Stroke Etiology and Outcomes after Endovascular Thrombectomy: Results from the SITS Registry and a Meta-Analysis

Population SITS thrombectomy registry

We collected data from the Safe Implementation of Treatment in Stroke (SITS) Thrombectomy register. Patients with acute ischemic stroke who underwent EVT treatment due to LAO stroke during January 1st, 2014 to September 3rd, 2019, with or without prior intravenous thrombolysis (IVT), were eligible for this study. Data collection and registration procedures have previously been presented [17,18]. Centers in the SITS Thrombectomy register commit to report all consecutive cases treated with EVT. This includes baseline and demographic characteristics, stroke severity per National Institutes of Health Stroke Scale (NIHSS), medical history, imaging data, and follow-up evaluation. Imaging studies, neurological status, and follow-up evaluation is performed according to local routines at each individual study center. Symptomatic intracranial hemorrhage (SICH) events were established based on locally read imaging data and clinical data entered into the registry. Study centers with less than 10 registered patients or more than 30% missing data on 3-month outcomes were excluded from this study.

We assigned stroke etiology using clinical and radiological characteristics and International Statistical Classification of Diseases and Related Health Problems version 10 (ICD-10) diagnosis, employing the criteria for stroke etiology from the TOAST definitions [1], based on steps that have previously been described [19]. We defined stroke etiology as LAA if the presence of a stenosis was observed during radiological examination at baseline, or if significant stenosis was observed during conventional angiography. Additionally, if no information on stenosis was available, we still defined stroke etiology as LAA if the patient was given an ICD-10 diagnosis of I63.0 (cerebral infarction due to thrombosis of precerebral arteries) or I63.3 (cerebral infarction due to thrombosis of cerebral arteries). We defined stroke etiology as CE if a patient had a known atrial fibrillation or the presence of atrial fibrillation was observed at baseline, 24 hours, discharge, or by 3 months. Additionally, if no information on atrial fibrillation was available, we still defined stroke etiology as CE if the patient was given an ICD-10 diagnosis of I63.4 (cerebral infarction due to embolism of cerebral arteries). Finally, if patients fulfilled both a LAA and CE according to our criteria, we defined those patient as having multiple etiologies, and excluded them from the analyses of the specific etiologies.

Outcomes

The primary outcome was successful reperfusion after EVT treatment, defined as a score of 2b-3 according to the modified Treatment in Cerebral Infarction (mTICI) sale [20]. The assessment of mTICI score was performed locally at study center. The mTICI scale is defined as: no perfusion (0); antegrade reperfusion past the initial occlusion, but limited distal branch filling with little or slow distal reperfusion (1); antegrade reperfusion of less than half of the occluded target artery previously ischemic territory (2a); antegrade reperfusion of more than half of the occluded target artery previously ischemic territory (2b); complete antegrade reperfusion of the previously occluded target artery ischemic territory, with absence of visualized occlusion in all distal branches (3).

Secondary outcomes were: complete reperfusion defined as mTICI 3, functional independence defined by a modified Ranking Scale (mRS) score [21] of 0–2 at 3 months, death within 3 months, SICH defined by the modified SITS Monitoring Study (SITS-MOST) criteria at the 24-hour control radiological examination (defined as a local or remote parenchymal hemorrhage type 2 or any subarachnoid hemorrhage on radiological findings on follow-up radiology at 22 to 36 hours post-stroke, in combination with a neurological deterioration ≥4 NIHSS-points or leading to death within 24 hours), number of EVT maneuvers, and procedural length of EVT (defined as arterial puncture to end of procedure time).

Statistical analysis

Baseline and demographic characteristics were presented with univariate analysis, comparing patients with LAA to patients with CE. Continuous variables were analyzed using Student’s t-test, while ordinal and categorical variables were analyzed using Mann-Whitney’s U test and Pearson’s chi-square test, respectively. For analyzing the dichotomous outcome parameters, multivariable logistic regression models were used with CE patients as reference. Variables included into the regression models were chosen if they showed a potential confounding effect according to direct acyclic graphs [22]. The variables included were: age, sex, hypertension, history of hyperlipidemia, history of diabetes mellitus, history of congestive heart failure, and history of smoking. For the continuous outcome parameters, Poisson regression models were used and adjusted for the same covariates as the logistic regression models. We performed secondary analyses by dichotomizing the study population based on (1) IVT treatment or not; (2) anterior or posterior circulation; and (3) more than two EVT maneuvers or not, where significance of P-values for interaction were tested for the subgroups. Statistical significance was defined as P<0.05. All analyses were performed in R v3.5.0 (https://cran.r-project.org).

Meta-analysis

The systematic review and meta-analysis was registered in the International prospective registry of systematic reviews PROSPERO (ID 167447). The meta-analysis followed Meta-Analysis of Observational Studies in Epidemiology (MOOSE) guidelines. The inclusion criteria for the meta-analysis were as follows: (1) published online or in print before February 3, 2020; (2) written in the English language; (3) published in peer-reviewed journal as an original scientific manuscript and based on a randomized controlled trial or observational data (both prospective and retrospective); (4) at minimum data on stroke etiology and one of the following outcomes: mTICI 2b-3, mRS 0–2 at 3 months, death by 3 months, and SICH by any definition. We allowed for one exception of rule #3, and that was the SITS registry which is presented in this manuscript.

The screening of studies for the meta-analysis was performed by one investigator (M.M.) who searched the titles and abstracts of studies in two databases of medical publications: PubMed, Web of Science and EMBASE. The search terms used were: (((stroke OR ischemic stroke OR ischaemic stroke)) AND thrombectomy) AND (etiology OR etiologies OR subtype OR subtypes). Eligible studies were further screened with full-text reads by two investigators (M.M. and C.C.) for inclusion into the meta-analysis. Additionally, the references of the eligible studies were screened for any additional publications that could be included into the meta-analysis. If two or more studies had data that could overlap the study with the highest number of patients was included. Any inconsistencies between investigators were resolved after discussing the studies in question with a senior investigator (N.A.). Baseline and demographic characteristics in addition to outcome measurements were recorded during screening. Measurements that were presented as median and interquartile range were converted to mean and standard deviation, based on the assumption of a log-normal distribution of the original measures [23]. Publication bias was checked using Egger’s test and Funnel plots. Risk of bias in the included studies was assessed using the Newcastle-Ottawa Scale for observational studies and Cochrane risk of bias tool for randomized trials. In order to maintain power, all eligible studies were included in the meta-analysis.

A random-effects meta-analysis was performed on the data from the included studies. Aggregate data was recorded for all studies. If it was not possible to extract all the necessary data from the manuscripts, the corresponding authors were contacted in order to receive the aggregated data. Tau estimators were calculated using the Sidik-Jonkman method, with the confidence interval (CI) calculated using Knapp-Hartung adustments. Forest plots were presented for the outcome variables. In order to attempt to correct for confounding effects on study-level differences, multiple meta-regression was performed. We used the same covariates as in our regression analyses for the meta-regression. All covariates that had data from at least 10 studies were included into the regression model. Aggregate data included in the meta-analysis is shown in supplementary data (Supplementary data 1).

Ethical committee approval

The current study was approved by the Stockholm Regional Ethical board through the framework of SITS-Monitoring Study II (approval number 2015/767-31). Ethical approval or written patient consent for participation in SITS differed among participating countries. Ethical approval or patient consent were obtained in countries that required this, while other countries approved the register for conduct as an anonymized audit.

Ethical and other approvals for sharing the aggregate data for the meta-analyses was the responsibility of the principle investigators of the studies asked to provide data.

Meta-analysis

We found 25 studies in the literature search that fit the criteria and could provide data for the meta-analysis (Supplementary Figure 1) [14,15,24-46]. The studies varied in study population sizes, from 19 to 1,134 patients with LAA or CE (Table 5). All studies that were included had an observational study design. In the risk of bias assessment, most studies fulfilled all or most of the criteria for selection and outcome while many studies did not fulfill the comparability criteria (Supplementary Table 2). The meta-analyses did not show any significant results for any of the outcomes with the unadjusted aggregate data (Figure 2). Heterogeneity between studies was moderate for the results of mTICI 2b-3 and death (I²=41%, I²=42%, respectively), high for the results of mRS 0–2 (I²=75%), and low for SICH (I²=9%). Nine studies [14,15,24-30] had aims that largely resembled that of our meta-analysis, while the remaining studies had other aims but still had the required data for the meta-analysis. There was no publication bias as per Egger’s test for mTICI 2b-3 (P=0.882), mRS 0–2 at 3 months (P=0.377), death by 3 months (P=0.902), and SICH (P=0.298). The funnel plots did not show major signs of asymmetry, but the results for death at 3 months and SICH showed S-shaped distributions of studies (Supplementary Figure 2).

Table 5.

Summary of studies included in the meta-analysis

Figure 2.

Meta-analysis of included studies for the outcomes, with aggregate, unadjusted data. (A) Modified Treatment in Cerebral Infarction (mTICI) 2b-3, (B) modified Rankin Scale (mRS) 0-2 at 3 months, (C) death by 3 months, and (D) symptomatic intracranial hemorrhage (SICH). LAA, large artery atherosclerosis; CE, cardiac emboli; OR, odds ratio; CI, confidence interval.

The meta-regression included all planned covariates for the analysis of all of the outcomes. For mTICI, meta-regression model was not significant (F_7,4=0.30, P=0.919) with no significant differences on study-level. The analysis of 3-month mRS 0–2 found that history of smoking showed a trend for association with the meta-analysis (P=0.061), but the meta-regression model was not significant (F_7,4=2.58, P=0.189). The meta-regression models for death and SICH did not show significant effects (F_7,3=0.28, P=0.924; and F_7,2=0.57, P=0.759, respectively).

Clinical implications

Despite data suggesting that stroke etiology may affect reperfusion and outcomes after EVT, it is still unclear how this knowledge could be transferred to the acute setting. Thrombi composition and associated underlying vascular changes, such as stenosis, may be reasons for differences in EVT outcomes between LAA and CE patients. Historically, LAA and CE thrombi have been categorized as white and red thrombi, respectively [3,4]. Yet a meta-analysis of thrombi composition only found a trend towards this association [48]. The main issue is that stroke etiology is not known before EVT treatment. However, radiological signs such as hyperdense artery sign in the non-contrast computed tomography could suggest the content of the thrombus [49,50]. The possibility of radiologically differentiating stroke etiology was further suggested by a recent consensus statement [51]. If this were to be confirmed in future studies, the information on etiology could be used for procedural choices given the difference in procedural outcomes based on stroke etiology. Currently, very limited data exists on etiology playing a significant role in procedural choices. One study found a trend for converting stent-retriever approach to aspiration with fibrin-rich thrombi [35], while other studies suggested that LAA thrombi had a lower chance of complete reperfusion on first pass [15], or more pre-interventional clot migration [52]. Additionally, fibrin rich thrombi with low erythrocyte percentage were associated with less successful reperfusion and higher risk of secondary embolism [49], while red blood cell rich thrombi where associated with fewer EVT maneuvers [35]. Combined, these results would suggest procedural differences based on thrombus composition and therefore etiology, with LAA thrombi requiring more demanding interventions and higher risk of secondary embolism.

Our study has some limitations. Firstly, the retrospective observational design may have led to unintended bias and an inability to correct for unknown confounders. Secondly, we could not validate or fill in any missing clinical or imaging data from the study centers due to the international multi-center setting of the SITS registry. This could potentially bias our results due to the missing data. For the vast majority of cases, we relied on ICD-10 diagnoses which are set by clinicians and added to the database. Despite the information and support provided by the SITS registry in guiding physicians to use the most recent guidelines when diagnosing etiology, this may lead to differences between clinicians as well as centers. Thirdly, almost a quarter of our patients had unspecified or unknown etiologies, which may have affected our results. Fourthly, our definition of LAA, similarly to the TOAST definition, did not differentiate in situ-occlusion from artery-to-artery embolism. This differentiation was not available in the SITS registry, which may influence our results. Finally, as the meta-analysis used unadjusted aggregate data, there is a risk that potential confounders may have affected the results.