Customize your JAMA Network experience by selecting one or more topics from the list below.
Yaghi S, Rostanski SK, Boehme AK, et al. Imaging Parameters and Recurrent Cerebrovascular Events in Patients With Minor Stroke or Transient Ischemic Attack. JAMA Neurol. 2016;73(5):572–578. doi:10.1001/jamaneurol.2015.4906
Copyright 2016 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
Neurological worsening and recurrent stroke contribute substantially to morbidity associated with transient ischemic attacks and strokes (TIA-S).
To determine predictors of early recurrent cerebrovascular events (RCVEs) among patients with TIA-S and National Institutes of Health Stroke Scale scores of 0 to 3.
Design, Setting, and Participants
A retrospective cohort study was conducted at 2 tertiary care centers (Columbia University Medical Center, New York, New York, and Tulane University Medical Center, New Orleans, Louisiana) between January 1, 2010, and December 31, 2014. All patients with neurologist-diagnosed TIA-S with a National Institutes of Health Stroke Scale score of 0 to 3 who presented to the emergency department were included.
Main Outcomes and Measures
The primary outcome (adjudicated by 3 vascular neurologists) was RCVE: neurological deterioration in the absence of a medical explanation or recurrent TIA-S during hospitalization.
Of the 1258 total patients, 1187 had no RCVEs and 71 had RCVEs; of this group, 750 patients (63.2%) and 39 patients (54.9%), respectively, were aged 60 years or older. There were 505 patients with TIA-S at Columbia University; 31 (6.1%) had RCVEs (15 patients had neurological deterioration only, 11 had recurrent TIA-S only, and 5 had both). The validation cohort at Tulane University consisted of 753 patients; 40 (5.3%) had RCVEs (24 patients had neurological deterioration only and 16 had both). Predictors of RCVE in multivariate models in both cohorts were infarct on neuroimaging (computed tomographic scan or diffusion-weighted imaging sequences on magnetic resonance imaging) (Columbia University: not applicable and Tulane University: odds ratio, 1.75; 95% CI, 0.82-3.74; P = .15) and large-vessel disease etiology (Columbia University: odds ratio, 6.69; 95% CI, 3.10-14.50 and Tulane University: odds ratio, 8.13; 95% CI, 3.86-17.12; P < .001). There was an increase in the percentage of patients with RCVEs when both predictors were present. When neither predictor was present, the rate of RCVE was extremely low (up to 2%). Patients with RCVEs were less likely to be discharged home in both cohorts.
Conclusions and Relevance
In patients with minor stroke, vessel imaging and perhaps neuroimaging parameters, but not clinical scores, were associated with RCVEs in 2 independent data sets. Prospective studies are needed to validate these predictors.
Recurrent cerebrovascular events (RCVEs) are one of the main determinants of functional outcomes in patients after minor strokes and transient ischemic attacks (TIA).1,2 The risk of recurrence is highest within 90 days and is particularly high in the first 48 hours.3 For example, in a recent clinical trial in China,4 the risk of recurrent stroke after high-risk TIA (ABCD2 score ≥4) or minor stroke (National Institutes of Health Stroke Scale [NIHSS] score, 0-3) was up to 12% at 90 days, with more than 50% of the recurrent events occurring in the first week. Several scores have been derived using retrospective cohorts to predict the early risk of stroke after TIA; the most widely used is the ABCD2 score.3,5 Subsequently developed scores have incorporated vascular and parenchymal imaging.6-9 Moreover, recent studies have challenged the value of the ABCD2 score for predicting stroke recurrence risk.10 However, the ABCD2 score and more recent scores are limited by their derivation from mostly nonneurologist-diagnosed TIA samples and by their development before the 2009 American Heart Association–endorsed TIA definition, which takes into account the presence of infarct on neuroimaging.11 Prediction scores have not been frequently applied to minor stroke,9 which may carry the same or higher risk of early recurrent stroke as TIA.4
Establishing which patients with TIA or strokes (TIA-S; NIHSS score, 0-3) are at highest risk of RCVEs allows for targeted clinical trials, some of which are now under way and focus on antithrombotic agents in a broad population of patients with TIA-S. Our study aimed to identify predictors of RCVEs in a contemporary population of patients with neurologist-diagnosed TIA-S. Because large-vessel disease (LVD) has the highest likelihood of early stroke recurrence among ischemic stroke subtypes,12 we hypothesized that patients with the LVD stroke subtype are at the highest risk of stroke recurrence or neurological deterioration.
Question What predicts early recurrent cerebrovascular events (RCVEs) among patients with transient ischemic attacks and minor strokes and National Institutes of Health Stroke Scale scores of 0 to 3?
Findings Recurrent cardiovascular event predictors on multivariate models were infarct on neuroimaging and large-vessel disease etiology. There was an increase in the percentage of patients with RCVEs when both predictors were present. When neither predictor was present, the rate of RCVE was extremely low (up to 2%).
Meaning In patients with minor stroke, vessel imaging and perhaps neuroimaging parameters but not clinical scores predicted RCVEs in 2 independent patient cohorts.
We retrospectively analyzed a prospective cohort of consecutive patients hospitalized and treated at the Columbia University Medical Center, New York, New York. Patients included those with (1) a diagnosis of TIA-S confirmed by a vascular neurologist between January 1, 2010, and December 31, 2014; (2) an evaluation at the emergency department; (3) an NIHSS score of 0 to 3 (similar to the cutoffs used in clinical trials4,13); and (4) who presented within 12 hours after symptom onset.13 As a general rule at the Columbia University Medical Center, we hospitalized all patients with suspected TIA-S during the study period. Patients who were not hospitalized with TIA-S were also included in our prospective database. Approval for this study was obtained from the Columbia University Medical Center institutional review board; informed consent was waived because the study reviewed medical records and retrospectively collected data.
The primary outcome in the study was RCVE during hospitalization, which included recurrent TIA-S and neurological deterioration. Recurrent TIA was included given its inclusion in prior risk-stratification scores11 as an important outcome; neurological deterioration was included given evidence that it can cause long-term disability14 and was used in prior TIA-S clinical trials.4 Recurrent TIA-S included new neurological symptoms in the absence of hemorrhage on brain imaging, not attributable to any other medical condition and confirmed clinically or by imaging. Neurological deterioration was defined as any worsening of presenting neurological deficits detected on a full neurological examination or recurrence of deficits not attributable to any other medical condition (such as fever, infection, or metabolic derangement).4 Owing to the potential overlap between recurrent TIA-S and neurological deterioration, we opted to combine both measures to constitute our primary outcome, similar to what previous investigators have done.4 Our primary outcome was adjudicated by 3 independent stroke neurologists (S.Y., S.K.R., and J.Z.W.).
The primary predictor was the LVD stroke subtype, which was determined by vascular imaging and defined as greater than 50% stenosis of extracranial or intracranial arteries in the territory of the syndrome identified by a stroke neurologist. The imaging modalities used were extracranial computed tomographic (CT) angiography, extracranial magnetic resonance angiography, carotid ultrasonography, intracranial CT angiography, and intracranial magnetic resonance angiography. As a general rule, all patients received intracranial and extracranial vessel imaging as part of their diagnostic evaluation, although data on the type of imaging modality were not systematically collected in our cohorts. The stroke subtype was determined by reviewing the medical records, imaging, and discharge summary of patients, and we used the Trial of ORG 10172 in Acute Stroke Treatment criteria to classify stroke subtype.15
Covariates were obtained from medical records at the time of evaluation and included demographic characteristics (age, sex, and race/ethnicity), vascular risk factors (hypertension, diabetes, hyperlipidemia, atrial fibrillation, coronary heart disease, congestive heart failure, peripheral vascular disease, smoking status, history of TIA-S, and active cancer), clinical variables (admission systolic blood pressure, admission diastolic blood pressure, admission NIHSS score, symptom duration [for transient deficits], symptom type [motor, speech, or other], and recent similar symptoms), and neuroimaging parameters (new infarct on brain CT or magnetic resonance imaging [MRI]). Both CTs and MRIs were performed on most patients with acute ischemic stroke in our institution unless there was a medical contraindication to MRI. We also assessed functional outcome (discharge-modified Rankin Scale score, discharge disposition [home vs not home], and length of hospital stay) and prediction scores (ABCD2, ABCD3, and ABCD3-I).3,5 Poor discharge outcome was defined by a discharge-modified Rankin Scale score of 2 or more; this definition was used because the patient population in our study had TIA-S, so we were more conservative in defining good outcomes, which was also done in a previous study.2
Patients were divided into 2 groups based on whether they had an RCVE within their hospitalization. Our primary predictor (LVD stroke subtype), baseline demographic characteristics, vascular risk factors, clinical variables, neuroimaging parameters, functional outcomes, and prediction scores were compared between the 2 groups using nonparametric tests for continuous variables and Fisher exact tests for categorical variables. A multivariate model was built that included the LVD subtype (compared with other subtypes) and other predictors on univariate analysis to determine the predictors of RCVEs in our cohort. P < .05 was considered statistically significant.
To confirm our findings, we used an independent data set of patients evaluated at Tulane University, New Orleans, Louisiana, with (1) a discharge diagnosis of TIA-S confirmed by a vascular neurologist between January 1, 2010, and December 31, 2014; (2) an evaluation at the emergency department; and (3) an NIHSS score at admission of 0 to 3. Similarly, almost all patients presenting to the emergency department with a suspected diagnosis of TIA-S were hospitalized. Neurological deterioration and recurrent stroke outcomes were captured in the Tulane data set, and our combined outcome, RCVE, was recorded. We used a similar analytical plan to confirm the predictors of RCVEs in this data set. The Tulane Stroke Registry is approved by the Tulane University institutional review board; informed consent for this study was waived because the registry contains deidentified data.
The Columbia University cohort included 505 patients (141 with TIA and 364 with minor stroke) with a mean (SD) length of hospital stay of 3.2 (4.1) days. The LVD stroke subtype was present in 64 patients (12.7%); of those, 21 (32.8%) were extracranial, 42 (65.6%) were intracranial, and 1 (1.6%) was both extracranial and intracranial. Thirty-one patients (6.1%) had RCVEs (15 patients had neurological deterioration only, 11 had recurrent TIA-S only, and 5 had both). An infarct was present on neuroimaging (MRI or CT) in 264 patients (52.2%), and 61 patients (12.1%) had positive CT scans.
In univariate analyses, predictors of RCVEs were the LVD stroke subtype (45.2%, n = 14 vs 10.5%, n = 50; P = .005) and the presence of infarct on neuroimaging (96.8%, n = 30 vs 49.4%, n = 234; P < .001). The median admission systolic blood pressure was higher in patients with vs without RCVEs (171 mm Hg vs 156 mm Hg; P = .04). The rest of the variables were not significantly associated with RCVEs and are listed in Table 1. Notably, the ABCD2 and ABCD3 scores, which do not include neuroimaging and vascular imaging, were not predictive of RCVEs, whereas the ABCD3-I score, which includes imaging, was predictive (P < .001); the result was driven by imaging alone (Table 1).
In a multivariate model that included LVD and systolic blood pressure, the LVD stroke subtype (compared with other subtypes) was the only predictor of RCVEs (odds ratio = 6.69; 95% CI = 3.10-14.50) (Table 2). Infarct presence on neuroimaging could not be included in the model because only 1 patient with negative neuroimaging had an RCVE.
The Tulane University cohort included 753 patients (272 with TIA and 481 with minor stroke) with a mean (SD) length of hospital stay of 4.5 (4.3) days. The LVD stroke subtype was present in 88 patients (11.7%). Forty patients (5.3%) had RCVEs (24 patients had neurological deterioration only and 16 had both neurological deterioration and recurrent TIA-S). An infarct was present on neuroimaging (MRI or CT) in 318 of 662 patients (48.0%), and 114 patients (17.2%) had positive CT scans.
In univariate analyses, predictors of RCVEs were the LVD stroke subtype (45.0%, n = 18 of 40 vs 16.1%, n = 70 of 436; P < .001) and the presence of infarct on neuroimaging (71.1%, n = 27 of 38 vs 46.6%, n = 291 of 624; P = .003). In addition, compared with patients without RCVEs, patients with RCVEs were younger (median age, 56.5 vs 62.0 years; P = .02) and had a higher median admission diastolic blood pressure (99 mm Hg vs 90 mm Hg; P = .02). The rest of the variables were not significantly associated with RCVEs and are listed in Table 1.
In a multivariate model that included the LVD stroke subtype, age, diastolic blood pressure, and infarct presence on neuroimaging, LVD was the only predictor significantly associated with RCVEs (odds ratio = 8.13; 95% CI = 3.86-17.12). There was a trend toward an association between infarct presence on neuroimaging and RCVEs (odds ratio = 1.75; 95% CI = 0.82-3.74) (Table 2). We noted that the ABCD3-I score was colinear with vascular and parenchymal imaging and therefore was not included owing to instability and poor fit in the model.
To further analyze the RCVE predictors, we determined the proportion of patients with RCVEs in each of the following categories: LVD stroke subtype, infarct presence on neuroimaging, both LVD stroke subtype and infarct presence on neuroimaging, and neither LVD stroke subtype nor infarct presence on neuroimaging. These results demonstrated an increase in the percentage of patients with RCVEs when both predictors were present. When neither predictor was present, the rate of RCVEs was extremely low (up to2%). These were similar in both cohorts, and results are summarized in the Figure.
Patients with RCVEs were more likely to have a poor outcome at discharge. When discharge to home was considered a good outcome, patients without RCVEs were more likely to be discharged home in the Columbia University data set (90.1%, n = 427 vs 38.7%, n = 12; P < .001) and the Tulane University data set (85.0%, n = 606 vs 32.5%, n = 13; P < .001). When a poor discharge functional outcome was considered a modified Rankin Scale score of 2 or more, patients with RCVEs were less likely to have a good discharge outcome in the Columbia University data set (84.3%, n = 381 of 452 vs 32.3%, n = 10; P < .001) and the Tulane University data set (62.9%, n = 437 of 695 vs 15.0%, n = 6 of 40; P < .001).
We performed sensitivity analysis on the Columbia University data set, excluding patients with an admission diagnosis of TIA because the event rate was low in patients with TIA (1 of 141; 0.7%), and the predictors of RCVEs remained unchanged. Sensitivity analysis could not be performed on the Tulane University data set because patients with an admission diagnosis of TIA who had recurrent in-hospital stroke were coded as having had stroke and not TIA.
Our findings demonstrate the importance of using imaging parameters to risk stratify patients with TIA-S, and these findings were confirmed in 2 independent, ethnically different, and diverse patient cohorts. Incorporating neuroimaging and vascular imaging into stroke prediction after TIA provides the most important information, which can be used to alter initial clinical management. For example, the use of aggressive stroke-prevention strategies, such as dual antiplatelet agents and high-intensity statin therapy after a TIA-S, may be particularly useful in patients with the LVD stroke subtype,16 who were shown to have the highest risk of early recurrence or deterioration. The detection of extracranial internal carotid stenosis also holds the potential for changing clinical management by allowing early revascularization. In addition, the use of early vessel imaging in this group of patients may identify patients with otherwise unsuspected large-vessel occlusion, which is associated with significant rates of clinical decline in the acute period.17
In our study, clinical scores, such as the ABCD2 and ABCD3 scores, were not useful in predicting RCVEs, but the ABCD3-I score, which includes neuroimaging and vascular imaging, was associated with RCVEs. Our findings on the poor performance of the ABCD2 score were not surprising given the work of others.10,18,19 The ABCD2 score performed better when used by nonneurologists compared with neurologists,20 which could be the reason why this score did not predict the risk of recurrent events in our cohort. Most importantly, the ABCD2 score was shown to poorly correlate with the presence of symptomatic extracranial carotid stenosis, which has a relatively high early stroke recurrence risk20-22 and can be reduced by urgent surgical or endovascular revascularization.16 However, our study differs from others in suggesting that when the tissue-based definition of TIA is used, the risk of early RCVEs is low.
Interestingly, in both cohorts, the rate of recurrent stroke after TIA-S was extremely low (up to 2%) in patients with nonatherosclerotic TIA-S subtype and absence of infarct on neuroimaging. The rate was close to 30% in patients with the LVD stroke subtype and detectable infarct on neuroimaging. Our study emphasizes the importance of urgent parenchymal and vascular imaging to risk stratify patients with TIA-S. We did find an association between atrial fibrillation and a lower risk of RCVEs, which was unexpected. On the other hand, patients with atrial fibrillation have a low risk of recurrence within 2 weeks after stroke,23 are likely to be treated with anticoagulation for small infarcts, and infrequently exhibit flow failure. An additional finding in our study was that RCVEs were an important driving force for functional outcomes in this patient population, and thus, aggressive measures to reduce the risk of RCVEs are needed to improve functional outcomes and cause less stroke disability in this patient population.
Our study had several limitations, including its retrospective nature and the relatively low number of RCVEs in both cohorts. Furthermore, owing to the extremely low recurrence rate in patients with TIA in our cohort, our findings may be more applicable to patients with minor stroke. Therefore, we may have been underpowered to detect more subtle effects from other predictors on risk of RCVEs. In addition, because our screening methods were imperfect, there is a possibility we missed patients with TIA-S during the study period, especially those discharged from the emergency department. However, it is a general rule in both institutions to hospitalize all patients with suspected TIA-S. Additionally, in the Columbia University cohort, patients who were diagnosed as having TIA-S and discharged from the emergency department were included in the data set. We acknowledge that some patients with TIA may have had a mimic instead, although we included only patients who had been diagnosed as having TIA by a board-certified vascular neurologist a priori in our study. Moreover, because the Tulane University cohort used a discharge rather than admission diagnosis of stroke or TIA, a small number of patients with TIA who had RCVEs were coded as having had a stroke and not a TIA. However, these patients were still included in our data set and analyzed in our study because their admission NIHSS score was 0. We included patients on admission with both TIA and minor ischemic stroke, which may represent different clinical entities because the recurrence rate in patients with TIA in our study was very low. On the other hand, TIA and minor ischemic stroke still have the same risk of recurrence in other studies, and ongoing clinical trials do not make a distinction between the 2. Furthermore, the study outcome, RCVEs, was assessed retrospectively; however, it was adjudicated by independent stroke neurologists.
Our study also had several strengths. Most importantly, we showed similar predictors of RCVEs in 2 independent cohorts with major differences in race/ethnic distributions, making our results more robust and generalizable.
In patients with minor stroke, vessel imaging and perhaps neuroimaging parameters predicted RCVEs in 2 independent, ethnically different patient cohorts. Recurrent cerebrovascular events were associated with poor functional outcome. The emphasis on early vascular imaging is generalizable and can be implemented widely in clinical practice; whether rapid outpatient evaluations can be streamlined by our predictors remains to be studied in a clinical trial. Prospective studies are needed to validate these predictors.
Corresponding Author: Shadi Yaghi, MD, Department of Neurology, Warren Alpert Medical School, Brown University, 593 Eddy St, APC 530, Providence, RI 02903 (firstname.lastname@example.org).
Accepted for Publication: December 15, 2015.
Published Online: March 21, 2016. doi:10.1001/jamaneurol.2015.4906.
Author Contributions: Dr Yaghi had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Yaghi, Blum, Marshall, Willey.
Acquisition, analysis, or interpretation of data: Yaghi, Rostanski, Boehme, Martin-Schild, Samai, Silver, Jayaraman, Siket, Khan, Furie, Elkind, Willey.
Drafting of the manuscript: Yaghi.
Critical revision of the manuscript for important intellectual content: Rostanski, Boehme, Martin-Schild, Samai, Silver, Blum, Jayaraman, Siket, Khan, Furie, Elkind, Marshall, Willey.
Statistical analysis: Boehme, Khan.
Administrative, technical, or material support: Rostanski, Martin-Schild, Samai, Furie, Willey.
Study supervision: Martin-Schild, Silver, Jayaraman, Elkind, Marshall, Willey.
Conflict of Interest Disclosures: Dr Elkind has received compensation for serving on advisory boards and consulting fees from Boehringer-Ingelheim Inc, Bristol-Myers Squibb–Pfizer Partnership, Daiichi-Sankyo, Janssen Pharmaceuticals, and BioTelemetry/Cardionet. Dr Silver has received compensation from the Joint Commission for surveyor activities, from the University of California, San Francisco, for adjudication in the SOCRATES Trial, from the Fred Hutchinson Cancer Center for adjudication in the Women’s Health Initiative, and for his expert review of medicolegal cases and received honoraria for authorship in Ebix, Medlink, and Medscape. No other disclosures were reported.
Funding/Support: Dr Yaghi was funded by the National Institutes of Health StrokeNet. Dr Boehme was funded by grant T32 NS007153-31 from the National Institute of Neurological Disorders and Stroke of the National Institutes of Health.
Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.