Cerebral Microbleeds and the Risk of Incident Ischemic Stroke in CADASIL (Cerebral Autosomal Dominant Arteriopathy With Subcortical Infarcts and Leukoencephalopathy)
Background and Purpose—Cerebral microbleeds are associated with an increased risk of intracerebral hemorrhage. Recent data suggest that microbleeds may also predict the risk of incident ischemic stroke. However, these results were observed in elderly individuals undertaking various medications and for whom causes of microbleeds and ischemic stroke may differ. We aimed to test the relationship between the presence of microbleeds and incident stroke in CADASIL (Cerebral Autosomal Dominant Arteriopathy With Subcortical Infarcts and Leukoencephalopathy)—a severe monogenic small vessel disease known to be responsible for both highly prevalent microbleeds and a high incidence of ischemic stroke in young patients.
Methods—We assessed microbleeds on baseline MRI in all 378 patients from the Paris–Munich cohort study. Incident ischemic strokes were recorded during 54 months. Survival analyses were used to test the relationship between microbleeds and incident ischemic stroke.
Results—Three hundred sixty-nine patients (mean age, 51.4±11.4 years) were followed-up during a median time of 39 months (interquartile range, 19 months). The risk of incident ischemic stroke was higher in patients with microbleeds than in patients without (35.8% versus 19.6%, hazard ratio, 1.87; 95% confidence interval, 1.16–3.01; P=0.009). These results persisted after adjustment for history of ischemic stroke, age, sex, vascular risk factors, and antiplatelet agents use (hazard ratio, 1.89; 95% confidence interval, 1.10–3.26; P=0.02).
Conclusions—The presence of microbleeds is an independent risk marker of incident ischemic stroke in CADASIL, emphasizing the need to carefully interpret MRI data.
The presence of cerebral microbleeds on brain MRI has become a major focus in vascular neurology because they are thought to reflect hemorrhagic-prone states. For instance, whether alteplase or antithrombotic drugs can be used safely in patients with ischemic strokes and multiple microbleeds has become strongly debated.1 In fact, the mechanisms leading to the presence of microbleeds on MRI remain incompletely understood,2 and a recent meta-analysis showed that the presence of microbleeds in patients with a history of stroke or transient ischemic attack is not only associated with an increased risk of first-ever hemorrhagic stroke as expected but also with an increased risk of recurrent ischemic events.3 However, in elderly individuals, microbleeds and incident ischemic strokes may be related to different but concurrent mechanisms with shared risk factors. Additionally, these individuals are likely exposed to various medications that may differently impact the prevalence of microbleeds and the incidence of ischemic stroke. A key argument supporting the link between microbleeds and incident ischemic stroke would be to observe similar results in patients having both microbleeds and incident ischemic stroke related to the same disorder, while minimizing the influence of medications.
CADASIL (Cerebral Autosomal Dominant Arteriopathy With Subcortical Infarcts and Leukoencephalopathy) is a pure, monogenic form of cerebral small vessel disease, which is characterized by both a high prevalence of microbleeds on MRI and a high frequency of ischemic strokes.4 Given the young age of onset of the disease, the probability of having different associated causes of microbleeds or of ischemic events is low.4 Moreover, no specific treatment is known to alter the course of the disease, and the vast majority of patients are treated with antiplatelet agents.4 In the present study, we aimed to determine the relationship between the presence of microbleeds and the risk of incident ischemic stroke in a large prospective cohort of CADASIL patients.
CADASIL patients were prospectively recruited between September 2003 and April 2011 in Lariboisière, Paris (n=249) and Munich (n=129). Included individuals were >18 years of age and had a genetically confirmed diagnosis. Details of the study protocol have been previously reported.5 Briefly, at inclusion and every 18 months until 54 months, clinical, biological, and MRI evaluations were obtained using standardized protocols. Clinical and demographic data were collected including age, sex, antithrombotic medication, and vascular risk factors: hypertension (defined as previous diagnosis of hypertension [>140/90] or use of antihypertensive treatment), hypercholesterolemia, and active smoking. History of stroke (rapidly evolving focal symptoms lasting ≥24 hours with no other cause than vascular origin) was recorded at baseline and at follow-up. Informed consent was obtained from each subject or from a close relative if the subject was too severely disabled. This study was approved by an independent ethics committee in both participating centers.
MRI Acquisitions and Processing
MRI scans were obtained on 1.5-T systems (General Electric Medical Systems Signa, Paris and Munich, or Siemens Magnetom Vision, Munich). Three-dimensional sequences (Paris: repetition time/echo time [TR/TE], 9.1/2 ms; slice thickness, 0.8 mm; no interslice gap; in-plane resolution, 1.02×1.02 mm2; Munich: TR/TE, 11.4/4.4 ms; slice thickness, 1.2 mm; no gap; in-plane resolution, 0.9×0.9 mm2), Fluid Attenuated Inversion Recovery (Paris: TR/TE/inversion time, 8402/161/2002 ms; slice thickness, 5.5 mm; no gap; in-plane resolution, 0.94×0.94 mm2; Munich: TR/TE/inversion time, 4284/110/1428 ms; slice thickness, 5 mm; no gap; in-plane resolution, 0.98×0.98 mm2), and T2*-weighted gradient echo imaging (Paris: TR/TE, 500/15 ms; slice thickness, 5.5 mm; no gap; in-plane resolution, 0.98×0.98 mm2; Munich: TR/TE, 1056/22 ms; slice thickness, 5 mm; no gap; in-plane resolution, 0.98×0.98 mm2) were performed.
We used validated methods1 following the STRIVE criteria (Standards for Reporting Vascular Changes on Neuroimaging) to extract the number of microbleeds from T2* sequences.6 Identification and quantification of white matter hyperintensities (WMH) and lacunes were performed according to STRIVE criteria.6 Masks of WMH were automatically determined on all axial Fluid Attenuated Inversion Recovery slices using adaptative intensity thresholding.5 Lacunes and dilated perivascular spaces were manually identified on 3-dimensional T1 images by experienced readers (the differentiation of lacunes from dilated perivascular spaces was based on current recommendations).6 In difficult cases, a consensus was obtained after reviewing all MRI data available. All masks were manually validated by double reading, as previously reported.5 Volumes of WMH and lacunes were obtained by multiplying the number of voxels in the masks with the voxel size depending on the sequence. The interrater reliability of the quantification of the volume of WMH, lacunes, and the number of microbleeds is known and all superior to 0.8.5 Determination of brain volumes from 3-dimensional T1 sequences was performed using Brainvisa software (http://brainvisa.info) after field inhomogeneities correction. Brain parenchymal fraction—a marker of brain atrophy—was defined as the ratio of brain tissue volume to the intracranial cavity volume as previously reported.7 To ease the interpretation of our analyses, volumes of WMH and of lacunes and brain parenchymal fraction were divided into quartiles.
Patients with or without a history of ischemic stroke were compared at inclusion according to the presence of microbleeds and vascular risk factors. We tested the relationship between the presence of microbleeds at inclusion and the occurrence of ischemic stroke during follow-up using survival analyses. The primary end point was the incidence of ischemic stroke. Patients were followed until date of ischemic stroke occurrence, date of death, date of last contact in the case of loss to follow-up, or at 54 months (end of the study period), whichever came first.
Kaplan–Meier incident stroke-free survival curves were computed for groups with or without microbleeds and were compared with the log-rank test. Cox proportional hazards regression models were then fitted to obtain the estimated hazard ratios and 95% confidence intervals (CIs). We used the corrected group prognosis method to build adjusted survival curves from proportional hazards models. Validity of Cox proportional hazard regression models was ensured by visual inspection of log cumulative hazard plots and of Shoenfeld residuals related to all covariates.
Missing data (2.2%) were imputed using a multiple-imputation method (Monte Carlo Markov Chain) including all variables in the analysis. We chose to generate 5 imputed datasets based on simulation studies demonstrating little gain in statistical power for higher number of imputations. All statistical analyses were performed using the R software (http://www.r-project.org/). For all tests, a P value of <0.05 was considered significant.
Among 378 patients, 9 were excluded because MRI data were not available at baseline. At inclusion, 227 (61.5%) individuals had experienced at least 1 prior ischemic event (stroke or transient ischemic attack). The demographic, clinical, biological, and MRI characteristics of the 369 included patients are summarized Table 1. Mean age at first stroke was 46±9.7 years. Two individuals (0.5%) had had an intracerebral hemorrhage. Microbleeds were identified in 131 of 369 patients (35.5%), with multiple microbleeds in 87 of 131 patients (66.4%). Three hundred and twenty-six patients (88.6%) were under antiplatelet agents.
The mean follow-up time was 39±17 months, and the median follow-up time was 39 months (25th–75th percentile, 35–54 months). One hundred and forty-one patients (38%) were lost to follow-up. Their mean follow-up time was 26±15 months. A total of 69 patients experienced an incident ischemic stroke during the follow-up period (31 patients with microbleeds and 38 without). Among these 69 patients, none had an intracerebral hemorrhage. The average incidence rate of ischemic stroke was 4.4 per 100 person-years. There were 27 deaths (7.3%) during the follow-up.
The Kaplan–Meier curves (Figure) highlighted a significant association between microbleeds at baseline and ischemic stroke incidence. The 54-month ischemic stroke-free survival rate was lower in patients with microbleeds than in patients without (64.2% versus 80.4%; log-rank test P=0.009). This corresponded to a 1.87-fold increase (95% confidence interval [CI], 1.16–3.01) of the risk of ischemic stroke in patients with microbleeds (35.8% versus 19.6%; P=0.009). In the Cox regression model (Table 2), patients with microbleeds had ≈2-fold higher risk to have an incident ischemic stroke independently of age, history of stroke, sex, vascular risk factors, and antiplatelet agents use (hazard ratio, 1.89; 95% CI, 1.10–3.26; P=0.02). We further adjusted the survival analyses for the volume of lacunes, volume of WMH, and brain parenchymal fraction (Table 3). In this fully adjusted model, the predictive value of microbleeds tended toward significance (hazard ratio, 1.71; 95% CI, 0.97–2.99; P=0.06).
In the present study, we found a significant association between the presence of microbleeds at inclusion and the risk of incident ischemic stroke in a large cohort of CADASIL patients. The risk increase was independent of known predictors of ischemic stroke. Roughly, the presence of microbleeds was associated with a 2-fold increased risk of incident ischemic stroke. Our results extend those of the meta-analysis evaluating the links between the presence of microbleeds and the risk of ischemic stroke after a stroke or a transient ischemic attack3 by controlling several important biases unavoidable in other stroke populations, particularly those related to different underlying causes for microbleeds and for ischemic events8 and the variable use of antithrombotic medications.
Indeed, it is important to understand that in most stroke populations, various mechanisms can artificially create a positive association between the presence of microbleeds and the risk of ischemic stroke. For instance, antiplatelet agents may be under prescribed in patients with microbleeds leading to higher measured rates of ischemic stroke. This was not the case in the present study. Patients were relatively young, all affected by the same disease, with few if any variability in undertaken medication. Additionally, a vast majority of patients was under antiplatelet agents whether or not they had a history of ischemic stroke, given the suspected high risk of ischemic events at time of inclusion in the cohort, whereas antiplatelet agents are now less often prescribed in primary prevention. We anyway forced the adjustment of our models for antiplatelet agent intake, which did not alter our results.
The results of the present study largely persisted after further adjustments for other MRI markers of small vessel disease, namely the volume of lacunes, the volume of WMH, and brain parenchymal fraction. Although the threshold for statistical significance only tended to be reached (P=0.06), the amplitude of the hazard ratio (1.71; 95% CI, 0.97–2.99) was close to that obtained in the adjusted model not including MRI markers as covariates. This suggests that significance was not reached because of a lack of power and that the link between microbleeds and incident stroke is not only mediated by the co-occurrence of lacunes, WMH, or undetected microlesions of ischemic origin that may translate into brain atrophy.9
In the present study, we did not include microbleed localization in our analyses. In the elderly, microbleed localization is clearly associated with the underlying microangiopathy. Hence, the importance to differentiate the potential impact of lobar microbleeds, strongly suggesting cerebral amyloid angiopathy with higher risks of intracerebral bleeds when compared with deep microbleeds, which are more associated with hypertension-related microangiopathies.10 Here, although we could not formally exclude that some microbleeds were actually related to associated microangiopathies, most of them were most likely secondary to CADASIL-specific mechanisms and as such, we had no reason to suspect that the relationships between the presence of microbleeds and the risk of ischemic stroke would be spatially determined.
In the present study, we did not either find a dose–response relationship between microbleed number or risk of ischemic stroke (data not shown). Although this could suggest that the links between microbleed number and the risk of ischemic stroke are indirect, an alternative explanation could be a lack of power to detect such an association. Indeed, although such dose–response relationships can be detected in samples with similar numbers of ischemic events in other stroke populations, CADASIL patients are likely to have, despite a higher risk of ischemic stroke, a far lower variability in this risk, thus, hampering the detection of dose–response relationships.
Surprisingly, we found in adjusted models that age was inversely related to the risk of ischemic stroke. Although we could not exclude unexpected characteristics of the natural history of CADASIL, the most likely explanation could be that ischemic events are more difficult to detect in patients with the most severe forms of the disease, which in most cases are also the oldest.
Our results and those of the abovementioned meta-analysis question the pathophysiology of microbleeds, in line with several reports that recently emphasized the possible heterogeneity of microbleeds that may in certain cases reflect ischemic mechanisms.11–14 Taken together, current and previously available data on microbleeds and stroke risk are compatible with the hypothesis that the presence of microbleeds helps identifying a subgroup of patients with severe forms of the disease rather than only quantifying the risk of intracranial bleeding.
Our study has some limitations. First, although CADASIL is considered a model of cerebral small vessel disease, our results may not be generalizable to other stroke populations. Although we systematically ensured that all strokes were actually ischemic and related to small vessel disease according to medical records, we did not systematically collect brain MRI obtained at the acute phase, which would have been mandatory for central control of stroke subtypes. However, in the 69 patients who had a stroke during the prospective follow-up study, we systematically checked on their subsequent MRI evaluation that no cortical lesion or intracerebral hemorrhage had appeared. In addition, they all had at least 1 new lacune. Altogether, this strongly supports that ischemic strokes were actually related to CADASIL. The period of recruitment was relatively long, and a potential heterogeneity in the inclusion of patients cannot be excluded. Given that our results were obtained from a cohort study, we could not exclude formally potential selection biases. Additionally, a substantial number of patients were lost to follow-up. However, one advantage of survival analysis is that randomly censored observations are considered until the time individuals ceased to be followed. Finally, although some patients had missing clinical data (8 patients; 2.2%), we also tested the models with multiple-imputation procedure without any significant alterations of our results.
Our study also has multiple strengths. First, we studied the impact of microbleeds on the prediction of incident stroke in a pure vascular disorder, affecting young patients unlikely to have comorbidities responsible for microbleeds or ischemic strokes. Also, given the high frequency of ischemic stroke in the disease, we could observe a large number of events in a small number of patients and adjust our analyses for known potential confounders.
- Received April 25, 2017.
- Revision received August 2, 2017.
- Accepted August 4, 2017.
- © 2017 American Heart Association, Inc.
- Jouvent E,
- Puy L,
- Chabriat H
- Fazekas F,
- Kleinert R,
- Roob G,
- Kleinert G,
- Kapeller P,
- Schmidt R,
- et al
- Charidimou A,
- Kakar P,
- Fox Z,
- Werring DJ
- Wardlaw JM,
- Smith EE,
- Biessels GJ,
- Cordonnier C,
- Fazekas F,
- Frayne R,
- et al
- Jouvent E,
- Viswanathan A,
- Mangin JF,
- O’Sullivan M,
- Guichard JP,
- Gschwendtner A,
- et al
- Jouvent E,
- Poupon C,
- Gray F,
- Paquet C,
- Mangin JF,
- Le Bihan D,
- et al
- Janaway BM,
- Simpson JE,
- Hoggard N,
- Highley JR,
- Forster G,
- Drew D,
- et al
- Grutzendler J,
- Murikinati S,
- Hiner B,
- Ji L,
- Lam CK,
- Yoo T,
- et al