictive of recurrent events, although home DBP variability weakly predicted all-cause mortality (Table III in the online-only Data Supplement). In contrast to beat-to-beat and home monitoring, BPV on ABPM did not predict any recurrent events (Table 2), but mean SBP on all 3 methods of measurement predicted the risk of future events (Table IV in the online-only Data Supplement).

eakly predicted all-cause mortality (Table III in the online-only Data Supplement). In contrast to beat-to-beat and home monitoring, BPV on ABPM did not predict any recurrent events (Table 2), but mean SBP on all 3 methods of measurement predicted the risk of future events (Table IV in the online-only Data Supplement). Table 2. Risk of Cardiovascular Events During Follow-Up, According to Variability on Each Method of Blood Pressure Measurement There was a significant increase in the absolute risk of recurrent stroke or all cardiovascular events across quartiles of BPV (Figure). Furthermore, beat-to-beat and day-to-day BPV were both moderately reproducible in 50 and 100 patients, respectively (intraclass correlation coefficient HBPM, 0.614; P<0.001; beat-to-beat, 0.503; P<0.001; Figure I in the online-only Data Supplement), resulting in a similar increase in the association between usual BPV on beat-to-beat and home monitoring after correction for regression dilution bias (Figure II in the online-only Data Supplement). Figure. Absolute risks of recurrent stroke or major cardiovascular events by quartile of blood pressure variability (BPV) on each form of monitoring. The percentage risk of a recurrent stroke or major cardiovascular event (cardiovascular death, stroke, myocardial infarction, or acute peripheral vascular disease) during follow-up for 405 patients undergoing all forms of blood pressure monitoring is shown, subdivided by quartile of each method of monitoring. ABPM indicates ambulatory blood pressure monitoring.

r major cardiovascular event (cardiovascular death, stroke, myocardial infarction, or acute peripheral vascular disease) during follow-up for 405 patients undergoing all forms of blood pressure monitoring is shown, subdivided by quartile of each method of monitoring. ABPM indicates ambulatory blood pressure monitoring. In models including both beat-to-beat and HBPM BPV, beat-to-beat BPV was more predictive of the risk of recurrent stroke, whereas BPV on HBPM was more predictive of the risk of all cardiovascular events (Table V in the online-only Data Supplement). Similarly, mean BPV on beat-to-beat monitoring was significantly lower in patients unaffected by stroke than affected patients, whereas BPV on home monitoring was significantly lower compared with patients dying or experiencing outcome event (Table VI in the online-only Data Supplement). Two hundred nineteen abstracts of 960 search responses were potentially relevant, with 34 articles reviewed in full. No study reported the risk of recurrent cardiovascular events per change in beat-to-beat BPV. As in our previous meta-analysis,14 the risk of a poor outcome after acute stroke was associated with both SBP variability (hazard ratio, 1.07 [0.9–1.2]) and DBP variability (hazard ratio, 1.33 [1.1–1.7]),8,15 whereas a reduced baroreceptor sensitivity was associated with poor outcome after stroke9 or myocardial infarction.16,17

previous meta-analysis,14 the risk of a poor outcome after acute stroke was associated with both SBP variability (hazard ratio, 1.07 [0.9–1.2]) and DBP variability (hazard ratio, 1.33 [1.1–1.7]),8,15 whereas a reduced baroreceptor sensitivity was associated with poor outcome after stroke9 or myocardial infarction.16,17 Discussion BPV predicted the risk of recurrent stroke and all cardiovascular events on 5 minutes of beat-to-beat BP monitoring, with a ≈4-fold increase in risk between the lowest and highest quartile of the population, with broadly similar predictive power to BPV on day-to-day monitoring.

previous meta-analysis,14 the risk of a poor outcome after acute stroke was associated with both SBP variability (hazard ratio, 1.07 [0.9–1.2]) and DBP variability (hazard ratio, 1.33 [1.1–1.7]),8,15 whereas a reduced baroreceptor sensitivity was associated with poor outcome after stroke9 or myocardial infarction.16,17 Discussion BPV predicted the risk of recurrent stroke and all cardiovascular events on 5 minutes of beat-to-beat BP monitoring, with a ≈4-fold increase in risk between the lowest and highest quartile of the population, with broadly similar predictive power to BPV on day-to-day monitoring. Residual visit-to-visit variability in BP on antihypertensive treatment has a poor prognosis, despite good control of mean BP, with an increased risk of stroke and all cardiovascular events,1,2 and benefits of some antihypertensive drugs in the prevention of stroke seem to be due partly to reduced variability in SBP.3,4 However, more rapid assessment and control of BPV would be clinically useful, especially in the acute phase after transient ischemic attack or stroke. BPV on home BP monitoring is also predictive of recurrent strokes and all cardiovascular events5,6 and can be assessed for days but still poses practical challenges in retrieving and analyzing equipment and readings. Our study shows that a rapid, 5-minute assessment of beat-to-beat BPV may have similar prognostic significance compared with HBPM. If affected by antihypertensive medication in the same way as visit-to-visit and HBPM BPV, beat-to-beat BPV could be a useful index to guide antihypertensive treatment decisions. However, we found only a weak correlation between BPV on different methods of measurement, yet they were independently related to outcomes. This is consistent with the weak relationship between within-visit and between-visit BPV in previous analyses of the ASCOT trial (Anglo-Scandinavian Cardiac Outcomes Trial).2 Therefore, BPV on beat-to-beat and home monitoring may well be a complementary measure, potentially reflecting different pathophysiological mechanisms leading to stroke.

the weak relationship between within-visit and between-visit BPV in previous analyses of the ASCOT trial (Anglo-Scandinavian Cardiac Outcomes Trial).2 Therefore, BPV on beat-to-beat and home monitoring may well be a complementary measure, potentially reflecting different pathophysiological mechanisms leading to stroke. We have demonstrated previously that home and beat-to-beat BPV are associated with a similar underlying physiological phenotype,7 including increased arterial stiffness, aortic pulsatility, reduced baroreceptor gain, and increased cardiovascular reactivity to stress. Furthermore, patients with an acute stroke have increased beat-to-beat BPV and reduced baroreceptor gain,18 which is associated with increased mortality9 and is partly dependent on stroke location.19 However, the precise mechanism by which BPV is associated with an increased risk of recurrent stroke is unclear. This may reflect either the effects of associated physiological indices (arterial stiffness, pulsatility, and cerebrovascular reactivity) or direct effects of beat-to-beat BPV itself. However, beat-to-beat BPV is a composite measure of multiple physiological processes, including irregular episodic components and rhythmic components related to breathing and to underlying autonomic rhythms (ie, low frequency oscillations at 0.04–0.15 Hz),20 and its prognostic significance may also reflect multiple pathophysiological processes.

is a composite measure of multiple physiological processes, including irregular episodic components and rhythmic components related to breathing and to underlying autonomic rhythms (ie, low frequency oscillations at 0.04–0.15 Hz),20 and its prognostic significance may also reflect multiple pathophysiological processes. Our study has some limitations. First, some patients were excluded because of poor-quality recordings, either because of poor peripheral circulation, excess ectopy, or atrial fibrillation during the recording. However, this reflects the strength of study, which included a consecutively recruited, unselected elderly population with acute events. Second, although statistical power to compare different measures of BPV was limited by the relatively small number of recurrent vascular events, the study is nevertheless the largest study of the prognostic significance of beat-to-beat SBP variability in patients with stroke. Third, BPV was estimated after initiation of antihypertensive treatment, which may affect BPV and its association with recurrent events. However, this also largely removes the confounding effects of inadequate mean BP control. Finally, repeated assessments for calculation of reproducibility of measures were performed after initiation of treatment. However, this would be expected to cause an underestimate of reproducibility.

iation with recurrent events. However, this also largely removes the confounding effects of inadequate mean BP control. Finally, repeated assessments for calculation of reproducibility of measures were performed after initiation of treatment. However, this would be expected to cause an underestimate of reproducibility. Beat-to-beat BPV is, therefore, an appealing measure to increase our understanding of both physiology and cerebrovascular risk prediction, but its potential use in clinical practice is limited by the need for continuous BP monitoring, specialist analysis, a need for validation in other cohorts, and a lack of normative values and thresholds for pathologically relevant BPV. These questions will require further research before the application of beat-to-beat BPV in practice. Furthermore, its use will ultimately depend on its capacity to alter management through improved risk prediction or the identification and monitoring of a novel treatment target. Conclusions Beat-to-beat BPV was a novel predictor of the risk of recurrent stroke and may be complementary to BPV on day-to-day home BP monitoring, may aid in risk stratification, and may help identify independently treatable mechanisms to reduce the risk of stroke. Acknowledgments We acknowledge the use of facilities of the Acute Vascular Imaging Centre and the Cardiovascular Clinical Research Facility, University of Oxford.

Conclusions Beat-to-beat BPV was a novel predictor of the risk of recurrent stroke and may be complementary to BPV on day-to-day home BP monitoring, may aid in risk stratification, and may help identify independently treatable mechanisms to reduce the risk of stroke. Acknowledgments We acknowledge the use of facilities of the Acute Vascular Imaging Centre and the Cardiovascular Clinical Research Facility, University of Oxford. Sources of Funding The Oxford Vascular Study is funded by the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre, Wellcome Trust, Wolfson Foundation, British Heart Foundation, and the European Unions Horizon 2020 Programme (grant 666881, SVDs@target). P.M. Rothwell is in receipt of an NIHR Senior Investigator award. A.J.S. Webb is funded by a Wellcome Trust Clinical Research Development Fellowship and British Heart Foundation Project grant. Disclosures None. Supplementary Material Guest Editor for this article was Natalia Rost, MD, MPH. The views expressed in this article are those of the author(s) and not necessarily of the National Health Service, the National Institute for Health Research, or the Department of Health. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.019107/-/DC1.

Although the associations between dementia and ischemic stroke have been comprehensively described,1 fewer data are available for spontaneous intracerebral hemorrhage (ICH), in part because of its high case fatality.2,3 Cognitive impairment often develops in survivors of ICH who were previously dementia free, particularly if the ICH is lobar, and has been associated with baseline neuroimaging markers of cerebral amyloid angiopathy (CAA).2 In those presenting with ICH, cognitive impairment before the event is common, with an estimated pooled incidence of 16.7%,4 suggesting that the underlying neurovascular and neuropathological processes that result in cognitive impairment after ICH might already be present at the time of initial presentation with ICH.2,4,5 However, it is not clear to what extent subsequent cognitive impairment after ICH is mediated by direct damage from the index ICH, the effects of recurrent ICH, or the impact of the underlying small vessel disease (SVD)2,4; understanding the contribution of these mechanisms is potentially important in developing rational dementia prevention strategies. We therefore investigated whether neuroimaging evidence of CAA (specifically, meeting the modified Boston criteria for probable CAA6 at presentation, and increases in a composite CAA score7) was associated with the presence of cognitive impairment before ICH. We then performed further analyses investigating the associations between individual magnetic resonance imaging (MRI) neuroimaging markers of SVD and cognitive impairment before ICH.

bable CAA6 at presentation, and increases in a composite CAA score7) was associated with the presence of cognitive impairment before ICH. We then performed further analyses investigating the associations between individual magnetic resonance imaging (MRI) neuroimaging markers of SVD and cognitive impairment before ICH. Materials and Methods Patient Selection We included patients recruited to a prospective multicentre observational cohort study of symptomatic patients with confirmed ICH (The Clinical Relevance of Microbleeds In Stroke Study; CROMIS-2). Those aged ≥18 years with an ICH confirmed on brain imaging (either computed tomography or MRI) were eligible, providing that there was no evidence that the ICH was because of an underlying structural cause or secondary to head trauma. This study has been preregistered, and the full details of the study protocol have been published previously.8 The study was approved by the National Research Ethics Service (IRAS reference 10/H0716/61). Written informed consent was obtained from each patient. The primary and substudy analyses for the CROMIS-2 study are ongoing; once all of these analyses are completed, the CROMIS-2 Steering Committee will consider applications from other researchers for access to anonymized source data.

e (IRAS reference 10/H0716/61). Written informed consent was obtained from each patient. The primary and substudy analyses for the CROMIS-2 study are ongoing; once all of these analyses are completed, the CROMIS-2 Steering Committee will consider applications from other researchers for access to anonymized source data. The Informant Questionnaire for Cognitive Decline in the Elderly (IQCODE) is a validated questionnaire given to a patient’s family member or caregiver which aims to establish whether there have been specific changes in cognitive and functional performance over the preceding 10-year time period.9–11 Specifically, the informant was asked to compare the patient’s performance from 10 years ago with their performance just before their stroke. The 16-item IQCODE was requested for all participants (score range, 1.0–5.0); this version of the IQCODE has been reported to have similar accuracy to the original 26-item version.10 We defined pre-ICH cognitive impairment as an IQCODE score of >3.3, based on previously reported pooled sensitivity and specificity values for detecting cognitive impairment from a meta-analysis investigating IQCODE accuracy in a general hospital setting.10 For inclusion in the final analysis, it was necessary for patients to have an IQCODE from the time of their admission, together with the MRI sequences needed for imaging analysis (described below).

The Informant Questionnaire for Cognitive Decline in the Elderly (IQCODE) is a validated questionnaire given to a patient’s family member or caregiver which aims to establish whether there have been specific changes in cognitive and functional performance over the preceding 10-year time period.9–11 Specifically, the informant was asked to compare the patient’s performance from 10 years ago with their performance just before their stroke. The 16-item IQCODE was requested for all participants (score range, 1.0–5.0); this version of the IQCODE has been reported to have similar accuracy to the original 26-item version.10 We defined pre-ICH cognitive impairment as an IQCODE score of >3.3, based on previously reported pooled sensitivity and specificity values for detecting cognitive impairment from a meta-analysis investigating IQCODE accuracy in a general hospital setting.10 For inclusion in the final analysis, it was necessary for patients to have an IQCODE from the time of their admission, together with the MRI sequences needed for imaging analysis (described below). Imaging Acquisition and Analysis Imaging was undertaken at each study center according to local protocols, and all brain imaging performed as part of the participant’s standard clinical care was sent to the study’s coordinating center in anonymized DICOM format.

For inclusion in the final analysis, it was necessary for patients to have an IQCODE from the time of their admission, together with the MRI sequences needed for imaging analysis (described below). Imaging Acquisition and Analysis Imaging was undertaken at each study center according to local protocols, and all brain imaging performed as part of the participant’s standard clinical care was sent to the study’s coordinating center in anonymized DICOM format. Imaging analysis was performed by 2 clinical research associates (D.W., G.B.) and 2 MSc students (K.O.-B.A, S.L.), all of whom were trained in neuroimaging rating and blinded to the participant clinical details. All structural imaging markers of cerebral SVD were rated in accordance with the Standards for Reporting Vascular Changes on Neuroimaging consensus criteria.12 Only those with an available MRI and all of the necessary sequences for cerebral SVD rating (ie, axial T2, axial or coronal fluid-attenuated inversion recovery (FLAIR), and a blood-sensitive sequence) were included in the neuroimaging marker analysis.

for Reporting Vascular Changes on Neuroimaging consensus criteria.12 Only those with an available MRI and all of the necessary sequences for cerebral SVD rating (ie, axial T2, axial or coronal fluid-attenuated inversion recovery (FLAIR), and a blood-sensitive sequence) were included in the neuroimaging marker analysis. Lacunes were identified and counted (D.W.) on T2 and FLAIR sequences.12 Cerebral microbleeds were rated (D.W.) using blood-sensitive (T2* weighted or susceptibility weighted images) sequences and the validated Microbleed Anatomical Rating Scale.13 MRI-visible perivascular spaces (PVS) in the centrum semiovale (CSO-PVS) and basal ganglia (BG-PVS) were defined and rated (G.B.) on T2 and FLAIR sequences using a validated 4-point visual rating scale12,14,15on a single predefined slice (first slice above the anterior commissure for the basal ganglia, and the first slice above the level of the lateral ventricles for the centrum semiovale). The hemisphere contralateral to the ICH was preferentially rated. White matter hyperintensities (WMH; also termed leukoaraiosis) were rated (K.O.-B.A.) on T2 and FLAIR sequences using the Fazekas scale.16,17 Cortical superficial siderosis (cSS) was identified on blood-sensitive sequences and classified (D.W.) as either focal (involving ≤3 sulci) or disseminated (involving ≥4 sulci), in keeping with previously described terminology.18 Medial temporal atrophy (MTA) was rated (G.B.) on T1 or FLAIR coronal images using the Scheltens visual scale.19,20 Global cortical atrophy (GCA) was rated (G.B.) using the Pasquier scale on axial T1 or FLAIR images. In cases where these sequences were not available, T2 images were used. For both MTA and GCA, there was good agreement between all sequences used (MTA κ=0.77; GCA κ=1.00). For both MTA and GCA, the hemisphere contralateral to the ICH was preferentially rated.

(G.B.) using the Pasquier scale on axial T1 or FLAIR images. In cases where these sequences were not available, T2 images were used. For both MTA and GCA, there was good agreement between all sequences used (MTA κ=0.77; GCA κ=1.00). For both MTA and GCA, the hemisphere contralateral to the ICH was preferentially rated. ICH location was defined as infratentorial, deep, or lobar, with the latter in cortical or cortical–subcortical regions and not involving any of the deep grey matter structures. Hematoma volume was calculated (S.L.) using a previously described validated semiautomated planimetric method.21 A clinico-radiological diagnosis of probable CAA was based on meeting the modified Boston criteria.6 The CAA score was calculated from a previously described 6-point scale.7 This scale awards 1 point for CSO-PVS rating of frequent-to-severe grades (ie, presence of >20 CSO-PVS) and WMH that is either Fazekas grade 3 if periventricular, or Fazekas grade ≥2 if deep.22 Additional points are awarded for the presence of lobar microbleeds (1 point if 2–4 are present; 2 points if there are ≥5) and cSS (1 point if focal; 2 points if disseminated).7 The SVD score was determined using a previously described 4-point scale.22,23 This scale awards 1 point for the presence of lacunes, microbleeds, BG-PVS rating of moderate-to-severe grades (ie, presence of >10 BG-PVS), and WMH that is either Fazekas grade 3 if periventricular or Fazekas grade ≥2 if deep.22

The CAA score was calculated from a previously described 6-point scale.7 This scale awards 1 point for CSO-PVS rating of frequent-to-severe grades (ie, presence of >20 CSO-PVS) and WMH that is either Fazekas grade 3 if periventricular, or Fazekas grade ≥2 if deep.22 Additional points are awarded for the presence of lobar microbleeds (1 point if 2–4 are present; 2 points if there are ≥5) and cSS (1 point if focal; 2 points if disseminated).7 The SVD score was determined using a previously described 4-point scale.22,23 This scale awards 1 point for the presence of lacunes, microbleeds, BG-PVS rating of moderate-to-severe grades (ie, presence of >10 BG-PVS), and WMH that is either Fazekas grade 3 if periventricular or Fazekas grade ≥2 if deep.22 Statistics We investigated for selection bias within our final cohort by comparing the characteristics of people with appropriate MRI and those without. IQCODE was dichotomized using a cutoff of 3.3, and baseline characteristics were compared (Table 1) for those with scores >3.3 (ie, with cognitive impairment) and those with scores ≤3.3. Continuous data were reviewed for normality, and if normally distributed we used the independent t test. Where continuous variables were not normally distributed, we used the (nonparametric) Mann–Whitney U test. We used the χ2 tests for categorical variables. The independent t test (normally distributed continuous data) and the 2-sample test of proportion (categorical data) were used to compare means and proportions, respectively. Table 1. Baseline Demographic and Clinical Characteristics

Statistics We investigated for selection bias within our final cohort by comparing the characteristics of people with appropriate MRI and those without. IQCODE was dichotomized using a cutoff of 3.3, and baseline characteristics were compared (Table 1) for those with scores >3.3 (ie, with cognitive impairment) and those with scores ≤3.3. Continuous data were reviewed for normality, and if normally distributed we used the independent t test. Where continuous variables were not normally distributed, we used the (nonparametric) Mann–Whitney U test. We used the χ2 tests for categorical variables. The independent t test (normally distributed continuous data) and the 2-sample test of proportion (categorical data) were used to compare means and proportions, respectively. Table 1. Baseline Demographic and Clinical Characteristics Univariate comparisons were used to identify potential confounders for inclusion in the multivariable models; all variables with P<0.05 were included. We then performed adjusted logistic regression analyses, adjusting for significant associations identified in univariate analyses (Table 2). In further analyses (Table 3), we investigated associations with other neuroimaging markers suggestive of CAA (the presence of strictly lobar microbleeds, and presentation with lobar ICH), as well as a composite SVD score and its component elements. In these analyses, each neuroimaging marker was considered individually (ie, each adjusted model included only 1 neuroimaging marker at a time). Given that these analyses were exploratory, we did not make an adjustment for multiple testing.

n with lobar ICH), as well as a composite SVD score and its component elements. In these analyses, each neuroimaging marker was considered individually (ie, each adjusted model included only 1 neuroimaging marker at a time). Given that these analyses were exploratory, we did not make an adjustment for multiple testing. Table 2. Univariable and Adjusted Logistic Regression Models, Investigating Associations Between Cognitive Impairment Before ICH and Evidence of CAA Table 3. Logistic Regression Models (Univariable and Adjusted), Reviewing Associations Between Cognitive Impairment Before ICH and Individual Structural Markers of Cerebral SVD, and a Composite SVD Score Statistical analysis was performed (G.B.) using Stata (Version 11.2).

Table 2. Univariable and Adjusted Logistic Regression Models, Investigating Associations Between Cognitive Impairment Before ICH and Evidence of CAA Table 3. Logistic Regression Models (Univariable and Adjusted), Reviewing Associations Between Cognitive Impairment Before ICH and Individual Structural Markers of Cerebral SVD, and a Composite SVD Score Statistical analysis was performed (G.B.) using Stata (Version 11.2). Results Cohort Characteristics The demographic and imaging characteristics of those included (n=166) are shown in Table 1. Patients without MRI (n=588) and those with MRI but with missing or uninterpretable sequences (n=43) were excluded (online-only Data Supplement). When compared with the excluded patients (online-only Data Supplement), those included were younger (mean, 68.9 versus 75.0 years; P<0.00001), less likely to have hypertension (58.2% versus 70.9%; P=0.002), hypercholesterolemia (35.8% versus 47.9%; P=0.006), diabetes mellitus (12.1% versus 19.8%; P=0.024), and atrial fibrillation (12.3% versus 43.5%; P<0.0001), and more likely to have previously had an ischemic stroke or transient ischemic attack (24.7% versus 18.1%; P=0.081), lower Glasgow Coma Scale at presentation (interquartile range, 13–15 versus 14–15; P=0.003) and pre-ICH cognitive decline (38.2% versus 24.7%; P=0.001).

rial fibrillation (12.3% versus 43.5%; P<0.0001), and more likely to have previously had an ischemic stroke or transient ischemic attack (24.7% versus 18.1%; P=0.081), lower Glasgow Coma Scale at presentation (interquartile range, 13–15 versus 14–15; P=0.003) and pre-ICH cognitive decline (38.2% versus 24.7%; P=0.001). When comparing those with and without pre-ICH cognitive decline, those with (n=41) were older (mean difference, 7.5 years; P<0.0012) and more likely to have hypercholesterolemia (51.2% versus 30.6%; P=0.017), diabetes mellitus (22.0% versus 8.9%; P=0.026), previous ischemic stroke or transient ischemic attack (29.0% versus 14.8%; P=0.047), and previous ICH (12.5% versus 3.2%; P=0.025). Associations With Pre-ICH Cognitive Decline: Univariate and Multivariate Analyses Univariate logistic regression analyses showed that pre-ICH cognitive decline was associated with meeting the modified Boston criteria for probable CAA at presentation and increasing CAA score (Table 2). In our multivariable analysis, we adjusted for age at event, hypercholesterolemia, presence of diabetes mellitus, previous ischemic stroke or transient ischemic attack, and previous ICH, which were statistically significant in univariate analyses (Table 1). Meeting the modified Boston criteria for probable CAA at presentation (odds ratio [OR], 4.01; 95% confidence interval [CI], 1.53–10.51); P=0.005) and increasing CAA score (for each point increase, OR, 1.42; 95% CI, 1.03–1.97; P=0.033) remained associated with pre-ICH cognitive decline (Table 2).

ate analyses (Table 1). Meeting the modified Boston criteria for probable CAA at presentation (odds ratio [OR], 4.01; 95% confidence interval [CI], 1.53–10.51); P=0.005) and increasing CAA score (for each point increase, OR, 1.42; 95% CI, 1.03–1.97; P=0.033) remained associated with pre-ICH cognitive decline (Table 2). We then performed further analyses investigating the associations between individual neuroimaging markers of SVD and cognitive impairment before ICH. In univariable analyses (Table 3), we identified associations between pre-ICH cognitive decline and increasing SVD score, WMH, the presence of cSS, presence of strictly lobar microbleeds, and lobar ICH at presentation. In analyses adjusted for clinical and demographic variables identified in the univariate analysis (as above), the presence of cSS (OR, 4.08; 95% CI, 1.28–13.05; P=0.018), strictly lobar microbleeds (OR, 2.47; 95% CI, 0.95–6.37; P=0.062), and lobar ICH at presentation (OR, 2.29; 95% CI, 0.99–5.31; P=0.053) showed associations with pre-ICH cognitive impairment. The previous associations with increasing SVD score and WMH were no longer statistically significant, although for WMH a large effect size remained (OR, 2.03). Discussion Our main new finding is that MRI neuroimaging markers of CAA are associated with pre-ICH cognitive impairment. This suggests that cognitive impairment in CAA is not only because of brain injury caused directly by ICH but also independently related to the underlying small vessel disruption associated with CAA.

Discussion Our main new finding is that MRI neuroimaging markers of CAA are associated with pre-ICH cognitive impairment. This suggests that cognitive impairment in CAA is not only because of brain injury caused directly by ICH but also independently related to the underlying small vessel disruption associated with CAA. Our findings add to growing evidence that CAA plays an important role in the development of cognitive impairment and dementia in those with ICH. The prevalence of pre-ICH dementia in lobar ICH is near double that in deep ICH,24 and structural imaging markers of CAA (cSS, cerebral microbleeds) present at the time of ICH are associated with later progression to dementia.2 Our results show that a composite CAA score has a per point association with cognitive decline; further studies could help establish whether such a score might be useful in patients with milder CAA (including those not fulfilling Boston criteria, or without macrohemorrhage). We found a strong association between cSS and pre-ICH cognitive impairment, suggesting that leptomeningeal hemorrhage, rather than parenchymal microbleeds, might be an especially important pathological process impairing cognition in CAA. Our findings also contribute to our understanding of the mechanisms by which CAA disrupts cognition, which include hematoma damage (via direct effects on cortical integrity and function2) and small vessel mechanisms. The latter may include effects on brain network efficiency,25 which correlates with cognitive performance and shows disturbances in the non-ICH hemisphere.26 Our finding that CAA is associated with cognitive impairment before ICH shows that hematoma damage cannot be the only mechanism contributing to cognitive disruption and supports the hypothesis that small vessel mechanisms are important.

ates with cognitive performance and shows disturbances in the non-ICH hemisphere.26 Our finding that CAA is associated with cognitive impairment before ICH shows that hematoma damage cannot be the only mechanism contributing to cognitive disruption and supports the hypothesis that small vessel mechanisms are important. A further possibility is that cognitive impairment before ICH is because of coincident Alzheimer’s disease.4 Although the co-occurrence of CAA and Alzheimer’s disease pathology is well recognized,27 CAA seems to have a cognitive profile distinct from that seen in Alzheimer’s disease, characterized primarily by deficits in processing speed and executive function.28,29 Recent neuropathological work30 found that CAA makes an independent contribution to cognitive performance in Alzheimer’s disease. Together, this evidence suggests that CAA has a specific neurovascular impact on cognitive performance, independent of coexistent Alzheimer’s pathology. Although we did not find an association between MTA or GCA (as putative imaging markers of Alzheimer’s pathology31) and pre-ICH cognitive impairment, we acknowledge that our sample size is small and so we cannot rule out missing subtle effects.

on cognitive performance, independent of coexistent Alzheimer’s pathology. Although we did not find an association between MTA or GCA (as putative imaging markers of Alzheimer’s pathology31) and pre-ICH cognitive impairment, we acknowledge that our sample size is small and so we cannot rule out missing subtle effects. The main strength of this study is our detailed neuroimaging description of the structural markers of cerebral SVD in the context of pre-ICH cognitive decline, in a richly phenotyped prospective nationwide cohort of patients. However, our work also has limitations. Those included in our study were younger, with fewer comorbidities and a lower IQCODE than those who did not have an interpretable MRI; additionally, we acknowledge that a suspicion of CAA could increase the likelihood of an MRI being performed (50% of our included patients presented with lobar ICH), and so our final cohort might not be representative of those presenting with a spontaneous ICH to an acute stroke service. Brain imaging at each study center was completed according to local protocols, and so there are unavoidable variations in the nature and manner of the sequences obtained, which could influence our results. In particular, the use of susceptibility-weighted versus T2*-weighted gradient echo sequences may result different microbleed counts, as the former is more sensitive to this; we did not adjust for this in our analyses. There are inherent limitations of using the IQCODE, including variations in the threshold used to define cognitive impairment and the lack of validation against a reference standard for prestroke cognitive impairment. Finally, we acknowledge that our study size is small, and so our results should be interpreted cautiously, particularly the adjusted analyses. As detailed, we chose not to apply an adjustment for multiple testing in order not to miss potential associations of interest. In addition, although our study is powered to detect moderate effect sizes, it may have missed smaller effects.

our results should be interpreted cautiously, particularly the adjusted analyses. As detailed, we chose not to apply an adjustment for multiple testing in order not to miss potential associations of interest. In addition, although our study is powered to detect moderate effect sizes, it may have missed smaller effects. Cognitive impairment before ICH is common and is associated with imaging findings consistent with an important contribution from CAA. This suggests that any future strategy aiming to reduce the impact of poststroke dementia in ICH will need to extend beyond stroke prevention and include strategies that address the small vessel impact of CAA. Further work on the natural history of when and how CAA may influence an individual’s cognitive profile is a priority for future research.

e strategy aiming to reduce the impact of poststroke dementia in ICH will need to extend beyond stroke prevention and include strategies that address the small vessel impact of CAA. Further work on the natural history of when and how CAA may influence an individual’s cognitive profile is a priority for future research. Appendix The CROMIS-2 Collaborators: Louise Shaw, MD; Jane Sword, MD; Azlisham Mohd Nor, MD; Pankaj Sharma, PhD; Roland Veltkamp MD; Deborah Kelly, MD; Frances Harrington, MD; Marc Randall, MD; Matthew Smith, MD; Karim Mahawish, MD; Abduelbaset Elmarim, MD; Bernard Esisi, MD; Claire Cullen, MD; Arumug Nallasivam, MD; Christopher Price, MD; Adrian Barry, MD; Christine Roffe, MD; John Coyle, MD; Ahamad Hassan, MD; Caroline Lovelock, DPhil; Jonathan Birns, MD; David Cohen, MD; L. Sekaran, MD; Adrian Parry-Jones, PhD; Anthea Parry, MD; David Hargroves, MD; Harald Proschel, MD; Prabel Datta, MD; Khaled Darawil, MD; Aravindakshan Manoj, MD; Mathew Burn, MD; Chris Patterson, MD; Elio Giallombardo, MD; Nigel Smyth, MD; Syed Mansoor, MD; Ijaz Anwar, MD; Rachel Marsh, MD; Sissi Ispoglou, MD; Dinesh Chadha, MD; Mathuri Prabhakaran, MD; Sanjeevikumar Meenakishundaram, MD; Janice O’Connell, MD; Jon Scott, MD; Vinodh Krishnamurthy, MD; Prasanna Aghoram, MD; Michael McCormick, MD; Paul O’Mahony, MD; Martin Cooper, MD; Lillian Choy, MD; Peter Wilkinson, MD; Simon Leach, MD; Sarah Caine, MD; Ilse Burger, MD; Gunaratam Gunathilagan, MD; Paul Guyler, MD; Hedley Emsley, MD; Michelle Davis, MD; Dulka Manawadu, MD; Kath Pasco, MD; Maam Mamun, MD; Robert Luder, MD; Mahmud Sajid, MD; Ijaz Anwar, MD; James Okwera, MD; Julie Staals, PhD; Elizabeth Warburton, MD; Kari Saastamoinen, MD; Timothy England, MD; Janet Putterill, MD; Enrico Flossman, MD; Michael Power, MD; Krishna Dani, MD; David Mangion, MD; Appu Suman, MD; John Corrigan, MD; Enas Lawrence, MD; and Djamil Vahidassr, MD.

D; Mahmud Sajid, MD; Ijaz Anwar, MD; James Okwera, MD; Julie Staals, PhD; Elizabeth Warburton, MD; Kari Saastamoinen, MD; Timothy England, MD; Janet Putterill, MD; Enrico Flossman, MD; Michael Power, MD; Krishna Dani, MD; David Mangion, MD; Appu Suman, MD; John Corrigan, MD; Enas Lawrence, MD; and Djamil Vahidassr, MD. Sources of Funding The CROMIS-2 study is funded by the Stroke Association and British Heart Foundation. G. Banerjee receives funding from the Rosetrees Trust. Dr Ambler receives funding from the National Institute for Health Research University College London Hospitals Biomedical Research Centre. Dr Al-Shahi Salman is funded by an Medical Research Council senior clinical fellowship. M.M. Brown’s Chair in Stroke Medicine is supported by the Reta Lila Weston Trust for Medical Research. Dr Werring receives research support from the Stroke Association, the British Heart Foundation, and the Rosetrees Trust. This work was undertaken at University College London Hospitals and University College London which receive a proportion of funding from the Department of Health National Institute for Health Research (NIHR) Biomedical Research Centres funding scheme.

Association, the British Heart Foundation, and the Rosetrees Trust. This work was undertaken at University College London Hospitals and University College London which receive a proportion of funding from the Department of Health National Institute for Health Research (NIHR) Biomedical Research Centres funding scheme. Disclosures Dr Cohen has received institutional research support from Bayer; honoraria for lectures and an Advisory Board from Bayer, diverted to a local charity; and travel/accommodation expenses for participation in scientific meetings covered by Bayer and Boehringer Ingelheim. G.H.Y. Lip has served as a consultant for Bayer, Astellas, Merck, AstraZeneca, Sanofi, BMS/Pfizer, Biotronik, Portola, and Boehringer Ingelheim and has been on the speakers’ bureau for Bayer, BMS/Pfizer, Boehringer Ingelheim, and Sanofi-Aventis. The other authors report no conflicts. Supplementary Material Continuing medical education (CME) credit is available for this article. Go to http://cme.ahajournals.org to take the quiz. * A list of all the CROMIS-2 Collaborators is given in the Appendix. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.019409/-/DC1.

The Wellcome Trust Case Control Consortium 2 genome-wide association study reported an association between a genetic variant on chromosome 7p21.1 and an increased risk of ischemic stroke because of large artery disease.1 The association was confined to large artery stroke and not present with cardioembolic or lacunar stroke. The association has been replicated in other cohorts of patients with stroke.2,3 The same genetic variant has also been associated with increased carotid intima–media thickness and asymptomatic carotid plaque,4 and less strongly, with coronary artery disease,5 suggesting an action via increasing atherosclerosis. The underlying gene is thought to be histone deacetylase 9 (HDAC9 ).4 Mice with a deficiency of the HDAC9 gene (HDAC9−/− apolipoprotein E–deficient) exhibit reduced aortic atherosclerosis compared with HDAC9+/+ apolipoprotein E–deficient mice that do not have a deficiency.6 Furthermore, HDAC9 expression is upregulated in symptomatic carotid atherosclerotic plaques in man.4

e 9 (HDAC9 ).4 Mice with a deficiency of the HDAC9 gene (HDAC9−/− apolipoprotein E–deficient) exhibit reduced aortic atherosclerosis compared with HDAC9+/+ apolipoprotein E–deficient mice that do not have a deficiency.6 Furthermore, HDAC9 expression is upregulated in symptomatic carotid atherosclerotic plaques in man.4 The antiepileptic drug (AED) sodium valproate (SVA) is a nonspecific inhibitor of HDAC9 activity7 and has been shown to attenuate atherosclerosis in apolipoprotein E–deficient mice.8 A large Danish study suggested that although epilepsy was associated with an increased risk of incident stroke, the extent of this effect varied with the type of AED that was prescribed. SVA was associated with a decreased risk of both stroke and myocardial infarction compared with carbamazepine.9,10 A further large community study found a dose–response relationship with higher doses of SVA being associated with lower risks of incident stroke, but similar associations were also seen with some other AEDs, raising the possibility of survivor bias.11 These findings raise the hypothesis that inhibiting HDAC9 activity might offer a novel preventative treatment for large artery atherosclerotic ischemic stroke. We sought to indirectly test this hypothesis by exploring the association between exposure to SVA and subsequent risk of recurrent stroke in 3 large cohorts of patients with prior stroke or transient ischemic attack (TIA).

vity might offer a novel preventative treatment for large artery atherosclerotic ischemic stroke. We sought to indirectly test this hypothesis by exploring the association between exposure to SVA and subsequent risk of recurrent stroke in 3 large cohorts of patients with prior stroke or transient ischemic attack (TIA). Methods Data Sources This project used data provided to us by 3 long-term follow-up stroke studies. Access to these separate data sources is therefore not available via this project. Data were collected, and pooled, from 3 prospective studies recruiting patients with previous stroke or TIA and with long-term follow-up: The SLSR (South London Stroke Register; n=4972) was a prospective population-based cohort study to record first-ever strokes in Lambeth and Southwark, London, United Kingdom.12 The final data set included data collected for patients with first-ever strokes between January 01, 1995, and September 30, 2014. Stroke diagnosis was confirmed by a study physician within 1 week of the event. Face-to-face follow-up took place at 3 months and then annually after the index event. For patients reaching at least 1 follow-up, the mean time from initial stroke to final follow-up was 4.6 years (SD=4.4; range=0–19);

September 30, 2014. Stroke diagnosis was confirmed by a study physician within 1 week of the event. Face-to-face follow-up took place at 3 months and then annually after the index event. For patients reaching at least 1 follow-up, the mean time from initial stroke to final follow-up was 4.6 years (SD=4.4; range=0–19); The VITATOPS (Vitamins to Prevent Stroke Study; n=8164) was a clinical trial recruiting patients based on any stroke or TIA within the 7 months preceding randomization.13 Randomization took place between January 17, 1997, and December 29, 2008. Follow-up took place every 6 months from randomization to trial completion, either face-to-face or by telephone. For patients reaching at least 1 follow-up, the mean time from initial stroke or TIA to final follow-up was 3.4 years (SD=2.4; range=0–11); The OXVASC (Oxford Vascular study; n=2113) is a population-based study of acute vascular events in Oxfordshire.14,15 The data set included here comprised all recruits ascertained between April 03, 2002, and March 31, 2012, with any first ischemic stroke or TIA in the study period. Multiple methods of follow-up were used, including face-to-face follow-up. Follow-up took place at 1, 6, 12, 24, 60, and 120 months. The mean time from initial stroke or TIA to final follow-up in this subset was 4.3 years (SD=3.4; range=0–12). Ethics SLSR was approved by the following ethics committees: St Thomas’ Hospital, King’s College Hospital, Wandsworth, Riverside, and National Hospital for Neurology and Neurosurgery and the Institute of Neurology.

The OXVASC (Oxford Vascular study; n=2113) is a population-based study of acute vascular events in Oxfordshire.14,15 The data set included here comprised all recruits ascertained between April 03, 2002, and March 31, 2012, with any first ischemic stroke or TIA in the study period. Multiple methods of follow-up were used, including face-to-face follow-up. Follow-up took place at 1, 6, 12, 24, 60, and 120 months. The mean time from initial stroke or TIA to final follow-up in this subset was 4.3 years (SD=3.4; range=0–12). Ethics SLSR was approved by the following ethics committees: St Thomas’ Hospital, King’s College Hospital, Wandsworth, Riverside, and National Hospital for Neurology and Neurosurgery and the Institute of Neurology. VITATOPS received ethics approval in the United Kingdom from the Multicentre Research Ethics Committee for Scotland, in New Zealand from the Multi-region Ethics Committee, and from local research ethics committees applicable to each participating center. The Oxford Vascular Study was approved by the local research ethics committee (OREC A: 05/Q1604/70). Data Extracted Study Populations Study entry date was recorded as the date of index stroke or TIA. Classification of initial stroke pathology as ischemic or hemorrhagic was taken from the Oxfordshire Community Stroke Project (OSCP) or TOAST classifications (Trial of ORG 10172 in Acute Stroke Treatment) according to which classification had the least missing data per study. Where this was not available, the cases were categorized as unclassified.

ke pathology as ischemic or hemorrhagic was taken from the Oxfordshire Community Stroke Project (OSCP) or TOAST classifications (Trial of ORG 10172 in Acute Stroke Treatment) according to which classification had the least missing data per study. Where this was not available, the cases were categorized as unclassified. Study End Point-Stroke Recurrence Recurrent stroke occurrence and TOAST classification of recurrent ischemic stroke were collected across all studies. Stroke recurrence was captured at follow-up and subsequently checked by study physicians. These were then subtyped based on the TOAST classification.16 Data were censored if, without recurrence, the patient died, reached their final follow-up, or the study ended. For the SLSR, stroke recurrence was defined as a new neurological deficit >24 hours after incident stroke and not considered to be because of edema, hemorrhagic transformation, or intercurrent illness. Recurrence within 21 days of the index stroke was only included if a different location was clearly indicated. For VITATOPS, recurrent stroke was defined as a new disturbance of focal neurological change lasting >24 hours or resulting in death and confirmed on imaging; all of the recorded recurrences were >24 hours from the index event. For OXVASC, recurrent stroke was recorded as any new neurological event lasting >24 hours or resulting in death and confirmed by a study physician who reviewed the surviving patients, the case records, and imaging.

g in death and confirmed on imaging; all of the recorded recurrences were >24 hours from the index event. For OXVASC, recurrent stroke was recorded as any new neurological event lasting >24 hours or resulting in death and confirmed by a study physician who reviewed the surviving patients, the case records, and imaging. Antiepileptic Treatment Prescription data for antiepileptic medication were available for all 3 studies. Prescription data were recorded at baseline and then each follow-up. For OXVASC, AED data were available at baseline and 1-year follow-up. For the SLSR records of SVA, carbamazepine, phenobarbital, phenytoin, gabapentin, lamotrigine, levetiracetam, and occasional less frequently prescribed medications (other) were available. For VITATOPS records of SVA, carbamazepine, phenobarbital, and phenytoin were available. For OXVASC, records of SVA, carbamazepine, phenobarbital, phenytoin, gabapentin, and lamotrigine were available.

gabapentin, lamotrigine, levetiracetam, and occasional less frequently prescribed medications (other) were available. For VITATOPS records of SVA, carbamazepine, phenobarbital, and phenytoin were available. For OXVASC, records of SVA, carbamazepine, phenobarbital, phenytoin, gabapentin, and lamotrigine were available. Other Variables Other variables included in the analyses were study, age, sex, and diagnosis of epilepsy. Diagnoses of epilepsy were captured separately to antiepileptic medication. This was defined either as an existing diagnosis at baseline, made by a qualified physician, or a diagnosis recorded at follow-up. For VITATOPS, seizures were recorded as adverse events at each follow-up. These were classified as epileptic seizures where they met the criteria for the International League Against Epilepsy, and no baseline data were available on this. For the SLSR, baseline diagnoses were taken from patient medical records, and subsequent diagnoses were collected via self-report at follow-ups. For the OXVASC study, this included people with a history of >1 seizure in later childhood or adult life recorded at baseline by self-report. Statistical Analysis Figure 1 shows the selection of patients, exposure groups, and recurrent events. Patients were excluded from analyses if they had a hemorrhagic qualifying event or were lost to first follow-up.

Other Variables Other variables included in the analyses were study, age, sex, and diagnosis of epilepsy. Diagnoses of epilepsy were captured separately to antiepileptic medication. This was defined either as an existing diagnosis at baseline, made by a qualified physician, or a diagnosis recorded at follow-up. For VITATOPS, seizures were recorded as adverse events at each follow-up. These were classified as epileptic seizures where they met the criteria for the International League Against Epilepsy, and no baseline data were available on this. For the SLSR, baseline diagnoses were taken from patient medical records, and subsequent diagnoses were collected via self-report at follow-ups. For the OXVASC study, this included people with a history of >1 seizure in later childhood or adult life recorded at baseline by self-report. Statistical Analysis Figure 1 shows the selection of patients, exposure groups, and recurrent events. Patients were excluded from analyses if they had a hemorrhagic qualifying event or were lost to first follow-up. Figure 1. Flow cart of cases included in analysis. *Where prescription is before recurrent stroke or study end only. AED indicates antiepileptic drug; OXVASC, Oxford Vascular Study; SLSR, South London Stroke Register; SVA, sodium valproate; and VITATOPS, Vitamins to Prevent Stroke Study.

Statistical Analysis Figure 1 shows the selection of patients, exposure groups, and recurrent events. Patients were excluded from analyses if they had a hemorrhagic qualifying event or were lost to first follow-up. Figure 1. Flow cart of cases included in analysis. *Where prescription is before recurrent stroke or study end only. AED indicates antiepileptic drug; OXVASC, Oxford Vascular Study; SLSR, South London Stroke Register; SVA, sodium valproate; and VITATOPS, Vitamins to Prevent Stroke Study. The SVA exposure population was defined based on any prescription of SVA before recurrent stroke or study end. This was calculated based on any SVA-recorded exposure in the period preceding the study outcome, death, or final follow-up. We used the date of the first follow-up where there was a prescription of SVA to calculate this. SVA prescribed after recurrence was not considered SVA exposure.

re recurrent stroke or study end. This was calculated based on any SVA-recorded exposure in the period preceding the study outcome, death, or final follow-up. We used the date of the first follow-up where there was a prescription of SVA to calculate this. SVA prescribed after recurrence was not considered SVA exposure. Our protocol specified analysis was as follows: The SVA exposure populations were compared with 2 minimally selective control populations to avoid bias. (1) all other patients; this included everyone other than those receiving SVA; and (2) all other AED prescriptions: defined as any record of AED prescription at any time from study entry to final follow-up, including those with dates after recurrence but excluding those with concurrent prescriptions of SVA. Because the cohorts we used were from non-AED studies, the precise start date for AED use was not available and recorded instead as the first follow-up visit at which their use was recorded. For this reason, we initially included AED use at any time as our control population

those with concurrent prescriptions of SVA. Because the cohorts we used were from non-AED studies, the precise start date for AED use was not available and recorded instead as the first follow-up visit at which their use was recorded. For this reason, we initially included AED use at any time as our control population It follows that our minimally selective criteria for the control populations could result in the other AED population, including some patients for whom AED prescriptions might be given after recurrent stroke. We, therefore, performed further secondary analyses with more restrictive comparison populations. We first compared patients receiving SVA with patients receiving no AEDs only, defined as patients without any AED prescriptions recorded before the study end or recurrent stroke. Second, we compared patients receiving SVA with patients only receiving other AED before stroke or study end based on the date of follow-up that these prescriptions were recorded.

A with patients receiving no AEDs only, defined as patients without any AED prescriptions recorded before the study end or recurrent stroke. Second, we compared patients receiving SVA with patients only receiving other AED before stroke or study end based on the date of follow-up that these prescriptions were recorded. Survival time was calculated as the number of days from the date of the index event (stroke or TIA) to the date of recurrent stroke occurrence or censoring. Life tables were calculated to describe the cumulative stroke-free survival for the exposure groups at 1, 5, 10, and 15 years. Survival curves were estimated with Kaplan–Meier and groups compared using the log-rank test in SPSS version 22. Cox regression was carried to calculate adjusted risk of a recurrent stroke. SVA exposure (SVA versus control), age, sex, history of epileptic symptoms, initial event type (TIA or stroke), and study were included in the model. A secondary prespecified analysis was to determine whether any protective associations with SVA were confined to patients with large artery stroke, as might be expected from the genetic association data. Therefore, we repeated analyses in patients with large artery stroke. Post hoc Cox regression analyses were performed for each study population individually. Exposure to SVA was compared with all patients without SVA exposure. Covariates were as above.

A secondary prespecified analysis was to determine whether any protective associations with SVA were confined to patients with large artery stroke, as might be expected from the genetic association data. Therefore, we repeated analyses in patients with large artery stroke. Post hoc Cox regression analyses were performed for each study population individually. Exposure to SVA was compared with all patients without SVA exposure. Covariates were as above. Results Descriptive Data The study cohort and patients included in the analysis are shown in the flow chart and in Table 1. A total of 11 949 patients were included in the pooled analyses, all of whom had a confirmed ischemic event at entry and follow-up available. The number in the SVA group was 168, and for those on other AEDs was 530. The total number of outcome events were 17 of 168 for patients prescribed SVA; 1470 of 11 781 for patients never prescribed SVA; 105 of 530 for patients prescribed other AEDs at any time; 1426 of 11 312 for patients not prescribed AEDs; and 44 of 469 for patients prescribed other AED when selected as prestroke/study end. Table 1. Pooled Background Data for OXVASC, VITATOPS, and SLSR Survival Models The cumulative stroke-free survival, based on yearly data, was greater for the SVA group than for patients not prescribed SVA. Data are shown in Table 2. Table 2. Cumulative Survival for All Entry Events (% Stroke Free)

Results Descriptive Data The study cohort and patients included in the analysis are shown in the flow chart and in Table 1. A total of 11 949 patients were included in the pooled analyses, all of whom had a confirmed ischemic event at entry and follow-up available. The number in the SVA group was 168, and for those on other AEDs was 530. The total number of outcome events were 17 of 168 for patients prescribed SVA; 1470 of 11 781 for patients never prescribed SVA; 105 of 530 for patients prescribed other AEDs at any time; 1426 of 11 312 for patients not prescribed AEDs; and 44 of 469 for patients prescribed other AED when selected as prestroke/study end. Table 1. Pooled Background Data for OXVASC, VITATOPS, and SLSR Survival Models The cumulative stroke-free survival, based on yearly data, was greater for the SVA group than for patients not prescribed SVA. Data are shown in Table 2. Table 2. Cumulative Survival for All Entry Events (% Stroke Free) For the nonselective control populations, Kaplan–Meier estimates were calculated for exposure to SVA, no exposure to SVA, and exposure to AED medication at any time. Log-rank tests showed that the difference between SVA exposure and no SVA exposure was not significant (χ2[1]=2.7; P=0.1) although there was a graphical trend indicating a difference in survival based on exposure at later time points; and the difference between SVA exposure and any other AED exposure was significant (χ2[1]=9.6; P=0.002). For the survival plots, see Figures 2 and 3.

nd no SVA exposure was not significant (χ2[1]=2.7; P=0.1) although there was a graphical trend indicating a difference in survival based on exposure at later time points; and the difference between SVA exposure and any other AED exposure was significant (χ2[1]=9.6; P=0.002). For the survival plots, see Figures 2 and 3. Figure 2. Survival curve comparing proportion of population free of recurrent stroke, in patients on sodium valproate (SVA) compared with those not on SVA. For the selective control populations, Kaplan–Meier estimates were calculated for exposure to SVA, no exposure to AED medication, and exposure to any other AED medication prestroke/study end. A log-rank test showed that the difference between SVA exposure and no AED exposure was not significant (χ2[1]=2.91; P=0.088) although there was a graphical trend indicating a difference in survival at later time points, and the difference between SVA exposure and other AED exposure was not significant (χ2[1]=0.01; P=0.937).

rank test showed that the difference between SVA exposure and no AED exposure was not significant (χ2[1]=2.91; P=0.088) although there was a graphical trend indicating a difference in survival at later time points, and the difference between SVA exposure and other AED exposure was not significant (χ2[1]=0.01; P=0.937). Hazard Models Adjusted for Covariates Cox hazard models were calculated and adjusted for covariates. For the nonselective control comparisons, 2 models were created to account for the overlap in the control groups. Exposure to SVA was associated with a reduced risk of stroke compared with all patients without SVA exposure (hazard ratio [HR]=0.50; 95% confidence interval [CI], 0.31–0.82; Wald test, P=0.006). Exposure to SVA was associated with a reduced risk of stroke compared with the group prescribed other AEDs at any time (HR=0.41; 95% CI, 0.25–0.71; Wald test, P=0.001). For the selective control comparison, a single model was created. Group status was significant (Wald=27.9; P<0.0001). SVA was associated with a reduced risk of stroke compared with no AED exposure (HR=0.48; 95% CI, 0.35–0.67; Wald test, P<0.0001). Exposure to SVA was not associated with change in risk of stroke compared with other AED exposure group when selected as prestroke/study end (HR=1.2; 95% CI, 0.66–2.02; Wald test, P=0.62).

. SVA was associated with a reduced risk of stroke compared with no AED exposure (HR=0.48; 95% CI, 0.35–0.67; Wald test, P<0.0001). Exposure to SVA was not associated with change in risk of stroke compared with other AED exposure group when selected as prestroke/study end (HR=1.2; 95% CI, 0.66–2.02; Wald test, P=0.62). Large Artery Stroke Analyses Subgroup analyses were performed to determine whether any associations were specific to large artery stroke. For those patients with large artery index events, 17 patients were in the exposed to SVA group, 2830 were in the never exposed to SVA, of which 108 were exposed to AED medication other than SVA. Survival data are given in Table 3. Table 3. Cumulative Survival for Cases With Large Artery Entry Events Kaplan–Meier estimates were computed. Although survival curves diverged for the groups, with a trend to better outcomes for those exposed to SVA, the data were limited by the small sample sizes. For a comparison between the SVA-exposed group and all other non-SVA–exposed patients, a log-rank test showed that the difference between the 2 survival curves was not significant (χ2[1]=0.073; P=0.787). For a comparison between the SVA-exposed group and the other AED group, a log-rank test showed that the difference between the 2 survival curves was not significant (χ2[1]=1.16; P=0.281).

posed patients, a log-rank test showed that the difference between the 2 survival curves was not significant (χ2[1]=0.073; P=0.787). For a comparison between the SVA-exposed group and the other AED group, a log-rank test showed that the difference between the 2 survival curves was not significant (χ2[1]=1.16; P=0.281). Post Hoc Analyses Cox hazard models were calculated per study. Exposure to SVA was associated with a trend toward a reduced risk of stroke compared with all patients without SVA exposure for VITATOPS (HR=0.52; 95% CI, 0.26–1.0; Wald test, P=0.052) and the SLSR (HR=0.33; 95% CI, 0.12–0.92; Wald test, P=0.033) but not OXVASC (HR=1.2; 95% CI, 0.40–3.4; Wald test, P=0.781). Discussion This analysis of 3 large cohorts of patients with prior stroke or TIA was undertaken to explore an a priori hypothesis (ie, the hypothesis was generated from published studies1–11 before the epidemiological data were analyzed) that exposure to an inhibitor of HDAC, in the form of SVA, may be associated with a lower risk of recurrent stroke compared with nonexposure or to exposure to other AEDs. Although the design of our study is prone to systematic and random error and cannot infer causality, the results provide some evidence for the prestudy hypothesis and suggest that SVA, a nonspecific HDAC inhibitor, may be associated with a reduced stroke recurrence rate.

d with nonexposure or to exposure to other AEDs. Although the design of our study is prone to systematic and random error and cannot infer causality, the results provide some evidence for the prestudy hypothesis and suggest that SVA, a nonspecific HDAC inhibitor, may be associated with a reduced stroke recurrence rate. Previously, data have suggested that SVA reduces stroke risk in a stroke-free population, but this analysis provides new data suggesting that such an effect can also be found in patients who have already presented with ischemic stroke. This is particularly relevant if HDAC9 inhibition is to be considered as a potential secondary preventative treatment for stroke. The results are broadly consistent with the results of 2 large population-based studies in Denmark and the United Kingdom. In a Danish study, SVA was associated with a lower risk of both myocardial infarction and stroke when compared with other AEDs.9,10 In the British study, which used the Clinical Practice Research Database, SVA exposure was associated with a reduced risk of myocardial infarction but not ischemic stroke.11 However, when a dose–response analysis was performed, longer exposure to SVA was associated with a reduced risk of ischemic stroke although similar associations were found with other AEDs raising questions about the specificity of the association.

with a reduced risk of myocardial infarction but not ischemic stroke.11 However, when a dose–response analysis was performed, longer exposure to SVA was associated with a reduced risk of ischemic stroke although similar associations were found with other AEDs raising questions about the specificity of the association. In our study when comparing SVA use with all other patients with ischemic stroke and those patients with ischemic stroke on no AED, there was a highly significant reduced risk of stroke. This was replicated in our preplanned analysis comparing patients with SVA with those taking any other AED. Because of the use of cohorts collected for other study purposes, we did not have a precise start date for AEDs, and we therefore used the date of the follow-up at which the medication was recorded. For this reason, we initially included AED use at any time as our control population. However, when we performed further exploratory analyses in the data where the AED record dates were before recurrent stroke, the difference was no longer significant. This raises the possibility that the significant risk reduction seen in the SVA group compared with the 3 control groups might be because of some difference if patient characteristics between the 2 groups. Such bias are impossible to exclude in a cohort study, such as this, and confirming whether SVA does indeed reduce stroke risk will require a randomized trial design. One potential bias might be if stroke subtype differs in patients on SVA, for example, lacunar stroke might have a lower risk of epilepsy requiring AED therapy and a lower risk of stroke recurrence. However, there was no evidence of any difference in lacunar stroke frequency between groups (Table 1). A second possible issue is related to the collection of prescription dates which were based on follow-up rather than prescription date. Alternatively, it may be a possibility that other AEDs also have some inhibitory effect on HDAC9 although evidence for this is less clear. Notwithstanding, SVA exposure appeared overall to have a positive effect compared with no SVA exposure.

escription dates which were based on follow-up rather than prescription date. Alternatively, it may be a possibility that other AEDs also have some inhibitory effect on HDAC9 although evidence for this is less clear. Notwithstanding, SVA exposure appeared overall to have a positive effect compared with no SVA exposure. Strengths of this study include the prospective design, standardized diagnostic criteria for the qualifying TIA and stroke, and prolonged follow-up. Two of the 3 cohorts were population-based studies, reducing the risk of any selection bias. The third was a large randomized clinical trial that again incorporated prolonged follow-up.

his study include the prospective design, standardized diagnostic criteria for the qualifying TIA and stroke, and prolonged follow-up. Two of the 3 cohorts were population-based studies, reducing the risk of any selection bias. The third was a large randomized clinical trial that again incorporated prolonged follow-up. However, the data sets also had some limitations. When examined separately as a post hoc analysis, 2 of the data sets yielded the same trend toward and effect for SVA, OXVASC did not. However, OXVASC had a small number of patients with confirmed SVA exposure (n=11), making it difficult to interpret in isolation. Sufficient data were not available to assess dose–response relationships, and therefore analyses were restricted to SVA prescribed at any time during follow-up. Because of the nature of data collection, often at annual follow-up, determining the exact start date of AED was not possible, and these were recorded as starting at the time the patient was first followed-up on that AED. Despite the large sample size of the >10 000 patients with stroke, the number of participants taking SVA and other AEDs was relatively few (in the hundreds) which limited statistical power. Furthermore, the number of cases of large artery stroke was too small to reliably test whether any protective effect was specific to this subtype, as hypothesized from the genetic association data.1 It is also inherent in this type of data that the outcomes are more heavily weighted in the early years (because of early recurrence and long follow-up), but this is consistently the case across comparison groups. It may be that data sets with more power would benefit for an analysis splitting short- and long-term outcomes in relation to AED medication.

data that the outcomes are more heavily weighted in the early years (because of early recurrence and long follow-up), but this is consistently the case across comparison groups. It may be that data sets with more power would benefit for an analysis splitting short- and long-term outcomes in relation to AED medication. It is also acknowledged that SVA is a nonspecific HDAC inhibitor (inhibiting a wide range of HDACs) and has other actions, independent of HDAC9 inhibition. Hence, a more specific inhibitor of HDAC9 might have a stronger effect in reducing the risk of recurrent stroke. HDACs are a class of enzymes that remove acetyl groups from an ε-N-acetyl lysine amino acid on a histone, allowing the histones to wrap the DNA more tightly. There are 18 HDACs in humans. Eleven of the HDACs are zinc dependent, classified on the basis of homology to yeast HDACs: class I includes HDACs 1, 2, 3, and 8; class IIA includes HDACs 4, 5, 7, and 9; class IIB, HDACs 6 and 10; and class IV, HDAC11.17 This study provides some support for the hypothesis that HDAC9 is important in the pathogenesis of ischemic stroke and that its inhibition, by SVA or a more specific HDAC9 inhibitor, is worthy of evaluation as a treatment to prevent recurrent ischemic stroke. However, because of limitations in a cohort study of this design, and possible unidentified bias, determining whether HDAC9 inhibition does reduce stroke risk requires randomized controlled trials of SVA or other HDA9 inhibitors.

C9 inhibitor, is worthy of evaluation as a treatment to prevent recurrent ischemic stroke. However, because of limitations in a cohort study of this design, and possible unidentified bias, determining whether HDAC9 inhibition does reduce stroke risk requires randomized controlled trials of SVA or other HDA9 inhibitors. Figure 3. Survival curve comparing proportion of population free of recurrent stroke, in patients on sodium valproate (SVA) compared with those on other antiepileptic drug (AED). Acknowledgments We thank Adina Feldman for assistance in data management and analysis. Dr Markus is supported by a National Institute for Health Research (NIHR) Senior Investigator award and the NIHR Biomedical Research Centre at Cambridge. Dr Rothwell is in receipt of an NIHR Senior Investigator Award and a Wellcome Trust Senior Investigator Award. Dr Wolfe is supported by the NIHR Biomedical Research Centre at Guy’s and St Thomas’ National Health Service Foundation Trust and King’s College London. The views expressed are those of the author(s) and not necessarily those of the National Health Service, the National Institutes of Health Research, or the Department of Health.

s supported by the NIHR Biomedical Research Centre at Guy’s and St Thomas’ National Health Service Foundation Trust and King’s College London. The views expressed are those of the author(s) and not necessarily those of the National Health Service, the National Institutes of Health Research, or the Department of Health. Sources of Funding This project was funded by a British Heart Foundation (PG/13/30005). The VITATOPS trial (Vitamins to Prevent Stroke Study) was funded by the National Health and Medical Research Council of Australia Project Grants 110267 (2000–2004) and 403913 (2006–2008) and Program Grants 251525 (2003–2007) and 454417 (2007–2011). The Oxford Vascular Study has been funded by Wellcome Trust, Wolfson Foundation, UK Stroke Association, British Heart Foundation, and NIHR Oxford Biomedical Research Centre and supported by the facilities of the Acute Vascular Imaging Centre, Oxford. Disclosures Drs Markus and Wolfe received funding from the British Heart Foundation as above to fund this project. The other authors report no conflicts. Supplementary Material Guest Editor for this article was Emmanuel Touze, PhD.

Cerebral small-vessel disease (SVD) is a term used to describe a group of pathological processes that affect the perforating cerebral arterioles and capillaries. Many brain parenchymal pathologies can occur, including small infarcts, microbleeds, ischemic demyelination with axonal loss, and diffuse brain atrophy.1 Clinically SVD presents with lacunar strokes, which represent ≈20% of all ischemic strokes, and it is the major cause of vascular cognitive impairment.2 Enlarged perivascular spaces (PvS) visible on magnetic resonance imaging (MRI) are a feature of SVD and vascular dementia2 and are associated with lacunar stroke and T2 white matter hyperintensities (WMH).3 The significance of MRI-visible PvS in SVD, however, remains controversial.2 PvS are pial-lined interstitial fluid-filled cavities that surround penetrating arteries, arterioles, veins, and venules.4 Their relationship with cognitive impairment in SVD remains uncertain, and the few studies that have investigated this have yielded inconsistent results.5–8 This inconsistency may be because of the difficulty in distinguishing between lacunes and PvS. Lacunes are subcortical cavities which occur in the region of a previous acute small deep brain infarct or hemorrhage in the territory of a perforating arteriole2 and often look very similar to PvS. Lacunes are thought to impair cognition by disrupting white matter pathways.9 A failure to distinguish between them adequately may reduce the sensitivity of studies and clinical trials investigating cognitive impairment in SVD.

hemorrhage in the territory of a perforating arteriole2 and often look very similar to PvS. Lacunes are thought to impair cognition by disrupting white matter pathways.9 A failure to distinguish between them adequately may reduce the sensitivity of studies and clinical trials investigating cognitive impairment in SVD. Another reason for the inconsistent association between PvS and cognition may be because most previous studies have used rating scales which are operator dependant and preferentially use T2-weighted MRI to assess PvS.6,7,10 PvS are difficult to rate in patients with moderate-to-severe SVD,10 as they are often obscured by WMH leading to an under or overestimation of PvS load. In addition, the rating scales currently used have a relatively narrow range (scores between 0 and 4) making longitudinal evaluation difficult. Measuring volumes using T1-weighted imaging is an alternative method to assess PvS load11 and may be more accurate in patients with SVD and WMH. In addition, it makes the longitudinal evaluation of PvS possible in a way that will be more sensitive to change than rating scales. To date, there are little data on whether PvS volumes change over time and whether they are associated with cognition.

ss PvS load11 and may be more accurate in patients with SVD and WMH. In addition, it makes the longitudinal evaluation of PvS possible in a way that will be more sensitive to change than rating scales. To date, there are little data on whether PvS volumes change over time and whether they are associated with cognition. The aim of this study is to further understand the clinical significance of PvS in symptomatic SVD. We carefully distinguished between lacunes and PvS and then investigated the relationship between PvS at baseline and cognitive change over a 5-year follow-up period in symptomatic SVD using both validated rating scales and computational measurements of PvS volumes. We compared this to the relationship between lacunes at baseline and cognitive change in the same population. In addition, we examined the change in PVS volume over time. Methods The data that support the findings of this study are available from the corresponding author on reasonable request.

The aim of this study is to further understand the clinical significance of PvS in symptomatic SVD. We carefully distinguished between lacunes and PvS and then investigated the relationship between PvS at baseline and cognitive change over a 5-year follow-up period in symptomatic SVD using both validated rating scales and computational measurements of PvS volumes. We compared this to the relationship between lacunes at baseline and cognitive change in the same population. In addition, we examined the change in PVS volume over time. Methods The data that support the findings of this study are available from the corresponding author on reasonable request. Participants Patients with SVD were recruited as part of the prospective SCANS study (St George’s Cognition and Neuroimaging in Stroke).12 Recruitment was from 3 hospitals covering a contiguous catchment area in South London (St George’s, King’s College, and St Thomas’ Hospitals). Inclusion criteria comprised a clinical lacunar stroke syndrome2 with an anatomically corresponding lacunar infarct on MRI in addition to confluent WMH on MRI (Fazekas grade ≥2).13 Exclusion criteria were any cause of stroke mechanism other than SVD (eg, cardioembolic source or extra- or intracerebral artery stenosis of >50%), other major central nervous system disorders, major psychiatric disorders, any other cause of white matter disease, contraindications to MRI, or nonfluent in English. The study was approved by the local ethics committee, and all patients gave written informed consent. MRI acquisitions and cognitive assessments were performed at least 3 months after the last stroke to exclude acute effects on cognition. All patients were also screened for cardiovascular risk factors including hypertension (defined as systolic blood pressure >140 mm Hg or diastolic >90 mm Hg or treatment with antihypertensive drugs), hypercholesterolemia (defined as a serum total cholesterol >5.2 mmol/L or treatment with a statin), diabetes mellitus, and smoking.

s were also screened for cardiovascular risk factors including hypertension (defined as systolic blood pressure >140 mm Hg or diastolic >90 mm Hg or treatment with antihypertensive drugs), hypercholesterolemia (defined as a serum total cholesterol >5.2 mmol/L or treatment with a statin), diabetes mellitus, and smoking. Participants were invited back annually for repeated cognitive testing and MRI scanning for a period of 3 years. Subsequently, 2 further annual assessments of cognitive function were conducted at years 4 and 5. A total of 121 subjects were recruited. Of these, 103 attended >1 cognitive assessment. Eighteen subjects only attended 1 assessment because of death (n=7), formal study withdrawal (n=6), house move (n=1), lost to follow-up (n=2), and withdrawal from full neuropsychological testing (n=2). Of the 103 subjects who attended cognitive assessments more than once, MRI data at multiple time points were available for 99, 98 at year 1, 77 at year 2, and 71 at year 3. One subject attended the baseline and missed the year 1 follow-up, but attended all subsequent sessions. Four subjects missed the year 2 follow-up, but subsequently attended at year 3. Four subjects withdrew from imaging but remained in the study for neuropsychological testing.

t year 1, 77 at year 2, and 71 at year 3. One subject attended the baseline and missed the year 1 follow-up, but attended all subsequent sessions. Four subjects missed the year 2 follow-up, but subsequently attended at year 3. Four subjects withdrew from imaging but remained in the study for neuropsychological testing. MRI Acquisition Images were acquired on a 1.5-T scanner (General Electric, Milwaukee, WI). All image sequences were acquired across the whole brain, and total imaging time was ≈45 minutes. The imaging protocol included T2-weighted, fluid-attenuated inversion recovery, gradient echo, and 3-dimensional T1-weighted sequences. Further details on the imaging protocol are given in the online-only Data Supplement and Lawrence at al.12 Image Processing Image processing was performed using the (SPM)8 software package (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/). A summary is provided in Figure 1.14 Further details are given in the online-only Data Supplement. Figure 1. Summary of preprocessing pipeline. Please refer to Lambert et al14 for further details. FLAIR indicates fluid-attenuated inversion recovery; SPM, statistical parametric mapping; TPM, tissue probability map; and WMH, white matter hyperintensities.

Image Processing Image processing was performed using the (SPM)8 software package (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/). A summary is provided in Figure 1.14 Further details are given in the online-only Data Supplement. Figure 1. Summary of preprocessing pipeline. Please refer to Lambert et al14 for further details. FLAIR indicates fluid-attenuated inversion recovery; SPM, statistical parametric mapping; TPM, tissue probability map; and WMH, white matter hyperintensities. Identification of Lacunes and PvS Lacunes were identified in native subject space by a consultant neuroradiologist (A.D.M.; blinded to the patient and cognitive data), using a multimodality view with T1-weighted, T2-weighted, and fluid-attenuated inversion recovery images. Lacunes were defined as subcortical, fluid-filled (similar signal as cerebrospinal fluid [CSF]) cavities (<15 mm in diameter) thought to be present in the region of a previous acute small deep brain infarct or hemorrhage in the territory of a perforating arteriole.2 Lacunes were distinguished from PvS using several criteria outlined in Table 1. A few cavities were excluded from the analysis if it was uncertain whether it was a PvS or a lacune. Table 1. The Differences Between Lacunes and Perivascular Spaces on Conventional Clinical Imaging

Identification of Lacunes and PvS Lacunes were identified in native subject space by a consultant neuroradiologist (A.D.M.; blinded to the patient and cognitive data), using a multimodality view with T1-weighted, T2-weighted, and fluid-attenuated inversion recovery images. Lacunes were defined as subcortical, fluid-filled (similar signal as cerebrospinal fluid [CSF]) cavities (<15 mm in diameter) thought to be present in the region of a previous acute small deep brain infarct or hemorrhage in the territory of a perforating arteriole.2 Lacunes were distinguished from PvS using several criteria outlined in Table 1. A few cavities were excluded from the analysis if it was uncertain whether it was a PvS or a lacune. Table 1. The Differences Between Lacunes and Perivascular Spaces on Conventional Clinical Imaging PvS Rating Scale Images were rated by a trained observer using a validated visual rating scale10 using the recommended user guide (http://www.sbirc.ed.ac.uk/documents/epvs-rating-scale-user-guide.pdf).10 T2-weighted magnetic resonance scans (with T1-weighted and fluid-attenuated inversion recovery imaging also available) were used for analysis. Basal ganglia and centrum semiovale PvS were rated from 0 (none), 1 (1–10), 2 (11–20), 3 (21–40), and 4 (>40), by assessing and scoring each hemisphere separately and then using the hemisphere with the higher score. Midbrain PvS were rated 0 (none visible) or 1 (visible). PvS at the level of the anterior commissure were excluded from the overall rating. The scores from each region were added together to provide a total PvS score (EPVS) for each scan. Interrater reliability metrics were checked by 2 raters using 30 randomly selected scans. Both raters were blinded to the others ratings.

PvS at the level of the anterior commissure were excluded from the overall rating. The scores from each region were added together to provide a total PvS score (EPVS) for each scan. Interrater reliability metrics were checked by 2 raters using 30 randomly selected scans. Both raters were blinded to the others ratings. PvS Volumes Lacunes were first manually identified by an experienced neuroradiologist (A.D.M.) using criteria outlined above and manually delineated using ITK-SNAP (http://www.itksnap.org). The signal intensities of PvS spaces tend to be identical to or lower than those of CSF. Therefore, to create PvS maps, we used the already created CSF maps. The manually identified lacunes, the ventricles, and CSF surrounding the large vessels and outside the brain were removed from the CSF maps to create PvS maps. Each PvS map was manually inspected to ensure that only PvS were included. PvS volumes were calculated in individual subject space by summing these binarized corrected maps. Volumes were then normalized with respect to total brain volume. Inter- and intra-rater reliability metrics were checked by 2 raters using 20 randomly selected scans across all time points. Both raters were blinded with respect to subject and time point of each scan and results of cognitive testing.

zed corrected maps. Volumes were then normalized with respect to total brain volume. Inter- and intra-rater reliability metrics were checked by 2 raters using 20 randomly selected scans across all time points. Both raters were blinded with respect to subject and time point of each scan and results of cognitive testing. Neuropsychological Assessment Cognitive assessment was performed annually using well-established standardized tests to include measures sensitive to the pattern of cognitive impairment associated with SVD.12 Tasks were grouped into broad cognitive functions, and task performance was age scaled using published normative data, transformed into z scores, and aggregated to construct the cognitive indices of executive function and processing speed by averaging across the component test measures for each subject, with a Global Function index as an amalgam of all measures. Further details on the individual tasks used and the cognitive assessment are given in the online-only Data Supplement.

construct the cognitive indices of executive function and processing speed by averaging across the component test measures for each subject, with a Global Function index as an amalgam of all measures. Further details on the individual tasks used and the cognitive assessment are given in the online-only Data Supplement. Statistical Analysis Statistical analysis was performed in R version 3.1.2 (www.r-project.org).15 Because of the hierarchical nature of the data, we used linear mixed-effects modeling to estimate change over the follow-up period in our cognitive measures using the lme4 package (version 1.1–7) in R.16 Specifically, we used a random intercept and random slope model which permits the estimation of an average slope over 3 years for cognitive measures across the whole cohort while allowing for interindividual variability.17 The time of scan (in years from baseline) served as a within-subject variable. The neuropsychological scores (executive function, processing speed, and global function) were set as dependent variables separately. In all models, we controlled for age, sex, WMH volume, and number of cerebral microbleeds (CMBs). For each dependent variable, we first analyzed the main effects of baseline lacunes and PvS on baseline cognitive performance. In the same models, baseline MRI predictor×time interactions were explored to reveal the influence of baseline MRI markers on the rate of cognitive change. Our approach accommodates for patient dropout during follow-up with the assumption that unobserved measurements are missing at random.

seline cognitive performance. In the same models, baseline MRI predictor×time interactions were explored to reveal the influence of baseline MRI markers on the rate of cognitive change. Our approach accommodates for patient dropout during follow-up with the assumption that unobserved measurements are missing at random. Linear mixed-effect modeling (using a random intercept and random slope model) was also used to estimate change in PvS volume over a 3-year time period. To investigate the effect of change in PvS volume on cognition, PVS volumes were modeled as continuous time-varying variables, that is, decomposed into static (within-subject mean) and dynamic (residual between time-varying measurement and within-subject mean) components. This was done to assess the independent contribution of change in PVS volume on change in cognition. MRI predictor×time interactions were explored to reveal the influence of PvS volumes on the rate of cognitive change. Parameter estimates are summarized by their means and uncertainty, as expressed by the 95% confidence interval (95% CI). For PvS rating scales, we calculated interobserver weighted κ for agreement using the psych package in R.18 Estimates were considered statistically significant when CIs excluded zero.

Linear mixed-effect modeling (using a random intercept and random slope model) was also used to estimate change in PvS volume over a 3-year time period. To investigate the effect of change in PvS volume on cognition, PVS volumes were modeled as continuous time-varying variables, that is, decomposed into static (within-subject mean) and dynamic (residual between time-varying measurement and within-subject mean) components. This was done to assess the independent contribution of change in PVS volume on change in cognition. MRI predictor×time interactions were explored to reveal the influence of PvS volumes on the rate of cognitive change. Parameter estimates are summarized by their means and uncertainty, as expressed by the 95% confidence interval (95% CI). For PvS rating scales, we calculated interobserver weighted κ for agreement using the psych package in R.18 Estimates were considered statistically significant when CIs excluded zero. Results Demographics at baseline are given in Table 2. Patients who left the study (for any reason) had a significantly higher mean WMH load (P<0.004), a higher lacune load (P<0.013), a higher mean Rankin disability score (P<0.026), and a lower mean Mini-Mental Test Examination (P<0.004) score at baseline when compared with patients who attended all time points. There were, however, no significant differences in baseline brain volume or other demographic characteristics in patients that left the study Table 2. Patient Demographics at Baseline

Results Demographics at baseline are given in Table 2. Patients who left the study (for any reason) had a significantly higher mean WMH load (P<0.004), a higher lacune load (P<0.013), a higher mean Rankin disability score (P<0.026), and a lower mean Mini-Mental Test Examination (P<0.004) score at baseline when compared with patients who attended all time points. There were, however, no significant differences in baseline brain volume or other demographic characteristics in patients that left the study Table 2. Patient Demographics at Baseline Perivascular Spaces Of the 120 subjects, 1 had a total PvS score of 1, 7 had a score of 2, 34 had a score of 3, 23 had a score of 4, 39 had a score of 5, 12 had a score of 6, 3 had a score of 7, and 1 had a score of 8. Total PvS score at baseline was normally distributed with a mean (SD) of 4.2 (1.3). Baseline total PvS score was correlated with age (Spearman ρ=0.183; P=0.045), baseline lacunes (Spearman ρ=0.372; P<0.001), baseline CMBs (Spearman ρ=0.341; P<0.001), and baseline WMH volume (Spearman ρ=0.273; P=0.003). It was not associated with brain volume. Mean total PvS volume was 160.6 mm3 (SD). Baseline total PvS volume was correlated with baseline lacunes (Pearson r=0.365; P<0.001) and baseline CMBs (Pearson r=0.189, P=0.038) but not with age, WMH volume, or brain volume.

Perivascular Spaces Of the 120 subjects, 1 had a total PvS score of 1, 7 had a score of 2, 34 had a score of 3, 23 had a score of 4, 39 had a score of 5, 12 had a score of 6, 3 had a score of 7, and 1 had a score of 8. Total PvS score at baseline was normally distributed with a mean (SD) of 4.2 (1.3). Baseline total PvS score was correlated with age (Spearman ρ=0.183; P=0.045), baseline lacunes (Spearman ρ=0.372; P<0.001), baseline CMBs (Spearman ρ=0.341; P<0.001), and baseline WMH volume (Spearman ρ=0.273; P=0.003). It was not associated with brain volume. Mean total PvS volume was 160.6 mm3 (SD). Baseline total PvS volume was correlated with baseline lacunes (Pearson r=0.365; P<0.001) and baseline CMBs (Pearson r=0.189, P=0.038) but not with age, WMH volume, or brain volume. There was a good correlation between total PvS scores and PvS volume (Spearman ρ=0.582; P<0.001). The estimated κ statistic measurements (confidence boundaries) for PvS rating scores were 0.40 (0.18–0.63) for centrum semiovale PvS and 0.33 (0.06–0.60) for basal ganglia PvS. The interrater reliability metrics for PvS volumes were SEM=2 mm3, mean variability=4.32% (SD=4.19%), and Intraclass correlation coefficient=0.99.

estimated κ statistic measurements (confidence boundaries) for PvS rating scores were 0.40 (0.18–0.63) for centrum semiovale PvS and 0.33 (0.06–0.60) for basal ganglia PvS. The interrater reliability metrics for PvS volumes were SEM=2 mm3, mean variability=4.32% (SD=4.19%), and Intraclass correlation coefficient=0.99. No significant change in total PVS volume over the 3-year observational period (for which MRI data were available) was demonstrated. The estimate of average (95% CI) annual change was −1.93 mm3 (−8.82 to 4.94). There was no evidence that change in PvS volume predicted a change in cognitive indices over the same follow-up period. The estimated effects (CI) of change in PvS volume on the slope of executive function, processing speed, and global function are 1.4×10−3 (−6.0×10−5 to 2.9×10−3), −2.3×10−4 (−1.4×10−3 to 9.8×10−4), and 6.0×10−4 (1.7×10−4 to 1.4×10−3), respectively. Lacunes At baseline, cavitated lacunes were present in 99 (83%) of 120 subjects. Although all patients had clinical lacunar stroke syndrome (with corresponding MRI lacunar infarction), not all lesions detected on acute diffusion-weighted imaging subsequently cavitate on T1-weighted images.19 The mean (SD) number of lacunes was 4.18 (5.44). The mean (SD) total lacune volume as a percentage of total brain volume was 0.0754 (0.0979). The distributions of lacune count and volume were skewed and were therefore log10 transformed for statistical analysis. The number and distribution of lacunes at baseline has been described elsewhere.9

lacunes was 4.18 (5.44). The mean (SD) total lacune volume as a percentage of total brain volume was 0.0754 (0.0979). The distributions of lacune count and volume were skewed and were therefore log10 transformed for statistical analysis. The number and distribution of lacunes at baseline has been described elsewhere.9 Over the 3-year imaging observational period, 74 new lacunes were observed in 27 patients, of which 66 were supratentorial and 8 were in the cerebellum or brain stem. A single new lacune was found in 10 subjects, 2 in 9 subjects and ≥3 (maximum 9) in 8 subjects. Change in Cognitive Measures There was strong evidence of a decline in executive function, processing speed, and global function (z scores) over the course of the observational period. The average (95% CI) annual change for Executive Function was −4.2×10−2 (−4.2×10−2 to −0.4×10−2), for Processing Speed, −5.1×10−2 (−8.0×10−2 to −2.3×10−2), and for Global Function, −2.7×10−2 (−4.6×10−2 to −0.9×10−2). Associations Between PvS and Cognition Total PvS score at baseline was not associated with cognition at baseline or with longitudinal change in cognition over a 5-year follow-up period. PvS in the basal ganglia, centrum semiovale, or midbrain were also not associated with cognition. PvS volume at baseline was not associated with cognition (Table I in the online-only Data Supplement). To help visualize the effects of baseline PvS on cognition, Figures 2 and 3 show the estimated marginal effect of baseline PvS score and PvS volume on cognition over a 5-year follow-up period.

sociated with cognition. PvS volume at baseline was not associated with cognition (Table I in the online-only Data Supplement). To help visualize the effects of baseline PvS on cognition, Figures 2 and 3 show the estimated marginal effect of baseline PvS score and PvS volume on cognition over a 5-year follow-up period. Figure 2. The estimated marginal effect of baseline lacune number and perivascular spaces score (EPVS) on Executive and global function over a 5-year follow-up period. We chose the 10th, 50th, and 90th percentiles of baseline lacunes (0, 2, 10) and EPVS (3, 4, 6) to display effects. Values for all other covariates in the model were set to their sample average. Figure 3. The estimated marginal effect of baseline lacune volume and perivascular spaces (PvS) volume on Executive and global function over a 5-year follow-up period. We chose the 10th, 50th, and 90th percentiles of baseline lacune volume (0, 0.47, 1.07) and PVS volume (20, 82, 406) to display effects. Values for all other covariates in the model were set to their sample average. Associations Between Lacunes and Cognition Lacune number and volume at baseline were associated with all cognitive indices at baseline and were strongest for executive function, processing speed, and global function (Table I in the online-only Data Supplement). These associations survived inclusion of baseline WMH volume, brain volume, and CMBs in the model.

on Lacune number and volume at baseline were associated with all cognitive indices at baseline and were strongest for executive function, processing speed, and global function (Table I in the online-only Data Supplement). These associations survived inclusion of baseline WMH volume, brain volume, and CMBs in the model. The number of lacunes at baseline explained some of the variability of the slope of executive function with an estimated effect (CI) of −1.3×10−1 (−2.0×10−1 to −0.5×10−1) and global function with an estimated effect (CI) of −6.4×10−2 (−1.1×10−1 to −1.8×10−1) over time (Table I in the online-only Data Supplement). The volume of lacunes at baseline explained some of the variability in the slope of executive function with an estimated effect (CI) of −2.0×10−1 (−3.5×10−1 to −4.5×10−2; Table I in the online-only Data Supplement). The estimated effect of lacunes on the slope of executive function and global function remained significant after including WMH, CMBs, and brain volume in the model. Baseline, CMBs, WMH, and brain volume did not have a significant effect on the change of cognition over time. The model was not significantly affected by collinearity (variance inflation factors were all under 2). To help visualize the effects of baseline lacunes on cognition, Figures 2 and 3 show the estimated marginal effect of baseline lacune number and lacune volume volume on cognition over a 5-year follow-up period.

time. The model was not significantly affected by collinearity (variance inflation factors were all under 2). To help visualize the effects of baseline lacunes on cognition, Figures 2 and 3 show the estimated marginal effect of baseline lacune number and lacune volume volume on cognition over a 5-year follow-up period. Discussion We compared the impact of baseline PvS and lacunes in a population of patients with symptomatic SVD. We found that baseline lacunes had a significant effect on cognition, both at baseline and longitudinally, whereas PvS had no impact on cognition in the same cohort. PvS, however, do correlate with other MRI markers of SVD particularly lacunes and CMBs. This suggests that while PvS may be a marker of SVD severity, they are not associated with cognitive impairment.9

t effect on cognition, both at baseline and longitudinally, whereas PvS had no impact on cognition in the same cohort. PvS, however, do correlate with other MRI markers of SVD particularly lacunes and CMBs. This suggests that while PvS may be a marker of SVD severity, they are not associated with cognitive impairment.9 A few studies,7,8 however, have showed a positive association between PvS and cognitive impairment. Apart from confounding factors, a reason for this inconsistency may be because of the difficulty in differentiating between lacunes and PvS which may result in lacunes being included as PvS. In addition, the rating scales used in previous studies preferentially use T2-weighted images3 for analysis, which can often overestimate the burden of PvS in patients with severe white matter disease. Previous studies have suggested that the presence of lacunes and WMH was the main reason for discrepancy between observers.10 All our patients have moderate-to-severe SVD which may explain the only moderate reproducibility demonstrated for the visual rating scales. For this reason, in our study, we carefully differentiated between lacunes and PVS (Table 1) and used 2 different methods of quantifying PvS (a visual rating score and PvS volumes). Although the only moderate reproducibility may have reduced the ability to detect associations on visual rating, it would not affect the semiautomated volumetric measures.

carefully differentiated between lacunes and PVS (Table 1) and used 2 different methods of quantifying PvS (a visual rating score and PvS volumes). Although the only moderate reproducibility may have reduced the ability to detect associations on visual rating, it would not affect the semiautomated volumetric measures. Incident lacunes have been proposed as a disease biomarker in cerebral SVD.20 This study highlights the importance of carefully differentiating between lacunes and PvS, particularly in studies investigating vascular cognitive impairment. Failing to differentiate these MRI features will reduce the sensitivity to detect significant effects. Our results did not demonstrate a significant change in PVS volume over a 3-year observational period. This may be because PvS have a slow growth rate and a longer period of imaging follow-up may be necessary to determine change in PvS volumes. Detailed 3-dimensional tracing of PvS is now possible using high-field strength MRI21,22 which will no doubt enable us to better study PvS thereby improving our understanding of their pathophysiological significance in SVD and other neurodegenerative diseases.

low-up may be necessary to determine change in PvS volumes. Detailed 3-dimensional tracing of PvS is now possible using high-field strength MRI21,22 which will no doubt enable us to better study PvS thereby improving our understanding of their pathophysiological significance in SVD and other neurodegenerative diseases. A limitation of previous studies of PvS is that rating scales may not be sensitive enough to evaluate the true burden of PvS. For this reason, we also used a volumetric measure of PvS which also allowed us to assess for longitudinal change. We however only used T1-weighted images for our analysis of PvS volumes as these images had the highest resolution in our data. In future studies, the data might be improved by acquiring other sequences including T2-weighted imaging with isotropic voxel dimensions, at higher resolutions.

assess for longitudinal change. We however only used T1-weighted images for our analysis of PvS volumes as these images had the highest resolution in our data. In future studies, the data might be improved by acquiring other sequences including T2-weighted imaging with isotropic voxel dimensions, at higher resolutions. A limitation of this study was that we had a relatively high dropout rate, although this is consistent with previous longitudinal studies in aging.23 Our analysis, using linear mixed-effect models provides inferences under the missing at random assumption. Although the majority of longitudinal studies in neuroimaging including clinical trials make this assumption,24 the possibility of data missing not at random is difficult to rule out. For example, patients who did not complete follow-up tended to be older and more disabled. This may have led to an underestimation of the rate of change in MRI markers and cognition. The handling of missing data continues to be a subject of discussion and alternative methodologies, for example, multiple imputation models may provide a more flexible approach. Future studies with larger sample sizes may be able to detect a small effect that may have been missed on this study.

I markers and cognition. The handling of missing data continues to be a subject of discussion and alternative methodologies, for example, multiple imputation models may provide a more flexible approach. Future studies with larger sample sizes may be able to detect a small effect that may have been missed on this study. We also assume linearity of change over time. In our data with a relatively short follow-up period, cognitive change is more parsimoniously described by a linear fit rather than a quadratic fit (judged by the small sample corrected Akaike Information Criterion).25 However, it is possible that with fewer dropouts or a longer period of follow-up period, or a more homogeneous disease stage among participants, nonlinearities may be more apparent and a quadratic or cubic fit may prove more appropriate. It should be noted that the processing speed index is made of aggregate scores from the following tasks: Wechsler Adult Intelligence Scale-III (Wechsler,26 1997) Digit symbol substitution, Speed of Information Processing Task,25 and Grooved Pegboard Task,27 a combination which has good internal reliability.12 All of these tests have motor responses, in particular the Grooved Pegboard task which has a greater motor component than the other tests. A processing speed deficit may therefore include a component of motor slowing in SVD.

ssing Task,25 and Grooved Pegboard Task,27 a combination which has good internal reliability.12 All of these tests have motor responses, in particular the Grooved Pegboard task which has a greater motor component than the other tests. A processing speed deficit may therefore include a component of motor slowing in SVD. We tried to exclude patients with cerebral amyloid angiopathy by excluding patients with cortical hemorrhages or in whom the pattern of CMBs was in a lobar distribution, but it is possible that some patients did have coexistent cerebral amyloid angiopathy pathology as this becomes increasingly frequent with increasing age. To study a homogenous group of patients, we recruited only patients with symptomatic lacunar infarction confirmed on MRI and confluent leukoaraiosis. We recognize that this may limit the generalizability of the findings. However, the findings remain relevant to a large number of stroke patients—about 20% to 25% of all ischemic stroke is lacunar because of SVD and of these cases about half fall into the category of lacunar stroke and WMH. Conclusions In conclusion, PvS, although a feature of SVD, are not associated with cognitive decline over a 5-year follow-up period. In contrast, lacunes are an important predictor of future cognitive decline. This study underlines the importance of carefully differentiating between lacunes and PvS in studies investigating vascular cognitive impairment.

hough a feature of SVD, are not associated with cognitive decline over a 5-year follow-up period. In contrast, lacunes are an important predictor of future cognitive decline. This study underlines the importance of carefully differentiating between lacunes and PvS in studies investigating vascular cognitive impairment. Sources of Funding The SCANS (St Georges Cognition and Neuroimaging in Stroke) research study was supported by a Wellcome Trust grant (081589). Recruitment was supported by the English National Institute of Health Research (NIHR) Clinical Stroke Research Network. H.S. Mar research is supported by an NIHR Senior Investigator award and the Cambridge University Hospitals NIHR Comprehensive Biomedical Research Centre. Disclosures None. Supplementary Material * A.D. MacKinnon and H.S. Markus contributed equally. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.017526/-/DC1.

Evidence from randomized controlled trials and meta-analysis shows improved functional outcomes for acute ischemic stroke patients receiving intravenous tissue thrombolysis with the recombinant tissue-type plasminogen activator alteplase.1–3 However, it is still unclear whether intravenous thrombolysis has any effect on mortality, particularly in the long-term. This has led to some concerns of whether the early risks associated with thrombolysis (eg, intracranial hemorrhage) translate into better prognosis over time.4 Furthermore, most of the studies that have examined differences in outcomes between groups have had a limited follow-up time (5–17). Currently, information available about effects on survival after intravenous thrombolysis come from 2 randomized clinical trials and 3 observational studies.5–9 The National Institute of Neurological Disorders and Stroke Recombinant Tissue Plasminogen Activator Stroke Study assessed mortality at 12 months, without finding any significant difference between the 2 study arms.5 More recently, a study using participants from the IST-3 (Third International Stroke Trial) found a small reduction in 3-year mortality in the treatment arm which was nonsignificant for all study subjects but only for those who survived the first week.6 However, the IST-3 trial randomized patients who did not meet current eligibility criteria for intravenous thrombolysis with alteplase in standard practice (ie, patients after 4.5 hours from stroke onset). Three observational studies have examined long-term outcomes of intravenous alteplase; however, 27,8 are limited by the lack of a comparison group. The third study,9 and the only other propensity score-matched study on intravenous alteplase found a 34% decrease in mortality for treated stroke patients in Denmark. However, this study had a limited median follow-up of 1.4 years and did not examine differences in activities of daily living between groups.

ison group. The third study,9 and the only other propensity score-matched study on intravenous alteplase found a 34% decrease in mortality for treated stroke patients in Denmark. However, this study had a limited median follow-up of 1.4 years and did not examine differences in activities of daily living between groups. In this study, we use a propensity score-matched cohort study design to determine whether thrombolysis with intravenous alteplase, as given in standard daily clinical practice in the United Kingdom,10 improves long-term survival up to 10 years after an acute ischemic stroke. Propensity score methods are tools for the analysis of observational studies that allow reducing the effect of the confounding that can occur because differences in the distribution of baseline characteristics and allow to replicate the measures of effect commonly reported in randomized clinical trials.11 As secondary outcomes, we examine whether the benefits in functional status, as assessed by the Barthel Index (BI) and Frenchay Activities Index (FAI), persist at 5 years after a stroke, as well as if stroke recurrence is affected by intravenous alteplase. Methods The data that support the findings of this study are available from the corresponding author on reasonable request.

In this study, we use a propensity score-matched cohort study design to determine whether thrombolysis with intravenous alteplase, as given in standard daily clinical practice in the United Kingdom,10 improves long-term survival up to 10 years after an acute ischemic stroke. Propensity score methods are tools for the analysis of observational studies that allow reducing the effect of the confounding that can occur because differences in the distribution of baseline characteristics and allow to replicate the measures of effect commonly reported in randomized clinical trials.11 As secondary outcomes, we examine whether the benefits in functional status, as assessed by the Barthel Index (BI) and Frenchay Activities Index (FAI), persist at 5 years after a stroke, as well as if stroke recurrence is affected by intravenous alteplase. Methods The data that support the findings of this study are available from the corresponding author on reasonable request. Study Design The South London Stroke Register (SLSR) is an ongoing, prospective, population-based, stroke register. The SLSR started in January 1995 and documents all first-ever confirmed strokes (according to the World Health Organization Criteria12) in patients of all ages for an inner area of South London that includes 22 electoral wards in the Boroughs of Lambeth and Southwark.13 The total source population of the SLSR area is 357 308 inhabitants, as estimated in the 2011 census and comprises a distinctly multiethnic population with a significant proportion of black Caribbean and African residents.14

ea of South London that includes 22 electoral wards in the Boroughs of Lambeth and Southwark.13 The total source population of the SLSR area is 357 308 inhabitants, as estimated in the 2011 census and comprises a distinctly multiethnic population with a significant proportion of black Caribbean and African residents.14 Case Ascertainment All patients with a suspected diagnosis of first-ever stroke documented from hospital- and community-based sources were investigated for study eligibility.12 Completeness of case ascertainment has been estimated at 88% by a multinomial logit capture-recapture model using the methods described elsewhere.15 A more thorough discussion of the methods used to maximize completeness of case ascertainment is available elsewhere.12,16 Data Collection Specially trained study nurses and field workers collected all data prospectively. Patients were examined within 48 hours of referral to SLSR when possible.12 A study stroke physician verified the diagnosis of stroke and classified the cases according to the modified TOAST (Trial of ORG 10172 in Acute Stroke Treatment)17 and the Oxfordshire Community Stroke Project subtype.18 The Oxfordshire Community Stroke Project subtype, which was used to calculate the propensity score, consists of 4 defined subgroups total anterior circulation infarcts, partial anterior circulation infarcts, lacunar infarcts, and posterior circulation infarcts.

17 and the Oxfordshire Community Stroke Project subtype.18 The Oxfordshire Community Stroke Project subtype, which was used to calculate the propensity score, consists of 4 defined subgroups total anterior circulation infarcts, partial anterior circulation infarcts, lacunar infarcts, and posterior circulation infarcts. Age at stroke was calculated as the difference between the date of birth and date of symptoms onset; ethnicity was self-reported by the patient and then collated into 1 of the 3 main categories according to the UK 2001 census (White, Black, and Other).13 Vascular risk factors before stroke (self-reported and from medical notes) were collected, including smoking, hypertension, diabetes mellitus, atrial fibrillation, ischemic heart disease (angina pectoris or myocardial infarction), peripheral vascular disease, and history of transient ischemic attack. Information on thrombolysis was collected from medical charts from 2005 onwards. The window for thrombolysis with intravenous alteplase was up to 3 hours after symptoms’ onset at the start of the study period and then extended up to 4.5 hours from 2009 onwards.

ascular disease, and history of transient ischemic attack. Information on thrombolysis was collected from medical charts from 2005 onwards. The window for thrombolysis with intravenous alteplase was up to 3 hours after symptoms’ onset at the start of the study period and then extended up to 4.5 hours from 2009 onwards. The BI19 was used to assess functional status previous to stroke as well as at each follow-up visit. A cutoff of BI≥90 was used to reflect functional independence.20 The National Institutes of Health Stroke Scale (NIHSS) score was used to assess stroke severity21 during the acute phase. The degree of neurological deficit was classified according to the total NIHSS score as follows minor (NIHSS, 1–4), moderate (5–15), moderate-severe (16–20), and severe (≥21). The FAI is a frequently used measurement of Extended Activities of Daily Living in stroke and was assessed by trained fieldworker during follow-up visits.22 Follow-up visits were performed at 3 months, 1 year, and annually thereafter. Survival was ascertained by follow-up, which included contacting next of kin if the participant was unreachable, and by checking with the office of National Statistics. Because the registration of deaths in the United Kingdom is complete, all patients with no death record at the day of censoring were assumed to be alive.

rvival was ascertained by follow-up, which included contacting next of kin if the participant was unreachable, and by checking with the office of National Statistics. Because the registration of deaths in the United Kingdom is complete, all patients with no death record at the day of censoring were assumed to be alive. Study Population The study included patients recruited into the SLSR from 2005 to 2015, inclusive. Survival time was censored at the December 31, 2015. All patients receiving intravenous thrombolysis with complete data for matching variables (see below) were included in the analysis. None of the participants in this study were treated with intra-arterial thrombolysis or mechanical clot retrieval. Ethical Approval The study was approved by the ethics committees of Guy’s and St Thomas’ Hospital Trust, King’s College Hospital, Queen’s Square, and Westminster Hospital. Informed consent from patients or assent from next of kin was obtained for all participants for their inclusion into the South London Stroke Register.

Ethical Approval The study was approved by the ethics committees of Guy’s and St Thomas’ Hospital Trust, King’s College Hospital, Queen’s Square, and Westminster Hospital. Informed consent from patients or assent from next of kin was obtained for all participants for their inclusion into the South London Stroke Register. Propensity Score Matching Propensity scores were calculated for each patient based on a multivariable logistic regression model. This model included demographic variables (age, sex, and ethnicity), prestroke number of vascular risk factors, functional status previous to stroke (BI), stroke severity (NIHSS), stroke Oxfordshire Community Stroke Project subtype,23 and year of stroke. Figure I in the online-only Data Supplement shows the substantial overlap in propensity score distributions between both groups; this suggests a large area of common support for the eligible participants. We matched treated participants with controls in a 1:2 ratio using a greedy nearest neighbor method.24 Figure 1 presents the flowchart of participant selection and propensity score-matched set construction. The overall quality of the matched sample was assessed by comparing the standardized difference of means and the ratio of the variances between the propensity scores of both groups as well as by graphically inspecting the propensity scores between groups. Furthermore, we evaluated the balance between individual covariates between groups in the matched sample. Figure 1. Data attrition flowchart.

Propensity Score Matching Propensity scores were calculated for each patient based on a multivariable logistic regression model. This model included demographic variables (age, sex, and ethnicity), prestroke number of vascular risk factors, functional status previous to stroke (BI), stroke severity (NIHSS), stroke Oxfordshire Community Stroke Project subtype,23 and year of stroke. Figure I in the online-only Data Supplement shows the substantial overlap in propensity score distributions between both groups; this suggests a large area of common support for the eligible participants. We matched treated participants with controls in a 1:2 ratio using a greedy nearest neighbor method.24 Figure 1 presents the flowchart of participant selection and propensity score-matched set construction. The overall quality of the matched sample was assessed by comparing the standardized difference of means and the ratio of the variances between the propensity scores of both groups as well as by graphically inspecting the propensity scores between groups. Furthermore, we evaluated the balance between individual covariates between groups in the matched sample. Figure 1. Data attrition flowchart. Statistical Analysis Descriptive data are expressed in percentages, mean±SD or median and interquartile range (IQR) as appropriate. The primary outcome of this study was survival up to 10 years after the date of first-ever acute ischemic stroke; we report Kaplan–Meier survival estimates and the difference between survival curves tested using the log-rank test stratified to matched sets. We obtained the adjusted hazard ratio (HR) from a Cox regression model of proportional hazards with robust variance estimator. The Cox model was developed by iteratively adding clinical relevant variables to a model including only treatment arm (ie, treated or control) regressed by the propensity score, and used a log-likelihood test to evaluate whether the addition of the new predictor improved the fit of the previous model. To examine whether onset-to-arrival time modified the effect of alteplase, we then tested if a multiplicative interaction term between treatment with alteplase and arrival within 3 hours further improved the fit our model. Because we expected, based on the reviewed literature,6 that the proportionality of the hazards assumption would be violated, we further assessed survival time after stroke by comparing the restricted mean survival time (RMST) between groups. We then adjusted the RMST for the same covariates used in the Cox model with an analysis of covariance.25 A similar method was used to examine the difference in stroke recurrence between groups. Independence at 5-year BI (≥90) and FAI scores at 5-year follow-up were compared between groups and adjusted for age, sex, ethnicity, prestroke BI, acute phase NIHSS, and stroke subtype by performing multiple regression analysis. We conducted a sensitivity analysis with multiple imputation to examine how robust our results were to missing data. All analyses were performed using R26 version 3.2.2 (2015) on R-Studio27 version 1.0.136.

, sex, ethnicity, prestroke BI, acute phase NIHSS, and stroke subtype by performing multiple regression analysis. We conducted a sensitivity analysis with multiple imputation to examine how robust our results were to missing data. All analyses were performed using R26 version 3.2.2 (2015) on R-Studio27 version 1.0.136. Results A total of 2052 patients with their first-ever ischemic stroke were recruited between the January 1, 2005 and the December 31, 2015; 285 (13.9%) of these patients received intravenous thrombolysis with alteplase. From the total recruited, 334 (16.3%) had missing data for at least 1 of the variables used to calculate the propensity score and thus had to be excluded. Of the 1718 remaining subjects, we paired 246 treated patients with 492 controls (Figure 1). None of these patients received intra-arterial thrombolysis or underwent thrombectomy. Measures of balance diagnosis28 indicated that the sample was adequately matched, with a standardized difference of the means of propensity scores between groups of 0.14 (good balance<0.25) and a ratio of variances of propensity scores of 1.27 (good balance between 0.5–2). A comparison of the baseline characteristics further supports the good balance of our matched sample (Table 1; Table I in the online-only Data Supplement). Table 1. Baseline Characteristics The median follow-up time was 5.45 years (IQR=4.56; range, 0–10 years), and a total of 344 (46.6%) patients died during the study period.

Results A total of 2052 patients with their first-ever ischemic stroke were recruited between the January 1, 2005 and the December 31, 2015; 285 (13.9%) of these patients received intravenous thrombolysis with alteplase. From the total recruited, 334 (16.3%) had missing data for at least 1 of the variables used to calculate the propensity score and thus had to be excluded. Of the 1718 remaining subjects, we paired 246 treated patients with 492 controls (Figure 1). None of these patients received intra-arterial thrombolysis or underwent thrombectomy. Measures of balance diagnosis28 indicated that the sample was adequately matched, with a standardized difference of the means of propensity scores between groups of 0.14 (good balance<0.25) and a ratio of variances of propensity scores of 1.27 (good balance between 0.5–2). A comparison of the baseline characteristics further supports the good balance of our matched sample (Table 1; Table I in the online-only Data Supplement). Table 1. Baseline Characteristics The median follow-up time was 5.45 years (IQR=4.56; range, 0–10 years), and a total of 344 (46.6%) patients died during the study period. Primary Outcome: Survival up to 10 Years The Kaplan–Meier estimate shows a higher survival for patients treated with intravenous alteplase than for those in the control group at 5 and 10 years (Figure 2; log-rank test stratified by sets <0.001 for both). The median survival time for the treated group was 5.72 and 4.98 years for the control group. The absolute risk reduction at 5 years was 8.33% (95% confidence interval [CI], 8.19–8.47; number needed to treat, 12) and 5.07% (95% CI, 4.92–5.22) at 10 years (number needed to treat, 20).

tratified by sets <0.001 for both). The median survival time for the treated group was 5.72 and 4.98 years for the control group. The absolute risk reduction at 5 years was 8.33% (95% confidence interval [CI], 8.19–8.47; number needed to treat, 12) and 5.07% (95% CI, 4.92–5.22) at 10 years (number needed to treat, 20). Figure 2. Survival curves for intravenous alteplase treated group (darker) and control group (lighter) groups. Median follow-up time 5.45 years. Median survival for treated group 5.72 years. Median survival for control group 4.98 years. Stratified log-rank test: P<0.001. The unadjusted HR shows a 19% (HR, 0.81; 95% CI, 0.70–0.92) and 28% (HR, 0.72; 95% CI, 0.57–0.91) decrease in mortality risk for the treated group at 5 and 10 years, respectively. After adjusting for age, prestroke BI, prestroke use of anticoagulants, NIHSS during the acute phase, and poststroke treatment with antiplatelets, thrombolysis with intravenous alteplase was associated with a 28% (HR, 0.72; 95% CI, 0.60–0.87) decrease in mortality at 5 years and 37% (HR, 0.63; 95% CI, 0.48–0.82) at 10 years (Table II in the online-only Data Supplement). After including a multiplicative interaction term between thrombolysis with intravenous alteplase and arrival to the hospital within 3 hours, mortality reduction for those treated earlier was 32% (HR, 0.67; 95% CI, 0.52–0.88) at 5 years and 42% (HR, 0.58; 95% CI, 0.40–0.82) at 10 years (Table 2; Figure II in the online-only Data Supplement). Table 2. Multivariable Cox Regression on Survival Including Interaction Term

The unadjusted HR shows a 19% (HR, 0.81; 95% CI, 0.70–0.92) and 28% (HR, 0.72; 95% CI, 0.57–0.91) decrease in mortality risk for the treated group at 5 and 10 years, respectively. After adjusting for age, prestroke BI, prestroke use of anticoagulants, NIHSS during the acute phase, and poststroke treatment with antiplatelets, thrombolysis with intravenous alteplase was associated with a 28% (HR, 0.72; 95% CI, 0.60–0.87) decrease in mortality at 5 years and 37% (HR, 0.63; 95% CI, 0.48–0.82) at 10 years (Table II in the online-only Data Supplement). After including a multiplicative interaction term between thrombolysis with intravenous alteplase and arrival to the hospital within 3 hours, mortality reduction for those treated earlier was 32% (HR, 0.67; 95% CI, 0.52–0.88) at 5 years and 42% (HR, 0.58; 95% CI, 0.40–0.82) at 10 years (Table 2; Figure II in the online-only Data Supplement). Table 2. Multivariable Cox Regression on Survival Including Interaction Term Visual and formal testing revealed hazards to be nonproportional (P<0.0001) for the whole duration of the follow-up. To account for this, we calculated the RMST for each group. Patients receiving intravenous alteplase had an RMST of 6.06 years, whereas the control group RMST was 5.18 years. The estimated difference in RMST between groups was 0.88 years (95% CI, 0.18–1.59; P=0.015) over a 10-year follow-up period. After adjustment for the same covariates used in the Cox proportional hazards model, the estimated difference between RMST was 1.04 years (95% CI, 0.17–1.91; P=0.02).

l group RMST was 5.18 years. The estimated difference in RMST between groups was 0.88 years (95% CI, 0.18–1.59; P=0.015) over a 10-year follow-up period. After adjustment for the same covariates used in the Cox proportional hazards model, the estimated difference between RMST was 1.04 years (95% CI, 0.17–1.91; P=0.02). Secondary Outcomes: Functional Status at 5 Years Thrombolysis with intravenous alteplase was associated with improved functional status. After adjusting for age, sex, ethnicity, prestroke BI, acute phase NIHSS, and stroke subtype, treatment was significantly associated with independence (BI≥90) at discharge (odds ratio, 2.01; 95% CI, 1.27–3.20) and at 5 years (odds ratio, 3.76; 95% CI, 1.22–13.34). Intravenous alteplase was also associated with increased odds of higher FAI score (proportional odds ratio, 2.37; 95% CI, 1.16–4.91) at 5 years. There was no difference in stroke recurrence between groups.

(BI≥90) at discharge (odds ratio, 2.01; 95% CI, 1.27–3.20) and at 5 years (odds ratio, 3.76; 95% CI, 1.22–13.34). Intravenous alteplase was also associated with increased odds of higher FAI score (proportional odds ratio, 2.37; 95% CI, 1.16–4.91) at 5 years. There was no difference in stroke recurrence between groups. Discussion Key Findings In this study, which to our knowledge has the longest median follow-up time in the published literature, we found evidence in a real-world setting of reduced mortality after thrombolysis with intravenous alteplase. Our findings show that on average, and over a 10-year period, a patient treated with thrombolysis lives around 1 year longer than a similar nonthrombolyzed patient after adjustment for age, sex, prestroke BI, prestroke treatment with anticoagulants, acute phase NIHSS score, and poststroke treatment with antiplatelets. Our data shows that the number needed to treat to prevent 1 death in 5 years is equal to 12 patients and 20 to prevent 1 death at 10 years. These results complement and expand on previous data reported by the Danish Stroke Register9 and the subanalysis of the IST-3 trial6 by demonstrating that the benefits of thrombolysis in survival are appreciable even after a period as long as 10 years poststroke and the improved functional outcomes are still perceivable at 5 years. Furthermore, our findings also suggest that the observed benefit in survival is seemingly driven by older patients and those with NIHSS≥16 (moderate-severe and severe strokes; Figure 3).

appreciable even after a period as long as 10 years poststroke and the improved functional outcomes are still perceivable at 5 years. Furthermore, our findings also suggest that the observed benefit in survival is seemingly driven by older patients and those with NIHSS≥16 (moderate-severe and severe strokes; Figure 3). Figure 3. Forest plot showing the point estimates incidence rate ratio (IRR) with their respective CI for mortality per 100 person-year stratified by age and National Institutes of Health Stroke Scale (NIHSS) group. Median (interquartile range) NIHSS<16 group was 7 (5); NIHSS≥16 group was 21 (6). CI indicates confidence interval.

orest plot showing the point estimates incidence rate ratio (IRR) with their respective CI for mortality per 100 person-year stratified by age and National Institutes of Health Stroke Scale (NIHSS) group. Median (interquartile range) NIHSS<16 group was 7 (5); NIHSS≥16 group was 21 (6). CI indicates confidence interval. Compared with the Danish nationwide register study,9 our study had a slightly older mean age (treated, controls; SLSR, 68.0, 69.4 versus Danish Register, 65.8, 66.5) a significantly longer median follow-up time (5.45 versus 1.4 years), and consequently a higher proportion of deaths (46.6% compared with 14.7%). Furthermore, our study included a higher proportion of moderate (48.4%, 56.7% versus 33.8%, 33.9%) and severe strokes (16.7%, 15.2% versus 7.7%, 7.9%), a comparable proportion of moderate-severe strokes (17.5%, 12.2% versus 14.9%, 15.0%) and significantly fewer minor strokes (17.5%, 15.9% versus 41.4%, 41.5%) according to the scale used by each study (NIHSS versus Scandinavian Stroke Scale).The median stroke severity was also slightly higher in our cohort (10, 9 versus 8, 8, converted from Scandinavian Stroke Scale to NIHSS using the formula found elsewhere29). Our population also had a higher prevalence of hypertension (64.2%, 65.7% versus 50.9%, 47.8%), and diabetes mellitus (20.3%, 21.1% versus 9.6%, 12.6%; Table I in the online-only Data Supplement). Additionally, our population was comprised of first-ever strokes only. Despite these differences our adjusted HR for the whole duration of the study, without time to arrival interaction term, are similar (0.63 [95% CI, 0.48–0.82] versus 0.66 [95% CI, 0.49–0.88]) suggesting the benefits from intravenous alteplase are generalizable across populations.

s comprised of first-ever strokes only. Despite these differences our adjusted HR for the whole duration of the study, without time to arrival interaction term, are similar (0.63 [95% CI, 0.48–0.82] versus 0.66 [95% CI, 0.49–0.88]) suggesting the benefits from intravenous alteplase are generalizable across populations. The precise mechanism or mechanisms by which thrombolysis improves survival are unknown, although there is an evidence thrombolysis decreases infarct size30 and reduces the risk of readmissions because of pneumonia,31 although the impact on other factors is still unclear. Nevertheless, previous studies have shown that good functional outcomes in the short term (ie, within 6 months) are associated with improved long-term survival, in part because of fewer complications and more independence.9,32,33 In our cohort, patients in the treatment group had overall better scores in BI and FAI even at 5-years poststroke after adjustment, further suggesting the association between functional status and survival. Furthermore, better BI scores have been strongly associated with quality of life34 which means that the improved survival seen with alteplase is also accompanied by improved quality of life.

I and FAI even at 5-years poststroke after adjustment, further suggesting the association between functional status and survival. Furthermore, better BI scores have been strongly associated with quality of life34 which means that the improved survival seen with alteplase is also accompanied by improved quality of life. Strengths and Weaknesses The main strengths of this study include a long follow-up time (up to 10 years, median 5.45 year), a per protocol, prospectively collected data set with a wide range of variables that allowed a good balance in baseline variables between groups, an ethnically diverse study population of a well-defined area with a near-complete recording of deaths, achieved by linking with the Office of National Statistics and follow-up by the register fieldworkers. Additionally, we provide the adjusted difference in RMST, a summary statistic which has been shown to better estimate time-to-event than the HR when the proportionality of the hazards assumption is not met.25 Furthermore, the difference in RMST can be straightforwardly interpreted in clinical settings by both the treating physician, the patients and their next of kin. The main limitation of this study lies in its design as an observational rather than experimental study. However, we have used propensity score matching to reduce potential bias and strengthen our reported effect estimates.35 Additionally, we matched every treated patient with a similar control and included the calculated propensity score into the multivariable analysis (double propensity score adjustment), thus reducing as much as possible confounding because of incomplete matching and residual confounding. Another limitation is the effect of missing data in the propensity score matching as well as the analysis. This limitation is common to all studies, particularly those with long follow-up times. Nevertheless, a sensitivity analysis demonstrated that our results were robust to the influence of missing data. Additional limitations include that the SLSR does not collect information about the time of thrombolysis and that the modified Rankin Scale score is not available for follow-ups before 2014. However, although it is not possible for us to calculate the onset-to-treatment time, we have used onset-to-arrival times as a proxy instead; this value is not only correlated with the time from onset to treatment, but also is available for nontreated patients, and thus, it could be fitted into the Cox model.

ore 2014. However, although it is not possible for us to calculate the onset-to-treatment time, we have used onset-to-arrival times as a proxy instead; this value is not only correlated with the time from onset to treatment, but also is available for nontreated patients, and thus, it could be fitted into the Cox model. Although we do not have enough data on the modified Rankin Scale score at 5 years, we have used the BI with a threshold of ≥90 to define independence; using this cutoff point has been shown to be comparable to a modified Rankin Scale score of ≤220.

ore 2014. However, although it is not possible for us to calculate the onset-to-treatment time, we have used onset-to-arrival times as a proxy instead; this value is not only correlated with the time from onset to treatment, but also is available for nontreated patients, and thus, it could be fitted into the Cox model. Although we do not have enough data on the modified Rankin Scale score at 5 years, we have used the BI with a threshold of ≥90 to define independence; using this cutoff point has been shown to be comparable to a modified Rankin Scale score of ≤220. Implications for Clinical Practice Despite the amount of evidence provided by clinical trials demonstrating that thrombolysis with alteplase improves functional outcomes in ischemic strokes at 1 year, the adoption of thrombolysis has been slow by many centers. Although the reasons for this are multifactorial, including concerns with regards to costs and required experience, one of the main arguments has been the uncertainty on whether the risks associated with thrombolysis indeed result in better outcomes in the long-term. In this study, we have shown that not only patients treated with intravenous alteplase have better BI and FAI scores at 5 years but that they also experience a lower mortality risk, with relatively low numbers needed to treat to prevent a death at 10 years. Furthermore, this study was done using data from a real-world setting from a diverse population, and thus our results are generalizable. These findings should provide much-needed evidence to reassure treating clinicians and patients about the long-term benefits of intravenous thrombolysis therapy with alteplase following currently accepted guidelines.

using data from a real-world setting from a diverse population, and thus our results are generalizable. These findings should provide much-needed evidence to reassure treating clinicians and patients about the long-term benefits of intravenous thrombolysis therapy with alteplase following currently accepted guidelines. Acknowledgments We thank patients, their families, and the fieldworkers who have collected data for the South London Stroke Register since 1995. Sources of Funding We would like to acknowledge the support and funding from the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care South London at King’s College Hospital NHS Foundation Trust and the Royal College of Physicians, as well as the support from the NIHR Biomedical Research Centre based at Guy’s and St Thomas’ NHS Foundation Trust and King’s College London. Disclosures None. Supplementary Material Presented in part at the European Stroke Organization Conference, Prague, Czech Republic, May 16, 2017. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.019889/-/DC1.

See related article, p 513 Lifelong secondary prevention with antiplatelet agents is recommended in patients who experienced a transient ischemic attack (TIA) or ischemic stroke.1 Bleeding is a clinically important and potentially life-threatening side effect of antiplatelet drugs.2 Risk of bleeding increases steadily with age, and the gastrointestinal (GI) tract is shown to be the most common source of bleeding.3–5 Individualized prediction of bleeding risk may help physicians to identify patients at highest risk and may guide treatment decisions regarding initiation of gastroprotective agents. Recently, the S2TOP-BLEED score was developed to predict risk of major bleeding in patients with a TIA or ischemic stroke on antiplatelet agents.6 The model was derived from individual patient data from 6 randomized clinical trials (Table I in the online-only Data Supplement),7–12 including over 43 000 patients with a TIA or ischemic stroke, and was subsequently validated in the PERFORM trial (Prevention of Cerebrovascular and Cardiovascular Events of Ischaemic Origin With Terutroban in patients With a History of Ischaemic Stroke or Transient Ischaemic Attack Study),13 including another 19 000 patients with a recent TIA or ischemic stroke.

ischemic stroke, and was subsequently validated in the PERFORM trial (Prevention of Cerebrovascular and Cardiovascular Events of Ischaemic Origin With Terutroban in patients With a History of Ischaemic Stroke or Transient Ischaemic Attack Study),13 including another 19 000 patients with a recent TIA or ischemic stroke. A potential drawback of using trial data for development of a risk score is that participants may represent a selective subset of the population of interest, as frail and elderly patients are often excluded from trials. As a consequence, absolute risks may be underestimated in a real-world setting and associations between predictors and outcome may differ.14,15 External validation of a risk score in observational data could, therefore, provide valuable insight into the accuracy of the predicted risks and the generalizability to a wider range of patients. We aimed to externally validate the S2TOP-BLEED score in a population-based cohort and to assess its performance according to site and severity of bleeding. Subsequently, we compared its performance to other risk scores for bleeding in patients with a TIA or ischemic stroke.

neralizability to a wider range of patients. We aimed to externally validate the S2TOP-BLEED score in a population-based cohort and to assess its performance according to site and severity of bleeding. Subsequently, we compared its performance to other risk scores for bleeding in patients with a TIA or ischemic stroke. Methods Study Population The OXVASC (Oxford Vascular Study) is an ongoing population-based study on the incidence and outcome of all acute vascular events in Oxfordshire, United Kingdom. Methods and definition of events have been described previously.16 Briefly, the study population comprises 92 728 individuals, registered with 100 general practitioners in 9 general practices in Oxfordshire. Multiple overlapping methods of hot and cold pursuit are used for ascertainment of all acute vascular events in the study population, which has been shown to be near complete.17 For the current analysis, we studied patients with a TIA or ischemic stroke between 2002 and 2012, who were on antiplatelet drugs after their event. These included both patients who were on premorbid antiplatelet drugs, as well as patients who started antiplatelet drugs after the index event. Patients who switched to oral anticoagulants during follow-up were censored at the time of starting (Table I in the online-only Data Supplement).

elet drugs after their event. These included both patients who were on premorbid antiplatelet drugs, as well as patients who started antiplatelet drugs after the index event. Patients who switched to oral anticoagulants during follow-up were censored at the time of starting (Table I in the online-only Data Supplement). Information on patient demographics and vascular risk factors was collected during the initial assessment. Patients were followed-up face to face by a study nurse or physician at 1 month, 6 months, and 1, 5, and 10 years after the index event. Recurrent ischemic events, bleeding events that required medical attention, and disability (modified Rankin Scale) were recorded at each follow-up. Bleeding events were also identified by daily searches of all hospital admissions, by review of administrative diagnostic codes from hospital and primary care records, and by searches of blood transfusion records. Only bleeds that required medical attention or were fatal prior to medical attention could be sought were included. Bleeds secondary to trauma, surgery, or hematological malignancy were excluded.

iew of administrative diagnostic codes from hospital and primary care records, and by searches of blood transfusion records. Only bleeds that required medical attention or were fatal prior to medical attention could be sought were included. Bleeds secondary to trauma, surgery, or hematological malignancy were excluded. Bleeds were classified according to site of hemorrhage as either intracranial (intracerebral, subarachnoid, and subdural), upper GI, lower GI, epistaxis, genitourinary, or other. The severity of bleeds was recorded according to the CURE criteria (Clopidogrel in Unstable Angina to Prevent Recurrent Events).18 Major bleeds were bleeds that were substantially disabling with persistent sequelae, intraocular bleeds leading to significant loss of vision, or bleeds requiring transfusion of ≥2 units of blood. Major bleeds were classified as life-threatening if the bleeding episode was fatal, symptomatic intracranial, led to a reduction in hemoglobin level of at least 5 g/dL (3.1 mmol/L), led to substantial hypotension requiring use of intravenous inotropic agents, necessitated a surgical intervention, or necessitated transfusion of ≥4 units of blood. Bleeding events that required medical attention but did not fulfill the criteria of major bleeding were recorded as significant nonmajor bleeds. OXVASC has been approved by the local ethics committee, and all participants gave written informed consent. Requests for anonymized data will be considered by Professor Rothwell (peter.rothwell@clneuro.ox.ac.uk).

attention but did not fulfill the criteria of major bleeding were recorded as significant nonmajor bleeds. OXVASC has been approved by the local ethics committee, and all participants gave written informed consent. Requests for anonymized data will be considered by Professor Rothwell (peter.rothwell@clneuro.ox.ac.uk). Statistical Analysis Data were missing on body mass index in 79 patients (4%) and on smoking in 3 patients (<1%). These patients were excluded from the analysis. Variables of the S2TOP-BLEED score (Table II in the online-only Data Supplement) were matched to variables in OXVASC. A proxy was used if no direct match was available. The National Institutes of Health Stroke Scale was used to assess severity of the index event and was used as a proxy for the modified Rankin Scale score, where a National Institutes of Health Stroke Scale score ≤3 was considered a minor stroke and a score >3 a severe stroke. All patients who received a short course of aspirin plus clopidogrel (for the first 30–90 days) and were treated with aspirin (plus dipyridamole) thereafter were analyzed as if they were on aspirin (plus dipyridamole), as our interest was in long term risk of bleeding.

red a minor stroke and a score >3 a severe stroke. All patients who received a short course of aspirin plus clopidogrel (for the first 30–90 days) and were treated with aspirin (plus dipyridamole) thereafter were analyzed as if they were on aspirin (plus dipyridamole), as our interest was in long term risk of bleeding. The original regression equation was applied to the validation data to calculate 3-year risk of major bleeding. We assessed discriminatory performance of the model with the C statistic and calibration with the calibration slope and plots. Calibration at 3 years was examined by dividing patients in quintiles according to their predicted risk. The mean predicted risk per quintile group was subsequently plotted against the observed risk per quintile group. Calibration over time was assessed across risk groups that were predefined as low risk (0–10 points on the S2TOP-BLEED score), medium risk (11–15 points), and high risk (>15 points).6 Model performance was also assessed separately by severity of bleeding (nonmajor, major and life-threatening, or fatal) and by site of bleeding (intracranial, upper GI, lower GI, epistaxis, genitourinary, or other). We performed a sensitivity analysis excluding patients with an established high risk of bleeding or reduced life expectancy (patients with renal failure, liver failure, cancer, or a prior peptic ulcer) who are generally not included in trials.

ding (intracranial, upper GI, lower GI, epistaxis, genitourinary, or other). We performed a sensitivity analysis excluding patients with an established high risk of bleeding or reduced life expectancy (patients with renal failure, liver failure, cancer, or a prior peptic ulcer) who are generally not included in trials. We compared performance of the S2TOP-BLEED score with performance of the REACH score for major bleeding,19 and the Intracranial-B2LEED3S score (low BMI, high blood pressure, lacune, elderly, Asian ethnicity, cardiovascular disease, cerebrovascular disease, dual antithrombotic treatment or anticoagulant, sex) for intracranial hemorrhage after TIA or ischemic stroke (Table III in the online-only Data Supplement),20 by means of the C statistic, integrated discrimination improvement, and net reclassification improvement.21,22 Another risk score for intracranial hemorrhage after TIA or stroke could not be validated as it required postacute blood glucose levels, which were not available in the validation cohort. To study the influence of the different age categories used in the different risk scores for major bleeding on the performance, we assessed the C statistic of the models containing age only and compared it to the C statistic of the remainder of the model.

ucose levels, which were not available in the validation cohort. To study the influence of the different age categories used in the different risk scores for major bleeding on the performance, we assessed the C statistic of the models containing age only and compared it to the C statistic of the remainder of the model. As risk factors for bleeding events are also known to be risk factors for recurrent ischemic events, we assessed the discriminatory ability of the S2TOP-BLEED score for recurrent ischemic events at 3 years (defined as recurrent ischemic stroke, myocardial infarction, or sudden cardiac death). Next, we assessed the cumulative incidence of bleeding events and recurrent ischemic events at 3 years and their ratio across risk groups of the S2TOP-BLEED score. Results are reported in accordance with the TRIPOD statement (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis).23 All analyses were performed with R version 3.3.2.

events and recurrent ischemic events at 3 years and their ratio across risk groups of the S2TOP-BLEED score. Results are reported in accordance with the TRIPOD statement (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis).23 All analyses were performed with R version 3.3.2. Results Between 2002 and 2012, 2072 patients with a TIA or ischemic stroke on antiplatelet drugs were included in OXVASC. Median follow-up was 3.5 years (interquartile range 1.5–6.3). Baseline characteristics of patients in the development and validation cohort are shown in Table 1. Patients in OXVASC were older than patients in the CAT trials (Cerebrovascular Antiplatelet Trialists; mean age 73 years [SD, 13.4] versus 66 years [SD 9.7]). Two hundred fifty-four bleeds occurred during follow-up, of which 117 (46%) were major bleeds. Upper GI bleeds were the most common type of bleeding (32%; Table IV in the online-only Data Supplement). Four hundred sixty-one patients (22%) were classified as having an established high risk of bleeding, and 39% of all major bleeds occurred within this group. Risk of major bleeding was higher in the validation cohort than in the development cohort (Figure 1). Table 1. Baseline Characteristics of Patients in Development (CAT) and Validation Cohort (OXVASC) Figure 1. Cumulative risk of major bleeding in development cohort (CAT) and validation cohort (OXVASC). CAT indicates Cerebrovascular Antiplatelet Trialists; and OXVASC, Oxford Vascular Study.

Results Between 2002 and 2012, 2072 patients with a TIA or ischemic stroke on antiplatelet drugs were included in OXVASC. Median follow-up was 3.5 years (interquartile range 1.5–6.3). Baseline characteristics of patients in the development and validation cohort are shown in Table 1. Patients in OXVASC were older than patients in the CAT trials (Cerebrovascular Antiplatelet Trialists; mean age 73 years [SD, 13.4] versus 66 years [SD 9.7]). Two hundred fifty-four bleeds occurred during follow-up, of which 117 (46%) were major bleeds. Upper GI bleeds were the most common type of bleeding (32%; Table IV in the online-only Data Supplement). Four hundred sixty-one patients (22%) were classified as having an established high risk of bleeding, and 39% of all major bleeds occurred within this group. Risk of major bleeding was higher in the validation cohort than in the development cohort (Figure 1). Table 1. Baseline Characteristics of Patients in Development (CAT) and Validation Cohort (OXVASC) Figure 1. Cumulative risk of major bleeding in development cohort (CAT) and validation cohort (OXVASC). CAT indicates Cerebrovascular Antiplatelet Trialists; and OXVASC, Oxford Vascular Study. The C statistic of the S2TOP-BLEED score for major bleeding was 0.69 (95% confidence interval [CI], 0.64–0.73) and calibration at 3 years was accurate (calibration slope 1.13, P=0.48; Figure 2A). Early risk of bleeding was underestimated by the model, but calibration across risk groups was accurate for long term risk of bleeding (Figure 2B). The S2TOP-BLEED score was much more predictive for fatal and major bleeding (C statistic 0.77 and 0.69) than for nonmajor bleeds (C statistic 0.50; Table 2). Discriminatory ability was higher for intracranial and upper GI bleeds than for lower GI bleeds, genitourinary bleeds, and epistaxis (Table 2). A sensitivity analysis excluding patients with an established high risk of bleeding or reduced life expectancy showed comparable discriminatory performance of the S2TOP-BLEED score 0.70 (0.64–0.77)

y was higher for intracranial and upper GI bleeds than for lower GI bleeds, genitourinary bleeds, and epistaxis (Table 2). A sensitivity analysis excluding patients with an established high risk of bleeding or reduced life expectancy showed comparable discriminatory performance of the S2TOP-BLEED score 0.70 (0.64–0.77) Table 2. C Statistic (95% CI) of S2TOP-BLEED Score in Validation Cohort Figure 2. Calibration plots for the S2TOP-BLEED score. Calibration plots: 3-year major bleeding-free survival (A) and calibration across risk groups (B). Correspondence between observed and predicted 3-year major bleeding-free survival across quintile groups (A). Observed risk (solid line) and predicted risk (dotted line) across predefined risk groups of the S2TOP-BLEED score (B). The REACH score showed a C statistic of 0.63 (95% CI, 0.58–0.69) for major bleeding at 2 years and systematically underestimated risk of bleeding (Figure IA in the online-only Data Supplement). The Intracranial-B2LEED3S score had a C statistic of 0.60 (95% CI, 0.51–0.70) for intracranial bleeding at 2 years and showed accurate calibration (Figure IB in the online-only Data Supplement). The S2TOP-BLEED score showed improved reclassification and integrated discrimination as compared with the REACH and Intrancranial-B2LEED3S scores (Table V in the online-only Data Supplement).

0.51–0.70) for intracranial bleeding at 2 years and showed accurate calibration (Figure IB in the online-only Data Supplement). The S2TOP-BLEED score showed improved reclassification and integrated discrimination as compared with the REACH and Intrancranial-B2LEED3S scores (Table V in the online-only Data Supplement). A model with 5 age categories only as defined in the S2TOP-BLEED score (45–54, 55–64, 65–74, 75–85, and 85+) showed a C statistic of 0.66 (0.62–0.71), and a model containing 4 age categories as defined in the REACH score (45–54, 55–64, 65–74, and 75+) had a C statistic of 0.64 (0.60–0.69; Table VI in the online-only Data Supplement). The predictive performance of the models without age was 0.57 (0.51–0.64) for S2TOP-BLEED and 0.52 (0.45–0.58) for REACH. Four hundred thirty-eight patients experienced a recurrent ischemic event during follow-up, and the overall observed 3-year risk was 19% (95% CI, 17%–21%). The C statistic of the S2TOP-BLEED score for predicting recurrent ischemic events was 0.58 (95% CI, 0.55–0.61). Three-year risk of recurrent ischemic events was 15% (95% CI, 12%–17%) in the low bleeding risk group and 23% (95% CI, 16%–30%) in both the medium- and high-risk group (Figure 3; P for trend =0.22). The ratio of ischemic events versus bleeds decreased from 7.5:1 in the low-risk group to 2.9:1 in the intermediate-risk group and 1.8:1 in the high-risk group (P for trend <0.001).

I, 12%–17%) in the low bleeding risk group and 23% (95% CI, 16%–30%) in both the medium- and high-risk group (Figure 3; P for trend =0.22). The ratio of ischemic events versus bleeds decreased from 7.5:1 in the low-risk group to 2.9:1 in the intermediate-risk group and 1.8:1 in the high-risk group (P for trend <0.001). Figure 3. Cumulative 3-year risk of recurrent ischemic events and major bleeding events across risk groups of the S2TOP-BLEED score. Observed 3-year risk of major bleeds and recurrent ischemic events across predefined risk groups of the S2TOP-BLEED score. Discussion We externally validated the S2TOP-BLEED score for major bleeding in patients with a TIA or ischemic stroke in a population-based cohort and found modest discriminatory performance and calibration. Compared with the REACH and Intracranial-B2LEED3S scores, the S2TOP-BLEED score showed best performance, both for prediction of intracranial and major bleeds. Although high bleeding risks were also associated with high risks of recurrent ischemic events, risk stratification may still be useful to identify a group of patients at particularly high risk of bleeding, in whom preventive measures are indicated.

est performance, both for prediction of intracranial and major bleeds. Although high bleeding risks were also associated with high risks of recurrent ischemic events, risk stratification may still be useful to identify a group of patients at particularly high risk of bleeding, in whom preventive measures are indicated. Discriminatory performance of the S2TOP-BLEED score slightly improved compared with the original development study (C statistic 0.69; 95% CI, 0.64–0.73 versus 0.63; 95% CI, 0.61–0.64). This is likely explained by the fact that the validation cohort is more heterogeneous than the development cohort, as patients were not selected on the basis of strict inclusion and exclusion criteria. In general, external validation studies tend to show a drop in performance of models, often because of overfitting of risk scores in the development data.15,24 The observation that performance is maintained in a broader setting underlines the robustness of the model and confirms its generalizability to a wide range of stroke patients. Also, performance of the model is maintained after excluding patients with an established high bleeding risk or reduced life expectancy, showing that the model can help to stratify patients in the group with most uncertainty about the risk of bleeding. Of note, the S2TOP-BLEED score performed particularly well for prediction of major and fatal bleeds, which are of clinical importance and may substantially offset the benefit of antiplatelet drugs.

tancy, showing that the model can help to stratify patients in the group with most uncertainty about the risk of bleeding. Of note, the S2TOP-BLEED score performed particularly well for prediction of major and fatal bleeds, which are of clinical importance and may substantially offset the benefit of antiplatelet drugs. The REACH score systematically underestimated risk of bleeding, which is likely because of the fact that the model was derived from patients with or at risk of atherothrombosis. It has been shown previously that patients with symptomatic vascular disease have higher risks of bleeding than patients with risk factors only.25 The slightly lower discriminatory performance of REACH compared with S2TOP-BLEED can partly be explained by differences in the representation of age in both models, as shown by differences in C statistics for models containing age only. In the REACH score, the weights assigned to age groups imply a linear association between age and bleeding, while the risk of bleeding tends to increase more rapidly at older ages.5 Also, the elderly patients were not represented separately in the REACH score (the highest category was >75 years), whereas nearly half of all patients with a TIA or stroke are over 75 years of age.5 Although age was the most important factor in predicting risk of bleeding, other variables in the S2TOP-BLEED score do have a relevant contribution to risk prediction, as is shown in Figure II in the online-only Data Supplement; younger patients with multiple risk factors may have higher predicted risk of bleeding than patients in older age groups without additional risk factors.

of bleeding, other variables in the S2TOP-BLEED score do have a relevant contribution to risk prediction, as is shown in Figure II in the online-only Data Supplement; younger patients with multiple risk factors may have higher predicted risk of bleeding than patients in older age groups without additional risk factors. Although the C statistic improved slightly compared with the development cohort, values below 0.7 are still considered moderately discriminative. However, similar C statistics are seen for bleeding risk scores in other domains, such as for the HAS-BLED (hypertension, abnormal renal/liver function, stroke, bleeding history or predisposition, labile international normalized ratio, elderly, drugs/alcohol concomitantly) and ORBIT (older age, reduced haemoglobin/haematocrit/anaemia, bleeding history, insufficient kidney function, treatment with antiplatelets) scores in atrial fibrillation.26,27 Furthermore, calibration of a risk score is as important as its discrimination or may be considered even more important in the current setting, where risk of bleeding has to be weighed against the risk of recurrent ischemic events. We showed that long-term predicted risks accurately corresponded with observed risks. The fact that the model showed good calibration in the validation cohort despite differences in baseline risk and case-mix indicates that variables in the model accounted for most of the differences between the 2 cohorts.

emic events. We showed that long-term predicted risks accurately corresponded with observed risks. The fact that the model showed good calibration in the validation cohort despite differences in baseline risk and case-mix indicates that variables in the model accounted for most of the differences between the 2 cohorts. As shown previously, high bleeding risks are associated with high risks of recurrent ischemic events.28 As such, high estimated bleeding risks cannot easily guide treatment decisions of antiplatelet therapy and should always be accompanied by the assessment of ischemic event risk. However, our results do show that risk of ischemic events stabilizes while risk of bleeding increases in patients in medium- and high-risk groups of the S2TOP-BLEED score. Risk stratification may therefore be useful to identify patients in the high-risk group in whom caution seems warranted before starting aggressive dual antiplatelet therapy. Also, estimation of bleeding risk may help to identify patients in whom gastroprotective agents might be indicated. Trials have shown that proton pump inhibitors (PPI) effectively reduce the risk of upper GI bleeding by 70% to 90%,29 but in clinical practice, proton pump inhibitors are not routinely prescribed, possibly because of concerns over side effects associated with long-term use.30,31 A recent study has shown that the numbers needed to treat to prevent one upper GI bleed in patients on aspirin are reasonable, particularly in elderly patients (numbers needed to treat 23 to prevent one upper GI bleed at 5 years in patients aged ≥75 years).5 Co-prescription of proton pump inhibitors may be an effective intervention to lower the risk of GI bleeds, but safety of long-term proton pump inhibitor treatment has not been established in a randomized trial yet. Furthermore, high predicted bleeding risks may trigger physicians to treat and monitor hypertension more closely, aiming to reduce risk of intracerebral hemorrhages.32

intervention to lower the risk of GI bleeds, but safety of long-term proton pump inhibitor treatment has not been established in a randomized trial yet. Furthermore, high predicted bleeding risks may trigger physicians to treat and monitor hypertension more closely, aiming to reduce risk of intracerebral hemorrhages.32 Strengths of our study include the population-based nature of the study, the thorough ascertainment of bleeding events through multiple overlapping sources and the long-term follow-up. However, there are also some limitations. Not all variables included in the risk scores were available in the validation cohort, but suitable proxies could be found for most variables. Furthermore, the number of bleeds in the validation cohort was moderate, particularly for the assessment of performance according to site and severity. Last, a small proportion of patients were excluded as they were not prescribed antiplatelet drugs because of recent bleeding or intolerance. However, this reflects clinical practice.

he number of bleeds in the validation cohort was moderate, particularly for the assessment of performance according to site and severity. Last, a small proportion of patients were excluded as they were not prescribed antiplatelet drugs because of recent bleeding or intolerance. However, this reflects clinical practice. In conclusion, the current study shows that the S2TOP-BLEED score can be used to estimate the risk of major bleeding in patients with a TIA or ischemic stroke on antiplatelet drugs. Although the risk of recurrent ischemic events will outweigh the risk of bleeding in the majority of patients, the risk score identifies patients at particularly high risk of bleeding in whom preventive measures should be taken. Future studies may focus on refinement of the S2TOP-BLEED score for major bleeding by including results from laboratory tests, such as renal failure and anemia, or radiological characteristics, such as microbleeds. Also, a more thorough assessment of the balance between benefits and risks of long-term antiplatelet drugs is required, incorporating risk estimates on risk of recurrent ischemic events, as well as risk of bleeding.

aboratory tests, such as renal failure and anemia, or radiological characteristics, such as microbleeds. Also, a more thorough assessment of the balance between benefits and risks of long-term antiplatelet drugs is required, incorporating risk estimates on risk of recurrent ischemic events, as well as risk of bleeding. Acknowledgments We are grateful to all the staff in the general practices that collaborated in OXVASC (Oxford Vascular Study): Abingdon Surgery, Stert St, Abingdon; Malthouse Surgery, Abingdon; Marcham Road Family Health Centre, Abingdon; The Health Centre, Berinsfield; Key Medical Practice; Kidlington; 19 Beaumont St, Oxford; East Oxford Health Centre, Oxford; Church Street Practice, Wantage. We also acknowledge the use of the facilities of the Acute Vascular Imaging Centre, Oxford. Sources of Funding The Oxford Vascular Study is funded by the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre (BRC), Wellcome Trust, Wolfson Foundation, and British Heart Foundation. Dr Rothwell is in receipt of a NIHR Senior Investigator award. Drs Greving and Hilkens are supported by a grant from the Dutch Heart Foundation (grant number 2013T128). The views expressed are those of the author(s) and not necessarily those of the National Health Service, the NIHR, or the Department of Health. Disclosures None. Supplementary Material Guest Editor for this article was Tatjana Rundek, MD, PhD. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.019259/-/DC1.

It has been hypothesized that cerebral small vessel disease (SVD) may be part of a multisystem disorder affecting other vascular beds, such as the kidney.1,2 However, previous studies of the associations of chronic renal impairment and imaging makers of SVD were conflicting.3–8 While some studies suggested that the association of renal impairment and SVD was explained by shared risk factors such as hypertension, others proposed that genetic factors might contribute to shared susceptibility. Any association caused by shared genetic susceptibility would usually be strongest at younger ages and should remain even after detailed adjustment for shared vascular risk factors. Most previous studies of the association of renal impairment and SVD were hospital based, many had small numbers, the majority were confined to lacunar stroke patients, and all previous studies only adjusted for history of hypertension or a single recent blood pressure, which does not allow adequate adjustment for the potential confounding by long-term blood pressure burden. Moreover, few studies stratified the analyses by age, and no study has focused on the associations of renal impairment and cerebral SVD at younger ages.7–10

story of hypertension or a single recent blood pressure, which does not allow adequate adjustment for the potential confounding by long-term blood pressure burden. Moreover, few studies stratified the analyses by age, and no study has focused on the associations of renal impairment and cerebral SVD at younger ages.7–10 Therefore, in a population-based study, the OXVASC (Oxford Vascular Study), we studied patients with transient ischemic attack (TIA) and minor ischemic stroke to determine the age-specific associations of renal impairment and the overall burden of SVD (total SVD score),9 as well as individual SVD markers, with adjustment for hypertension based on the average premorbid blood pressure level over many years, and by using both the premorbid and baseline creatinine measurement for the diagnosis of renal impairment. Methods Requests for access to data from OXVASC will be considered by the corresponding author.

Therefore, in a population-based study, the OXVASC (Oxford Vascular Study), we studied patients with transient ischemic attack (TIA) and minor ischemic stroke to determine the age-specific associations of renal impairment and the overall burden of SVD (total SVD score),9 as well as individual SVD markers, with adjustment for hypertension based on the average premorbid blood pressure level over many years, and by using both the premorbid and baseline creatinine measurement for the diagnosis of renal impairment. Methods Requests for access to data from OXVASC will be considered by the corresponding author. We studied consecutive patients with TIA or ischemic stroke who underwent cerebral magnetic resonance imaging in OXVASC from 2004 to 2014. OXVASC is an ongoing population-based study of the incidence and outcome of all acute vascular events in a population of 92 728 individuals, registered with 100 general practitioners in 9 general practices in Oxfordshire, United Kingdom. The multiple overlapping methods used to achieve near complete ascertainment of all individuals with TIA and ischemic stroke and the imaging protocol of OXVASC are detailed in Methods in the online-only Data Supplement and have been reported previously.11,12 All cases were reviewed by the senior study neurologist (Dr Rothwell), and TIA/stroke etiology was classified according to the modified Trial of ORG 10172 in Acute Stroke Treatment criteria.11 For the current analyses, patients with cerebral or systemic vasculitis, Cerebral Autosomal Dominant Arteriopathy With Subcortical Infarcts and Leukoencephalopathy, or Fabry disease were excluded.

, and TIA/stroke etiology was classified according to the modified Trial of ORG 10172 in Acute Stroke Treatment criteria.11 For the current analyses, patients with cerebral or systemic vasculitis, Cerebral Autosomal Dominant Arteriopathy With Subcortical Infarcts and Leukoencephalopathy, or Fabry disease were excluded. Demographic data, vascular risk factors (hypertension, diabetes mellitus, known atrial fibrillation, history of smoking, hyperlipidemia), history of previous TIA/stroke, and history of ischemic heart disease were collected from face-to-face interview and cross-referenced with primary care records. Patients routinely had creatinine measured after the acute event as part of the standard protocol. We also collected all premorbid blood pressure readings with dates up to at least 15 years before the event from patient records held in primary care. The most recent premorbid creatinine measurement within 1 year of the index event was also obtained from the regional biochemistry database. The MDRD (Modification of Diet in Renal Disease) Study Group equation was used to calculate estimated glomerular filtration rate (eGFR) for each patient, and renal impairment was defined as estimated eGFR <60 mL/min per 1.73 m2.13 To minimize the potential impact of acute renal injury after TIA and ischemic stroke, we used creatinine taken at the time of the index event in the primary analysis, and the most recent premorbid creatinine taken within 1 year of the index event was also used for sensitivity analysis.

ed eGFR <60 mL/min per 1.73 m2.13 To minimize the potential impact of acute renal injury after TIA and ischemic stroke, we used creatinine taken at the time of the index event in the primary analysis, and the most recent premorbid creatinine taken within 1 year of the index event was also used for sensitivity analysis. One neuroradiologist (Dr Kuker) provided ongoing supervision of interpretation of the magnetic resonance images throughout the study period, and the independently derived and proposed total SVD score was used to assess the overall burden of SVD.9 One point is allocated to each of the following: (1) presence of lacunes; (2) presence of cerebral microbleeds (CMB); (3) moderate–severe (>10) basal ganglia (BG) perivascular spaces (PVS), and (4) severe periventricular or moderate–severe deep white matter hyperintensity (WMH). Lacunes were defined as rounded or ovoid lesions, >3 and <20 mm in diameter, in the BG, internal capsule, centrum semiovale, or brain stem, of cerebrospinal fluid signal density on T2 and fluid-attenuated inversion recovery and no increased signal on diffusion-weighted imaging.14 CMBs were defined as rounded, hypodense foci up to 10 mm in size and were differentiated from microbleed mimics based on current guidelines.15 PVSs were defined as small (<3 mm) punctate (if perpendicular to the plane of scan) or linear (if longitudinal to the plane of scan) hyperintensities on T2 images in BG or centrum semiovale based on a previously validated scale,16 and only BG-PVS were used in the total SVD score. The severity of white matter disease was determined for periventricular versus deep WMH, respectively, according to the Fazekas scale.17

longitudinal to the plane of scan) hyperintensities on T2 images in BG or centrum semiovale based on a previously validated scale,16 and only BG-PVS were used in the total SVD score. The severity of white matter disease was determined for periventricular versus deep WMH, respectively, according to the Fazekas scale.17 Statistical Analysis Categorical variables are reported as absolute numbers with percentages, and continuous variables are reported as means with SD. χ2 and analysis of variance tests were performed to compare categorical and continuous variables between groups. We first used ordinal regression to determine the age-specific (overall/stratified by age groups: <60, 60–79, and ≥80 years) associations of renal impairment and the total SVD score. We then used logistic regression to study the age-specific associations of renal impairment and individual SVD markers, including presence of lacunes, presence of CMBs, BG-PVS, moderate–severe periventricular WMH, and moderate–severe deep WMH. All analyses were adjusted for age (continuous/per year), sex, history of hypertension, diabetes mellitus, and 15-year premorbid systolic blood pressure. All analyses were performed using SPSS version 20 (SPSS Inc, Chicago, IL). Standard Protocol Approvals, Registrations, and Patient Consents Written informed consent or assent from relatives was obtained for all participants. OXVASC was approved by the local research ethics committee (OREC A: 05/Q1604/70).

We first used ordinal regression to determine the age-specific (overall/stratified by age groups: <60, 60–79, and ≥80 years) associations of renal impairment and the total SVD score. We then used logistic regression to study the age-specific associations of renal impairment and individual SVD markers, including presence of lacunes, presence of CMBs, BG-PVS, moderate–severe periventricular WMH, and moderate–severe deep WMH. All analyses were adjusted for age (continuous/per year), sex, history of hypertension, diabetes mellitus, and 15-year premorbid systolic blood pressure. All analyses were performed using SPSS version 20 (SPSS Inc, Chicago, IL). Standard Protocol Approvals, Registrations, and Patient Consents Written informed consent or assent from relatives was obtained for all participants. OXVASC was approved by the local research ethics committee (OREC A: 05/Q1604/70). Results Among 1080 consecutive patients with TIA or ischemic stroke who underwent magnetic resonance brain imaging, 1028 (95.2%) had a full magnetic resonance imaging protocol completed and creatinine measured at baseline and were, thus, included in the analyses. Mean (SD) age was 68.4 (14.1) years, and 261 (26.7%) were <60 years. The baseline characteristics of patients stratified by the total SVD score are shown in Table 1. Table 1. Baseline Characteristics of Patients Included in Analyses Stratified by the Total SVD Score*

Results Among 1080 consecutive patients with TIA or ischemic stroke who underwent magnetic resonance brain imaging, 1028 (95.2%) had a full magnetic resonance imaging protocol completed and creatinine measured at baseline and were, thus, included in the analyses. Mean (SD) age was 68.4 (14.1) years, and 261 (26.7%) were <60 years. The baseline characteristics of patients stratified by the total SVD score are shown in Table 1. Table 1. Baseline Characteristics of Patients Included in Analyses Stratified by the Total SVD Score* As expected, patients with higher total SVD score were older, more likely to have history of hypertension, atrial fibrillation, TIA/stroke prior to the index event and history of ischemic heart disease (Table 1), and had higher blood pressure measured both acutely and during the 15 years before the index event (Table 1).

As expected, patients with higher total SVD score were older, more likely to have history of hypertension, atrial fibrillation, TIA/stroke prior to the index event and history of ischemic heart disease (Table 1), and had higher blood pressure measured both acutely and during the 15 years before the index event (Table 1). The eGFR decreased with increasing total SVD score using creatinine measured either at the index event or 1 year before the index event (both Ptrend<0.001; Table 1). In an ordinal regression, renal impairment (eGFR<60 mL/min per 1.73 m2) was associated with total SVD score (odds ratio [OR], 2.16, 95% confidence interval [CI], 1.69–2.75; P<0.001; Table 2), but this association was only apparent at age <60 years (<60 years: OR, 3.97; 95% CI, 1.69–9.32; P=0.002; 60–79 years: OR, 1.01; 95% CI, 0.72–1.41; P=0.963; ≥80 years: OR, 0.95; 95% CI, 0.59–1.54; P=0.832). The overall association of renal impairment and total SVD score was lost after adjustment for age and sex (OR, 0.94; 95% CI, 0.72–1.23; P=0.639; Table 2), and with additional adjustment for history of hypertension (OR, 0.85; 95% CI, 0.65–1.12; P=0.247; Table 2), and tended to be reversed when also adjusting for premorbid average systolic blood pressure (OR, 0.76; 95% CI, 0.56–1.02; P=0.067; Table 2). However, although the similar attenuation was observed for all age groups, the independent association of renal impairment and total SVD score was maintained in the multivariate analyses at age <60 years (adjusted OR, 3.11; 95% CI, 1.21–7.98; P=0.018; Table 2). Results were similar in patients with lacunar events and in those with nonlacunar events (Table II in the online-only Data Supplement).

e independent association of renal impairment and total SVD score was maintained in the multivariate analyses at age <60 years (adjusted OR, 3.11; 95% CI, 1.21–7.98; P=0.018; Table 2). Results were similar in patients with lacunar events and in those with nonlacunar events (Table II in the online-only Data Supplement). Table 2. Associations of Renal Impairment and Total SVD Score Stratified by Age and Adjusted for Age/Sex and for Known Vascular Risk Factors When we looked at the associations of renal impairment and individual SVD markers, the results were also consistent, with attenuation of apparent associations after adjustment for known risk factors but with independent associations of renal impairment and SVD remaining at younger ages (Figure and Table III in the online-only Data Supplement), particularly for the presence of CMB (OR, 5.84; 95% CI, 1.45–23.53; P=0.013; Figure; Table III in the online-only Data Supplement) and for moderate–severe periventricular WMH (OR, 6.28; 95% CI, 1.54–25.63; P=0.010; Figure; Table III in the online-only Data Supplement). Figure. Associations of renal impairment (adjusted odds ratio and 95% CI) and the presence of individual small vessel disease markers stratified by age. Analyses were adjusted for age, sex, history of hypertension, diabetes mellitus, and premorbid mean systolic blood pressure. Renal impairment is defined as eGFR<60 mL/min per 1.73 m2. CI indicates confidence interval; CMB, cerebral microbleeds; eGFR, estimated glomerular filtration rate; PVS, perivascular spaces; and WMH, white matter hyperintensity.

sex, history of hypertension, diabetes mellitus, and premorbid mean systolic blood pressure. Renal impairment is defined as eGFR<60 mL/min per 1.73 m2. CI indicates confidence interval; CMB, cerebral microbleeds; eGFR, estimated glomerular filtration rate; PVS, perivascular spaces; and WMH, white matter hyperintensity. Sensitivity analyses using creatinine measured 1 year prior to the index event also showed consistent results (Table IV and V in the online-only Data Supplement). Discussion In this population-based study of patients with TIA and ischemic stroke, we found that the associations of renal impairment (eGFR<60 mL/min per 1.73 m2) and SVD were attenuated after adjustment for age, sex, known risk factors, and premorbid average blood pressure and disappeared at older ages. However, the association was maintained at age <60 years, both for the overall SVD burden and for individual SVD markers.

ations of renal impairment (eGFR<60 mL/min per 1.73 m2) and SVD were attenuated after adjustment for age, sex, known risk factors, and premorbid average blood pressure and disappeared at older ages. However, the association was maintained at age <60 years, both for the overall SVD burden and for individual SVD markers. Our findings of age-specific associations of renal impairment and cerebral SVD in TIA and minor stroke are in line with previous studies done predominantly in lacunar stroke or in the nonstroke population using individual SVD markers. A rigorous and comprehensive systematic review and meta-analysis showed that the 4-fold increased risk of renal impairment in lacunar versus nonlacunar stroke was only observed at younger ages.8 Similarly, in the nonstroke population, compared with studies of a mean age of 70 years, there was a stronger relationship between renal impairment and WMH in studies with an average age of 50 to 60 years.8 The same pattern was also seen for renal impairment and CMB or enlarged PVS, where studies including younger patients tended to report a strong association5,7 and studies of older cohorts tended to find no association.4 However, one hospital-based study in TIA and ischemic stroke of a similar mean age to OXVASC (70 years versus 68.4 years) reported a strong association of proteinuria and CMB,18 but only adjusted for history of diagnosed hypertension.

a strong association5,7 and studies of older cohorts tended to find no association.4 However, one hospital-based study in TIA and ischemic stroke of a similar mean age to OXVASC (70 years versus 68.4 years) reported a strong association of proteinuria and CMB,18 but only adjusted for history of diagnosed hypertension. The reason why renal impairment correlates with SVD independent of hypertension and other vascular risk factors at younger ages is uncertain. One explanation is that rather than being the end organ damage from vascular risk factors such as hypertension in 2 different systems, renal impairment and cerebral SVD could be part of a multisystem disease directly affecting small vessels more generally. Our findings that the independent association of renal impairment and SVD seemed to be strongest in those presenting with acute small vessel disease (ie, acute lacunar event) also supported this hypothesis and are consistent with the previous systematic review of renal impairment and lacunar stroke.8 Multisystem pathogenesis is also supported by associations of cerebral SVD with transforming growth factor-β signaling, which has also been associated with cancer, inflammation, and autoimmune diseases.19,20 Alternatively, the independent associations of renal impairment and SVD at younger ages could similarly suggest shared susceptibility to vascular risk factors, most likely at the genetic level, leading to premature disease.

gnaling, which has also been associated with cancer, inflammation, and autoimmune diseases.19,20 Alternatively, the independent associations of renal impairment and SVD at younger ages could similarly suggest shared susceptibility to vascular risk factors, most likely at the genetic level, leading to premature disease. Notably, we did not observe any apparent associations of renal impairment and SVD at older ages. Moreover, after adjustment for age, sex, vascular risk factors, and premorbid blood pressure level, the association of renal impairment and SVD even showed a trend of reversed relationship at older ages. Given that both renal impairment and SVD are associated with premature death,21 the nonassociation of renal impairment and SVD could be explained by a survival bias at older ages, where patients with stronger associations of renal impairment and SVD might have died at a younger age and were not therefore “available” to be recruited into the study. Similarly, patients with multiple comorbidities might also die prematurely, leaving those with fewer risk factors or less susceptibility to risk factors in the cohort, leading to a reverse association after adjustment for these risk factors.

younger age and were not therefore “available” to be recruited into the study. Similarly, patients with multiple comorbidities might also die prematurely, leaving those with fewer risk factors or less susceptibility to risk factors in the cohort, leading to a reverse association after adjustment for these risk factors. A strength of our study is that we were able to adjust associations for long-term premorbid mean blood pressure, but there are also limitations. First, we used the creatinine-based MDRD calculation for eGFR. The MDRD equation was derived from a population with a mean age of 50.6±12.7 years.22 Therefore, the eGFR calculation might not be sensitive enough to differentiate between normal aging-related renal impairment versus pathological renal impairment at older ages. However, the current clinical diagnosis of renal impairment is based on the same eGFR cutoff irrespective of age. Second, we did not measure cystatin C, which may be a more sensitive marker when the creatinine-based eGFR is between 45 and 59 mL/min per 1.73 m2.13 Therefore, we might have overestimated the true prevalence of renal impairment. Even so, we still found an independent association of renal impairment and SVD at younger ages. Third, we did not have data on proteinuria and used eGFR measurement after the acute event for the diagnosis of renal impairment. However, our sensitivity analyses using the eGFR prior to the index event showed consistent results. Fourth, we used the total SVD score to assess the burden of cerebral SVD. However, quantitative measurements of SVD markers might be more accurate in measuring the overall burden particularly for the more severe end. Fifth, although we adjusted for known confounding factors, the possibility of residual confounding could not be excluded. Moreover, we did not adjust for long-term blood pressure variability, although preliminary analyses do not suggest any significant confounding. Sixth, the multiple subgroup analyses were mainly for hypothesis generating and should therefore be interpreted with caution. Finally, our study is based in a predominantly White population and might not be generalizable to the Asian population, where there seems to be stronger association of renal impairment and SVD.

Sixth, the multiple subgroup analyses were mainly for hypothesis generating and should therefore be interpreted with caution. Finally, our study is based in a predominantly White population and might not be generalizable to the Asian population, where there seems to be stronger association of renal impairment and SVD. Our study has several implications. First, the independent associations of renal impairment and SVD at younger ages highlight the importance of effective renal function monitoring and management for young patients. Second, young patients with renal impairment and SVD could be potentially an interesting group for future genetic studies of small vessel disease, and future studies should stratify analyses by age. Finally, further research is needed to understand if there is age-specific treatment effect of renal impairment management on reducing the overall burden of cerebral SVD.

SVD could be potentially an interesting group for future genetic studies of small vessel disease, and future studies should stratify analyses by age. Finally, further research is needed to understand if there is age-specific treatment effect of renal impairment management on reducing the overall burden of cerebral SVD. Acknowledgments We are grateful to all the staff in the general practices that collaborated in the Oxford Vascular Study: Abingdon Surgery, Stert St, Abingdon; Malthouse Surgery, Abingdon; Marcham Road Family Health Centre, Abingdon; The Health Centre, Berinsfield; Key Medical Practice; Kidlington; 19 Beaumont St, Oxford; East Oxford Health Centre, Oxford; Church Street Practice, Wantage. We also acknowledge the use of the facilities of the Acute Vascular Imaging Centre, Oxford. Dr B Liu collected data, did the statistical analysis and interpretation, wrote, and revised the manuscript. Dr KK Lau, Dr L Li, Dr C Lovelock, and Dr W Kuker collected data and revised the manuscript. Dr M Liu provided study design and supervision. Dr PM Rothwell conceived and designed the overall study, provided study supervision and funding, acquired, analyzed, and interpreted data, and wrote and revised the manuscript.

t. Dr KK Lau, Dr L Li, Dr C Lovelock, and Dr W Kuker collected data and revised the manuscript. Dr M Liu provided study design and supervision. Dr PM Rothwell conceived and designed the overall study, provided study supervision and funding, acquired, analyzed, and interpreted data, and wrote and revised the manuscript. Sources of Funding The Oxford Vascular Study is funded by the National Institute for Health Research (NIHR), Oxford Biomedical Research Centre (BRC), Wellcome Trust, Wolfson Foundation, British Heart Foundation, and the European Union’s Horizon 2020 research and innovation programme under grant agreement 666881, SVDs@target. Professor Rothwell is in receipt of an NIHR Senior Investigator award. Dr Liu is funded by the China Scholarship Council. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health. Disclosures None. Supplementary Material Continuing medical education (CME) credit is available for this article. Go to http://cme.ahajournals.org to take the quiz. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.019650/-/DC1.

Stroke remains a leading cause of death and disability worldwide.1 Despite decades of research, tissue-type plasminogen activator and endovascular devices are the only available treatment options.2 However, because of a narrow therapeutic time window and potential contraindications, only 3% to 5% of stroke patients are able to benefit from these interventions.3 This highlights the need for a broader understanding of tissue injury mechanisms to develop more effective treatments. The loss of cerebral blood flow leads to decreased oxygen levels, impairment of mitochondrial oxidative phosphorylation and energy failure in the ischemic area, initiating a sequence of pathophysiological events that after reoxygenation lead to ischemia/reperfusion (I/R) damage.4 Mitochondria play a key role in ischemic brain injury, both through impairment of mitochondrial ATP production with bioenergetic dysfunction and oxidative stress and by mediating cell death pathways.5,6 The lack of oxygen resulting from ischemia leads to impaired mitochondrial ATP production (primary energy failure), collapse of the mitochondrial membrane potential, and, consequently, activation of intrinsic cell death pathways.7,8

bioenergetic dysfunction and oxidative stress and by mediating cell death pathways.5,6 The lack of oxygen resulting from ischemia leads to impaired mitochondrial ATP production (primary energy failure), collapse of the mitochondrial membrane potential, and, consequently, activation of intrinsic cell death pathways.7,8 After reperfusion, there is a transient restoration of bioenergetic state, which is followed by a second phase of energy depletion (secondary energy failure) leading to delayed tissue damage.9,10 This sequence of events has been confirmed by several laboratories, which have also ruled out microcirculatory failure or changes in substrate availability as the cause of the secondary energy depletion and cell death.11,12 In fact, data indicate that secondary energy failure after transient ischemia might be the result of delayed mitochondrial damage, likely because of oxidative stress.11,12 Mitochondrial electron transport chain (ETC) enzymes are known to become rapidly over-reduced in the absence of oxygen and to be damaged by subsequent reoxygenation.11,13,14 However, despite intensive research, the molecular mechanisms of mitochondria damage in I/R remain to be elucidated.

use of oxidative stress.11,12 Mitochondrial electron transport chain (ETC) enzymes are known to become rapidly over-reduced in the absence of oxygen and to be damaged by subsequent reoxygenation.11,13,14 However, despite intensive research, the molecular mechanisms of mitochondria damage in I/R remain to be elucidated. Here, we used a mouse model of middle cerebral artery occlusion (MCAO) to investigate acute I/R-induced changes of mitochondrial function, focusing on the molecular and biochemical mechanisms of primary and secondary energy failure. Our results suggest a central role of mitochondrial complex I (C-I) impairment in the development of bioenergetic failure after acute I/R brain injury. Protection of C-I enzymatic function during ischemia and the initial stages of reperfusion could be an effective approach to prevent subsequent detrimental events in the I/R cascade, ultimately preserving neuronal integrity and reducing brain damage after stroke. Materials and Methods All data and materials have been made publicly available at the https://pure.qub.ac.uk/portal/ repository, and a detailed Methods section is available in the online-only Data Supplement.

Here, we used a mouse model of middle cerebral artery occlusion (MCAO) to investigate acute I/R-induced changes of mitochondrial function, focusing on the molecular and biochemical mechanisms of primary and secondary energy failure. Our results suggest a central role of mitochondrial complex I (C-I) impairment in the development of bioenergetic failure after acute I/R brain injury. Protection of C-I enzymatic function during ischemia and the initial stages of reperfusion could be an effective approach to prevent subsequent detrimental events in the I/R cascade, ultimately preserving neuronal integrity and reducing brain damage after stroke. Materials and Methods All data and materials have been made publicly available at the https://pure.qub.ac.uk/portal/ repository, and a detailed Methods section is available in the online-only Data Supplement. MCAO Model All procedures were approved by the Institutional Animal Care and Use Committee of Weill Cornell Medicine and performed in accordance with the ARRIVE guidelines (Animals in Research: Reporting In Vivo Experiment).15 Transient MCAO was induced using an intraluminal filament as described.16 In brief, 7- to 9-week-old, male mice were anesthetized with 1.5% to 2.0% isoflurane and rectal temperature was maintained at 37.3±0.3°C. Cerebral blood flow was measured with laser-Doppler flowmetry (Periflux System 5010; Perimed) in the ischemic center (2 mm posterior, 5 mm lateral to bregma). After 35 minutes, the filament was retracted and cerebral blood flow reestablished. This duration of cerebral ischemia has been used extensively by us17,18 and others19 and leads to reproducible infarct volumes of 50 to 60 mm3 and measurable neurological deficits. Only animals that exhibited a reduction in cerebral blood flow 85% during MCAO and in which cerebral blood flow recovered by 80% after 10 minutes of reperfusion were included in the study.20,21 Three days after, MCAO functional impairment was assessed and infarct volume was quantified in cresyl violet–stained sections and corrected for swelling, as previously described.16

ood flow 85% during MCAO and in which cerebral blood flow recovered by 80% after 10 minutes of reperfusion were included in the study.20,21 Three days after, MCAO functional impairment was assessed and infarct volume was quantified in cresyl violet–stained sections and corrected for swelling, as previously described.16 Administration of Glutathione-Ester and Glutathione Content Measurement Reduced glutathione-ethyl ester (G1404; Sigma Aldrich) was administered immediately after the initiation of reperfusion via jugular vein (400 mg/kg). Saline injections served as control. Total glutathione content was determined using Glutathione Assay Kit (703002; Cayman). Mitochondrial Measurements After MCAO alone or MCAO with a period of recirculation as indicated, mice were decapitated. Brains were removed and a standardized 4 mm MCA area tissue sample dissected using a mouse brain matrix (Zivic Instruments). The brain sample was homogenized in ice-cold isolation buffer (in mmol/L: 210 mannitol, 70 sucrose, 1 ethylene glycol-bis(β-aminoethyl ether)-N,N,N’,N’-tetraacetic acid, 5 HEPES, pH 7.4) with 80 strokes of a Dounce homogenizer. The homogenate was centrifuged at 1000g for 5 minutes at 4°C and the supernatant was collected and used for respiration analysis. Respiration was measured using Oxygraph-2k (Oroboros Instruments).

hylene glycol-bis(β-aminoethyl ether)-N,N,N’,N’-tetraacetic acid, 5 HEPES, pH 7.4) with 80 strokes of a Dounce homogenizer. The homogenate was centrifuged at 1000g for 5 minutes at 4°C and the supernatant was collected and used for respiration analysis. Respiration was measured using Oxygraph-2k (Oroboros Instruments). For isolation of mitochondria, brain homogenates were centrifuged for 15 minutes at 20 000g. The obtained membrane pellet was rinsed twice with (in mmol/L): 250 sucrose, 50 Tris-HCl (pH 7.5), 0.2 EDTA medium, and subsequently resuspended in the same medium. Frozen aliquots were stored at −80°C until use. Protein content was determined by bicinchoninic acid assay (Sigma) with 0.1% deoxycholate for solubilization of mitochondrial membranes. Mitochondria and Respiratory Chain Analysis Activities of respiratory chain enzyme and citrate synthase were measured spectrophotometrically as described.22 Flavin mononucleotide (FMN) was determined fluorometrically.23 Immunoblot analyses were performed using OXPHOS antibody cocktail (ab110413; Abcam).22 Experimental Design and Statistical Analysis Mice were randomly assigned to the experimental groups, and analyses were performed by an investigator blinded to the treatment protocol. Data are expressed as mean±SEM. Differences were considered statistically significant when *P<0.05. Details of statistical analyses are indicated in the Figure legends and online-only Data Supplement.

assigned to the experimental groups, and analyses were performed by an investigator blinded to the treatment protocol. Data are expressed as mean±SEM. Differences were considered statistically significant when *P<0.05. Details of statistical analyses are indicated in the Figure legends and online-only Data Supplement. Results Multiphasic Impairment of Mitochondrial Respiration in I/R We studied I/R-induced changes of mitochondrial function in a mouse model of focal ischemia after transient MCAO. Figure 1A shows representative traces of malate/glutamate-supported respiration of brain homogenates of sham and after 35 minutes ischemia. ADP-stimulated mitochondrial respiration showed multiphasic impairment after I/R (Figure 1B). A decline (59.0±5.9% of sham control; P<0.05; n=4 per group) was observed during ischemia, followed by a partial recovery (79.6±5.4% of control; P>0.05; n=5 per group) at 10 minutes of reperfusion, and by a subsequent profound decline in respiration (50.7±6.2% of control; P<0.05; n=5 per group) at 30 minutes of reperfusion. These early changes in mitochondrial function were followed by a recovery of respiration at 1 hour of reperfusion (84.7±2.3% of control; P>0.05; n=5 per group) and then by a progressive decline in respiration, occurring 2 to 24 hours (55±7.8% of control; P<0.05; n=5 per group, at 24 hours) after reperfusion (Figure 1B).

hese early changes in mitochondrial function were followed by a recovery of respiration at 1 hour of reperfusion (84.7±2.3% of control; P>0.05; n=5 per group) and then by a progressive decline in respiration, occurring 2 to 24 hours (55±7.8% of control; P<0.05; n=5 per group, at 24 hours) after reperfusion (Figure 1B). Figure 1. Multiphasic pattern of mitochondrial respiratory decline after ischemia/reperfusion (I/R). A, Representative traces of mitochondrial oxygen consumption from sham (red) and 35 minute ischemia (black) in whole tissue homogenates. Addition of 1 mM cyanide (KCN) almost fully inhibited respiration in homogenates. B, Effect of I/R injury on ADP-stimulated oxygen consumption (sham: n=19; different time points n as indicated by the dots; Kruskal–Wallis test with Dunn multiple comparisons test; C) citrate synthase (C) and NADH oxidase (D) activity in the same preparations. Citrate synthase activity, an indicator of mitochondrial content, did not significantly differ from control at any time point (P>0.05; n=3–7 per group; Figure 1C). Further, we did not detect significant changes in the protein levels of ETC complexes I to V and mitochondrial respiratory control ratio, during the ischemic phase or within 24 hours after reperfusion (Figure I in the online-only Data Supplement).

fer from control at any time point (P>0.05; n=3–7 per group; Figure 1C). Further, we did not detect significant changes in the protein levels of ETC complexes I to V and mitochondrial respiratory control ratio, during the ischemic phase or within 24 hours after reperfusion (Figure I in the online-only Data Supplement). Activities of Individual Mitochondrial Membrane Complexes Are Differently Affected in I/R We note that brain homogenates included nonsynaptic mitochondria but also synaptosomes containing synaptic mitochondria. The synaptic mitochondria, however, do not contribute to ADP-stimulated respiration, because of restricted ADP access to synaptosomes. Respiration measured in whole tissue homogenates is a product of several processes including transport of substrates, activities of NAD-dependent dehydrogenases, and ETC. For ETC activity measurements, substrate delivery into all mitochondrial populations was ensured by addition of the membrane-permeabilizing agent alamethicin. To specifically assay for I/R-induced changes in ETC complexes, we assessed the overall activity of the respiratory chain by measuring NADH oxidase (complexes I+III+IV; Figure 1D). The temporal profile of NADH oxidase activity changes strongly corresponded to the multiphasic pattern observed for the mitochondrial respiration (Figure 1B), suggesting that the observed mitochondrial dysfunction is a result of I/R-induced ETC impairment.

hain by measuring NADH oxidase (complexes I+III+IV; Figure 1D). The temporal profile of NADH oxidase activity changes strongly corresponded to the multiphasic pattern observed for the mitochondrial respiration (Figure 1B), suggesting that the observed mitochondrial dysfunction is a result of I/R-induced ETC impairment. Next, we measured the activities of individual ETC complexes: succinate dehydrogenase (C-II), ferrocytochrome c oxidase (C-IV), and NADH:ubiquinone oxidoreductase (C-I), as well as succinate:cytochrome c reductase (C-II+C-III). C-II–linked activities were not affected at any time point after I/R, indicating that C-II and C-III were not responsible for the I/R-induced mitochondrial dysfunction (P>0.05; n=4–7 per group; Figure 2A; C-II+III data not shown). Figure 2. Enzymatic activities of respiratory chain complexes are differently affected after ischemia/reperfusion (I/R). A, Complex II (C-II), (B) C-IV, and (C) C-I NADH:Q1 and (D) C-I NADH:hexaammineruthenium (HAR) reductase activities were measured in whole tissue homogenates; n=4 to 12 per group; Kruskal–Wallis test. C-IV activity was lower at all time points compared with sham (63.7±9.1% at 24 hours; P<0.05; n=4; Figure 2B). However, the decline of C-IV did not follow the multiphasic pattern observed in ADP-stimulated respiration and NADH oxidase activity, suggesting that the mechanism of C-IV and NADH oxidase impairment is different and that C-IV is not responsible for the I/R-induced changes.

.7±9.1% at 24 hours; P<0.05; n=4; Figure 2B). However, the decline of C-IV did not follow the multiphasic pattern observed in ADP-stimulated respiration and NADH oxidase activity, suggesting that the mechanism of C-IV and NADH oxidase impairment is different and that C-IV is not responsible for the I/R-induced changes. Complex I Impairment Is Associated With the Multiphasic Pattern of Respiratory Decline in I/R To elucidate the mechanisms of C-I impairment, we measured the physiological activity and the relative amount of C-I using two different approaches. The physiological activity of C-I was assessed as NADH:Q1 reductase. The relative content of C-I (proportional to flavin [FMN] content in the enzyme) was determined as oxidation of NADH by hexaammineruthenium (HAR). The NADH:HAR reaction occurs only at the head of the enzyme, where HAR accepts electrons from the FMN, the first redox center of C-I.24 The physiological activity of C-I (Figure 2C) followed the same pattern as ADP-stimulated respiration (Figure 1B) and NADH oxidase activity (Figure 1D), indicating that the impairment of oxidative phosphorylation in the ischemic tissue was because of a specific dysfunction of C-I. Interestingly, NADH:HAR reductase showed an apparent decrease in the relative content of C-I after 35 minutes of ischemia (68.7±1.3%; P=0.0001; n=6 per group), followed by a rapid recovery after reoxygenation (97.0±3.9%; P=0.99; n=6 per group) and a slow gradual decline at subsequent time points after I/R injury (78.7±4.3%; P=0.0003; n=8 per group; Figure 2D).

n apparent decrease in the relative content of C-I after 35 minutes of ischemia (68.7±1.3%; P=0.0001; n=6 per group), followed by a rapid recovery after reoxygenation (97.0±3.9%; P=0.99; n=6 per group) and a slow gradual decline at subsequent time points after I/R injury (78.7±4.3%; P=0.0003; n=8 per group; Figure 2D). On the basis of the results above, we identified 35 minutes ischemia and 30 minutes, 1 hour, and 24 hours of reperfusion as critical time points for the development of mitochondrial dysfunction in I/R injury. To further elucidate the mechanism of C-I impairment, we assayed NADH:HAR and NADH:Q1 reductase activity in preparations of mitochondrial membranes isolated at these time points. As shown in Figure 3A, 35 minutes ischemia resulted in a robust decline of NADH:Q1 reductase and NADH:HAR activity (74.2±4.2% and 80.5±2.7%, n=5 per group, respectively). A significant decrease in NADH:Q1 activity at 30 minutes (71.2±2.3%; n=4 per group), recovery at 1 hour (92.7±2.1%; n=6 per group), and another activity decline at 24 hours (58.9±3.3%; n=4 per group) of reperfusion was observed, confirming the results in whole tissue homogenates. Conversely, NADH:HAR activity showed a transient recovery at 30 minutes and 1 hour (92.9±2.9%; n=4 per group; 90.9±3.5%, n=6 per group, respectively) followed by a gradual decline at 24 hours after reoxygenation (77.2±2.0%; n=4 per group; Figure 3A).

ion was observed, confirming the results in whole tissue homogenates. Conversely, NADH:HAR activity showed a transient recovery at 30 minutes and 1 hour (92.9±2.9%; n=4 per group; 90.9±3.5%, n=6 per group, respectively) followed by a gradual decline at 24 hours after reoxygenation (77.2±2.0%; n=4 per group; Figure 3A). Figure 3. Complex I (C-I) impairment is associated with the multiphasic pattern of respiratory decline observed in ischemia/reperfusion (I/R). A, Overall activity of C-I at critical time points after I/R in mitochondrial membranes. B, In vitro time course of the reductive inactivation of NADH:hexaammineruthenium (HAR) reductase activity in mitochondrial membranes. Ischemic over-reduction of the ETC resulted in a decrease of the relative amount of C-I (red line) compared with control (black line). Addition of reduced FMN restored NADH:HAR reductase activity (arrow). C, Decrease of FMN in mitochondrial membranes obtained from the ischemic area after 35 minutes of middle cerebral artery occlusion (MCAO) compared with sham controls (n=6–12 per group; P=0.0048; ANOVA). n.s. indicates not signficant.

ol (black line). Addition of reduced FMN restored NADH:HAR reductase activity (arrow). C, Decrease of FMN in mitochondrial membranes obtained from the ischemic area after 35 minutes of middle cerebral artery occlusion (MCAO) compared with sham controls (n=6–12 per group; P=0.0048; ANOVA). n.s. indicates not signficant. The drop in physiological NADH:Q1 reductase activity could be explained by two fundamentally different mechanisms: a decline in C-I content or a decrease in the catalytic efficiency of C-I (number of NADH molecules oxidized by 1 enzyme molecule per minute). To estimate the relative catalytic efficiency of C-I, the ratio of NADH:Q1/NADH:HAR reductase (Q1/HAR) was calculated as previously described.25 No significant reduction in the catalytic efficiency of C-I after 35 minutes of ischemia (92.1±4.2%; P>0.05; n=5 per group) was observed. After reperfusion, a substantial decline in the efficiency of the enzyme was found at 30 minutes (76.8±2.8%; P<0.05; n=4 per group), followed by a complete recovery at 1 hour (102.6±4.0%; P>0.05; n=6 per group), and another decline at 24 hours (79.4±6.0%; P>0.05; n=4 per group) after I/R (Figure 3A). Note that although there was a significant decrease in NADH:Q1 and NADH:HAR reductase activities at 35 minutes ischemia, the Q1/HAR ratio did not change. This could be interpreted as decrease in the number of functional C-I molecules in the membrane with no change in the individual C-I enzyme catalytic efficiency (Q1/HAR).

that although there was a significant decrease in NADH:Q1 and NADH:HAR reductase activities at 35 minutes ischemia, the Q1/HAR ratio did not change. This could be interpreted as decrease in the number of functional C-I molecules in the membrane with no change in the individual C-I enzyme catalytic efficiency (Q1/HAR). Functional Impairment C-I in Ischemia Is Because of a Loss of FMN To further explore the transient decrease of NADH:HAR reductase after 35 minutes of ischemia, we performed in vitro experiments. Brain mitochondrial membranes from naive animals were incubated in conditions of metabolic reductive hypoxia,26 mimicking the over-reduction of the ETC in ischemia (Figure 3B). The first redox center of C-I, noncovalently bound FMN, is capable of dissociating from the enzyme.27 We found that incubation of mitochondrial membranes in reductive conditions resulted in a rapid decline of the HAR reductase activity over time, which seemed as a decrease of C-I content. To confirm these in vitro findings, we determined the content of noncovalently bound FMN in mitochondrial membranes isolated at the critical time points after I/R. We found a significant decline of FMN content in the samples obtained after 35 minutes of ischemia (Figure 3C). This drop correlated with the decrease in NADH:HAR reductase activity in the same samples (Figure 3A, red bar) indicating ischemia-induced loss of FMN from the enzyme without a decrease in C-I content.

r I/R. We found a significant decline of FMN content in the samples obtained after 35 minutes of ischemia (Figure 3C). This drop correlated with the decrease in NADH:HAR reductase activity in the same samples (Figure 3A, red bar) indicating ischemia-induced loss of FMN from the enzyme without a decrease in C-I content. Glutathione Improves C-I Dysfunction and Neurological Outcome After I/R Reperfusion-induced oxidative stress is one of the main contributors to tissue injury in I/R.28,29 Intracellular glutathione-dependent enzymatic systems regulate the thiol-based redox homeostasis and play a major role in the protection against oxidative stress. As shown in Figure 4A, I/R resulted in a significant decline of total glutathione content in the affected area in comparison to the contralateral hemisphere or sham. To test if reduced glutathione is able to confer a C-I–linked neuroprotection in vivo, we administered membrane-permeable glutathione-ethyl ester at the onset of reperfusion. Glutathione-ethyl ester restored total glutathione content in the ipsilateral hemisphere to control values (Figure 4A), indicating that it is able to penetrate into the brain tissue and interact with cellular glutathione pool. Furthermore, administration of glutathione-ethyl ester led to a 61% reduction in infarct volume (glutathione: 25.9±4.4 mm3 versus control: 66.8±7.0 mm3; P=0.0002; n=8–9 per group; Figure 4B and 4C), which correlated with decreased body weight loss (glutathione: 7.0±2.4% versus control: 19.8±2.9%; P=0.0037; n=8–9 per group; Figure IIA in the online-only Data Supplement). Overall functional outcome, assessed by the hanging wire test (Figure 4D) and modified Bederson score (Figure IIB in the online-only Data Supplement), was also improved compared with saline-treated controls.

: 7.0±2.4% versus control: 19.8±2.9%; P=0.0037; n=8–9 per group; Figure IIA in the online-only Data Supplement). Overall functional outcome, assessed by the hanging wire test (Figure 4D) and modified Bederson score (Figure IIB in the online-only Data Supplement), was also improved compared with saline-treated controls. Figure 4. Restoration of total glutathione (GSH) tissue content attenuates ischemic injury and improves neurological outcome 72 hours after middle cerebral artery occlusion (MCAO). A, GSH ester treatment at the onset of reperfusion restored total GSH content in the ischemic brain area 30 minutes after reperfusion (n=3–4 per group; ANOVA). B, Reduced infarct volume in GSH-treated mice compared with controls 72 hours after MCAO (n=8–9 per group; P=0.0002; t test). C, Representative images of corresponding Nissl-stained brain sections of a GSH-treated mouse compared with control 3 days after MCAO. The red dashed line indicates the infarct area. D, GSH-treated mice show a significantly reduced motor impairment indicated by hanging wire test (6–8 per group; P<0.0001; t test).

est). C, Representative images of corresponding Nissl-stained brain sections of a GSH-treated mouse compared with control 3 days after MCAO. The red dashed line indicates the infarct area. D, GSH-treated mice show a significantly reduced motor impairment indicated by hanging wire test (6–8 per group; P<0.0001; t test). Glutathione Prevents Mitochondrial Dysfunction and C-I Activity Decline Early After I/R To test the effect of glutathione-ethyl ester administration on mitochondrial function in vivo, we measured mitochondrial respiration 30 minutes after the onset of reperfusion comparing glutathione-treated mice and saline-treated controls. A significant increase in respiration was observed in tissue homogenates prepared from the ischemic area of glutathione-treated mice compared with saline-treated animals subjected to MCAO (P=0.03; n=5–6 per group; Figure 5A). These findings were associated with an increase in NADH:Q1 activity (P=0.0002; n=6 per group; Figure 5B). NADH:HAR activity was not affected (P>0.05; n=6 per group; Figure 5C) at 30 minutes of reperfusion.

tathione-treated mice compared with saline-treated animals subjected to MCAO (P=0.03; n=5–6 per group; Figure 5A). These findings were associated with an increase in NADH:Q1 activity (P=0.0002; n=6 per group; Figure 5B). NADH:HAR activity was not affected (P>0.05; n=6 per group; Figure 5C) at 30 minutes of reperfusion. Figure 5. Glutathione (GSH) ester treatment improves complex I (C-I)–mediated bioenergetic failure early after reperfusion. A, GSH ester treatment ameliorates mitochondrial respiratory decline at 30 minutes of reperfusion (n=5–6 per group; P=0.03; Mann–Whitney U test). B, C-I activity is significantly improved in GSH-treated mice compared with controls (n=5–6 per group; P<0.05; Mann–Whitney U test). C, No change in the relative amount of C-I was observed. D, In vitro pre-incubation of whole tissue homogenates with GSH was able to partially recover ischemia/reperfusion (I/R)-induced C-I activity decline 30 minutes after reperfusion (n=4 per group; P=0.0009; t tests). GSH treatment did not affect C-I activity in sham, 1 or 24 hours after reperfusion. MCAO indicates middle cerebral artery occlusion.

of whole tissue homogenates with GSH was able to partially recover ischemia/reperfusion (I/R)-induced C-I activity decline 30 minutes after reperfusion (n=4 per group; P=0.0009; t tests). GSH treatment did not affect C-I activity in sham, 1 or 24 hours after reperfusion. MCAO indicates middle cerebral artery occlusion. Mitochondrial membranes isolated from the ischemic area of untreated animals at the critical time points after MCAO were preincubated ex vivo with thiol-reducing agent glutathione. NADH:Q1 and NADH:HAR activity were measured before and after glutathione incubation (Figure 5D). Pre-incubation with glutathione was able to recover NADH:Q1 activity in membranes obtained at 30 minutes of reperfusion (Figure 5D), indicating that reversible oxidation of C-I thiols is the underlying post-translational modification early after reperfusion. In contrast, glutathione treatment did not affect NADH:Q1 activity at 24 hours of reperfusion, pointing to an irreversible decline of C-I catalytic efficiency at later time points.

(Figure 5D), indicating that reversible oxidation of C-I thiols is the underlying post-translational modification early after reperfusion. In contrast, glutathione treatment did not affect NADH:Q1 activity at 24 hours of reperfusion, pointing to an irreversible decline of C-I catalytic efficiency at later time points. Discussion In the present study, we established a spatiotemporal profile of biochemical mechanisms contributing to the evolution of mitochondrial bioenergetic failure in I/R using a mouse model of transient MCAO. Using brain homogenates, we demonstrate an I/R-induced, multiphasic pattern of mitochondrial respiratory dysfunction in the brain, which to our knowledge has not been described before (Figure 6). We observed an initial decline in respiration after 35 minutes of ischemia, which is in agreement with previously published studies.28,30 The rapid partial recovery of mitochondrial respiration after 10 minutes of reperfusion followed by a first reflow-induced respiratory decline at 30 minutes of reoxygenation has never been reported. This decline in tissue respiration was followed by an almost full recovery at 1 hour with a slow decrease at later reperfusion time points (4–24 hours). In the samples from all time points, citrate synthase activity was similar to the sham controls, indicating preservation of mitochondrial mass, for 24 hours post-ischemia.

decline in tissue respiration was followed by an almost full recovery at 1 hour with a slow decrease at later reperfusion time points (4–24 hours). In the samples from all time points, citrate synthase activity was similar to the sham controls, indicating preservation of mitochondrial mass, for 24 hours post-ischemia. Figure 6. Proposed biochemical mechanisms contributing to complex I (C-I)–mediated energy failure in brain ischemia/reperfusion (I/R). A, Electron transfer within C-I during normoxia. B, Ischemic over-reduction of the ETC results in a reduction of C-I FMN. Reduced flavin (FMNH2) loses the affinity for its binding site and dissociates from the enzyme. C, Reflow-induced reoxidation of FMNH2 by molecular oxygen is associated with the generation of reactive oxygen species (ROS) and likely contributes to oxidative stress in the mitochondrial matrix. D, Recovery of C-I function at the early stage of reperfusion is followed by oxidation of critical C-I thiol residues (-SH) at later stages. MCAO indicates middle cerebral artery occlusion. The rate of mitochondrial respiration can be used as a predictor of tissue survival after I/R.31 Our tissue preparations from the MCA area include a mixture of different brain cells so the observed changes in mitochondrial respiration cannot be exclusively attributed to only one cell type. Several publications have described a glia/neuron ratio of 0.4 to 0.35 in mouse brain,32,33 suggesting that neuronal mitochondria may contribute the major fraction of respiratory activity in brain homogenates.

s so the observed changes in mitochondrial respiration cannot be exclusively attributed to only one cell type. Several publications have described a glia/neuron ratio of 0.4 to 0.35 in mouse brain,32,33 suggesting that neuronal mitochondria may contribute the major fraction of respiratory activity in brain homogenates. We identified C-I as the key respiratory enzyme responsible for the multiphasic pattern of mitochondrial dysfunction in I/R. C-I has a high degree of flux control over oxidative phosphorylation and is considered to be the rate-limiting component of NADH oxidase activity within the ETC.34 Supporting our results, a comparable pattern of rotenone-sensitive C-I activity decline within 4 hours after reperfusion has been reported previously.35 The observed progressive decline in the enzymatic activity of C-IV after I/R injury is likely because of a different mechanism.11 C-II and C-III were not significantly affected in I/R, which is in agreement with previous in vivo studies investigating mitochondrial membrane complexes in stroke.7,28,34,35

We identified C-I as the key respiratory enzyme responsible for the multiphasic pattern of mitochondrial dysfunction in I/R. C-I has a high degree of flux control over oxidative phosphorylation and is considered to be the rate-limiting component of NADH oxidase activity within the ETC.34 Supporting our results, a comparable pattern of rotenone-sensitive C-I activity decline within 4 hours after reperfusion has been reported previously.35 The observed progressive decline in the enzymatic activity of C-IV after I/R injury is likely because of a different mechanism.11 C-II and C-III were not significantly affected in I/R, which is in agreement with previous in vivo studies investigating mitochondrial membrane complexes in stroke.7,28,34,35 Inhibition of NAD+-dependent respiration after ischemia has been observed in many stroke studies,7,13,28,34,35 but the mechanism was never established. Our results strongly suggest that ischemia induces a reversible release of FMN from C-I that caused the robust decrease of enzyme activity, which was rapidly restored within 10 minutes of reflow. C-I contains 1 molecule of noncovalently bound FMN per molecule of the enzyme,36 and it is the main source of membrane-associated flavin in mitochondria.37 FMN release is likely to occur in ischemia because of complex I over-reduction via reverse electron transfer.38 Reductive dissociation of C-I FMN has been reported in vitro27,38 but has not been shown in physiological settings. The release of a significant amount of reduced FMN (30–40 µmol/L) to the mitochondrial matrix is potentially harmful for the cell. On reperfusion, reduced FMN can be quickly reoxidized by oxygen,39 generating an equimolar amount of H2O2 in the matrix and significantly contributing to I/R-induced oxidative stress and tissue injury.

of a significant amount of reduced FMN (30–40 µmol/L) to the mitochondrial matrix is potentially harmful for the cell. On reperfusion, reduced FMN can be quickly reoxidized by oxygen,39 generating an equimolar amount of H2O2 in the matrix and significantly contributing to I/R-induced oxidative stress and tissue injury. Mitochondrial function depends strongly on the maintenance of a cellular redox balance. Reperfusion triggers a burst of reactive oxygen species formation directly damaging cells via several different mechanisms.40 A critical component in the mitochondrial antioxidant defense system is endogenous glutathione. Reduced glutathione prevents or repairs oxidative damage generated by reactive oxygen species. Glutathione homeostasis is severely affected after I/R, therefore, making protein thiols a major target of oxidative damage.41–43 Mitochondrial respiratory enzymes are particularly susceptible to reactive oxygen species-mediated modulation of the thiol redox systems.44,45

ve damage generated by reactive oxygen species. Glutathione homeostasis is severely affected after I/R, therefore, making protein thiols a major target of oxidative damage.41–43 Mitochondrial respiratory enzymes are particularly susceptible to reactive oxygen species-mediated modulation of the thiol redox systems.44,45 We demonstrate that restoring total glutathione levels in the ischemic area at the onset of reperfusion is associated with protection of mitochondrial C-I activity and a robust neuroprotective effect. This is in agreement with previous studies showing a cytoprotective action of membrane-permeable thiol antioxidants against I/R-induced brain injury,46–48 but the mechanisms were not completely understood. Here, we presented evidence suggesting a reversible oxidation of critical thiols of C-I early after reperfusion, which is associated with a significant decrease in the enzymatic activity. Reconstitution of glutathione levels in vivo prevents mitochondrial bioenergetic dysfunction and C-I activity decline at 30 minutes of reperfusion. It should be noted that, in addition to protecting mitochondrial C-I, the antioxidant action of glutathione could also have beneficial impact through other cellular pathways, including inhibition of apoptosis48 and prevention of cytokine release.46 The ex vivo treatment of post-I/R mitochondrial membranes with glutathione recovered C-I activity at early, but not at late time points after reperfusion. Our data suggest that early reversible post-translational modifications of C-I are followed by an irreversible enzyme damage. On the basis of the neuroprotection of glutathione-ethyl ester and its positive effect on mitochondrial bioenergetics at 30 minutes of reperfusion, it is fair to speculate that this time point is particularly critical for the evolution of tissue infarction in our I/R model.

re followed by an irreversible enzyme damage. On the basis of the neuroprotection of glutathione-ethyl ester and its positive effect on mitochondrial bioenergetics at 30 minutes of reperfusion, it is fair to speculate that this time point is particularly critical for the evolution of tissue infarction in our I/R model. Conclusions We provide the first evidence that focal cerebral ischemia induces a C-I–mediated pattern of mitochondrial respiratory decline early after I/R. The ischemia-induced impairment of C-I activity is because of the reversible dissociation of reduced flavin from the enzyme (Figure 6). Because FMNH2 is a strong reactive oxygen species generator, this might be an important mechanism for the development of transient oxidative stress after reintroduction of oxygen on reperfusion. Administration of ethyl ester of glutathione at the onset of reperfusion reduces infarction volume by 61% and improves neurological outcomes. This neuroprotective effect is associated with an increase of mitochondrial respiration and C-I activity. Thus, we conclude that reperfusion-induced C-I decline at 30 minutes after reperfusion is the result of a reversible modification of critical thiols of the enzyme. These findings indicate that preventing oxidative thiol modification of ETC early after the onset of reperfusion may be a viable approach to ameliorate mitochondrial dysfunction after I/R injury, ultimately reducing brain damage after stroke.

is the result of a reversible modification of critical thiols of the enzyme. These findings indicate that preventing oxidative thiol modification of ETC early after the onset of reperfusion may be a viable approach to ameliorate mitochondrial dysfunction after I/R injury, ultimately reducing brain damage after stroke. Sources of Funding This study was supported by Medical Research Council UK grant MR/L007339/1 (Dr Galkin) and National Institutes of Health grants R01NS34179 (Dr Iadecola) and R01NS095692 (Drs Manfredi and Iadecola). Dr Kahl was recipient of a postdoctoral research grant from the Deutsche Forschungsgemeinschaft (KA3810/1-1). Disclosures None. Supplementary Material Guest Editor for this article was Miguel A Perez-Pinzon, PhD. * Dr Kahl and A. Stepanova contributed equally. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.019687/-/DC1. Correspondence to Alexander Galkin, PhD, Division of Neonatology, Department of Pediatrics, Columbia University, 3959 Broadway, CHN 1201, New York, NY 10032. E-mail ag4003@cumc.columbia.edu

Cerebral small vessel disease (cSVD) is a pathological process involving small perforating vessels in the brain.1 It may cause symptomatic lacunar infarcts but also silent brain damage that is visible on structural magnetic resonance imaging (MRI), including white matter hyperintensities (WMHs) and lacunes.2 Insights into the radiological evolution of cSVD features are important because they might increase our knowledge of vascular disease and neurodegeneration and could be used as surrogate markers for therapeutic studies. However, there is limited information on long-term tissue damage in sporadic cSVD. Although central cavitation (lacune formation) of the sporadic symptomatic lacunar infarct has been reported in several studies,3–6 only 1 study investigated long-term perilesion morphological changes of sporadic symptomatic lacunar infarcts, finding new WMH (caps) lateral or superior to 18% (15 of 82) of sporadic symptomatic lacunar infarcts.7 Other studies have suggested that sporadic symptomatic infarcts are associated with widespread secondary degeneration in adjacent white matter tracts.8 In contrast, in patients with cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL)—a genetic cSVD—several studies document long-term perilesional and remote morphological changes of cSVD lesions, including development of new lacunes at the edge of WMH or spread of WMH around incident lacunes.9,10

with cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL)—a genetic cSVD—several studies document long-term perilesional and remote morphological changes of cSVD lesions, including development of new lacunes at the edge of WMH or spread of WMH around incident lacunes.9,10 The aim of the present study was to investigate further the natural disease course of sporadic cSVD by assessing the long-term, morphological lesional and perilesional changes in symptomatic lacunar infarcts and adjacent white matter on structural MRI. Materials and Methods The data that support the findings of this study are available from the corresponding author on reasonable request.

The aim of the present study was to investigate further the natural disease course of sporadic cSVD by assessing the long-term, morphological lesional and perilesional changes in symptomatic lacunar infarcts and adjacent white matter on structural MRI. Materials and Methods The data that support the findings of this study are available from the corresponding author on reasonable request. Study Population and Recruitment We used longitudinal data from 2 nonoverlapping, prospective observational studies in patients with minor (ie, nondisabling) ischemic stroke who were all scanned on 1 research-dedicated MRI scanner (MSS [Mild Stroke Study]-1, 2003–2007; MSS-2, 2010–2012). The design of both studies has been described before.11–13 Briefly, both studies recruited patients who presented with minor ischemic stroke, including symptoms of lacunar stroke, to the regional stroke service. Patients with contraindications to MRI and unstable medical conditions were excluded. For the present study, we selected all patients with (1) a clinical diagnosis of lacunar stroke and a relevant (to symptoms) acute small subcortical infarct (lacunar infarct; index lesion) on diffusion MRI at presentation and (2) a follow-up structural MRI between 1 and 5 years after the index stroke. We recorded age at stroke onset, sex, and vascular risk factors, as defined earlier.11 Both studies were approved by the Scotland and Lothian Research Ethics Committee, and all patients gave written informed consent.

on MRI at presentation and (2) a follow-up structural MRI between 1 and 5 years after the index stroke. We recorded age at stroke onset, sex, and vascular risk factors, as defined earlier.11 Both studies were approved by the Scotland and Lothian Research Ethics Committee, and all patients gave written informed consent. MR Imaging All patients had a brain MRI at baseline (median, 4; range, 0–57; days after stroke onset) on a research-dedicated 1.5 Tesla MR scanner (Signa LX; General Electric, Milwaukee, WI). Sequences included axial diffusion-weighted imaging, fluid-attenuated inversion recovery, T2-weighted and T2*-imaging, and sagittal T1-weighted sequences (details were described elsewhere11,13). An experienced neuroradiologist (J.M.W.) assessed the MR images for presence of the acute small subcortical (lacunar) infarct, cSVD lesions, and brain atrophy, all according to the STRIVE criteria (Standards for Reporting Vascular Changes on Neuroimaging).2 We defined an acute small subcortical (lacunar) infarct as a round or ovoid (axial diameter, <20 mm) lesion in the basal ganglia, internal capsule, centrum semiovale, or brain stem, which was hyperintense on diffusion-weighted imaging, had reduced signal on apparent diffusion coefficient imaging, with or without increased signal on fluid-attenuated inversion recovery or T2-weighted imaging.2 We refer to this symptomatic acute lesion as the index lacunar lesion. The diameter of each index lacunar infarct was measured in 3 directions on fluid-attenuated inversion recovery (we report maximum diameter). Baseline MRI was also rated separately and blind to the index lesion progression for presence of lacunes (number and location), WMH (periventricular and deep; Fazekas score14), cerebral microbleeds (location and number; modified brain observer microbleed scale15), and perivascular spaces (5-point score in basal ganglia and centrum semiovale separately11).

y and blind to the index lesion progression for presence of lacunes (number and location), WMH (periventricular and deep; Fazekas score14), cerebral microbleeds (location and number; modified brain observer microbleed scale15), and perivascular spaces (5-point score in basal ganglia and centrum semiovale separately11). We selected those patients who had a follow-up brain MRI at 1 to 5 years after index stroke (median, 403; range, 315–1781; days after index stroke) as part of the original studies (MSS-1 patients had a follow-up MRI at 3 years after stroke and MSS-2 patients had a follow-up MRI at 1 year after stroke; some had longer intervals). The follow-up MR protocol was the same as the baseline MR protocol and performed at the same MR scanner. We visually scored the follow-up appearance of the index lacunar lesion and adjacent white matter on MRI, blind to the other cSVD features, by examining the index lesion and at least 3 MRI slices superior (ie, in the direction of the cortex) and 3 MRI slices inferior (ie, in the direction of the brain stem) from the index lacunar infarct. We assessed the appearance of the index lacunar lesion itself by measuring the diameter in 3 planes as at baseline (we report on maximum diameter) and the appearance, including different degrees of cavitation (no cavitation, partial cavitation, or complete cavitation), as defined previously.3,5 Definitions and imaging characteristics of different degrees of cavitation are in Methods in the online-only Data Supplement.

s at baseline (we report on maximum diameter) and the appearance, including different degrees of cavitation (no cavitation, partial cavitation, or complete cavitation), as defined previously.3,5 Definitions and imaging characteristics of different degrees of cavitation are in Methods in the online-only Data Supplement. Inferior and superior from index lacunar lesion, we assessed the presence of new WMH (WMH track and WMH cap, respectively). Figure 1 shows an example of a WMH cap adjacent (superior) to the index lesion. These WMH caps were single, mostly round or ovoid with indistinct margins, were not visible at baseline, and could be clearly outlined from diffuse WMH (if present). Figure 2 shows an example of a WMH track, located inferior from the index lacunar infarct. WMH tracks were round and small (often narrower than caps), were elongated (visible ≥2 MRI slices inferior to the index lacunar lesion), and followed the descending white matter tract, similar to Wallerian degeneration that is common after larger territorial infarcts. WMH tracks could be clearly outlined from extensive WMH (if present) and were not present at baseline. WMH progression was assessed by using a validated visual WMH change scale (modified Rotterdam progression scale16), not including the WMH cap or track in this assessment.

t is common after larger territorial infarcts. WMH tracks could be clearly outlined from extensive WMH (if present) and were not present at baseline. WMH progression was assessed by using a validated visual WMH change scale (modified Rotterdam progression scale16), not including the WMH cap or track in this assessment. Figure 1. The occurrence of a white matter hyperintensity (WMH) cap adjacent to index lacunar lesion. A, Baseline magnetic resonance imaging (MRI), performed 1 d after stroke onset, sporadic symptomatic lacunar infarct in the right internal capsule (white arrow; fluid-attenuated inversion recovery [FLAIR]). B, Baseline FLAIR image on MRI slice superior to the index lesion. C, Follow-up MRI at 1 y (353 d) after index stroke, partial cavitated lesion (lacey-like appearance) on FLAIR (white arrow). D, Follow-up FLAIR image on MRI slice superior to the cavitated index lesion, showing a WMH cap adjacent to the index lacunar lesion (black arrow). Figure 2. The occurrence of a white matter hyperintensity (WMH) track adjacent to index lacunar lesion. A and B, Baseline magnetic resonance imaging (MRI), performed 11 d after stroke onset, sporadic symptomatic lacunar infarct in the right pons (white arrow; fluid-attenuated inversion recovery; T2-weighted imaging). C, Follow-up MRI at 1 y (412 d) after index stroke, cavitated lacunar lesion in the right pons on T2-weighted imaging (white arrow). D, Follow-up MRI, WMH track proximal to index lacunar lesion in the right pons, visible on >2 MRI slices inferior to the index lacunar lesion (black arrow; T2-weighted imaging).

imaging). C, Follow-up MRI at 1 y (412 d) after index stroke, cavitated lacunar lesion in the right pons on T2-weighted imaging (white arrow). D, Follow-up MRI, WMH track proximal to index lacunar lesion in the right pons, visible on >2 MRI slices inferior to the index lacunar lesion (black arrow; T2-weighted imaging). Statistical Analysis Data differences between groups were tested using independent samples t test (normally distributed variables), Mann–Whitney U test (nonparametric data), and Pearson χ2 and Fisher test (categorical variables). Associations with occurrence of WMH caps and tracks were tested by univariable logistic regression analysis (including age, sex, vascular risk factors [hypertension, smoking, diabetes mellitus, and hypercholesterolemia], baseline MRI features [extensive basal ganglia perivascular spaces (grade 2–4), extensive WMH (Fazekas periventricular grade 3 or deep grade 2 or 3)], and follow-up MRI features [time between index stroke and follow-up MRI, and WMH progression], which are known variables associated with the presence and progression of WMH). Statistical significance was set at P<0.05 (2-tailed). Analyses were performed using SPSS statistical software package (SPSS, version 23.0; SPSS, Inc, Chicago, IL).

eatures [time between index stroke and follow-up MRI, and WMH progression], which are known variables associated with the presence and progression of WMH). Statistical significance was set at P<0.05 (2-tailed). Analyses were performed using SPSS statistical software package (SPSS, version 23.0; SPSS, Inc, Chicago, IL). Results Of 517 patients with stroke in the original studies, 169 (33%) had a lacunar stroke with an acute symptomatic (index) small subcortical (lacunar) infarct on MRI. Of these, 79 patients had a 1- to 5-year follow-up MRI and met the inclusion criteria for the present study. Recruitment details are shown in Figure 3. Table 1 shows baseline patient characteristics. The majority of patients were first ever strokes. Five (6%) patients had a history of transient ischemic attack, 5 (6%) had a history of an ischemic stroke, and 3 (4%) had both. Five (6%) patients had a recurrent stroke during follow-up: 3 were lacunar stroke (all in the contralateral hemisphere) and 2 were cortical stroke (1 in the ipsilateral hemisphere). Twenty-two (28%) patients had WMH progression (≥1 point on Rotterdam progression scale). Table 1. Baseline Characteristics Figure 3. Patients’ recruitment characteristics. FU indicates follow-up; MRI, magnetic resonance imaging; and MSS, Mild Stroke Study.

Results Of 517 patients with stroke in the original studies, 169 (33%) had a lacunar stroke with an acute symptomatic (index) small subcortical (lacunar) infarct on MRI. Of these, 79 patients had a 1- to 5-year follow-up MRI and met the inclusion criteria for the present study. Recruitment details are shown in Figure 3. Table 1 shows baseline patient characteristics. The majority of patients were first ever strokes. Five (6%) patients had a history of transient ischemic attack, 5 (6%) had a history of an ischemic stroke, and 3 (4%) had both. Five (6%) patients had a recurrent stroke during follow-up: 3 were lacunar stroke (all in the contralateral hemisphere) and 2 were cortical stroke (1 in the ipsilateral hemisphere). Twenty-two (28%) patients had WMH progression (≥1 point on Rotterdam progression scale). Table 1. Baseline Characteristics Figure 3. Patients’ recruitment characteristics. FU indicates follow-up; MRI, magnetic resonance imaging; and MSS, Mild Stroke Study. Evolution of the Index Lacunar Infarct on Follow-Up Imaging On follow-up imaging, some degree of cavitation was seen in 72 of 79 (91%) index lacunar infarcts, partial in 40 of 79 (51%), and complete in 32 of 79 (41%) index lacunar infarcts. Five of 79 (6%) index lesions had disappeared during follow-up, although one of these had developed a WMH cap (below). Two (3%) index lacunar lesions resembled a noncavitated WMH on follow-up imaging. No patient-related or imaging-related variables were associated with any degree of cavitation.

index lacunar infarcts. Five of 79 (6%) index lesions had disappeared during follow-up, although one of these had developed a WMH cap (below). Two (3%) index lacunar lesions resembled a noncavitated WMH on follow-up imaging. No patient-related or imaging-related variables were associated with any degree of cavitation. WMH Adjacent to Index Lacunar Infarct on Follow-Up Imaging We observed a new WMH adjacent to the index lacunar infarct in 42 of 79 (53%) patients: in 17 of 79 (22%) patients, the new WMH was superior (WMH cap); in 13 of 79 (16%) patients, the new WMH was inferior (WMH track); and in 12 of 79 (15%) patients, it was both superior and inferior from the index lacunar infarct. Table 2 shows characteristics of patients with WMH caps and tracks. WMH caps were most frequent on index lesions in the centrum semiovale (62%) but were also present in the internal and external capsule or nucleus lentiformis (24%), thalamus (10%), and brain stem (4%). We observed WMH tracks mostly in the centrum semiovale (64%) but also in the brain stem (16%), in the internal and external capsule and nucleus lentiformis (20%). Table 2. Occurrence of WMH Caps and Tracks Adjacent to Index Lacunar Lesions

WMH Adjacent to Index Lacunar Infarct on Follow-Up Imaging We observed a new WMH adjacent to the index lacunar infarct in 42 of 79 (53%) patients: in 17 of 79 (22%) patients, the new WMH was superior (WMH cap); in 13 of 79 (16%) patients, the new WMH was inferior (WMH track); and in 12 of 79 (15%) patients, it was both superior and inferior from the index lacunar infarct. Table 2 shows characteristics of patients with WMH caps and tracks. WMH caps were most frequent on index lesions in the centrum semiovale (62%) but were also present in the internal and external capsule or nucleus lentiformis (24%), thalamus (10%), and brain stem (4%). We observed WMH tracks mostly in the centrum semiovale (64%) but also in the brain stem (16%), in the internal and external capsule and nucleus lentiformis (20%). Table 2. Occurrence of WMH Caps and Tracks Adjacent to Index Lacunar Lesions WMH caps were associated with presence of extensive deep WMH at baseline (odds ratio, 3.31; 95% confidence interval, 1.27–8.59; P<0.05) and with overall WMH progression (odds ratio, 3.83; 95% confidence interval, 1.33–11.02; P<0.05). WMH tracks were associated with moderate–extensive basal ganglia perivascular spaces at baseline (odds ratio, 3.71; 95% confidence interval, 1.22–11.34; P<0.05). No other patient-related or imaging-related variables, including index lacunar infarct cavitation and time between baseline to follow-up imaging, were associated with occurrence of WMH caps or tracks.

xtensive basal ganglia perivascular spaces at baseline (odds ratio, 3.71; 95% confidence interval, 1.22–11.34; P<0.05). No other patient-related or imaging-related variables, including index lacunar infarct cavitation and time between baseline to follow-up imaging, were associated with occurrence of WMH caps or tracks. Discussion We assessed the long-term radiological evolution of sporadic symptomatic lacunar infarcts and perilesional white matter tissue on structural MRI. We demonstrate that, during follow-up of between 1 and 5 years, more than half of these sporadic symptomatic lacunar infarcts showed secondary perilesional morphological changes (WMH caps and tracks). WMH caps were associated with baseline deep WMH and WMH progression. Because severe WMH at baseline is the strongest predictor of WMH progression, similar risk factors might also be linked to the development of these WMH caps, although we could not find any association with conventional risk factors, such as age and hypertension. The present study confirms the results of a prior longitudinal MRI study that reported new WMH caps superior from sporadic symptomatic lacunar infarcts in 15 of 82 patients during 2-year follow-up.7

these WMH caps, although we could not find any association with conventional risk factors, such as age and hypertension. The present study confirms the results of a prior longitudinal MRI study that reported new WMH caps superior from sporadic symptomatic lacunar infarcts in 15 of 82 patients during 2-year follow-up.7 The new WMH tracks that appeared inferior from the index lacunar lesion seemed to be following a descending white matter tract similar to Wallerian degeneration. Wallerian degeneration of descending tracts is a well-known phenomenon after territorial stroke and reflects a pathological process with disintegration of axons, macrophage infiltration, degradation of myelin, and finally gliosis and atrophy of the affected tracts.17,18 Although we found that many index lacunar infarcts also underwent some degree of central cavitation, there was no association between progressive perilesional (WMH caps or tracks) and central lesion tissue damage.

crophage infiltration, degradation of myelin, and finally gliosis and atrophy of the affected tracts.17,18 Although we found that many index lacunar infarcts also underwent some degree of central cavitation, there was no association between progressive perilesional (WMH caps or tracks) and central lesion tissue damage. We suggest that secondary changes in white matter surrounding a sporadic index lacunar infarct should be considered a separate MRI feature of worsening brain damage and so disease progression. Their possible clinical and prognostic value requires further study. Wallerian degeneration has been associated with poor motor function recovery after territorial stroke.19 Longitudinal MRI studies with tractography have shown that subcortical infarcts in patients with CADASIL cause focal thinning in the remote cortex by degeneration of connected white matter tracts.9,20 Further, diffusion tensor imaging studies have shown secondary tract degeneration in white matter tracts, remote from sporadic symptomatic lacunar infarcts, the severity of which was independently related to worse cognitive functioning.8

hinning in the remote cortex by degeneration of connected white matter tracts.9,20 Further, diffusion tensor imaging studies have shown secondary tract degeneration in white matter tracts, remote from sporadic symptomatic lacunar infarcts, the severity of which was independently related to worse cognitive functioning.8 Several studies have assessed long-term cavitation of sporadic symptomatic lacunar infarcts.3–6 As in 2 former studies,5,6 we found a high partial or complete cavitation rate. Although some of the apparent difference in cavitation rates between studies likely reflects differences in definitions or interpretation of cavitation, the formation of even a partial cavity over time could reflect more interruption to white matter connectivity and might lead to worse clinical outcome. Therefore, it could be worthwhile to compare connectivity of white matter tracts (with diffusion tensor imaging) in cavitated versus noncavitated lesions. Some symptomatic lacunar lesions disappear, at least macroscopically, which could imply that conventional MRI underestimates the total cSVD-related brain damage.

me. Therefore, it could be worthwhile to compare connectivity of white matter tracts (with diffusion tensor imaging) in cavitated versus noncavitated lesions. Some symptomatic lacunar lesions disappear, at least macroscopically, which could imply that conventional MRI underestimates the total cSVD-related brain damage. The main strengths of our study are a relatively long follow-up time, standardized MRI protocols, and use of 1 carefully monitored MRI scanner. We used standardized international consensus criteria2 to describe cSVD imaging findings, and all assessments were blinded. However, our study also has limitations. Although this is one of the largest neuroimaging follow-up studies on this topic, our sample is relatively small for a common disease like lacunar stroke. The original studies only included patients with a (nondisabling) stroke, and for the present analysis, we selected patients with a follow-up MRI at 1 to 5 years after stroke for the original studies, both of which could have introduced bias. A few patients had clinically indicated MRI scans, which could also have led to a potential selection bias and affected the time to follow-up scanning. Three patients did not have a positive diffusion-weighted imaging, which could have caused that the wrong lesion was counted for the symptomatic lacunar infarct. However, we only included patients with a definitive clinical diagnosis of lacunar stroke, assessed by a panel of stroke experts, and the lesion had to be compatible with clinical signs. Our study describes imaging findings and is too small to correlate these with clinical features like cognition or stroke recovery. Further, we acknowledge the lack of statistical power to identify possible predictors of WMH caps and tracks. Larger prospective studies in an independent cohort are necessary to confirm our findings.

escribes imaging findings and is too small to correlate these with clinical features like cognition or stroke recovery. Further, we acknowledge the lack of statistical power to identify possible predictors of WMH caps and tracks. Larger prospective studies in an independent cohort are necessary to confirm our findings. Summary In conclusion, many sporadic symptomatic lacunar infarcts developed secondary changes in the adjacent inferior or superior white matter during follow-up and showed some degree of cavitation over time. Adjacent WMH caps and tracks may reflect another aspect of cSVD-related disease progression and neurodegeneration, with possible clinical and prognostic value. Larger prospective studies are necessary to confirm this hypothesis. Acknowledgments We thank the patients, their families, and the staff of the Brain Research Imaging Centre, Edinburgh, where magnetic resonance imaging scanning was performed.

Summary In conclusion, many sporadic symptomatic lacunar infarcts developed secondary changes in the adjacent inferior or superior white matter during follow-up and showed some degree of cavitation over time. Adjacent WMH caps and tracks may reflect another aspect of cSVD-related disease progression and neurodegeneration, with possible clinical and prognostic value. Larger prospective studies are necessary to confirm this hypothesis. Acknowledgments We thank the patients, their families, and the staff of the Brain Research Imaging Centre, Edinburgh, where magnetic resonance imaging scanning was performed. Sources of Funding The contributing studies were funded by the Chief Scientist Office of the Scottish Executive (grant 217 NTU R37933), the Wellcome Trust (grants 075611 and WT088134/Z/09/A), and Row Fogo Charitable Trust. The imaging was performed at the Brain Research Imaging Centre Edinburgh, which is supported by the SINAPSE (Scottish Imaging Network, A Platform for Scientific Excellence) collaboration and the Chief Scientist Office of the Scottish Government (http://www.bric.ed.ac.uk/). This work was supported by European Union Horizon 2020 (EU H2020), PHC- 03 to 15, project No. 666881, SVDs@Target, and the Fondation Leducq Transatlantic Network of Excellence for Study of Perivascular Spaces in Small Vessel Disease, ref No. 16 CVD 05. Dr Loos was supported by the Dutch Alzheimer Foundation. Disclosures None. Supplementary Material This work reflects the views of the authors and not of the funders.

Sources of Funding The contributing studies were funded by the Chief Scientist Office of the Scottish Executive (grant 217 NTU R37933), the Wellcome Trust (grants 075611 and WT088134/Z/09/A), and Row Fogo Charitable Trust. The imaging was performed at the Brain Research Imaging Centre Edinburgh, which is supported by the SINAPSE (Scottish Imaging Network, A Platform for Scientific Excellence) collaboration and the Chief Scientist Office of the Scottish Government (http://www.bric.ed.ac.uk/). This work was supported by European Union Horizon 2020 (EU H2020), PHC- 03 to 15, project No. 666881, SVDs@Target, and the Fondation Leducq Transatlantic Network of Excellence for Study of Perivascular Spaces in Small Vessel Disease, ref No. 16 CVD 05. Dr Loos was supported by the Dutch Alzheimer Foundation. Disclosures None. Supplementary Material This work reflects the views of the authors and not of the funders. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.020495/-/DC1.

Cerebral small vessel disease (CSVD) is an age-related disease affecting the small blood vessels of the brain.1,2 It accounts for at least 25% of all strokes and is the most common cause of vascular dementia.1,2 Several neuroimaging features are associated with CSVD, including lacunar infarcts, white matter hyperintensities (WMH), enlarged perivascular spaces, microbleeds, and brain atrophy. Intracerebral hemorrhages (ICH), particularly those arising from deep perforating small vessels, are also believed to be caused by CSVD.2 In addition, diffusion tensor imaging measures, such as fractional anisotropy (FA) and mean diffusivity (MD), are thought to capture microstructural changes of the white matter related to CSVD.3 However, the pathogenesis of CSVD is still uncertain,2 and consequently few effective and mechanism-based treatments for CSVD are available, aside from management of vascular risk factors of CSVD.2 Understanding which vascular risk factors are truly causal and improved understanding of how these pathological processes relate to different neuroimaging features of CSVD has the potential to improve treatment and prevention of CSVD.

ents for CSVD are available, aside from management of vascular risk factors of CSVD.2 Understanding which vascular risk factors are truly causal and improved understanding of how these pathological processes relate to different neuroimaging features of CSVD has the potential to improve treatment and prevention of CSVD. Type 2 diabetes mellitus (T2D) is an established risk factor for ischemic stroke and cognitive decline.4,5 Epidemiological studies have suggested that T2D is associated with lacunar stroke,6,7 but the relationships of T2D with WMH, ICH, and other radiological markers of CSVD have been inconsistent.4,6–9 Such studies are also limited by the study design, which can be confounded. Therefore, a definitive causal association between T2D and CSVD is yet to be established. In addition, few studies have specifically studied the relationship between higher insulin resistance and fasting glucose levels and risk of CSVD,10–12 leaving a gap in knowledge regarding whether either is the driving force behind increased risk of CSVD.

ausal association between T2D and CSVD is yet to be established. In addition, few studies have specifically studied the relationship between higher insulin resistance and fasting glucose levels and risk of CSVD,10–12 leaving a gap in knowledge regarding whether either is the driving force behind increased risk of CSVD. Mendelian randomization (MR), using genetic variants as instrumental variables, is a method that enables stronger claims to be made about the causality of risk factors in disease pathogenesis.13 It is based on the theory that genetic variants are randomly allocated at meiosis, similar to a randomized controlled trial.14 Therefore, genetic variants are independent of many other factors that bias observational studies, such as confounding and reverse causation. In the absence of pleiotropy, a significant association in an MR study between an exposure and outcome implies causality. In the present study, we aimed first to use MR to determine whether T2D is causally associated with clinical outcomes associated with CSVD; lacunar stroke and ICH; as well as intermediate radiological markers of CSVD; WMH, FA and MD. Second, we performed exploratory analyses investigating the relationship between higher insulin resistance and fasting glucose levels and risk of CSVD. All analyses are based on the aggregate effects of genetic variants rather than clinically diagnosed T2D. Methods The data that support the findings of this study are available from the corresponding author on reasonable request.

Mendelian randomization (MR), using genetic variants as instrumental variables, is a method that enables stronger claims to be made about the causality of risk factors in disease pathogenesis.13 It is based on the theory that genetic variants are randomly allocated at meiosis, similar to a randomized controlled trial.14 Therefore, genetic variants are independent of many other factors that bias observational studies, such as confounding and reverse causation. In the absence of pleiotropy, a significant association in an MR study between an exposure and outcome implies causality. In the present study, we aimed first to use MR to determine whether T2D is causally associated with clinical outcomes associated with CSVD; lacunar stroke and ICH; as well as intermediate radiological markers of CSVD; WMH, FA and MD. Second, we performed exploratory analyses investigating the relationship between higher insulin resistance and fasting glucose levels and risk of CSVD. All analyses are based on the aggregate effects of genetic variants rather than clinically diagnosed T2D. Methods The data that support the findings of this study are available from the corresponding author on reasonable request. Study Design, Data Sources, and Ethical Approval We performed a MR analysis, testing the causal relationship of T2D, fasting glucose, and fasting insulin with 5 manifestations of CSVD (magnetic resonance imaging–confirmed lacunar stroke, ICH [alone and stratified by the location of hemorrhage: deep ICH and lobar ICH], WMH, FA, and MD). Analyses of all CSVD phenotypes were based on subjects of European ancestry only.

ationship of T2D, fasting glucose, and fasting insulin with 5 manifestations of CSVD (magnetic resonance imaging–confirmed lacunar stroke, ICH [alone and stratified by the location of hemorrhage: deep ICH and lobar ICH], WMH, FA, and MD). Analyses of all CSVD phenotypes were based on subjects of European ancestry only. For ICH, we used a data set composed of 2254 cases and 8195 controls from 3 studies. One thousand five hundred forty-five cases and 1481 controls were from the Intracerebral Hemorrhage Genetics Collaboration and downloaded from the Cerebrovascular Disease Knowledge Portal (http://cerebrovascularportal.org).15 Five hundred seventy-five cases and 5750 propensity score–matched controls (matched on age, sex, and ancestry-informative principal components) were from UK Biobank, based on algorithmically defined ICH in White British subjects.16 One hundred thirty four cases and 964 controls were from the Cambridge ICH Genetics Study. For full details, Methods section in the online-only Data Supplement. The magnetic resonance imaging–confirmed lacunar stroke data were derived from 2191 lacunar stroke cases with magnetic resonance imaging confirmation from the SIGN-NINDS,17 WTCCC2,18 and DNA lacunar genome-wide association studies and 27 297 controls.19 All cases were obtained based on hospital admissions.

For ICH, we used a data set composed of 2254 cases and 8195 controls from 3 studies. One thousand five hundred forty-five cases and 1481 controls were from the Intracerebral Hemorrhage Genetics Collaboration and downloaded from the Cerebrovascular Disease Knowledge Portal (http://cerebrovascularportal.org).15 Five hundred seventy-five cases and 5750 propensity score–matched controls (matched on age, sex, and ancestry-informative principal components) were from UK Biobank, based on algorithmically defined ICH in White British subjects.16 One hundred thirty four cases and 964 controls were from the Cambridge ICH Genetics Study. For full details, Methods section in the online-only Data Supplement. The magnetic resonance imaging–confirmed lacunar stroke data were derived from 2191 lacunar stroke cases with magnetic resonance imaging confirmation from the SIGN-NINDS,17 WTCCC2,18 and DNA lacunar genome-wide association studies and 27 297 controls.19 All cases were obtained based on hospital admissions. Radiological markers of CSVD were derived from UK Biobank. Procedures for brain imaging acquisition and initial quality check have been described previously and are available on the UK Biobank website (Brain Imaging Documentation V1.3, http://www.ukbiobank.ac.uk).20 For FA and MD, we analyzed the first principal component of mean FA or MD values across 48 standard space tracts in 8357 individuals. WMH data were based on an analysis of 8429 subjects from UK Biobank. Additional details are provided in the Methods section in the online-only Data Supplement.

://www.ukbiobank.ac.uk).20 For FA and MD, we analyzed the first principal component of mean FA or MD values across 48 standard space tracts in 8357 individuals. WMH data were based on an analysis of 8429 subjects from UK Biobank. Additional details are provided in the Methods section in the online-only Data Supplement. Study characteristics for each of the data sets are provided in Table 1. Table 1. Cohort Characteristics An IRB or regional review board has approved the use of human subjects in each of the study populations. All patients gave informed consent. UK Biobank received ethical approval from the research ethics committee (REC reference 11/NW/0382). The present analyses were conducted under UK Biobank application number 19463. Single-Nucleotide Polymorphism Selection We selected single-nucleotide polymorphisms (SNPs) associated with T2D, fasting insulin, and fasting glucose, which reached genome-wide significance in the largest genome-wide association meta-analyses to date.21–23 Eighty-four SNPs were included for T2D, 36 for fasting glucose, and 18 for fasting insulin (Table I in the online-only Data Supplement). In instances where SNPs were not available in a data set because of poor imputation quality, we replaced them with proxy SNPs if available (r2>0.7). We included only 1 SNP from any associated locus. We verified that SNPs were uncorrelated by performing LD clumping (r2>0.1, 100 kb) using PLINK.24

-only Data Supplement). In instances where SNPs were not available in a data set because of poor imputation quality, we replaced them with proxy SNPs if available (r2>0.7). We included only 1 SNP from any associated locus. We verified that SNPs were uncorrelated by performing LD clumping (r2>0.1, 100 kb) using PLINK.24 Statistical Analyses We performed 2 complementary analyses to evaluate the impact of T2D-associated variants on CSVD phenotypes. For our primary analysis of the association of T2D with all CSVD phenotypes using 2-sample MR on summary statistics, we used a significance threshold of P<0.0071, equivalent to Bonferroni correction for 7 independent tests. For radiological markers of CSVD, we performed a confirmatory analysis using individual-level data in UK Biobank: Two-Sample MR Using Summary Statistics We assessed the impact of risk factor–associated SNPs on each CSVD phenotype using MR approaches. Our primary analysis used an inverse-variance weighted meta-analysis approach (conventional MR). We then performed secondary analyses using weighted median and penalized weighted median approaches. We assessed the potential role of directional pleiotropy by testing if the intercept from MR-Egger regression was significantly different from zero.

an inverse-variance weighted meta-analysis approach (conventional MR). We then performed secondary analyses using weighted median and penalized weighted median approaches. We assessed the potential role of directional pleiotropy by testing if the intercept from MR-Egger regression was significantly different from zero. As sensitivity analysis, we performed a look-up of all the SNPs used in our study in Phenoscanner (http://www.phenoscanner.medschl.cam.ac.uk/phenoscanner) to evaluate whether these SNPs were associated with other traits at genome-wide significance level which may affect our results. We found 8 SNPs related to T2D (rs2925979, rs2943640, rs3794991, rs3923113, rs429358, rs459193, rs635634, and rs780094), which were also associated with lipids and kidney function, as well as 4 SNPs for fasting glucose (rs174550, rs17762454, rs780094, and rs983309) and 9 SNPs for fasting insulin (rs10195252, rs1530559, rs2126259, rs2745353, rs2943645, rs3822072, rs459193, rs731839, and rs780094). We then reassessed the results after excluding these SNPs. Additionally, for significant findings, we assessed the potential that the association was mediated by body mass index (BMI) by performing a sensitivity analysis, removing 8 SNPs, which are also associated with BMI at genome-wide significance (rs2943640, rs11671664, rs12970134, rs8050136, rs10146997, rs5215, and rs7903146). All analyses were performed using the Mendelian Randomization and gtx libraries in R version 3.3.2 (https://www.R-project.org/).

ming a sensitivity analysis, removing 8 SNPs, which are also associated with BMI at genome-wide significance (rs2943640, rs11671664, rs12970134, rs8050136, rs10146997, rs5215, and rs7903146). All analyses were performed using the Mendelian Randomization and gtx libraries in R version 3.3.2 (https://www.R-project.org/). Two-Sample MR Using Genetic Risk Scores on Individual-Level Data For each of the 84 SNPs associated with T2D, we constructed a genetic risk score for each individual in UK Biobank by multiplying the log of the odds ratio (OR) for association with T2D by the number of risk alleles and summing this value for each individual. We then used a linear regression model to evaluate the effect of the genetic score on WMH, FA, and MD, with adjustment for genotyping batch, age, sex, BMI, blood pressure, and ancestry-informative principal components to control for these potential confounding factors. We also assessed their associations by constructing quartiles of the genetic score and calculated the ORs and 95% confidence interval (CI) for the score quartiles using quantile 1 as a reference, thus quantile 2 versus quantile 1, quantile 3 versus quantile 1, and quantile 4 versus quantile 1. Results MR Results: Associations of T2D With CSVD Phenotypes Conventional MR estimates for the effect of T2D on clinical outcomes associated with CSVD (lacunar stroke, ICH, deep ICH, and lobar ICH) and radiological markers of CSVD (WMH, FA, and MD) are displayed in the Figure and Table 2.

We also assessed their associations by constructing quartiles of the genetic score and calculated the ORs and 95% confidence interval (CI) for the score quartiles using quantile 1 as a reference, thus quantile 2 versus quantile 1, quantile 3 versus quantile 1, and quantile 4 versus quantile 1. Results MR Results: Associations of T2D With CSVD Phenotypes Conventional MR estimates for the effect of T2D on clinical outcomes associated with CSVD (lacunar stroke, ICH, deep ICH, and lobar ICH) and radiological markers of CSVD (WMH, FA, and MD) are displayed in the Figure and Table 2. Table 2. Mendelian Randomization Estimates for the Effect of T2D on CSVD Phenotypes Using Inverse-Variance Weighted, Weighted Median, and Penalized Weighted Median Methods Figure. Mendelian randomization estimates for the effect of type 2 diabetes mellitus (T2D) on cerebral small vessel disease (CSVD) phenotypes. Analyses were performed with conventional Mendelian randomization analysis (inverse-variance weighted method). CI indicates confidence interval; and OR, odds ratio.

Table 2. Mendelian Randomization Estimates for the Effect of T2D on CSVD Phenotypes Using Inverse-Variance Weighted, Weighted Median, and Penalized Weighted Median Methods Figure. Mendelian randomization estimates for the effect of type 2 diabetes mellitus (T2D) on cerebral small vessel disease (CSVD) phenotypes. Analyses were performed with conventional Mendelian randomization analysis (inverse-variance weighted method). CI indicates confidence interval; and OR, odds ratio. T2D was associated with higher risk of lacunar stroke (OR, 1.15; 95% CI, 1.04–1.28; P=0.007) and lower mean FA (OR, 0.78; 95% CI, 0.66–0.92; P=0.004). Conversely, we did not find any significant associations of T2D with WMH volume (OR, 1.01; 95% CI, 0.97–1.04; P=0.626) and higher mean MD (OR, 1.04; 95% CI, 0.89–1.23; P=0.612). Regarding the effect of T2D on ICH, lobar ICH, and deep ICH, none were found to be significant (ICH risk: OR, 1.07; 95% CI, 0.95–1.20; P=0.269; lobar ICH risk: OR, 1.07; 95% CI, 0.89–1.28; P=0.466; deep ICH risk: OR, 1.16; 95% CI, 0.99–1.36; P=0.074). Although nonsignificant, the odds ratio for association with deep ICH was near significant with an OR and direction of effect similar to that seen for lacunar stroke.

H risk: OR, 1.07; 95% CI, 0.95–1.20; P=0.269; lobar ICH risk: OR, 1.07; 95% CI, 0.89–1.28; P=0.466; deep ICH risk: OR, 1.16; 95% CI, 0.99–1.36; P=0.074). Although nonsignificant, the odds ratio for association with deep ICH was near significant with an OR and direction of effect similar to that seen for lacunar stroke. Weighted median and penalized median weighted analysis yielded similar effect estimates of T2D on lacunar stroke and FA, but the CIs were wide, so were not significant (Table 2). To assess the robustness and consistency of the results, we also conducted a sensitivity analysis by excluding the SNPs associated with lipids and kidney function at genome-wide significance level. In this sensitivity analysis, the associations of T2D with lacunar stroke and FA remained significant (Table III in the online-only Data Supplement). To assess whether the results were influence by BMI, we repeated the analyses for T2D and lacunar stroke as well as T2D and FA, both of which showed only minor differences after removing the SNPs (lacunar stroke: OR, 1.14; 95% CI, 1.01–1.29; P=0.028 and FA: OR, 0.78; 95% CI, 0.64–0.94; P=0.0095). MR-Egger regression showed no evidence of directional pleiotropy for the effects of T2D on lacunar stroke (intercept=0.008; P=0.515), ICH (intercept=0.009; P=0.528), deep ICH (intercept=0.022; P=0.218), lobar ICH (intercept=0.014; P=0.539), WMH (intercept=0.003; P=0.518), FA (intercept=−0.003; P=0.901), and MD (intercept=0.004; P=0.869).

showed no evidence of directional pleiotropy for the effects of T2D on lacunar stroke (intercept=0.008; P=0.515), ICH (intercept=0.009; P=0.528), deep ICH (intercept=0.022; P=0.218), lobar ICH (intercept=0.014; P=0.539), WMH (intercept=0.003; P=0.518), FA (intercept=−0.003; P=0.901), and MD (intercept=0.004; P=0.869). Genetic Risk Score Analyses: Effects of T2D on WMH, FA, and MD Results of the T2D genetic risk score on WMH, FA, and MD are presented in Table 3. The genetic score including 84 T2D-associated SNPs was significantly associated with FA (OR, 0.63; 95% CI, 0.45–0.89; P=0.008) after adjustment for genotyping batch, age, sex, BMI, blood pressure, and ancestry-informative principal components. We note that this result is significant when correcting for the 3 phenotypes studied in this secondary analysis but does not reach P<0.0071, the threshold used in our primary analysis. Conversely, the risk score was not significantly associated with WMH (OR, 1.01; 95% CI, 0.94–1.09; P=0.727) or MD (OR, 1.09; 95% CI, 0.78–1.52; P=0.613). Table 3. Association of the T2D-Related SNPs With WMH, FA, and MD in UK Biobank Using Linear Regression* Stratifying by quartiles of the genetic score, we only found a nominally significant effect of quantile 4 compared with quantile 1 of the score on FA (OR, 0.73; 95% CI, 0.57–0.94; P=0.015).Odds ratios±SE for WMH, FA, and MD based on the quartiles of the genetic score were plotted (Figures I through III in the online-only Data Supplement).

iles of the genetic score, we only found a nominally significant effect of quantile 4 compared with quantile 1 of the score on FA (OR, 0.73; 95% CI, 0.57–0.94; P=0.015).Odds ratios±SE for WMH, FA, and MD based on the quartiles of the genetic score were plotted (Figures I through III in the online-only Data Supplement). Associations of Fasting Glucose and Insulin With CSVD Phenotypes Fasting glucose and insulin were not associated with any CSVD phenotypes (lacunar stroke, ICH, deep ICH, lobar ICH, WMH, FA, and MD) in the present study using inverse-variance weighted, weighted median, or MR-Egger regression methods (Table II in the online-only Data Supplement).

ucose and Insulin With CSVD Phenotypes Fasting glucose and insulin were not associated with any CSVD phenotypes (lacunar stroke, ICH, deep ICH, lobar ICH, WMH, FA, and MD) in the present study using inverse-variance weighted, weighted median, or MR-Egger regression methods (Table II in the online-only Data Supplement). Discussion Prevention and management of CSVD is limited by our understanding of causal factors underlying the disease process. T2D is an established risk factor for CSVD, but a causal relationship is yet to be determined, and its impact on different CSVD phenotypes is not well understood. Using genetic data via an MR approach, we assessed the causal relationship between T2D and different CSVD phenotypes. Our primary analysis showed that a genetic predisposition to T2D was related to lacunar stroke and FA. We performed secondary analyses using an alternative weighted median method, which showed similar effects but wider CIs. Evidence indicating that the associations are robust is provided by the fact that (1) the results remained in sensitivity analyses removing pleiotropic SNPs and (2) associations with FA were significant when performing an alternative analysis based on individual-level data in UK Biobank, in which we were able to adjust for potential confounding factors.

ations are robust is provided by the fact that (1) the results remained in sensitivity analyses removing pleiotropic SNPs and (2) associations with FA were significant when performing an alternative analysis based on individual-level data in UK Biobank, in which we were able to adjust for potential confounding factors. The results were consistent with previous observational studies showing a positive association of T2D with risk of lacunar stroke6,7 and an MR analysis using small artery occlusion strokes based on the TOAST classification (Trial of ORG 10172 in Acute Stroke Treatment).25 However, the OR values of the relationship between T2D and lacunar stroke from a meta-analysis published in 20066 ranged from 1.3 to 2.2 in different cohorts, which were larger than our MR analyses (OR, 1.15; 95% CI, 1.04–1.28). This may be explained by the fact that previous observational epidemiological associations may have been influenced by potentially important confounders such as sex, blood pressure, and dietary factors. It should also be noted that the OR in our study refers to the increased risk associated with genetic markers related to T2D. Therefore, as these do not capture the total variance associated with T2D, the estimate is likely to be smaller than those from epidemiological studies.

sex, blood pressure, and dietary factors. It should also be noted that the OR in our study refers to the increased risk associated with genetic markers related to T2D. Therefore, as these do not capture the total variance associated with T2D, the estimate is likely to be smaller than those from epidemiological studies. Studies of the association between T2D and pathogenic subtypes of ICH are limited, but a recent meta-analysis9 included 19 case–control studies has reported that hemorrhagic stroke was 1.23-fold more prevalent in patients with diabetes mellitus, whereas the association was not observed in 3 population-based cohort studies.26–28 Our study using an MR approach showed no significant associations of T2D on total ICH risk, and on different location of ICH risk (lobar and deep ICH29,30). Although not significant, the effect of T2D on deep ICH was similar to that seen in lacunar stroke, which is consistent with the idea that deep ICH is associated with CSVD. However, we had limited power because of the small sample size for ICH, and this needs to be studied in larger data sets.

lobar and deep ICH29,30). Although not significant, the effect of T2D on deep ICH was similar to that seen in lacunar stroke, which is consistent with the idea that deep ICH is associated with CSVD. However, we had limited power because of the small sample size for ICH, and this needs to be studied in larger data sets. To better assess the causality of the associations between T2D and cerebral white matter injury, we used macrostructural (WMH volume) and microstructural (FA and MD) brain magnetic resonance imaging measures for the total brain. First, both MR approach and genetic score analyses did not show any significant associations between T2D and WMH volume, which was similar to the results from 2 recent reviews6,7 showing uncertainty for the effect of T2D on WMH. In addition, data from a randomized controlled trial study31 found that intensive control of diabetes mellitus (hemoglobin A1c <6.0%) did not impede WMH progression and conversely caused more serious WMH for unclear reasons.

milar to the results from 2 recent reviews6,7 showing uncertainty for the effect of T2D on WMH. In addition, data from a randomized controlled trial study31 found that intensive control of diabetes mellitus (hemoglobin A1c <6.0%) did not impede WMH progression and conversely caused more serious WMH for unclear reasons. Second, FA and MD are 2 commonly used diffusion tensor imaging indices, which are highly sensitive to subtle white matter changes.32 A few observational studies33–36 have shown a decreased FA and increased MD in T2D patients compared with controls, and the results were largely independent of WMH volume, which suggested that microstructural integrity damage likely precedes WMH, and diffusion tensor imaging indices maybe early markers for brain damage in patients with T2D. The results from the current study partly confirmed the findings from these aforementioned observational studies. We found the significant association between T2D and FA using MR methods, which also remained consistent in genetic score analyses. Surprisingly, T2D was not significantly associated with increased MD, and the reason for this discrepancy is unclear. It has been suggested that FA decrease maybe modulated more directly by myelin alterations, whereas MD is more sensitive to cellularity, edema, and necrosis.37 Our results might, therefore, point to demyelination being an important factor in T2D patients.

th increased MD, and the reason for this discrepancy is unclear. It has been suggested that FA decrease maybe modulated more directly by myelin alterations, whereas MD is more sensitive to cellularity, edema, and necrosis.37 Our results might, therefore, point to demyelination being an important factor in T2D patients. Strengths of our MR analysis include the use of multiple T2D SNPs, which enable us to construct a polygenic score to increase the precision of the estimates. In combination, the SNPs explained between 5% and 10% of the variance of T2D. Ours is the first MR study to investigate the relationship between T2D and CSVD phenotypes, and the design of MR study can prevent reverse causation and potential confounding factors,13,14 such as lifestyle and dietary preference. However, there are some limitations in our MR study. First, in some of our MR analyses, the sample size was relatively small, which resulted in limited statistical power in the respective analyses, especially for the ICH analysis stratified by hemorrhage location. Second, effects of the genetic variants on T2D were obtained largely from European populations, and all the subjects included in our study were Europeans. Therefore, the results may not be generalizable to other populations. Additionally, we note that differences in baseline characteristics between the T2D and CSVD populations might have subtle influences on the effect estimates, meaning that OR values given here should be interpreted with this caveat. Limitations of MR include the potential for residual pleiotropy that could have influenced the results when T2D-associated SNPs also influence other traits. We note that this is often challenging to rule out with absolute certainty.38 One possible pleiotropic pathway in this analysis is through elevated BMI. From the results presented here, we cannot rule out partial mediation of the genetic effects through this pathway. Finally, our significant association of T2D with lacunar stroke and FA was not confirmed after conducting weighted median and penalized weighted median methods. However, the weighted median approaches showed a similar effect size. MR studies with larger samples are, therefore, needed to confirm this result in the future.

nally, our significant association of T2D with lacunar stroke and FA was not confirmed after conducting weighted median and penalized weighted median methods. However, the weighted median approaches showed a similar effect size. MR studies with larger samples are, therefore, needed to confirm this result in the future. Conclusions Our MR study is consistent with a causal association between T2D and CSVD. In particular, we found evidence of associations with lacunar stroke and FA. Further MR studies with larger sample sizes are required to determine this with more certainty and to rule out associations with other CSVD phenotypes. Acknowledgments The Genetics of magnetic resonance imaging–confirmed lacunar stroke project contains samples derived from the SIGN-NINDS study, the WTCCC2 stroke study, and DNA lacunar. The SiGN study was funded by a cooperative agreement grant from the US National Institute of Neurological Disorders and Stroke, National Institutes of Health (U01 NS069208). Collection of the UK Young Lacunar Stroke DNA Study (DNA Lacunar) was primarily supported by the Wellcome Trust (WT072952) with additional support from the Stroke Association (TSA 2010/01). Genotyping of the DNA Lacunar samples was supported by a Stroke Association Grant (TSA 2013/01). The principal funding for the WTCCC2 stroke study was provided by the Wellcome Trust, as part of the Wellcome Trust Case Control Consortium 2 project (085475/B/08/Z and 085475/Z/08/Z and WT084724MA). The present analyses were conducted under UK Biobank application number 19463.

a Stroke Association Grant (TSA 2013/01). The principal funding for the WTCCC2 stroke study was provided by the Wellcome Trust, as part of the Wellcome Trust Case Control Consortium 2 project (085475/B/08/Z and 085475/Z/08/Z and WT084724MA). The present analyses were conducted under UK Biobank application number 19463. Sources of Funding Dr J. Liu was sponsored by China scholarship Council. Dr M. Liu is supported by Major International (Regional) Joint Research Project, National Natural Science Foundation of China (grant number: 81620108009). This work was supported by a British Heart Foundation Programme Grant (RG/16/4/32218). H. S. Markus is supported by a National Institute for Health Research (NIHR) Senior Investigator award, and his work is supported by the Cambridge Universities NIHR Comprehensive Biomedical Research Centre. Disclosures None. Supplementary Material The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.020536/-/DC1.

Cerebral small vessel disease (cSVD) is the major cause of vascular dementia and the pathology underlying a quarter of all strokes in the form of small subcortical (lacunar) strokes or intracerebral hemorrhage (ICH). Despite these numbers, the pathogenesis of cSVD is largely unknown, and this knowledge gap is a major factor behind the lack of specific therapies to delay cSVD progression.1,2 The most commonly studied marker of cSVD is white matter hyperintensities (WMHs) on magnetic resonance imaging (MRI). The prevalence and severity of WMH increases with advancing age and is associated with cardiovascular risk factors, stroke, dementia, and depression.3,4 However, studies suggest that microstructural tissue alterations underlying the cerebral white matter (fractional anisotropy [FA] and mean diffusivity [MD]) detected using diffusion tensor imaging (DTI) are more predictive of cognitive decline.5

ted with cardiovascular risk factors, stroke, dementia, and depression.3,4 However, studies suggest that microstructural tissue alterations underlying the cerebral white matter (fractional anisotropy [FA] and mean diffusivity [MD]) detected using diffusion tensor imaging (DTI) are more predictive of cognitive decline.5 Small vessels in the brain are difficult to investigate in vivo. Genetic studies provide a way to obtain novel insights in the disease mechanism underlying cSVD. Previous genome-wide association studies (GWAS) in population-based individuals and patients with stroke identified 12 genome-wide significant loci associated with WMH.6,7 However, there are only few published GWAS with a limited sample size that investigate the microstructural integrity of the white matter.8–10 The aims of this analysis were 2-fold: (1) to identify genetic variants associated with microstructural integrity of the white matter (FA and MD) in 8448 population-based individuals in UK Biobank and (2) to elucidate the relationships of FA, MD, and white matter hyperintensity volumes (WMHV) with clinical end points (eg, stroke, major depressive disorder, and Alzheimer disease). Methods The genetic and phenotypic UK Biobank data are available on application to the UK Biobank (http://www.ukbiobank.ac.uk/).

Small vessels in the brain are difficult to investigate in vivo. Genetic studies provide a way to obtain novel insights in the disease mechanism underlying cSVD. Previous genome-wide association studies (GWAS) in population-based individuals and patients with stroke identified 12 genome-wide significant loci associated with WMH.6,7 However, there are only few published GWAS with a limited sample size that investigate the microstructural integrity of the white matter.8–10 The aims of this analysis were 2-fold: (1) to identify genetic variants associated with microstructural integrity of the white matter (FA and MD) in 8448 population-based individuals in UK Biobank and (2) to elucidate the relationships of FA, MD, and white matter hyperintensity volumes (WMHV) with clinical end points (eg, stroke, major depressive disorder, and Alzheimer disease). Methods The genetic and phenotypic UK Biobank data are available on application to the UK Biobank (http://www.ukbiobank.ac.uk/). Study Population and Ethical Approval UK Biobank is a prospective study that recruited 502 620 community-dwelling participants from across the United Kingdom between 2006 and 2010, aged 40 to 69 years (http://www.ukbiobank.ac.uk). The study collects extensive data from questionnaires, interviews, health records, physical measures, biological samples, and imaging.

ank is a prospective study that recruited 502 620 community-dwelling participants from across the United Kingdom between 2006 and 2010, aged 40 to 69 years (http://www.ukbiobank.ac.uk). The study collects extensive data from questionnaires, interviews, health records, physical measures, biological samples, and imaging. A subset of the participants also underwent brain MRI. In the present study, we used the second release of MRI data, which included 9066 subjects who underwent brain MRI, on average 6.6 years (SD, 1.0 years) after initial recruitment at mean age 55.5 years (SD, 7.4 years) and had usable T2-weighted fluid-attenuated inversion recovery or DTI images. Patients with a baseline diagnosis of stroke, multiple sclerosis, Parkinson disease, dementia, any other neurodegenerative problem (InternationalClassification of Diseases, Ninth Revision/Tenth Revision, or self-report or health-record linkage) or no genetic data were excluded (Table I in the online-only Data Supplement). Participants with consistently extreme outlying tract-averaged water diffusion biomarker values were removed after visual inspection of the data by the authors. UK Biobank received ethical approval from the research ethics committee (reference 11/NW/0382). All participants provided informed consent to participate. The present analyses were conducted under UK Biobank application number 19463.

A subset of the participants also underwent brain MRI. In the present study, we used the second release of MRI data, which included 9066 subjects who underwent brain MRI, on average 6.6 years (SD, 1.0 years) after initial recruitment at mean age 55.5 years (SD, 7.4 years) and had usable T2-weighted fluid-attenuated inversion recovery or DTI images. Patients with a baseline diagnosis of stroke, multiple sclerosis, Parkinson disease, dementia, any other neurodegenerative problem (InternationalClassification of Diseases, Ninth Revision/Tenth Revision, or self-report or health-record linkage) or no genetic data were excluded (Table I in the online-only Data Supplement). Participants with consistently extreme outlying tract-averaged water diffusion biomarker values were removed after visual inspection of the data by the authors. UK Biobank received ethical approval from the research ethics committee (reference 11/NW/0382). All participants provided informed consent to participate. The present analyses were conducted under UK Biobank application number 19463. Magnetic Resonance Imaging Procedures for brain imaging acquisition and initial quality check have been described previously and are available on the UK Biobank website (Brain Imaging Documentation V1.3; http://www.ukbiobank.ac.uk).11

UK Biobank received ethical approval from the research ethics committee (reference 11/NW/0382). All participants provided informed consent to participate. The present analyses were conducted under UK Biobank application number 19463. Magnetic Resonance Imaging Procedures for brain imaging acquisition and initial quality check have been described previously and are available on the UK Biobank website (Brain Imaging Documentation V1.3; http://www.ukbiobank.ac.uk).11 In brief, all brain MRI data were acquired on a single standard Siemens Skyra 3T scanner (Siemens Medical Solutions, Germany) using the standard Siemens 32-channel radiofrequency receiver head coil. Sagittal T1-weighted scans were acquired using a 3-dimensional magnetization-prepared rapid acquisition gradient-echo sequence (resolution, 1×1×1 mm; field of view, 208×256×256; inversion time/repetition time=880/2000 ms). Sagittal T2-weighted fluid-attenuated inversion recovery scans were obtained using a 3-dimensional SPACE sequence (resolution, 1.05×1.0×1.0 mm; field of view, 192×256×256; inversion time/repetition time=1800/5000 ms). DTI scans were acquired with a spin-echo echo-planar imaging sequence and multishell acquisition (b0=0 s/mm−2, b=1000 s/mm−2, and b=2000 s/mm−2; 100 distinct diffusion-encoding directions [50 in each shell]; 2-mm isotropic voxels; field of view, 104×104×72).

w, 192×256×256; inversion time/repetition time=1800/5000 ms). DTI scans were acquired with a spin-echo echo-planar imaging sequence and multishell acquisition (b0=0 s/mm−2, b=1000 s/mm−2, and b=2000 s/mm−2; 100 distinct diffusion-encoding directions [50 in each shell]; 2-mm isotropic voxels; field of view, 104×104×72). White Matter Hyperintensities WMHs were automatically segmented using the combined T1 and T2-weighted fluid-attenuated inversion recovery data as input in the Brain Intensity Abnormality Classification Algorithm tool.12 Brain Intensity Abnormality Classification Algorithm is a fully automated supervised method for WMH detection, based on the k-nearest neighbor algorithm, which gives the probability per voxel of being WMH. The total WMHV was calculated from the voxels exceeding a probability of 0.9 of being WMH and located within a white matter mask. Obtained values were adjusted for the total intracranial volume and log transformed because of their skewed distribution.

ghbor algorithm, which gives the probability per voxel of being WMH. The total WMHV was calculated from the voxels exceeding a probability of 0.9 of being WMH and located within a white matter mask. Obtained values were adjusted for the total intracranial volume and log transformed because of their skewed distribution. FA and MD After gradient distortion correction and further correction for head movement and eddy currents, diffusion tensors and scalar diffusion parameters (ie, FA and MD) were calculated using the b=1000 shell (50 directions) and DTIFIT from the FSL software.13 The FA maps produced were then fed into tract-based spatial statistics processing, which aligns the FA map onto a standard-space FA template; in this work, the standard FMRIMB_158_FA template was used as the target image. The standard space FA subject images were then skeletonized and the MD (and other DTI output) maps were projected onto the subject skeleton, using the FA-derived alignment parameters. A set of 48 standard space tracts have been defined previously,14 these are then used as masks to generate tract-specific masks from the skeletonized images. These are then used to produce a mean FA or MD from each tract for each subject. This is similar to the processing applied in the ENIGMA project (http://enigma.ini.usc.edu/protocols/dti-protocols).14,15 Principal component analysis was applied on the 48 tracts to extract a latent measure. The first principal component of FA (FA.PC1) and MD (MD.PC1) was used in subsequent analyses as dependent variable.

FA and MD After gradient distortion correction and further correction for head movement and eddy currents, diffusion tensors and scalar diffusion parameters (ie, FA and MD) were calculated using the b=1000 shell (50 directions) and DTIFIT from the FSL software.13 The FA maps produced were then fed into tract-based spatial statistics processing, which aligns the FA map onto a standard-space FA template; in this work, the standard FMRIMB_158_FA template was used as the target image. The standard space FA subject images were then skeletonized and the MD (and other DTI output) maps were projected onto the subject skeleton, using the FA-derived alignment parameters. A set of 48 standard space tracts have been defined previously,14 these are then used as masks to generate tract-specific masks from the skeletonized images. These are then used to produce a mean FA or MD from each tract for each subject. This is similar to the processing applied in the ENIGMA project (http://enigma.ini.usc.edu/protocols/dti-protocols).14,15 Principal component analysis was applied on the 48 tracts to extract a latent measure. The first principal component of FA (FA.PC1) and MD (MD.PC1) was used in subsequent analyses as dependent variable. Genetic Data We used the June 2017 release of the imputed genetic data from UK Biobank (downloaded on June 3, 2017). Details of the design of the arrays, sample processing, and stringent quality control have been described elsewhere.16 In brief, 2 closely related arrays from Affymetrix, the UK BiLEVE Axiom array (9.9% of individuals), and the UK Biobank Axiom array were used to genotype ≈805 426 markers with good genome-wide coverage. Phasing was performed using SHAPEIT3 and imputation to a merged Haplotype Reference Consortium reference panel (39 131 578 autosomal single-nucleotide polymorphisms [SNPs]) and UK10K and 1000 Genomes Phase 3 panel was performed using the IMPUTE4 package.16–18 Imputed genotypes were available for 487 442 individuals in this study.16 From the resulting data set, we excluded (1) individuals who did not segregate with European samples based on principal component analysis, (2) individuals with high-level heterozygosity and missingness (>5%), and (3) individuals whose reported sex was inconsistent with sex inferred from the genetic data. In addition, only SNPs imputed from the HRC panel were included in this analysis.

segregate with European samples based on principal component analysis, (2) individuals with high-level heterozygosity and missingness (>5%), and (3) individuals whose reported sex was inconsistent with sex inferred from the genetic data. In addition, only SNPs imputed from the HRC panel were included in this analysis. Statistical Analysis In this analysis, we first subset the genetic data on the individuals who also had MRI imaging data. We performed a GWAS of FA, MD, and log (WMHV), using SNPTEST v2.5.4-beta3, including age at MRI, sex, genotyping batch, and the first 10 ancestry informative principal components as covariates. We set the study-wide significance threshold at P<1.7e-8, accounting for the 3 phenotypes studied. At this threshold, we had 80% power to identify variants explaining >0.5% of the trait variance. Fine-Mapping Derived From Credible SNP Set Analyses For all SNPs in linkage disequilibrium (LD) with the lead SNP (r2>0.1), we calculated Bayes factors from the effect sizes and SEs using Wakefield approximation.19 We then used these Bayes factors to calculate the posterior probability that each variant is causal and the 95% credible set for each association (the smallest set of variants with posteriors that sum to at least 95%) as described in the study by Maller et al.20 To identify additional independent signals at genome-wide significant loci we performed a forward stepwise regression using SNPTEST.

Fine-Mapping Derived From Credible SNP Set Analyses For all SNPs in linkage disequilibrium (LD) with the lead SNP (r2>0.1), we calculated Bayes factors from the effect sizes and SEs using Wakefield approximation.19 We then used these Bayes factors to calculate the posterior probability that each variant is causal and the 95% credible set for each association (the smallest set of variants with posteriors that sum to at least 95%) as described in the study by Maller et al.20 To identify additional independent signals at genome-wide significant loci we performed a forward stepwise regression using SNPTEST. Functional Annotation To evaluate whether the genome-wide significant variants potentially influence gene expression, we examined genome-wide cis-expression quantitative trait loci (eQTL) data in multiple tissues from 3 major eQTL databases: the Blood eQTL Browser,21 the Genotype-Tissue Expression Project,22 and the Brain eQTL Almanac.23 We performed a lookup of the genome-wide significant SNPs in available GWAS summary statistics of relevant clinical end points (Alzheimer disease,24 major depressive disorder,25 ICH,26 and MRI-confirmed lacunar stroke27), of which details are provided in Table II in the online-only Data Supplement and Methods in the online-only Data Supplement.

he genome-wide significant SNPs in available GWAS summary statistics of relevant clinical end points (Alzheimer disease,24 major depressive disorder,25 ICH,26 and MRI-confirmed lacunar stroke27), of which details are provided in Table II in the online-only Data Supplement and Methods in the online-only Data Supplement. Heritability and Genetic Correlations With Related Traits We estimated the heritability of FA, MD, and WMHV and tested for genetic correlation between the white matter measures using LD score regression.28 Subsequently, we estimated the genetic correlation between the white matter measures and 4 clinical end points: Alzheimer disease, major depressive disorder, ICH-, and MRI-confirmed lacunar stroke. Concerning stroke subtypes, specifically ICH- and MRI-confirmed lacunar stroke, were selected because small vessel disease is presumed to be the most important in these stroke subtypes. These analyses were based on genome-wide summary statistics obtained from online repositories and locally available data (Table III in the online-only Data Supplement).

ically ICH- and MRI-confirmed lacunar stroke, were selected because small vessel disease is presumed to be the most important in these stroke subtypes. These analyses were based on genome-wide summary statistics obtained from online repositories and locally available data (Table III in the online-only Data Supplement). Polygenic Association of FA, MD, and WMHV With Clinical End Points We derived polygenic risk scores at 3 different levels of significance (P<0.0001, P<0.05, P<0.5) from the GWAS data of FA, MD, and WMHV and tested them for association with Alzheimer disease, major depressive disorder, ICH-, and MRI-confirmed lacunar stroke using an inverse-variance weighted method using summary statistics data. We derived an independent (r2 <0.1 or 500 Kb apart) set of SNPs at each threshold using an LD-clumping procedure used using plink v1.90b3.45. Risk score analysis was performed in R using the gtx package. Results In total, 8448 individuals were included in the present analysis (Figure I in the online-only Data Supplement). At the time of the MRI scan, mean age was 62.2 (7.4) years. Among the individuals who passed genetic quality control checks, FA and MD measures could be calculated in 8239 individuals, and WMHV were available in 8429. The 3 measures were heritable (h2=0.29 for FA, h2=0.17 for MD, and h2=0.18 for WMHV) and showed high phenotypic and genetic correlation (Table 1). Table 1. Estimated Heritability, Phenotypic (Below Diagonal), and Genetic (Above Diagonal) Correlations Between the UK Biobank Magnetic Resonance Imaging Variables

At the time of the MRI scan, mean age was 62.2 (7.4) years. Among the individuals who passed genetic quality control checks, FA and MD measures could be calculated in 8239 individuals, and WMHV were available in 8429. The 3 measures were heritable (h2=0.29 for FA, h2=0.17 for MD, and h2=0.18 for WMHV) and showed high phenotypic and genetic correlation (Table 1). Table 1. Estimated Heritability, Phenotypic (Below Diagonal), and Genetic (Above Diagonal) Correlations Between the UK Biobank Magnetic Resonance Imaging Variables Genome-Wide Association Analysis of FA and MD Table 2 shows the genome-wide significant loci (P<1.7×10−8) for FA, MD, and WMHV with their corresponding alleles and effect sizes. The inflation of test statistics (λ) was equal to the inflation expected for the sample size. Manhattan plots and Q-Q plots are displayed in Figure 1 and Figure II in the online-only Data Supplement, respectively. Regional plots for the genome-wide significant hits for FA, MD, and WMHV are provided in Figures III, IV, and V in the online-only Data Supplement. Table 2. Genome-Wide Significant Loci Associated With FA, MD, and WMHV Figure 1. Association of genome-wide single-nucleotide polymorphisms with fractional anisotropy (FA; A) and mean diffusivity (MD; B) by genomic position. Association results for the genome-wide association analysis for FA (A), MD (B), and white matter hyperintensity volumes (WMHV; C). The dashed line marks the threshold of statistical significance (P=1.7×10−8).

-nucleotide polymorphisms with fractional anisotropy (FA; A) and mean diffusivity (MD; B) by genomic position. Association results for the genome-wide association analysis for FA (A), MD (B), and white matter hyperintensity volumes (WMHV; C). The dashed line marks the threshold of statistical significance (P=1.7×10−8). We identified 1 region harboring genome-wide significant SNPs for FA or MD. This locus, which was shared between FA and MD, was on chr5q14 (top SNP for FA: rs67827860, P=1.3×10−14; top SNP for MD: rs13164785, P=3.7×10−18; LD r2=1 for these 2 SNPs). We constructed 95% credible sets for the chr5q14 locus, which return the set of SNPs in which the causal SNP is contained with 95% certainty. The 95% credible set for the chr5q14 association contained 6 SNPs (rs13164785, rs67827860, rs10052710, rs17205972, rs12653308, and rs3852188). To identify additional independent signals at the genome-wide significant loci, we performed a forward stepwise regression using SNPTEST on both regions, which yielded a second independent genome-wide signal for both FA and MD in the chr5q14 region (Figures VI and VII in the online-only Data Supplement). Results of the joint model containing both SNPs are given in Table IV in the online-only Data Supplement. The chr5q14 locus contained 33 genome-wide SNPs for FA and 116 genome-wide SNPs for MD, of which 31 overlapped. The top SNP at the chr5q14 locus maps to an intron of the gene VCAN.

To identify additional independent signals at the genome-wide significant loci, we performed a forward stepwise regression using SNPTEST on both regions, which yielded a second independent genome-wide signal for both FA and MD in the chr5q14 region (Figures VI and VII in the online-only Data Supplement). Results of the joint model containing both SNPs are given in Table IV in the online-only Data Supplement. The chr5q14 locus contained 33 genome-wide SNPs for FA and 116 genome-wide SNPs for MD, of which 31 overlapped. The top SNP at the chr5q14 locus maps to an intron of the gene VCAN. Functional annotation of the genome-wide significant SNPs in the chr5q14 region using the Genotype-Tissue Expression and Brain eQTL Almanac resources identified no significant eQTLs at genome-wide significance. However, there was significant cis association with expression of VCAN in blood (rs3852188; P=1.65×10−5). A nominally significant SNP for the chr5q14 locus (rs3852188; P=0.01) was found in the MRI-confirmed lacunar stroke data set, but the effect was in the opposite direction. No other significant associations for any of the traits were found. Two genome-wide significant loci were identified for WMHV located at 2p16 (top SNP rs146896516; P=3.91×10−12) and 17q25 (top SNP rs3744020; P=3.52×10−11; Table 2). Both loci are known WMH loci, reported previously in population-based and stroke population.

A nominally significant SNP for the chr5q14 locus (rs3852188; P=0.01) was found in the MRI-confirmed lacunar stroke data set, but the effect was in the opposite direction. No other significant associations for any of the traits were found. Two genome-wide significant loci were identified for WMHV located at 2p16 (top SNP rs146896516; P=3.91×10−12) and 17q25 (top SNP rs3744020; P=3.52×10−11; Table 2). Both loci are known WMH loci, reported previously in population-based and stroke population. Both SNPs showed the expected direction of effect in the association with both FA and MD, but only rs146896516 was nominally associated with MD (Table V in the online-only Data Supplement). The top SNPs for MD and FA at the chr5q14 locus were nominally associated with WMHV (P=3.51×10−6 and P=2.95×10−6, respectively). Genetic Correlation With Related Traits Using LD Score Regression Genetic correlation between the MRI phenotypes and related phenotypes was estimated using LD score regression. The Bonferroni-adjusted significance threshold for this analysis was set at P=0.004. There was significant positive genetic correlation between MRI-confirmed lacunar stroke and MD (rG [SE]=0.71 [0.22]; P=0.0013), FA (rG [SE]=0.52 [0.17]; P=0.0028), as well as WMHV (rG [SE]=0.90 [0.27]; P=0.0004). All other correlations were not significant (Figure 2).

nce threshold for this analysis was set at P=0.004. There was significant positive genetic correlation between MRI-confirmed lacunar stroke and MD (rG [SE]=0.71 [0.22]; P=0.0013), FA (rG [SE]=0.52 [0.17]; P=0.0028), as well as WMHV (rG [SE]=0.90 [0.27]; P=0.0004). All other correlations were not significant (Figure 2). Figure 2. LD score regression results of white matter measures and related clinical phenotypes. Shared genetic contribution between the white matter phenotypes and related clinical phenotypes as determined LD score regression analysis. Genetic correlation (rG) and 95% confidence intervals are shown. The correlation estimates for fractional anisotropy (FA) have been inverted to show trait raising risk (ie, a positive rG means that the other trait is positively associated with a reduction in FA). MD indicates mean diffusivity; and WMHV, white matter hyperintensity volume. Polygenic Association of FA and MD With Clinical End Points We further tested FA, MD, and WMHV polygenic risk scores (P<0.0001, P<0.05, and P<0.5) for association with clinical end points. There was a polygenic association of all 3 MRI traits with MRI-confirmed lacunar stroke, with a similar association across all traits (Figure 3). In addition, MD polygenic risk scores were significantly associated with major depressive disorder. WMHV polygenic risk scores showed a significant association with Alzheimer disease.

a polygenic association of all 3 MRI traits with MRI-confirmed lacunar stroke, with a similar association across all traits (Figure 3). In addition, MD polygenic risk scores were significantly associated with major depressive disorder. WMHV polygenic risk scores showed a significant association with Alzheimer disease. Figure 3. Association between polygenic risk scores of white matter measures and clinical end points. AD indicates Alzheimer disease; FA, fractional anisotropy; ICH, intracerebral hemorrhage; MD, mean diffusivity; MDD, major depressive disorder; SVS, magnetic resonance imaging–confirmed lacunar stroke; and WMHV, white matter hyperintensity volume. Discussion We performed a GWAS of brain white matter microstructural integrity as assessed on DTI, identifying a novel genome-wide significant locus on chr5q14. We found strong evidence for a shared genetic component between FA, MD, and WMHV. Furthermore, we demonstrated that genetic variants that influence all white matter measures studied confer risk of lacunar stroke, as well as demonstrating associations between MD—a marker of white matter ultrastructural damage and major depressive disorder—and between Alzheimer disease and WMHV.

ween FA, MD, and WMHV. Furthermore, we demonstrated that genetic variants that influence all white matter measures studied confer risk of lacunar stroke, as well as demonstrating associations between MD—a marker of white matter ultrastructural damage and major depressive disorder—and between Alzheimer disease and WMHV. The top SNP within the chr5q14 locus maps to an intron of VCAN, which encodes the extracellular matrix proteoglycan VCAN. The top SNPs were only significantly associated with expression of VCAN in blood. VCAN is a versatile protein, which plays a role in intercellular signaling and in connecting cells with the extracellular matrix.29 Furthermore, VCAN may play a role in the regulation of cell motility, growth, and differentiation. SNPs in chr5q14 have been linked to intracranial aneurysm of which one of the lead SNPs in the previous study is among our genome-wide significant SNPs (rs173686; P=3.32×10−10).30 Furthermore, mutations in chr5q14 have been linked to Wagner disease—a rare vitreoretinal degeneration inherited as an autosomal dominant trait. VCAN is a component of the vitreous and likely involved in the maintenance and structural integrity of the vitreous.31 Mutations linked to Wagner disease lead to alternative splicing of exons 7 and 8 of VCAN. In the present study, the lead SNPs in chr5q14 were inversely associated with small vessel stroke, and, therefore, our study does not provide support for any vascular-mediated mechanism underlying the association of VCAN with FA and MD.

Furthermore, mutations in chr5q14 have been linked to Wagner disease—a rare vitreoretinal degeneration inherited as an autosomal dominant trait. VCAN is a component of the vitreous and likely involved in the maintenance and structural integrity of the vitreous.31 Mutations linked to Wagner disease lead to alternative splicing of exons 7 and 8 of VCAN. In the present study, the lead SNPs in chr5q14 were inversely associated with small vessel stroke, and, therefore, our study does not provide support for any vascular-mediated mechanism underlying the association of VCAN with FA and MD. The consistent correlation between MD, FA, WMHV, and MRI-confirmed lacunar stroke further contributes to the evidence that there is a shared polygenic component in the disease mechanisms of small vessel disease and alterations of the white matter.

In the present study, the lead SNPs in chr5q14 were inversely associated with small vessel stroke, and, therefore, our study does not provide support for any vascular-mediated mechanism underlying the association of VCAN with FA and MD. The consistent correlation between MD, FA, WMHV, and MRI-confirmed lacunar stroke further contributes to the evidence that there is a shared polygenic component in the disease mechanisms of small vessel disease and alterations of the white matter. In addition, there was an association between MD and major depressive disorder in the polygenic risk score analysis. This finding is in line with previous imaging studies that found a direct association between reduced structural integrity and major depressive disorder.32 The correlation between MD and major depressive disorder did not reach genome-wide significance in the LD score regression, although the effect was in the same direction. A possible explanation for this discrepancy is that even though the current study is the largest study published on genetics of white matter microstructural integrity to date, the sample size is still relatively small for GWAS. The cases contributing to the major depressive disorder had a broad range of ages of disease onset. At this point, it is not possible to determine whether the shared genetic component with MD is because of age-related changes or otherwise. This will be an important point of further enquiry.

l relatively small for GWAS. The cases contributing to the major depressive disorder had a broad range of ages of disease onset. At this point, it is not possible to determine whether the shared genetic component with MD is because of age-related changes or otherwise. This will be an important point of further enquiry. In the current analysis, Alzheimer disease was significantly associated with WMHV in the polygenic risk score analysis but not with the white matter microstructural integrity measures. Furthermore, the top SNPs for WMHV were not statistically significant associated with FA and MD, although the effect estimates were in the expected directions. An explanation for the discrepancies might be that there are differences in the information FA, MD, and WMHV capture. FA and MD in the present study represent microstructural integrity within both normal-appearing white matter, and WMH and microstructural integrity is lower in the latter.33 Decreased microstructural integrity is known to precede WMH.34 Therefore, the WMH assessed in this relatively healthy population may reflect more severe disease than the microstructural integrity. On the other hand, pathological studies have shown that WMHs are heterogenic, that is, they represent different degrees of tissue damage, but this is not captured by the WMHV measure.35 Another explanation for the discrepancies might be the lack of statistical power because of a combination of only mild detectable disease in this relatively young and healthy population and a sample size that is still small for a GWAS.

different degrees of tissue damage, but this is not captured by the WMHV measure.35 Another explanation for the discrepancies might be the lack of statistical power because of a combination of only mild detectable disease in this relatively young and healthy population and a sample size that is still small for a GWAS. Strengths of the present study include the large sample size with high-quality standardized research MRI scans. This present study is limited to white participants of European genetic descent because only a small fraction (2%–3%) of the participants were of non-European descent. Thus, the results may not be applicable to other populations. Another limitation is that we were not able to replicate our findings in independent samples because unfortunately no large-scale replication resources are currently available. We identified a novel locus that is genome-wide associated with microstructural integrity of the white matter in the brain measured as FA and MD. The results contribute to the growing evidence that mechanisms underlying white matter alterations are shared with cerebrovascular disease and highlight that inherited differences in white matter microstructure, possibly age related, impact on multiple diseases.

y of the white matter in the brain measured as FA and MD. The results contribute to the growing evidence that mechanisms underlying white matter alterations are shared with cerebrovascular disease and highlight that inherited differences in white matter microstructure, possibly age related, impact on multiple diseases. Acknowledgments We thank the International Genomics of Alzheimer Project (IGAP), the Psychiatric Genomics Consortium, the genetics of magnetic resonance imaging–confirmed lacunar stroke collaboration, and the Intracerebral Hemorrhage Genetics collaboration for providing summary results data for these analyses. The investigators within IGAP contributed to the design and implementation of IGAP and provided data but did not participate in the analysis or writing of this report. IGAP was made possible by the generous participation of the control subjects, the patients, and their families.

summary results data for these analyses. The investigators within IGAP contributed to the design and implementation of IGAP and provided data but did not participate in the analysis or writing of this report. IGAP was made possible by the generous participation of the control subjects, the patients, and their families. Sources of Funding This project has received funding from the European Union Horizon 2020 research and innovation program under grant agreement number 667375. This work was, in part, supported by a British Heart Foundation Programme Grant (RG/16/4/32218). Dr Rutten-Jacobs was supported by a British Heart Foundation Immediate Research Fellowship (FS/15/61/31626). Dr Tozer is supported by the Cambridge Universities National Institute for Health Research (NIHR) Comprehensive Biomedical Research Centre. Dr Dichgans received funding from the European Union Horizon 2020 research and innovation program under grant agreement number 666881 (SVDs@target) and from the German Research Foundation through the Collaborative Research Centres 1123 (B3) and the Munich Cluster for Systems Neurology (EXC 1010 SyNergy). H.S. Markus is supported by an NIHR Senior Investigator award, and his work is supported by the Cambridge Universities NIHR Comprehensive Biomedical Research Centre. International Genomics of Alzheimer Project: the i–Select chips was funded by the French National Foundation on Alzheimer disease and related disorders. European Alzheimer Disease Initiative was supported by the LABEX (laboratory of excellence program investment for the future) Development of Innovative Strategies for a Transdisciplinary Approach to Alzheimer's Disease (DISTALZ) grant, Inserm, Institut Pasteur de Lille, Université de Lille 2, and the Lille University Hospital. Genetic and Environmental Risk in Alzheimer's Disease (GERAD) was supported by the Medical Research Council (grant No. 503480), Alzheimer Research UK (grant No. 503176), the Wellcome Trust (grant No. 082604/2/07/Z), the and German Federal Ministry of Education and Research: Competence Network Dementia grant numbers 01GI0102, 01GI0711, and 01GI0420.

in Alzheimer's Disease (GERAD) was supported by the Medical Research Council (grant No. 503480), Alzheimer Research UK (grant No. 503176), the Wellcome Trust (grant No. 082604/2/07/Z), the and German Federal Ministry of Education and Research: Competence Network Dementia grant numbers 01GI0102, 01GI0711, and 01GI0420. Cohorts for Heart and Aging Research in Genomic Epidemiology Consortium (CHARGE) was partly supported by the National Institutes of Health (NIH)/National Institute on Aging grant R01 AG033193 and the National Institute on Aging AG081220 and Age, Gene/Environment Susceptibility Study (AGES) contract N01–AG–12100, the National Heart, Lung, and Blood Institute grant R01 HL105756, the Icelandic Heart Association, and the Erasmus Medical Center and Erasmus University. Alzheimer's Disease Genetics Consortium was supported by the NIH/National Institute on Aging grants U01 AG032984, U24 AG021886, and U01 AG016976 and the Alzheimer Association grant ADGC–10 to 196728. Genetics of magnetic resonance imaging (MRI)–confirmed lacunar stroke collaboration: the genetics of MRI-confirmed lacunar stroke project contains samples derived from the National Institute of Neurological Disorders and Stroke Stroke Genetics Network (SIGN-NINDS) study, the Wellcome Trust Case Control Consortium 2 stroke study, and DNA Lacunar (UK Young Lacunar Stroke DNA Study). The Stroke Genetics Network (SiGN) study was funded by a cooperative agreement grant from the US National Institute of Neurological Disorders and Stroke, NIH (U01 NS069208). Collection of the DNA Lacunar was primarily supported by the Wellcome Trust (WT072952) with additional support from the Stroke Association (TSA 2010/01). Genotyping of the DNA Lacunar samples was supported by a Stroke Association Grant (TSA 2013/01). The principal funding for the WTCCC2 stroke study was provided by the Wellcome Trust, as part of the Wellcome Trust Case Control Consortium 2 project (085475/B/08/Z, 085475/Z/08/Z, and WT084724MA).

ssociation (TSA 2010/01). Genotyping of the DNA Lacunar samples was supported by a Stroke Association Grant (TSA 2013/01). The principal funding for the WTCCC2 stroke study was provided by the Wellcome Trust, as part of the Wellcome Trust Case Control Consortium 2 project (085475/B/08/Z, 085475/Z/08/Z, and WT084724MA). Disclosures H.S. Markus received personal compensation for lectures from AstraZeneca. The other authors report no disclosures. Supplementary Material The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.118.020811/-/DC1.

multiple-territory disease was associated with post 90-day long-term risks of recurrent cardiovascular events independent of age, male sex, history of hypertension, diabetes mellitus, hypercholesterolemia, atrial fibrillation, cardiac failure, and history of smoking (Tables IV and V in the online-only Data Supplement). Discussion In this population-based study, we showed that over a quarter of patients presenting with TIA or ischemic stroke also had known symptomatic disease in other vascular beds. As expected, the number of affected vascular beds increased with the numbers of atherosclerotic risk factors. Despite intensive secondary prevention, 10-year risks of recurrent vascular events increased steeply with the number of territories affected. Of particular note, the long-term risks of recurrent nonstroke acute vascular events approached the risks of recurrent ischemic stroke in patients with multiple-territory disease.

With the widespread adoption of mechanical thrombectomy into clinical practice, there is renewed interest in strategies to mitigate secondary injury for patients with acute ischemic stroke, including neuroprotection and minimizing vasogenic edema.1,2 Lesion expansion after the ischemic insult is because of a combination of infarct growth (IG), which is associated with poor long-term clinical outcomes3,4 and anatomic distortion (AD) because of cerebral edema and hemorrhagic transformation—the major cause of neurological deterioration and death in the days after the event.5–7 Quantifying IG and AD separately poses significant challenges and limits the opportunity for clinical trials to assess treatment efficacy. The most common approaches to defining IG in trials have used either differences in measured infarct volumes between time points or the identification of regions of new infarction after linear image registration.8,9 Both will be confounded by AD because of edema or hemorrhage included in the volume. To minimize confounding, IG has been defined using extension into new anatomic territories.10,11 However, moving away from a volume-based approach compromises the ability to demonstrate a potential treatment effect.

ge registration.8,9 Both will be confounded by AD because of edema or hemorrhage included in the volume. To minimize confounding, IG has been defined using extension into new anatomic territories.10,11 However, moving away from a volume-based approach compromises the ability to demonstrate a potential treatment effect. Quantification of edema is similarly challenging. Current strategies involve labor intensive and subjective methodology defining regions on a slice-by-slice basis,12–14 measures of midline shift insensitive to submassive distortions,2 or inferences about focal edema from changes in whole brain or cerebrospinal fluid (CSF) volumes, which have produced inconsistent results and a fixed error that limits their use in smaller infarcts.15,16 Despite these challenges, the importance of AD has been highlighted by manual quantification of brain swelling, which has identified >11 mL volume as the threshold with greatest sensitivity and specificity for predicting poor outcome.11 Image registration has been key in the interpretation and design of acute stroke trials.17–19 Although linear registration is a well-suited approach within a time point when there are no structural differences between the acquired images, it does not correct for distortion.20 In contrast, nonlinear registration corrects for distortion in follow-up imaging at 24 hours and 1 week in patients with stroke.20

.17–19 Although linear registration is a well-suited approach within a time point when there are no structural differences between the acquired images, it does not correct for distortion.20 In contrast, nonlinear registration corrects for distortion in follow-up imaging at 24 hours and 1 week in patients with stroke.20 This study investigates the use of nonlinear registration to a presenting magnetic resonance imaging (MRI) scan in the definition of IG in patients with acute ischemic stroke at 24 hours and 1 week. We use the mismatch of infarct volumes after linear and nonlinear registration to define AD, which is compared with preexisting methods. Given that MRI on presentation is not routine in clinical practice, alternative reference images to the presenting MRI in the definition of AD are evaluated. Exploratory analysis of the optimum volume of distortion at 24 hours to predict 11-mL distortion at 1 week is derived.11 Finally, the ability of distortion at 1 week to predict clinical deterioration was explored in this cohort. Methods The data that support the findings of this study are available from the corresponding author on reasonable request.

This study investigates the use of nonlinear registration to a presenting magnetic resonance imaging (MRI) scan in the definition of IG in patients with acute ischemic stroke at 24 hours and 1 week. We use the mismatch of infarct volumes after linear and nonlinear registration to define AD, which is compared with preexisting methods. Given that MRI on presentation is not routine in clinical practice, alternative reference images to the presenting MRI in the definition of AD are evaluated. Exploratory analysis of the optimum volume of distortion at 24 hours to predict 11-mL distortion at 1 week is derived.11 Finally, the ability of distortion at 1 week to predict clinical deterioration was explored in this cohort. Methods The data that support the findings of this study are available from the corresponding author on reasonable request. Patients Patients aged >18 years with nonlacunar ischemic stroke were recruited within 18 hours of symptom onset into a prospective observational imaging cohort study under research protocols agreed by the UK National Research Ethics Service committee (ref 12/SC/0292 and 13/SC/0362) and by the local institutional review board. Written or witnessed consent was obtained from patients or agreement sought from a representative. Inclusion criteria were nonlacunar ischemic stroke and unilateral infarct visible on follow-up imaging at 24 hours, at 1 week, or both. Patients with a contraindication to MRI or impaired conscious level at presentation (score >1 on question 1a of the National Institutes for Health Stroke Scale) were not included. National Institutes for Health Stroke Scale was performed at the time of each scan. Six healthy volunteers were recruited and imaged under an agreed technical development protocol approved by the institution’s research governance office.

question 1a of the National Institutes for Health Stroke Scale) were not included. National Institutes for Health Stroke Scale was performed at the time of each scan. Six healthy volunteers were recruited and imaged under an agreed technical development protocol approved by the institution’s research governance office. Imaging Patients were imaged on presentation using computed tomography (CT) and MRI as soon as possible after that. Follow-up MRI was performed the following day (24 hours) and at 3 to 9 days (1 week), whenever possible (Methods in the online-only Data Supplement). Healthy volunteers underwent T1-weighted MRI on 2 occasions, 1 week apart. Lesion Definition All image analyses were performed using the Oxford Centre for Functional MRI of the Brain (FMRIB) software library (FSL). Lesion masks to define infarct at presentation were generated using the apparent diffusion coefficient (ADC) imaging and a threshold of 620×10−6 mm2/s (Methods in the online-only Data Supplement).21

All image analyses were performed using the Oxford Centre for Functional MRI of the Brain (FMRIB) software library (FSL). Lesion masks to define infarct at presentation were generated using the apparent diffusion coefficient (ADC) imaging and a threshold of 620×10−6 mm2/s (Methods in the online-only Data Supplement).21 All lesion masks used to define infarct at 24 hours and 1 week were defined manually by 2 separate independent stroke clinicians (4 different individuals across the study) using the masking tool in FSLView.22 The diffusion-weighted b1000 image (b=1000 s/mm2) was used for the 24-hour outcome23 and the T2-weighted fluid attenuated inversion recovery image for the 1-week outcome.20 Agreement was quantified using the concordance correlation coefficient. All lesion masks were reviewed and discrepancies resolved by a neuroradiologist. Masks were restricted to voxels within tissue masks created using the FMRIB’s automated segmentation tool (FAST).24 Image Registration Within-time point image registration was performed using linear (also known as rigid body) registration of either the diffusion-weighted or T2-weighted fluid attenuated inversion recovery images to the corresponding T1-weighted structural scan using FMRIB linear registration tool.25,26 Across time point, image registration of the follow-up T1-weighted image was made to the reference image space using both linear (FMRIB linear registration tool) and nonlinear registration using FMRIB’s nonlinear registration tool.20,22 Full details of registration are described in Methods in the online-only Data Supplement.

time point, image registration of the follow-up T1-weighted image was made to the reference image space using both linear (FMRIB linear registration tool) and nonlinear registration using FMRIB’s nonlinear registration tool.20,22 Full details of registration are described in Methods in the online-only Data Supplement. Infarct masks were resampled directly into the reference image space using a concatenation of the within-time point linear registration matrix, and either the nonlinear warp or the linear matrix generated from the registration of the T1-structural to the reference image. Once in the reference image space, the masks had a threshold of 0.5 applied. Infarct Growth IG was defined at 24 hours and at 1 week. In keeping with the method most commonly used in stroke trials,9 uncorrected IG was calculated as the difference in volume between the follow-up infarct and the presenting ADC-defined lesion volumes. Corrected IG was calculated as the difference in volume between the follow-up infarct and the presenting ADC-defined lesion volumes after nonlinear registration to the presenting MRI (Figure 1).

ed IG was calculated as the difference in volume between the follow-up infarct and the presenting ADC-defined lesion volumes. Corrected IG was calculated as the difference in volume between the follow-up infarct and the presenting ADC-defined lesion volumes after nonlinear registration to the presenting MRI (Figure 1). Figure 1. Schematic demonstrating the registration processes to quantify IG and presenting magnetic resonance imaging (MRI) distortion. A, The presenting apparent diffusion coefficient (ADC) lesion is registered to the presenting T1-weighted MRI using a linear registration to define infarct core. The T2-weighted (FLAIR) lesion mask at 1 week (or b1000 diffusion image at 24 hours) is registered to the presenting T1-weighted image using both linear (yellow arrows) and nonlinear (blue arrow) registration. B, The corresponding images registered to the presenting MRI are shown for reference. The difference between infarct core and the nonlinearly registered FLAIR lesion mask represents infarct growth (blue mask). The difference between the linear and nonlinear lesion masks represents anatomic distortion (yellow mask).

on. B, The corresponding images registered to the presenting MRI are shown for reference. The difference between infarct core and the nonlinearly registered FLAIR lesion mask represents infarct growth (blue mask). The difference between the linear and nonlinear lesion masks represents anatomic distortion (yellow mask). Two of the stroke clinicians who defined infarction also scored the Alberta Stroke Program Early CT Score (ASPECTS) on all b1000 diffusion-weighted imaging MRI scans at a time independent of lesion definition, with discrepancies resolved by consensus. The change in ASPECTS score between scan time points was calculated to assess for IG into new anatomic territories.11 Corrected IG was compared between groups of patients categorized according to whether there was any deterioration in the ASPECTS.10 Anatomic Distortion AD was defined as a within-time point measure at 24 hours and at 1 week. AD is the difference in volume of the lesion mask volumes generated after linear and nonlinear registration to the reference image (Figure 1). Absolute AD was used as the primary measure because of the externally derived association of absolute edema volume (>11 mL) and poor long-term outcome.11 Relative AD (absolute AD relative to the final infarct volume) was also calculated to control AD for infarct size.

ear registration to the reference image (Figure 1). Absolute AD was used as the primary measure because of the externally derived association of absolute edema volume (>11 mL) and poor long-term outcome.11 Relative AD (absolute AD relative to the final infarct volume) was also calculated to control AD for infarct size. To quantify any systematic measurement error associated with AD, representative lesion masks (selected from the first [small] and third [large] quartiles of infarct volumes) were registered to the T1-weighted structural scans of 6 healthy volunteers who had been scanned on 2 separate occasions, 1 week apart. These representative infarcts are shown in Figure I in the online-only Data Supplement. The identical processing and analysis was followed to generate a pseudo MRI distortion volume as with the patient data. Four different reference images were used for defining AD: Presenting MRI distortion was considered as the benchmark comparator, given it is the closest approximation to the premorbid brain structure before the effects of edema and hemorrhage have manifested6,20; Mirror MRI distortion was created by reflecting the follow-up structural MRI along the midline, using the patient’s own contralateral hemisphere as an approximation of an undistorted reference for the affected hemisphere; Template MRI distortion was used as an external reference image with the 2-mm isotropic T1-weighted image in MNI152 space supplied with FSL as the standard22,27,28; and

Mirror MRI distortion was created by reflecting the follow-up structural MRI along the midline, using the patient’s own contralateral hemisphere as an approximation of an undistorted reference for the affected hemisphere; Template MRI distortion was used as an external reference image with the 2-mm isotropic T1-weighted image in MNI152 space supplied with FSL as the standard22,27,28; and Presenting CT distortion provided an alternative approximation to the premorbid brain structure, but relies on cross-modality registration and less spatial information with which to align anatomic regions. All follow-up scans were reviewed for the presence of edema according to set criteria.11 Edema was defined as present if ≥2 of the following criteria were met on 2 axial slices: direct evidence of mass effect of affected gyri, indirect evidence based on new distortion of adjacent tissue, new midline shift, or new effacement of sulci or lateral ventricle. Any disagreements were resolved by consensus. The performance of AD was compared with this classification of the presence of edema.

l slices: direct evidence of mass effect of affected gyri, indirect evidence based on new distortion of adjacent tissue, new midline shift, or new effacement of sulci or lateral ventricle. Any disagreements were resolved by consensus. The performance of AD was compared with this classification of the presence of edema. A CSF-defined metric of AD was also calculated from the diffusion-weighted image (ADC maps) as described by Tipirneni-Sajja et al.16 In summary, CSF volume was quantified at presentation and follow-up, and the difference was used to estimate the degree of CSF displacement because of AD. Within each scan, CSF volumes were quantified using the ADC value of each voxel to estimate the proportion of CSF within that voxel using the formula: C=(ADCvoxel−840)/2560, where ADC is measured in units of 10−6 mm2/s. The volumes of CSF in all voxels were then summed across the registered images to provide an estimate of total CSF volume. Analysis Uncorrected and corrected IGs were correlated to explore the relationship between these metrics using Spearman correlation coefficient (r). The gradient of the correlation was calculated by linear regression and used to estimate the proportion by which uncorrected IG overestimated corrected IG. Corrected IG volumes were then compared across groups of patients with different changes in ASPECTS score from presentation to follow-up using ANOVA and the range of values of IG that exist within the groups described.

ar regression and used to estimate the proportion by which uncorrected IG overestimated corrected IG. Corrected IG volumes were then compared across groups of patients with different changes in ASPECTS score from presentation to follow-up using ANOVA and the range of values of IG that exist within the groups described. The median and interquartile range (IQR) of the absolute and relative pseudo MRI distortion within healthy volunteers were quantified. Median and IQR values of absolute and relative presenting MRI distortion were also calculated from within the stroke population at 24 hours and 1 week. Correlation of both absolute and relative presenting MRI distortion with infarct volume was calculated using Spearman correlation coefficient. Presenting MRI distortion volumes were compared between patients with and without rater-categorized edema using the Mann-Whitney U test.

troke population at 24 hours and 1 week. Correlation of both absolute and relative presenting MRI distortion with infarct volume was calculated using Spearman correlation coefficient. Presenting MRI distortion volumes were compared between patients with and without rater-categorized edema using the Mann-Whitney U test. The concordance correlation coefficient was used to quantify the agreement of presenting MRI distortion with CSF-defined distortion in all patients and separately for those with infarct volumes above and below the median.29 The grouping assessed the effect on small infarct volumes of the fixed error observed when using CSF-defined AD.16 Agreements of template MRI distortion, mirror MRI distortion, and presenting CT distortion with presenting MRI distortion were also quantified at 24 hours and 1 week using the concordance correlation coefficient from the patients where all metrics were available at both time points. Bland-Altman plots were used to explore the differences between presenting MRI distortion and other measures of AD.

distortion with presenting MRI distortion were also quantified at 24 hours and 1 week using the concordance correlation coefficient from the patients where all metrics were available at both time points. Bland-Altman plots were used to explore the differences between presenting MRI distortion and other measures of AD. The ability of the AD metric that had the highest agreement with presenting MRI distortion was evaluated as a tool at 24 hours to predict clinically significant AD at 1 week (11 mL)—a value derived in an external cohort of patients.11 Optimum thresholds were chosen using receiver operating characteristic (ROC) curve analysis followed by calculation of the Youden statistic.30 To explore the threshold of presenting MRI distortion in this cohort that predicted a clinical deterioration at 1 week, ROC curve analyses were performed using volumes of presenting MRI distortion at 1 week and any deterioration in National Institutes for Health Stroke Scale. All statistical analyses were performed using Prism (GraphPad, CA) and Stata 15 (StataCorp LLC, TX).

The ability of the AD metric that had the highest agreement with presenting MRI distortion was evaluated as a tool at 24 hours to predict clinically significant AD at 1 week (11 mL)—a value derived in an external cohort of patients.11 Optimum thresholds were chosen using receiver operating characteristic (ROC) curve analysis followed by calculation of the Youden statistic.30 To explore the threshold of presenting MRI distortion in this cohort that predicted a clinical deterioration at 1 week, ROC curve analyses were performed using volumes of presenting MRI distortion at 1 week and any deterioration in National Institutes for Health Stroke Scale. All statistical analyses were performed using Prism (GraphPad, CA) and Stata 15 (StataCorp LLC, TX). Results Patient Details Of 57 consecutively enrolled patients, 37 met the criteria for inclusion, and patient demographics are presented in Table 1. Lack of follow-up imaging was the most common reason why patients did not meet the inclusion criteria for analysis: 9 patients for medical instability or death and 9 patients declined to undergo further MRI. Two patients had no lesion on imaging at follow-up. Thirty-six patients had a CT scan at presentation. All patients underwent MRI scanning at presentation (median delay CT to MRI, 1 hour 24 minutes; IQR, 54 minutes to 2 hours 11 minutes), 30 patients at 24 hours, 28 at 1 week, and 21 at all time points. Median ASPECTS were 7, 6, and 5 at presentation, 24-hour, and 1-week time points, respectively. Final infarct volumes ranged from 0.2 to 340 mL. The interrater concordance correlation coefficient was 0.99 at 24 hours and 0.98 at 1 week. Edema was categorized as present in 13 (43%) and 15 (52%) patients at 24 hours and 1 week, respectively. There was evidence of hemorrhagic transformation in 9 (29%, all hemorrhagic infarctions) and 7 (24%) scans at each time point. At 1 week and 24 hours, 6 and 5 patients exhibited both hemorrhagic transformation and edema. Midline shift was not seen within 24 hours of onset and in only 2 patients at 1 week.

respectively. There was evidence of hemorrhagic transformation in 9 (29%, all hemorrhagic infarctions) and 7 (24%) scans at each time point. At 1 week and 24 hours, 6 and 5 patients exhibited both hemorrhagic transformation and edema. Midline shift was not seen within 24 hours of onset and in only 2 patients at 1 week. Table 1. Patient Demographics Image Registration The default registration algorithms were successful (example patient is shown in Figure 2) with the exception of the nonlinear registration of a 1-week scan from a single patient to its mirror T1-weighted image. In this case, the extensive infarct volume (340 plus 104 mL edema, presenting MRI distortion) resulted in insufficient unaffected brain with which to reference the mirror image. Registrations to presenting, template, and CT images were successful in all patients. Figure 2. Representative imaging from a single patient. T2-weighted fluid attenuated inversion recovery imaging at 1 week and the associated infarct masks were registered to the different reference images. Imaging is displayed in the reference image space, accounting for the difference in image orientations. The anatomic distortion can be quantified as the difference between the linearly (yellow) and nonlinearly (blue) registered masks in the reference image space. CT indicates computed tomography.

rent reference images. Imaging is displayed in the reference image space, accounting for the difference in image orientations. The anatomic distortion can be quantified as the difference between the linearly (yellow) and nonlinearly (blue) registered masks in the reference image space. CT indicates computed tomography. Infarct Growth Whereas corrected IG correlated strongly with uncorrected IG (r=0.98; P<0.0001) at both time points (Figure II in the online-only Data Supplement), uncorrected IG consistently overestimated corrected IG at both time points with gradients of 1.20 (95% confidence interval, 1.15–1.26) and 1.36 (95% confidence interval, 1.31–1.41) at 24 hours and 1 week, respectively. Corrected IG was not significantly different between patients grouped according to change in ASPECTS (P=0.1 and 0.2 at 24 hours and 1 week, respectively; ANOVA). Corrected IG of those patients with no change in ASPECTS over time was a maximum of 27.5 mL at 24 hours and 43.1 mL at 1 week. Anatomic Distortion In healthy volunteers, the median absolute pseudo MRI distortion was −0.2 mL (IQR, −0.4 to −0.03) and −0.05 mL (IQR, −0.1 to −0.03) for large (32.5 mL) and small (9.2 mL) infarct volumes, respectively. The relative pseudo MRI Distortion values were −0.8% and −0.5% for large and small infarct volumes.

Anatomic Distortion In healthy volunteers, the median absolute pseudo MRI distortion was −0.2 mL (IQR, −0.4 to −0.03) and −0.05 mL (IQR, −0.1 to −0.03) for large (32.5 mL) and small (9.2 mL) infarct volumes, respectively. The relative pseudo MRI Distortion values were −0.8% and −0.5% for large and small infarct volumes. In patients with stroke, the median presenting MRI distortion values were 2.1 mL (IQR, 0.3–8.0) and 3.4 mL (IQR, 0.3–15.0) at 24 hours and 1 week, with median relative presenting MRI distortion values of 12% (IQR, 6%–23%) and 21% (IQR, 8%–29%). Absolute presenting MRI distortion values correlated with corrected infarct volume (Figure III in the online-only Data Supplement; 24 hours: r=0.85, P<0.0001; 1 week: r=0.94, P<0.0001), but relative presenting MRI distortion only correlated at 1 week (24 hours: r=0.32, P=0.1; 1 week: r=0.63, P=0.0004). Presenting MRI distortion differed significantly between patients with and without rater-defined edema classification at both 24 hours and 1 week (median: 10.5 versus 0.3 mL, P<0.0001, and 14.4 versus 0.1 mL, P<0.0001, respectively; Mann-Whitney U test). Where the scan was classified as not having edema present, the maximum AD was 6.3 mL at 24 hours and 3.5 mL at 1 week. The presenting MRI distortion did not differ significantly between those with and without hemorrhagic transformation (t test; P=0.9).

d 14.4 versus 0.1 mL, P<0.0001, respectively; Mann-Whitney U test). Where the scan was classified as not having edema present, the maximum AD was 6.3 mL at 24 hours and 3.5 mL at 1 week. The presenting MRI distortion did not differ significantly between those with and without hemorrhagic transformation (t test; P=0.9). CSF-defined AD was concordant with presenting MRI distortion at 24 hours and 1 week (Table 2). However, there was no correlation for infarct volumes below the median volume at either time point. CSF-defined AD correlated with corrected lesion volumes similarly to presenting MRI distortion (Figure III in the online-only Data Supplement). There was greater concordance of absolute mirror MRI distortion, template MRI distortion, and presenting CT distortion with presenting MRI distortion at 24 hours and 1 week (Figures 3 and 4; Table 2). Representative images using all 4 reference approaches from 1 patient are shown in Figure 2. Presenting CT distortion displayed the strongest agreement at both time points with the benchmark comparator, presenting MRI distortion. Table 2. Agreement Between Measures of Anatomical Distortion

CSF-defined AD was concordant with presenting MRI distortion at 24 hours and 1 week (Table 2). However, there was no correlation for infarct volumes below the median volume at either time point. CSF-defined AD correlated with corrected lesion volumes similarly to presenting MRI distortion (Figure III in the online-only Data Supplement). There was greater concordance of absolute mirror MRI distortion, template MRI distortion, and presenting CT distortion with presenting MRI distortion at 24 hours and 1 week (Figures 3 and 4; Table 2). Representative images using all 4 reference approaches from 1 patient are shown in Figure 2. Presenting CT distortion displayed the strongest agreement at both time points with the benchmark comparator, presenting MRI distortion. Table 2. Agreement Between Measures of Anatomical Distortion Figure 3. Agreement between registration-defined measures of anatomical distortion at 24 hours. Top, Template magnetic resonance imaging (MRI) distortion, mirror MRI distortion, and presenting computed tomography (CT) distortion compared with presenting MRI distortion at 24 hours. Continuous line: correlation; intermittent line: line of unity. Bottom, Bland-Altman plots showing the differences between the measurement techniques. Intermittent lines: 95% limits of agreement.

rror MRI distortion, and presenting computed tomography (CT) distortion compared with presenting MRI distortion at 24 hours. Continuous line: correlation; intermittent line: line of unity. Bottom, Bland-Altman plots showing the differences between the measurement techniques. Intermittent lines: 95% limits of agreement. ROC curve analysis using presenting CT distortion at 24 hours to predict the externally derived, clinically meaningful threshold of distortion (11 mL) at 1 week generated an area under the curve of 0.99. The Youden statistic defined an optimum presenting CT distortion threshold of 4.8 mL at 24 hours for predicting edema of 11 mL at 1 week, with a sensitivity of 100% and a specificity of 93%. The CSF-defined metric of AD produced an area under the curve of 0.79 for the prediction of 11 mL edema at 1 week. Exploratory ROC curve analysis to derive a threshold of AD that predicted any increase in the National Institute for Health Stroke Scale from presentation to 1 week produced an area under the curve of 0.77 and an optimum threshold of 15 mL (Figure V in the online-only Data Supplement).

diction of 11 mL edema at 1 week. Exploratory ROC curve analysis to derive a threshold of AD that predicted any increase in the National Institute for Health Stroke Scale from presentation to 1 week produced an area under the curve of 0.77 and an optimum threshold of 15 mL (Figure V in the online-only Data Supplement). Discussion Using nonlinear registration to correct for AD provides not only improved estimates of IG but also quantifies the lesion expansion associated with edema and hemorrhagic transformation. AD can be quantified automatically across a range of infarct volumes in the absence of presenting MRI. Of the alternate reference images, the presenting CT scan provided the best comparator when compared against the benchmark presenting MRI. In this study, AD, when measured at 24 hours, could predict a clinically significant volume of AD at 1 week. Lesion expansion over time comprises both IG and AD.11 In this study, uncorrected IG overestimated corrected IG by an average of 20% and 36% at 24 hours and 1 week. This overestimation represents the AD that occurs in the hours to days after stroke onset because of blood-brain barrier disruption that underlies both vasogenic edema and hemorrhagic transformation.5–7 That AD is a measure of edema is supported by the observation that AD is significantly different between patients with and without rater-categorized edema11 and, secondly, that presenting MRI distortion agreed with a similar measure of AD derived from CSF displacement.16

sogenic edema and hemorrhagic transformation.5–7 That AD is a measure of edema is supported by the observation that AD is significantly different between patients with and without rater-categorized edema11 and, secondly, that presenting MRI distortion agreed with a similar measure of AD derived from CSF displacement.16 Absolute AD correlated more closely with infarct volume than relative AD. This points to the presence of individual factors that influence edema and hemorrhagic transformation—the absolute volumes of which may be proportional to the final infarct volume. It will require larger cohorts of patients to explore which clinical factors contribute to this variability. The error attributable to this registration-defined measure of AD was <1% when quantifying pseudo MRI distortion in healthy volunteers regardless of infarct volume used. This ability to measure small volumes of AD accurately is important because it provides the opportunity to estimate the threshold of AD at 24 hours that predicts clinically meaningful edema at 1 week (11 mL),11 which in this study was a 24-hour presenting CT distortion of 4.8 mL. Exploratory ROC curve analysis in this cohort showed that an optimum AD volume of 15 mL was most closely associated with clinical deterioration at 1 week, similar to the independently derived threshold of 11 mL.11

eaningful edema at 1 week (11 mL),11 which in this study was a 24-hour presenting CT distortion of 4.8 mL. Exploratory ROC curve analysis in this cohort showed that an optimum AD volume of 15 mL was most closely associated with clinical deterioration at 1 week, similar to the independently derived threshold of 11 mL.11 CSF-defined measures of distortion broadly agreed with registration-defined AD, but the relationship failed in patients with small (below median) infarct volumes. CSF-defined distortion is a global approach that measures CSF displacement within the whole skull. This limits its use in patients with smaller infarct volumes, not least because of errors ≤10 mL seen in patients with little or no evidence of infarction seen on MRI.16 This is in keeping with previously identified challenges of global measures of whole brain volume introduced by effects in regions remote from infarction, including noise, resorption of parenchymal extracellular fluid, and displacement of cerebral blood volume.15,31 In contrast, AD defined by nonlinear registration is a direct estimation of local distortion, and, therefore, any errors are likely to be proportional to the infarct volume.

in regions remote from infarction, including noise, resorption of parenchymal extracellular fluid, and displacement of cerebral blood volume.15,31 In contrast, AD defined by nonlinear registration is a direct estimation of local distortion, and, therefore, any errors are likely to be proportional to the infarct volume. Unlike presenting MRI distortion and CSF-defined distortion, template MRI distortion, mirror MRI distortion, and presenting CT distortion are not bound by the necessity to have an MRI scan at presentation. However, mirror MRI distortion could be adversely affected by the presence of a previous contralateral infarct, and the presence of global atrophy could adversely affect template MRI distortion. This latter effect may explain why template MRI distortion did not perform, as well as the other measures, given the heterogeneous degrees of global atrophy seen in a population of patients with stroke.

a previous contralateral infarct, and the presence of global atrophy could adversely affect template MRI distortion. This latter effect may explain why template MRI distortion did not perform, as well as the other measures, given the heterogeneous degrees of global atrophy seen in a population of patients with stroke. The data show a strong agreement between using the presenting CT scan with the presenting MRI as the benchmark on which to define AD. Cross-modality (CT to MRI) registration requires only a slight modification of the approach used for MRI-to-MRI comparison. This method could enable the comparison of CT-based ischemic core volumes with MRI-defined final infarction to quantify IG in addition to AD.32,33 This is an important area of future research given the fact that, on pragmatic grounds, CT imaging will be the core imaging modality on acute presentation in clinical trials. Furthermore, this approach can be incorporated into automated image analysis software to provide objective measures of IG and AD. The differentiation of IG and AD provides the opportunity to test interventions aimed at either of these 2 specific processes in early-phase clinical trials. The same registration-based approach could also be implemented in preclinical stroke models, facilitating translation of novel approaches into early-phase trials.34 More consistent measurements improve statistical power in such studies, offering the potential to reduce sample size.20

ses in early-phase clinical trials. The same registration-based approach could also be implemented in preclinical stroke models, facilitating translation of novel approaches into early-phase trials.34 More consistent measurements improve statistical power in such studies, offering the potential to reduce sample size.20 This study has several limitations. Despite the minimal degree of measurement error seen in the pseudo MRI distortion metric, in a patient cohort, one might expect these errors of quantification to be greater. However, it is likely that any measurement error would remain proportional to the infarct volume. Further studies in larger cohorts are required to validate AD and IG derived from this registration approach by linking AD and IG to clinical outcomes and to explore the potential for defining IG using CT definitions of presenting infarct core. Limiting the immediate use of AD in historical clinical imaging cohorts is the requirement for a structural T1-weighted image acquired at the follow-up imaging time point. Conclusions Registration-defined measures of IG and AD allow the distinction of the 2 processes of secondary injury that constitute lesion expansion. These techniques are objective, can be automated, and can be used to measure a wide range of infarct volumes.

This study has several limitations. Despite the minimal degree of measurement error seen in the pseudo MRI distortion metric, in a patient cohort, one might expect these errors of quantification to be greater. However, it is likely that any measurement error would remain proportional to the infarct volume. Further studies in larger cohorts are required to validate AD and IG derived from this registration approach by linking AD and IG to clinical outcomes and to explore the potential for defining IG using CT definitions of presenting infarct core. Limiting the immediate use of AD in historical clinical imaging cohorts is the requirement for a structural T1-weighted image acquired at the follow-up imaging time point. Conclusions Registration-defined measures of IG and AD allow the distinction of the 2 processes of secondary injury that constitute lesion expansion. These techniques are objective, can be automated, and can be used to measure a wide range of infarct volumes. Figure 4. Agreement between registration-defined measures of anatomical distortion at 1 week. Top, Template magnetic resonance imaging (MRI) distortion, mirror MRI distortion, and presenting computed tomography (CT) distortion compared with presenting MRI distortion at 1 week. Continuous line: correlation; intermittent line: line of unity. Bottom, Bland-Altman plots showing the differences between the measurement techniques. Intermittent lines: 95% limits of agreement.

mirror MRI distortion, and presenting computed tomography (CT) distortion compared with presenting MRI distortion at 1 week. Continuous line: correlation; intermittent line: line of unity. Bottom, Bland-Altman plots showing the differences between the measurement techniques. Intermittent lines: 95% limits of agreement. Acknowledgments We wish to acknowledge the facilities and staff of the Oxford Acute Vascular Imaging Centre, the Oxford University Hospitals NHS Foundation Trust Acute Stroke Service, and the Acute Magnetic Resonance Imaging in Cerebral Ischemia group. Sources of Funding This study was supported by the National Institute for Health Research (NIHR) Biomedical Research Centre, Oxford, the National Institute for Health Research Clinical Research Network, the Dunhill Medical Trust (grant No. OSRP1/1006), the Centre of Excellence for Personalized Healthcare funded by the Wellcome Trust and Engineering and Physical Sciences Research Council under grant number WT088877/Z/09/Z, the Oxford University Clinical Academic Graduate School, and the Wellcome Trust Institutional Strategic Support Fund (2014–2015). Disclosures M. Jenkinson receives royalties from licensing of FSL to nonacademic, commercial entities. The other authors report no conflicts. Supplementary Material Presented in part at the International Stroke Conference, Houston, TX, February 21–24, 2017. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.118.020788/-/DC1.

Atherosclerotic vascular disease is a chronic disease of the arterial wall of different vascular territories, causing cerebrovascular, coronary, peripheral, and aortic vascular disease. It is the leading cause of death and morbidity worldwide.1 Vascular risk factors, such as hypertension and hypercholesterolemia are important drivers of atherosclerosis2–5 and blood pressure lowering, lipid-lowering, and antiplatelet treatment are effective in reducing acute vascular events both in primary and secondary prevention settings.6–9 Recent randomized trials showed that novel anti-inflammatory and lipid-lowering therapies reduced risk of recurrent cardiovascular events in patients with cardiovascular disease on current standard secondary prevention treatment.10–12 However, these new agents are expensive and are unlikely to be cost-effective in patients at low vascular risk.

howed that novel anti-inflammatory and lipid-lowering therapies reduced risk of recurrent cardiovascular events in patients with cardiovascular disease on current standard secondary prevention treatment.10–12 However, these new agents are expensive and are unlikely to be cost-effective in patients at low vascular risk. Patients in the secondary prevention setting with atherosclerotic disease affecting 2 or more vascular beds seem to be at high risk for future vascular events.13,14 However, previous studies were hospital-based, had a relatively short follow-up, and did not focus specifically on patients with transient ischemic attack (TIA) and stroke. To assess whether TIA/ischemic stroke patients with disease in other vascular beds were at particularly high risk of future vascular events as previously suggested,13,14 we studied patients presenting with TIA or ischemic stroke in relation to the number of other vascular beds (coronary, peripheral) affected by symptomatic disease to determine long-term prognosis on current standard secondary prevention in a population-based study. We hypothesized that the number of affected vascular beds could be used as a simple clinical rule in identifying patients who are at high risk of recurrent vascular events. Methods Requests for access to data from OXVASC will be considered by the corresponding author.

Patients in the secondary prevention setting with atherosclerotic disease affecting 2 or more vascular beds seem to be at high risk for future vascular events.13,14 However, previous studies were hospital-based, had a relatively short follow-up, and did not focus specifically on patients with transient ischemic attack (TIA) and stroke. To assess whether TIA/ischemic stroke patients with disease in other vascular beds were at particularly high risk of future vascular events as previously suggested,13,14 we studied patients presenting with TIA or ischemic stroke in relation to the number of other vascular beds (coronary, peripheral) affected by symptomatic disease to determine long-term prognosis on current standard secondary prevention in a population-based study. We hypothesized that the number of affected vascular beds could be used as a simple clinical rule in identifying patients who are at high risk of recurrent vascular events. Methods Requests for access to data from OXVASC will be considered by the corresponding author. We studied consecutive patients with a first-in-the-study-period TIA or ischemic stroke in OXVASC (Oxford Vascular Study) from 2002 to 2014. OXVASC is an ongoing population-based study of the incidence and outcome of all acute vascular events in a population of 92 728 individuals, registered with 100 general practitioners in 9 general practices in Oxfordshire, United Kingdom. The multiple overlapping methods used to achieve near complete ascertainment of all individuals with TIA and ischemic stroke have been reported previously.15 Briefly, these included (1) a daily, rapid-access TIA, and stroke clinic to which participating general practitioners and the local emergency department team referred individuals with suspected TIA or minor stroke; (2) daily searches of admissions to medical, stroke, neurology, and other relevant wards; (3) daily searches of the local emergency department attendance register; (4) daily searches of in-hospital death records via the bereavement office; (5) monthly searches of all death certificates and coroner’s reports for out-of-hospital deaths; (6) monthly searches of general practitioner diagnostic coding and hospital discharge codes; and (7) monthly searches of all brain and vascular imaging referrals.

hes of in-hospital death records via the bereavement office; (5) monthly searches of all death certificates and coroner’s reports for out-of-hospital deaths; (6) monthly searches of general practitioner diagnostic coding and hospital discharge codes; and (7) monthly searches of all brain and vascular imaging referrals. Demographic data, risk factors for atherosclerosis (eg, hypertension, diabetes mellitus, history of smoking, hypercholesterolemia), and history of vascular disease in other vascular beds (symptomatic coronary and peripheral vascular disease) were collected from face-to-face interview and cross-referenced with primary care records.

data, risk factors for atherosclerosis (eg, hypertension, diabetes mellitus, history of smoking, hypercholesterolemia), and history of vascular disease in other vascular beds (symptomatic coronary and peripheral vascular disease) were collected from face-to-face interview and cross-referenced with primary care records. Patients were considered to have concurrent coronary heart disease if they had at least one of the following conditions: previous myocardial infarction; unstable angina; angina; and history of percutaneous coronary intervention or coronary artery bypass graft surgery. Concurrent symptomatic peripheral vascular disease was defined as having at least one of the following conditions: previous aortic aneurysm rupture; aortic dissection; acute limb ischemia; critical limb ischemia; acute visceral ischemia; intermittent claudication; previous angioplasty or stenting; peripheral arterial bypass graft or amputation. Patients without a history of symptomatic coronary or peripheral vascular disease were classified as having single-territory disease (TIA/stroke only), whereas patients with concurrent coronary or peripheral vascular disease were classified as having double-territory disease and patients with diseases in both coronary and peripheral vascular beds were classified as having triple-territory symptomatic vascular disease.

having single-territory disease (TIA/stroke only), whereas patients with concurrent coronary or peripheral vascular disease were classified as having double-territory disease and patients with diseases in both coronary and peripheral vascular beds were classified as having triple-territory symptomatic vascular disease. All patients routinely had brain imaging (computed tomography or magnetic resonance imaging), intracranial and extracranial vascular imaging (carotid Doppler/CTA/MRA/DSA), 12-lead electrocardiography, and routine bloods (ie, full blood count, clotting, C-reactive protein, erythrocyte sedimentation rate, liver function, renal function, thyroid function, electrolytes, and lipid profile) after the event. Echocardiography, 24-hour electrocardiography, and 5-day electrocardiography event recorder (R test) were also done when clinically indicated. Standard secondary preventive treatment was continued or started on the day of the initial clinical assessment, which usually included antithrombotic treatment, antihypertensive drugs, and a statin. Notably, although we routinely prescribed a statin, the exact regime continued for long-term use (ranging from simvastatin 40 mg daily to atorvastatin 80 mg daily) was left to the patient’s primary care physician, who has a responsibility in the UK healthcare system for long-term management of patients.

, and a statin. Notably, although we routinely prescribed a statin, the exact regime continued for long-term use (ranging from simvastatin 40 mg daily to atorvastatin 80 mg daily) was left to the patient’s primary care physician, who has a responsibility in the UK healthcare system for long-term management of patients. Patients were followed-up face-to-face at 1, 6, 12, 60, and 120 months by a study nurse or physician to identify any recurrent stroke and other acute vascular events (myocardial infarction, peripheral vascular event), supplemented by review of primary care records. Patients who had moved out of the study area were followed-up via telephone at the same time points as face-to-face follow-up. We recorded all deaths during follow-up with the underlying causes by direct follow-up, via primary care records, and by centralized registration with Office for National Statistics. All recurrent events that occurred during follow-up would also be identified by the ongoing daily case ascertainment. Statistical Analysis Baseline characteristics were compared between all 3 groups (single-territory versus double-territory versus triple-territory), using χ2 test for categorical variables and 1-way ANOVA test for continuous variables.

Patients were followed-up face-to-face at 1, 6, 12, 60, and 120 months by a study nurse or physician to identify any recurrent stroke and other acute vascular events (myocardial infarction, peripheral vascular event), supplemented by review of primary care records. Patients who had moved out of the study area were followed-up via telephone at the same time points as face-to-face follow-up. We recorded all deaths during follow-up with the underlying causes by direct follow-up, via primary care records, and by centralized registration with Office for National Statistics. All recurrent events that occurred during follow-up would also be identified by the ongoing daily case ascertainment. Statistical Analysis Baseline characteristics were compared between all 3 groups (single-territory versus double-territory versus triple-territory), using χ2 test for categorical variables and 1-way ANOVA test for continuous variables. We compared the prevalence of atherosclerotic risk factors (overall number and individual risk factors) and also prevalence of asymptomatic carotid disease (based on vascular imaging performed routinely for all patients as part of the diagnostic workup) in patients with single- versus multiple-territory disease using χ2 test and logistic regression analysis adjusted for age and sex.

ll number and individual risk factors) and also prevalence of asymptomatic carotid disease (based on vascular imaging performed routinely for all patients as part of the diagnostic workup) in patients with single- versus multiple-territory disease using χ2 test and logistic regression analysis adjusted for age and sex. Kaplan-Meier survival analysis was used to calculate the 1-year, 5-year, and 10-year risks of vascular events during follow-up, censored at death or September 30, 2014, for single-, double-, and triple-territory disease. We compared the following outcomes in patients with single- versus multiple-territory disease using Cox-regression analysis adjusted for age and sex: first major cardiovascular event (any recurrent ischemic stroke, myocardial infarction, acute peripheral vascular event, or vascular death), vascular death, first recurrent ischemic stroke, and first nonstroke acute vascular event (myocardial infarction, acute peripheral vascular event, or sudden cardiac death). Exploratory analyses were performed with additional adjustment for other known vascular risk factors. Sensitivity analyses were also performed confined to patients with large artery disease according to TOAST classification (Trial of ORG 10172 in Acute Stroke Treatment), excluding patients with known atrial fibrillation at baseline and stratified by the type of the index event (TIA versus ischemic stroke). All analyses were done using SPSS version 22.

Sensitivity analyses were also performed confined to patients with large artery disease according to TOAST classification (Trial of ORG 10172 in Acute Stroke Treatment), excluding patients with known atrial fibrillation at baseline and stratified by the type of the index event (TIA versus ischemic stroke). All analyses were done using SPSS version 22. Standard Protocol Approvals, Registrations, and Patient Consents Written informed consent or assent from relatives was obtained in all participants. OXVASC was approved by the local research ethics committee (OREC A: 05/Q1604/70). Results Of 2554 patients with a first-in-the-study-period event (1606 ischemic stroke and 948 TIA), 1842 (72.1%) had single-territory disease (TIA/ischemic stroke only), 608 (23.8%) had double-territory, and 104 (4.1%) had triple-territory symptomatic vascular disease.

Standard Protocol Approvals, Registrations, and Patient Consents Written informed consent or assent from relatives was obtained in all participants. OXVASC was approved by the local research ethics committee (OREC A: 05/Q1604/70). Results Of 2554 patients with a first-in-the-study-period event (1606 ischemic stroke and 948 TIA), 1842 (72.1%) had single-territory disease (TIA/ischemic stroke only), 608 (23.8%) had double-territory, and 104 (4.1%) had triple-territory symptomatic vascular disease. As shown in Table 1, patients with double- or triple-territory disease were more likely to be on preventative agents for vascular disease before the index TIA/stroke. The proportion of patients on secondary prevention further increased after the index event for all patients (Table 1) but was more intensive in patients with multiple-territory disease (Table 1). At 1-month follow-up, 84 (95.5%) of the 88 patients with triple-territory disease were on antithrombotic agents, 74 (84.1%) on antihypertensive treatment, and 73 (83.0%) were on statins (Table 1). At 1-year follow-up, 72 (97.3%) of the 74 patients with triple-territory disease remained on antithrombotic agents, 63 (85.1%) on antihypertensive treatment, and 61 (82.4%) were still on statins (Table 1). Table 1. Demographics, Risk Factors, and Secondary Prevention Treatment in Patients With Baseline Single-, Double-, and Triple-Territory Diseases

As shown in Table 1, patients with double- or triple-territory disease were more likely to be on preventative agents for vascular disease before the index TIA/stroke. The proportion of patients on secondary prevention further increased after the index event for all patients (Table 1) but was more intensive in patients with multiple-territory disease (Table 1). At 1-month follow-up, 84 (95.5%) of the 88 patients with triple-territory disease were on antithrombotic agents, 74 (84.1%) on antihypertensive treatment, and 73 (83.0%) were on statins (Table 1). At 1-year follow-up, 72 (97.3%) of the 74 patients with triple-territory disease remained on antithrombotic agents, 63 (85.1%) on antihypertensive treatment, and 61 (82.4%) were still on statins (Table 1). Table 1. Demographics, Risk Factors, and Secondary Prevention Treatment in Patients With Baseline Single-, Double-, and Triple-Territory Diseases As shown in Figure 1, the number of affected vascular beds increased with the numbers of atherosclerotic risk factors (Ptrend<0.0001), with the highest prevalence in patients with triple-territory disease (Table 1). Compared with patients with TIA/stroke only, those with double- or triple-territory disease had more hypertension (age/sex-adjusted odds ratio [OR], 3.43; 95% confidence interval [CI], 2.76–4.27; P<0.0001; Table 2), diabetes mellitus (OR, 2.89; 95% CI, 2.27–3.66; P<0.0001; Table 2), hypercholesterolemia (OR, 4.67; 95% CI, 3.85–5.66; P<0.0001; Table 2), and history of smoking (OR, 1.52; 95% CI, 1.26–1.84; P<0.0001; Table 2). The same was observed when comparing patients with triple-territory disease to patients with TIA/stroke alone (Table 2) and triple-territory disease was particularly strongly associated with known hypercholesterolemia (OR, 6.80; 95% CI, 4.39–10.53; P<0.0001; Table 2), with a baseline mean/SD total cholesterol of 4.5/1.2 mmol/L despite 62% being on statin treatment before the index TIA/stroke (Table 1).

ts with TIA/stroke alone (Table 2) and triple-territory disease was particularly strongly associated with known hypercholesterolemia (OR, 6.80; 95% CI, 4.39–10.53; P<0.0001; Table 2), with a baseline mean/SD total cholesterol of 4.5/1.2 mmol/L despite 62% being on statin treatment before the index TIA/stroke (Table 1). Table 2. Crude and Age/Sex-Adjusted ORs of Different Atherosclerotic Risk Factors in Multiple-Territory vs Single-Territory Events Figure 1. Distribution of numbers of atherosclerotic risk factors (A) and severity of asymptomatic carotid bifurcation stenosis (B) in patients with single-, double-, and triple-territory disease at baseline. History of smoking data missing for 8 patients in A. Not only did the prevalence of vascular risk factors increase with the number of affected vascular beds, but patients with multiple-territory disease also had more severe asymptomatic carotid bifurcation stenosis (Figure 1). Twenty-six (33.3%) of 78 patients in the triple-territory group had at least 50% asymptomatic stenosis, compared with 69 (4.8%) of 1441 patients in the TIA/stroke only group (age/sex-adjusted OR, 7.39; 95% CI, 4.26–12.81; P<0.0001).

tory disease also had more severe asymptomatic carotid bifurcation stenosis (Figure 1). Twenty-six (33.3%) of 78 patients in the triple-territory group had at least 50% asymptomatic stenosis, compared with 69 (4.8%) of 1441 patients in the TIA/stroke only group (age/sex-adjusted OR, 7.39; 95% CI, 4.26–12.81; P<0.0001). During 10 679 patient-years of follow-up, there were 515 vascular deaths, 417 recurrent ischemic strokes, and 203 recurrent nonstroke acute vascular events (136 acute coronary events and 67 acute peripheral events). Despite more intensive secondary prevention in patients with multiple-territory disease, the 5-year risks of major cardiovascular event, vascular death, recurrent ischemic stroke, or recurrent nonstroke acute vascular events increased steeply with the number of territories affected (Figure 2; Figure I in the online-only Data Supplement). Although the risks were particularly front-loading (Figure 2), patients with multiple-territory disease also had higher post 90-day long-term risks of recurrent vascular events (double/triple versus single 10-year major cardiovascular events: 50.7% versus 29.0%; age/sex-adjusted hazard ratio [HR], 1.67; 95% CI, 1.37–2.01; P<0.0001; Table 3). The risks were higher in patients with TIA/stroke plus peripheral vascular disease than in patients with TIA/stroke plus coronary artery disease (10-year major cardiovascular events: 60.8% versus 46.1%; age/sex-adjusted HR, 1.58, 95% CI, 1.03–2.43; P=0.04), and were highest in those with triple-territory disease (triple versus single major cardiovascular events: 64.2% versus 29.0%; age/sex-adjusted HR, 2.68; 95% CI, 1.86–3.86; P<0.0001; Table 3). Moreover, compared with patients with TIA/stroke only, patients with triple-territory disease also had a 2-fold increase of recurrent ischemic stroke (10-year age/sex-adjusted HR, 2.32; 95% CI, 1.39–3.88; P=0.001; Table 3), and a 5-fold increase of recurrent nonstroke acute vascular events (HR, 4.62; 95% CI, 2.68–7.98; P<0.0001), with the risks of recurrent nonstroke acute vascular events approaching the risks of recurrent ischemic stroke (Figure 3).

ischemic stroke (10-year age/sex-adjusted HR, 2.32; 95% CI, 1.39–3.88; P=0.001; Table 3), and a 5-fold increase of recurrent nonstroke acute vascular events (HR, 4.62; 95% CI, 2.68–7.98; P<0.0001), with the risks of recurrent nonstroke acute vascular events approaching the risks of recurrent ischemic stroke (Figure 3). Table 3. Post 90-Day Cumulative Risks of Vascular Death, Recurrent Ischemic Stroke, or Recurrent Nonstroke Acute Vascular Event Stratified by Number of Affected Vascular Beds at Baseline Figure 2. Ten-year risks of recurrent vascular events in patients with baseline single, double, and triple-territory diseases. Panels are for different outcomes. A, Major cardiovascular event: any recurrent ischemic stroke, myocardial infarction, acute peripheral vascular event, or vascular death; B, vascular death; C, recurrent ischaemic stroke; D, recurrent nonstroke acute vascular event: myocardial infarction, acute peripheral vascular event, or sudden cardiac death. Figure 3. Five-year risk of recurrent ischemic stroke and nonstroke acute vascular event by the number of affected vascular beds. Overall risks (A) and risks excluding the first 90-days (B) are presented in panels. Error bar represents SE.

Figure 2. Ten-year risks of recurrent vascular events in patients with baseline single, double, and triple-territory diseases. Panels are for different outcomes. A, Major cardiovascular event: any recurrent ischemic stroke, myocardial infarction, acute peripheral vascular event, or vascular death; B, vascular death; C, recurrent ischaemic stroke; D, recurrent nonstroke acute vascular event: myocardial infarction, acute peripheral vascular event, or sudden cardiac death. Figure 3. Five-year risk of recurrent ischemic stroke and nonstroke acute vascular event by the number of affected vascular beds. Overall risks (A) and risks excluding the first 90-days (B) are presented in panels. Error bar represents SE. Sensitivity analysis confined to patients with large artery disease (Table I in the online-only Data Supplement), excluding patients with known atrial fibrillation at baseline (Table II in the online-only Data Supplement), or stratified by the type of the index event (Table III in the online-only Data Supplement) also showed consistent results. Exploratory multivariate analyses adjusting for other vascular risk factors also suggested that multiple-territory disease was associated with post 90-day long-term risks of recurrent cardiovascular events independent of age, male sex, history of hypertension, diabetes mellitus, hypercholesterolemia, atrial fibrillation, cardiac failure, and history of smoking (Tables IV and V in the online-only Data Supplement).

espite intensive secondary prevention, 10-year risks of recurrent vascular events increased steeply with the number of territories affected. Of particular note, the long-term risks of recurrent nonstroke acute vascular events approached the risks of recurrent ischemic stroke in patients with multiple-territory disease. Our findings are in line with previous studies showing that despite standard secondary prevention, patients with multiple-territory disease still had a ≈1.5-fold increase of recurrent vascular events or vascular death than patients with TIA or ischemic stroke alone.13,14,16–18 However, our estimates of the absolute risks were much higher than previous studies, even after excluding the acute phase post-TIA/ischemic stroke.13,17 For example, the 90-day to 1-year vascular death in patients with multiple-territory disease was 4.7% in OXVASC versus 2.8% in the REACH registry (Reduction of Atherothrombosis for Continued Health),17 and the risk of all major vascular events was 8.6 per 100 patient-years versus 5.0 in the SMART study (Second Manifestations of Arterial Disease).13 These differences probably reflect the larger number of elderly patients with multiple comorbidities in OXVASC owing to the population-based design and the longer period of follow-up.

of all major vascular events was 8.6 per 100 patient-years versus 5.0 in the SMART study (Second Manifestations of Arterial Disease).13 These differences probably reflect the larger number of elderly patients with multiple comorbidities in OXVASC owing to the population-based design and the longer period of follow-up. That the number of affected territories still predicts a poor outcome in a cohort on current standard secondary prevention highlights the unmet need for more effective treatment in TIA or ischemic stroke patients with symptomatic disease in multiple vascular beds. We found that the number of atherosclerotic risk factors increased with the number of affected vascular beds, with particularly strong associations with hyperlipidemia, reflecting the importance of lipids and smoking in coronary and peripheral vascular disease.19,20 Both the recent FOURIER (Further Cardiovascular Outcomes Research With PCSK9 Inhibition in Subjects With Elevated Risk) and the REVEAL (Randomized Evaluation of the Effects of Anacetrapib Through Lipid Modification) trials showed that in patients with atherosclerotic cardiovascular disease there is some additional benefit from lowering of cholesterol levels below current targets.11,12 Although our patients with multiple-territory disease were usually on premorbid statins, they still had total cholesterol of 4.5 mmol/L at baseline, and although lipid-lowering was intensified thereafter in the majority, additional treatments might be justified.21,22 Moreover, previous studies have shown that there is a systemic predisposition to atherosclerosis,23–26 and we found that in patients with multiple-territory disease, the long-term risks of nonstroke acute vascular events approached the risks of recurrent ischemic stroke.

, additional treatments might be justified.21,22 Moreover, previous studies have shown that there is a systemic predisposition to atherosclerosis,23–26 and we found that in patients with multiple-territory disease, the long-term risks of nonstroke acute vascular events approached the risks of recurrent ischemic stroke. In our exploratory analyses, we found that numbers of affected vascular beds were associated with long-term risks of recurrent major cardiovascular events independent of known vascular risk factors, and multiple-territory disease seemed to be a stronger predictor than vascular risk factors measured at baseline. This perhaps reflects the fact that crude prevalence of reported vascular risk factors is not always an adequate measure of risk because of measurement error, premorbid preventative treatment, and different individual susceptibility. Hence, the number of affected vascular beds is a perhaps more informative summative measure of risk.

eflects the fact that crude prevalence of reported vascular risk factors is not always an adequate measure of risk because of measurement error, premorbid preventative treatment, and different individual susceptibility. Hence, the number of affected vascular beds is a perhaps more informative summative measure of risk. Although we consider our conclusions to be valid, our study has limitations. First, we did not screen patients for asymptomatic coronary or peripheral vascular disease and will have underestimated the real burden of multiple-territory disease. However, screening for asymptomatic coronary or peripheral vascular disease is not routine in clinical practice. Moreover, even in patients with no known coronary heart disease, statins have been shown to reduce major coronary-related events in patients with TIA or ischemic stroke.6 Second, although the majority of the patients were on a statin during long-term follow-up, the exact regime varied (usually ranging from simvastatin 40 mg daily to atorvastatin 80 mg daily). However, this heterogeneity does reflect real-world clinical practice. Third, we did not routinely recheck lipid levels during follow-up, as this is the responsibility of primary care physicians in the UK healthcare system, and so we did not have systematic data on the quality of cholesterol control. Fourth, although we used several overlapping methods (ie, interviews, ongoing daily ascertainment, primary diagnostic coding, death certificates, and national hospital coding) to achieve near complete follow-up, a small proportion of patients (<1%) emigrated from the United Kingdom and could not always be followed-up. Finally, our results based on a predominantly white population in the United Kingdom might not be generalizable to other countries.

certificates, and national hospital coding) to achieve near complete follow-up, a small proportion of patients (<1%) emigrated from the United Kingdom and could not always be followed-up. Finally, our results based on a predominantly white population in the United Kingdom might not be generalizable to other countries. To conclude, in a population-based cohort of TIA and ischemic stroke patients treated with contemporary standard secondary prevention, we found that patients with multiple-territory disease had a very high risk of recurrent vascular events during long-term follow-up, suggesting that number of affected vascular beds could potentially be a simple clinical rule in identifying patients who are at high risk of recurrent vascular events. Our updated risk estimates with contemporary secondary prevention therapies could also help to inform the design of future randomized trials. Finally, patients with the multiple-territory disease might benefit from more intensive prevention with novel therapies and should be the focus of future clinical trials.

events. Our updated risk estimates with contemporary secondary prevention therapies could also help to inform the design of future randomized trials. Finally, patients with the multiple-territory disease might benefit from more intensive prevention with novel therapies and should be the focus of future clinical trials. Acknowledgments This article is dedicated to Rose Wharton, who provided statistical support but sadly died prior to publication. We are also grateful to all the staff in the general practices that collaborated in the Oxford Vascular Study: Abingdon Surgery, Stert St, Abingdon; Malthouse Surgery, Abingdon; Marcham Road Family Health Centre, Abingdon; The Health Centre, Berinsfield; Key Medical Practice; Kidlington; 19 Beaumont St, Oxford; East Oxford Health Centre, Oxford; Church Street Practice, Wantage. We also acknowledge the use of the facilities of the Acute Vascular Imaging Centre, Oxford.

thouse Surgery, Abingdon; Marcham Road Family Health Centre, Abingdon; The Health Centre, Berinsfield; Key Medical Practice; Kidlington; 19 Beaumont St, Oxford; East Oxford Health Centre, Oxford; Church Street Practice, Wantage. We also acknowledge the use of the facilities of the Acute Vascular Imaging Centre, Oxford. Sources of Funding Wellcome Trust, Wolfson Foundation, British Heart Foundation, and the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre. Dr Rothwell is in receipt of an NIHR Senior Investigator Award and Dr Heldner is in receipt of a Swiss National Science Foundation Advanced Postdoc Mobility Grant, of a Novartis Foundation for medical-biological Research Grant and of a B. Braun Foundation Grant. Drs Heldner and Li collected data, did the statistical analysis and interpretation, wrote and revised the article. Drs Lovett, Kubiak, and Lyons collected the data. Professor P.M. Rothwell conceived and designed the overall study, provided study supervision and funding, acquired, analyzed and interpreted data, and wrote and revised the article. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health. Disclosures None. Supplementary Material Guest Editor for this article was Gregory W. Albers, MD. The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.118.020913/-/DC1.

Cerebral small vessel disease (SVD), which accounts for 20% to 25% of all strokes and up to 45% of all dementias,1 is a slowly progressive disorder, often with subtle features initially,2 but frequently progressing into a chronic disabling vasculopathy with cognitive impairment, depression, and gait disturbances.1,3 Although the pathogenesis of SVD is incompletely understood,1 and novel mechanisms have been postulated,1,4 hypertension is one of the leading causes,1,3,5 and has been associated with all main neuroimaging biomarkers of SVD—lacunes,6 white matter hyperintensity (WMH),7 microbleeds,8 and magnetic resonance imaging (MRI)-visible enlarged perivascular spaces (PVSs),9 as well as the global burden of SVD as assessed by the total SVD score.10

is one of the leading causes,1,3,5 and has been associated with all main neuroimaging biomarkers of SVD—lacunes,6 white matter hyperintensity (WMH),7 microbleeds,8 and magnetic resonance imaging (MRI)-visible enlarged perivascular spaces (PVSs),9 as well as the global burden of SVD as assessed by the total SVD score.10 The majority of previous studies on the association between hypertension and SVD (or other novel risk factors where hypertension has been adjusted for) have been cross-sectional, or have based on single clinic or ambulatory blood pressure (BP) measurements, known history of hypertension and prior use of antihypertensive agents,6,7,9,10 potentially underestimating the effects of BP during the many years before clinical presentation. Since systolic BP (SBP) increases, and diastolic BP (DBP) decreases with age,11 and in view of the evidence that midlife hypertension may be an important determinant of later cerebrovascular disease and dementia,12,13 it may well be important to consider BP many years before the assessment of SVD. Indeed, in one recent prospective cohort study,14 the associations between both SBP and DBP and WMH were attenuated during 3-year follow-up as the mean age of the cohort increased from 70 to 73 years. We showed previously that recent premorbid BP may be a trigger for acute lacunar events15 and intracerebral hemorrhage,16 but analyses of much longer-term premorbid BP control may be required to reliably determine the association with chronic SVD, which is a slowly progressive disorder. To better understand the role of long-term prior BP in development of SVD before transient ischemic attack (TIA)/ischemic stroke, we determined the age-specific time-course of premorbid BP in relation to the total SVD score17 in the population-based OXVASC (Oxford Vascular Study).

hich is a slowly progressive disorder. To better understand the role of long-term prior BP in development of SVD before transient ischemic attack (TIA)/ischemic stroke, we determined the age-specific time-course of premorbid BP in relation to the total SVD score17 in the population-based OXVASC (Oxford Vascular Study). Methods Request for access to data will be considered by the corresponding author. We prospectively studied patients with TIA/ischemic stroke from OXVASC. In brief, OXVASC is an ongoing population-based study of all acute vascular events occurring within a population of 92 728 individuals, irrespective of age, who are registered with 100 general practitioners in 9 general practices of Oxfordshire, United Kingdom.18 The analysis herein includes 1080 consecutive cases of TIA/ischemic stroke recruited from November 1, 2004, to September 30, 2014, who had an MRI brain imaging. The imaging protocol of OXVASC has been described in detail elsewhere.19,20 Briefly, from April 1, 2002, to March 31, 2010 (phase 1), MRI and magnetic resonance angiography was performed in selected patients when clinically indicated. From April 1, 2010 onwards (phase 2), brain MRI and magnetic resonance angiography became the first-line imaging methods.

een described in detail elsewhere.19,20 Briefly, from April 1, 2002, to March 31, 2010 (phase 1), MRI and magnetic resonance angiography was performed in selected patients when clinically indicated. From April 1, 2010 onwards (phase 2), brain MRI and magnetic resonance angiography became the first-line imaging methods. We collected demographic data, atherosclerotic risk factors, details of hospitalization of index event during face-to-face interview and cross-referenced these with primary care and hospital records. All patients had their BPs measured during ascertainment using an oscillometric BP measurement device (A&D Medical, Japan). BPs were taken after 5 minutes of rest in the sitting or lying position, and a single BP reading was used for analysis. Hypertension was defined as known history of hypertension or prior use of antihypertensive agents. We also collected premorbid BP readings from the primary care records (both paper and electronic) for all patients during the preceding 20 years before ascertainment and calculated the mean of all readings, and readings taken between 1 to 5 years, 5 to 10 years, and 10 to 20 years before TIA/ischemic stroke were used for analysis.

ected premorbid BP readings from the primary care records (both paper and electronic) for all patients during the preceding 20 years before ascertainment and calculated the mean of all readings, and readings taken between 1 to 5 years, 5 to 10 years, and 10 to 20 years before TIA/ischemic stroke were used for analysis. Patients were scanned predominantly (856 out of 1009 patients) with either of 2 scanners—Achieva, Philips Healthcare (1.5T, n=481), and Magnetom Verio, Siemens Healthcare (3T, n=375).19,20 Details of scan parameters are provided in Table I in the online-only Data Supplement. MRI-visible enlarged PVSs were defined as small (<3 mm) punctate (if perpendicular to the plane of scan) or linear (if longitudinal to the plane of scan) hyperintensities on T2 images in the basal ganglia based on a previously validated scale.21 Burden of PVSs was then stratified into 3 groups: <11, 11 to 20, and >20. The severity of WMH was determined for each patient according to the Fazekas scale.22 Cerebral microbleeds were defined as rounded, hypodense foci up to 10 mm in size and were differentiated from microbleed mimics based on current guidelines.23 The location and number of microbleeds were scored according to the Microbleed Anatomical Rating Scale.24 Lacunes were defined as rounded or ovoid lesions, >3 and <20 mm in diameter, in the basal ganglia, internal capsule, centrum semiovale, or brain stem, of cerebrospinal fluid signal density on T2 and fluid-attenuated inversion recovery and no increased signal on diffusion-weighted imaging.25 The total burden of SVD was represented by calculating the total SVD score where one point is allocated to each of the following: (1) presence of lacunes, (2) presence of microbleeds, (3) moderate-severe (>10) MRI-visible enlarged basal ganglia-PVSs, and (4) severe periventricular and moderate-severe deep WMH.17

ing.25 The total burden of SVD was represented by calculating the total SVD score where one point is allocated to each of the following: (1) presence of lacunes, (2) presence of microbleeds, (3) moderate-severe (>10) MRI-visible enlarged basal ganglia-PVSs, and (4) severe periventricular and moderate-severe deep WMH.17 One senior neuroradiologist (Dr Küker), who was blinded to the premorbid and baseline BP readings, provided ongoing supervision of interpretation of the MRI images throughout the study period. Definitions of neuroimaging biomarkers were based on Standards for Reporting Vascular Changes on Neuroimaging (STRIVE).25 The intrarater κ for 50 randomly selected scans was: lacunes=0.85, microbleed burden (0, 1, 2–4, ≥5)=0.88, periventricular WMH burden (Fazekas grade 0, 1, 2, 3)=0.66, subcortical WMH burden (Fazekas grade 0, 1, 2, 3)=0.75, and basal ganglia-PVS burden (<11, 11–20, >20)=0.86. Patients gave written informed consent after an event or assent was obtained from relatives for patients who were unable to provide consent. The study was approved by the local research ethics committee.

One senior neuroradiologist (Dr Küker), who was blinded to the premorbid and baseline BP readings, provided ongoing supervision of interpretation of the MRI images throughout the study period. Definitions of neuroimaging biomarkers were based on Standards for Reporting Vascular Changes on Neuroimaging (STRIVE).25 The intrarater κ for 50 randomly selected scans was: lacunes=0.85, microbleed burden (0, 1, 2–4, ≥5)=0.88, periventricular WMH burden (Fazekas grade 0, 1, 2, 3)=0.66, subcortical WMH burden (Fazekas grade 0, 1, 2, 3)=0.75, and basal ganglia-PVS burden (<11, 11–20, >20)=0.86. Patients gave written informed consent after an event or assent was obtained from relatives for patients who were unable to provide consent. The study was approved by the local research ethics committee. Statistical Analysis We determined by binary and ordinal logistic regression, the relationships of hypertension, baseline BP (top versus bottom quartile as referent) and mean premorbid BP (top versus bottom quartile) with the presence of lacunes and the total SVD score, in univariate analysis and analyses adjusted for age and sex. Test of parallel lines was performed to examine the equal slope assumption on ordinal logistic regression. We also determined by ordinal logistic regression, the relationships of mean premorbid BP taken within 1 year, 1 to 5 years, 5 to 10 years, and 10 to 20 years before TIA/ischemic stroke with the total SVD score overall, and also stratified analysis by age (<70 versus ≥70—chosen as the approximate median age) and by premorbid use of antihypertensive agents. Given the interrelation between years before event and age at the time of BP measurement, we also determined the relationships of total SVD score with mean premorbid BP of measurements taken when patients were aged ≤55, 56 to 65, 66 to 75, and >75. All analyses were done with SPSS version 22.

f antihypertensive agents. Given the interrelation between years before event and age at the time of BP measurement, we also determined the relationships of total SVD score with mean premorbid BP of measurements taken when patients were aged ≤55, 56 to 65, 66 to 75, and >75. All analyses were done with SPSS version 22. Role of the Funding Source The funding source had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had the final responsibility for the decision to submit for publication.

f antihypertensive agents. Given the interrelation between years before event and age at the time of BP measurement, we also determined the relationships of total SVD score with mean premorbid BP of measurements taken when patients were aged ≤55, 56 to 65, 66 to 75, and >75. All analyses were done with SPSS version 22. Role of the Funding Source The funding source had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had the final responsibility for the decision to submit for publication. Results One thousand eighty patients were recruited during the study period. After excluding 71 patients (6.6%) with missing clinical, premorbid BP, or imaging data, 1009 patients (TIA, n=528; ischemic stroke, n=481) were included in the final analysis. Details of baseline clinical and imaging characteristics are shown in Table 1. The mean (SD) age of the study population was 68.6 (13.8) years and 52% were male. Fifty-five percent of the study population had a history of hypertension or were on antihypertensive agents. A total of 22 096 premorbid BP readings (median, 15 readings/patient; interquartile range [7–33]; 9 [4–21] in age <70 and 23 [12–41] in aged ≥70) were obtained. The mean (SD) premorbid BP was 139 (14)/80 (8) mm Hg, whereas the mean BP on assessment was 150 (24)/84 (13) mm Hg. The mean (SD) total SVD score was 1.12 (1.11). Compared with patients ≥70 years, those aged <70 were more likely to be men, smokers, had fewer vascular risk factors, better renal function, lower premorbid and baseline mean SBP, and higher premorbid and baseline mean DBP (Table 1). The overall prevalence of individual neuroimaging markers and burden of SVD was also lower in patients aged <70 (Table 1).