Introduction

A potentially severe complication of reperfusion therapies in patients with acute ischemic stroke (AIS) is intracranial hemorrhage (ICH). The concern of extensive hemorrhages was the main factor for strict therapeutic time windows in medical reperfusion therapies, with late administration of IV rtPA being associated with a higher risk of large intracerebral hemorrhages and death [1]. Several preceding studies tried to establish predictive models to estimate the risk of ICH after IV rtPA and could find and validate certain parameters, such as age, systolic blood pressure, stroke onset to treatment time and blood glucose as applicable predictors [2, 3].

By now, the superior efficacy and safety of endovascular treatment via mechanical thrombectomy (MT) was validated by multiple randomized controlled studies without showing an increase incidence of clinically relevant ICH in comparison to exclusive conservative therapy [1, 4]. Still, there is limited understanding about possible risk factors and predictors for clinically relevant ICH after mechanical thrombectomy, which significantly aggravate the patient’s outcome [3,4,5,6]. ICH may be further distinguished to assess clinical relevance, e. g., with the National Institute of Neurological Disorders and Stroke (NINDS [7]) and European Cooperative Acute Stroke Study (ECASS [8]), which were impaired by the lack of a generally accepted definition [9]. The recently published Heidelberg Bleeding Classification (HBC), proposed on a consensus meeting of leading stroke researchers, provides a possible solution for this issue [10] and allows a more differentiated bleeding assessment than previously used classifications [11]. To provide better understanding of the risk factors and clinical relevance of periprodecural hemorrhages after endovascular treatment in patients with acute ischemic stroke of the anterior circulation, we analyzed our single-center database to possibly identify clinically relevant predictors of ICH and specifically symptomatic ICH.

Materials and methods

Study population

All consecutive patients with acute ischemic stroke due to large-vessel occlusion in the anterior circulation and consecutive treatment with mechanical thrombectomy were selected from a database of a tertiary care university hospital in the period from January 2012 to December 2016 (n = 740). Patients without adequate baseline or follow-up imaging or insufficient case history (e.g., missing 90-day clinical follow-up) were excluded (n = 128). A total of 43 clinical and procedural parameters were collected retrospectively by reviewing each patient’s records. Last-well-seen times were used to calculate time-to-groin periods for patients with wake-up strokes.

Imaging and endovascular procedures

All patients were examined with non-contrast-enhanced CT (NCCT) and sometimes MRI of the neurocranium, supplemented by angiography and mostly perfusion of the brain (67.5%). ASPECT scores were evaluated for triage assessment [12]. In case of large artery occlusion and absence of contraindications, patients were immediately transferred to angiography for endovascular treatment. Thrombectomy was performed with modern stent-like retriever devices (Solitaire and Solitaire Platinum, Medtronic, Irvine, CA; Trevo, Stryker, Fremont, CA). If necessary, additional therapeutic approaches such as (stent-) angioplasty were performed. Whenever stent angioplasty was performed, we treated patients with antithrombotic medication with different therapeutic regimens. The majority of patients were medicated with 500 mg Aspirin i. v. during the intervention and loaded with Clopidogrel or Tirofiban afterwards.

Additionally, we treated all eligible patients with IV rtPA according to national and international guidelines. Endovascular treatment was performed either under conscious sedation or, if necessary, general anesthesia (GA).

Image evaluation

Routine follow-up brain imaging with NCCT or in rare cases with cranial MRI was performed within 18 to 36 h or earlier in case of clinical deterioration and independently reviewed by neuroradiology staff (U.N., resident with four years of experience in neuroradiology; J.P., board-certified radiologist and neuroradiologist with eight years of experience in neuroradiology), with a following consensus read on differing results. Final infarct volumes were determined on a visual basis with categories for no infarct volume, infarct volume < 1/3 of the affected territory and infarct volume > 1/3 of the affected territory. Reported ICH were categorized according to the HBC classification [10], see Table 1. Correlation of neurological deterioration with imaging findings was assessed by reviewing clinical data (e.g., medical records) to further classify ICH as symptomatic ICH or asymptomatic ICH, in compliance with operational guidelines of HBC. Asymptomatic ICH was not further considered separately. Clinical outcomes were measured by collecting the National Institutes of Health Stroke Scale (NIHSS) score and modified Rankin Scale (mRS) and discharge and 90 days after onset. A NIHSS score of 42 was assigned to patients who deceased during their stay in hospital.

Table 1 Overview of bleeding events, categorized with HBC and the ECASS III classification, according to anatomical, descriptive, and clinical features. Patients with sICH are a subgroup of all patients with any ICH
Full size table

Ethical considerations

The stroke database was approved by the local ethics committee. As this was a retrospective analysis, additional written informed consent was waived.

Statistics

Due to the retrospective study design, few data were missing. As simple exclusion of cases with missing values can potentially produce seriously misleading results, we used multiple imputations as the method of choice to complement missing clinical data [13, 14]. Data imputation was performed with the “missForest” package of R, which is used particularly to impute missing values of mixed-type (continuous and/or categorical) and non-normally distributed data. It uses a random forest model trained on the observed values of the existing data to predict the missing values [15]. The relative maximum for any missing parameter was 3.9% (24/613 for systolic and diastolic blood pressure) with a mean of 1.3% and a median of 0.4% for all parameters. No patient had more than 4 missing values of all 43 parameters.

Quantitative data was tested for normal distribution with the Shapiro-Wilk test and showed that all quantitative data was distributed non-parametrically. Univariate analyses were performed to compare groups based on their outcome (any ICH compared to no ICH or symptomatic ICH compared to no symptomatic ICH) on baseline characteristics, baseline measurements, procedure related factors and outcome parameters using Wilcoxon rank-sum test and Chi-squared test for nonparametric quantitative variables and categorical variables, respectively. P values of univariate analyses were adjusted for multiple testing with the false discovery rate method. Parameters with p < 0.25 entered multivariate logistic regression models for either the first outcome any ICH compared to no ICH or the second outcome symptomatic ICH compared to no symptomatic ICH to evaluate their capacity as independent predictors for these events. Parameters of clinical outcome were not included in the multivariate regression models. Results are represented as odds ratios (OR) and their respective 95% confidence interval (CI).

Results

Overall, there were 612 patients who met the inclusion criteria. Patient characteristics of all patients, as well as clinical and procedural parameters, are shown in Table 2. Any form of ICH was detected in 195 patients (31.8%), of which 27 were considered symptomatic (4.4%) according to HBC criteria. For six of this symptomatic ICH (1.0%), the cause of hemorrhage could be related to periprocedural complications, e. g., vessel perforation. Bleeding classification specifics are shown in Table 1.

Table 2 Baseline characteristics, clinical and periprocedural parameters, and clinical outcome parameters for all patients
Full size table

Three hundred eighty patients (62.1%) received IV rtPA. In 143 patients (23.4%), extra- or intracranial stenting was performed with concomitant dual antiplatelet therapy.

Risk factors of hemorrhage

For the development of any ICH, multivariate logistic regression showed that infarct size > 1/3 of vascular territory in control imaging (OR 2.18, 95% CI 1.45–3.21), occlusion of the right hemisphere (OR 1.54, 95% CI 1.06–2.24), higher serum glucose levels (OR 1.23 for increments of 30 units change in mg/dL, 95% CI 1.10–1.39) and higher thrombectomy maneuver count (OR 1.21, 95% CI 1.11–1.32) as independent risk factors.

No infarct demarcation (OR 0.10, 95% CI 0.01–0.80), mTICI 3 (OR 0.57, 95% CI 0.37–0.86), and higher age (OR 0.97 for increments of 5 units change in years, 95% CI 0.91–1.03) were associated with a lower risk of any ICH.

For patients with symptomatic ICH, wake-up strokes (OR 3.99, 95% CI 1.38–11.60), transfers from an external clinic (OR 3.04, 95% CI 1.24–7.48), and higher serum glucose levels (OR 1.22 for increments of 30 units change in mg/dL, 95% CI 1.05–1.42) were revealed as statistically significant independent risk factors in multivariate logistic regression.

Results of univariate analysis and multivariate logistic regressions for all parameters are summarized in Table 3 for any ICH versus no hemorrhage and Table 4 for symptomatic ICH versus no symptomatic ICH.

Table 3 Results of univariate analysis (p values) and multivariate logistic regression (odds ratios and 95% confidence intervals) for the outcome of any ICH compared with no ICH after mechanical thrombectomy. Predictors are baseline characteristics, baseline measures, and periprocedural parameters
Full size table
Table 4 Results of univariate analysis (p values) and multivariate logistic regression (odds ratios and 95% confidence intervals) for the outcome of symptomatic ICH compared with no symptomatic ICH after mechanical thrombectomy. Predictors are baseline characteristics, baseline measures, and periprocedural parameters
Full size table

Parameters of clinical outcome of all patient groups are collected in Table 5.

Table 5 Clinical outcome of patients with any ICH versus no ICH and symptomatic ICH versus no symptomatic ICH after mechanical thrombectomy. Statistical differences between groups are given in univariate p values
Full size table

Discussion

Intracerebral hemorrhage following recanalization therapies of acute ischemic stroke represents a serious complication in many cases. Although the rates of symptomatic ICH with endovascular therapy are not significantly higher than with best medical therapy alone, the identification of a subgroup of patients with particularly increased risk of hemorrhage might assist in improving their clinical outcome. This study revealed that patients with AIS who are treated with MT are particularly prone to suffer from symptomatic ICH (a) in the presence of a wake-up stroke and (b) if they are transferred from a peripheral clinic in a drip-and-ship model.

Recently, Bücke et al. published a report on patients with wake-up strokes who were treated with mechanical thrombectomy. In their analysis, they did not find a higher incidence of symptomatic ICH or a significant difference in functional outcome compared to controls [16]. Opposed to their finding, the most predictive factor for development of symptomatic ICH in the course of treatment was the presence of a wake-up stroke. However, the selection process of patients who were deemed eligible for mechanical thrombectomy is different in every institution, which might explain the deviating findings. In our analysis, patients with wake-up strokes had significantly longer time-to-groin periods than patients with a known onset. This might make them particularly prone to larger and clinically relevant ICH, speculatively on an “all or nothing” basis by damaging the blood-brain barrier.

Although studies with smaller patient cohorts and models did not report a higher rate of hemorrhagic transformation in patients that were treated in a “drip-and-ship” concept [17, 18], transfer from an external clinic was the second most predictive factor for the development of a symptomatic ICH in our study. Even though the “drip-and-ship” concept empirically provides adequate results in our institute, longitudinal trials of larger patient cohorts are needed to assess the possible implications of this patient collective more thoroughly. The RACECAT study in Barcelona, Spain, is a prospective, randomized controlled trial, which aims to address this question by evaluating the effectiveness of direct transfer to an endovascular stroke center based on identification of suspected large-vessel occlusion [19]. The results of the RACECAT trial are still pending, but our findings suggest that we might improve care of stroke patients, if they are transferred directly to an endovascular stroke center. Similarly to patients with wake-up strokes, we noticed significantly longer time-to-groin periods in patients who were transferred from an external clinic in contrast to patients who were directly admitted to our hospital, which might be the underlying causative factor. But longer time-to-groin periods by itself was not revealed as a statistically significant independent risk factor in the multivariate logistic regression.

Antithrombotic medication has to be administered in carotid artery stent placement to prevent acute stent thrombosis. This however entails the potential risk of ICH. It has been demonstrated that combined antiplatelet therapy was associated with a higher risk for in-hospital mortality due to ICH [20]. This finding is reflected in our results as an increased risk for both any ICH and symptomatic ICH was found in patients with PTA and stenting of ICA stenosis or occlusion, which results in concomitant treatment with dual platelet inhibition in our institution. An increased risk for ICH was also described due to posttherapeutic hyperperfusion syndrome after recanalization of ICA occlusions, which might have also contributed to this finding [21].

The importance of blood glucose levels as a prognostic factor has been postulated before. For example, Strbian et al. included this parameter in the SEDAN score for assessing risk of symptomatic ICH in patients with AIS undergoing IV rtPA [2]. Our results confirm the capability to use serum glucose levels as a predictive factor for both any ICH and symptomatic ICH in patients undergoing MT (and additionally IV rtPA if applicable). In this context, higher HbA1c levels and diagnosis of diabetes mellitus were also both identified as independent risk factors for any ICH, which reinforces the validity of this finding. In contrast to the results of Strbian et al., higher age was revealed as a protective factor against hemorrhagic transformation in our study. This was also reported in a recently published study of patients who were treated with MT [22].

Complete reperfusion (TICI 3) in comparison to “only” successful recanalization (TICI 2b) was shown to be associated with substantially better neurologic recovery, shorter stay during the initial hospitalization and lower ICH rate [23, 24]. We can support this observation as in our study, patients with complete reperfusion of mTICI 3 were attributable to a lower risk of developing any ICH in comparison to all patients with successful recanalization (mTICI 2b, mTICI 2c, and mTICI 3 combined). Partially restricted territorial perfusion, as implied by mTICI 2b, will aggravate the infarct severity in these areas with subsequent territorial damaging of blood-brain barrier, thus presumably causing hemorrhagic transformation [25]. In cases with only partially successful or unsuccessful recanalizations, the quantity of thrombectomy maneuvers was also higher than in cases with complete reperfusion (4.3 versus 2.3), which itself was revealed as a risk factor of hemorrhagic transformation in our study. This reinforces the notion to achieve complete recanalization preferably with a less number of thrombectomy maneuvers [26].

Recently, multiple meta-analyses compared the effect of conscious sedation versus GA in patients treated with AIS undergoing MT and found that conscious sedation could be the anesthesia technique of choice with reduced overall mortality and better functional outcome in comparison to GA [27, 28]. Our results underpin their position, as intubation anesthesia was revealed as a predictor for symptomatic ICH in our study. An important advantage of GA is less patient movement during MT, which can be crucial, as patient movement can compromise safe device maneuvering with possible vessel injury and hemorrhage. However, delay of treatment for patients receiving GA is considered to be a main factor influencing outcome, as reported in the GA versus non-GA subgroup analysis of the MR CLEAN trial [29].

Image evidence of extended hypoattenuation of greater than one-third of the vascular territory is a contraindication for eligibility of treatment IV rtPA in the current AHA/ASA guidelines for the early management of patients with AIS because of increased risk of hemorrhage [30]. Our findings support this notion, as larger final infarct size in the postinterventional control imaging with either cranial CT or MRI was also associated with a higher risk for any ICH and symptomatic ICH in our study.

Notably, treatment with IV rtPA in addition to mechanical thrombectomy alone or medication with oral anticoagulants or antiplatelet drugs did not increase the risk for any ICH or symptomatic in our analysis.

We acknowledge some limitations to our study. A significant proportion of patients (128 of 740 cases) could not be considered in the study because of lacking adequate imaging or insufficient case history, which might have caused a bias in our results. As some of the data was not retrievable at the time of the analysis, multiple imputation for a small fraction of missing data was performed, which is not uncommon in medical research but has to be handled with care [13]. Our results regarding GA vs. conscious sedation might be impaired by not considering conversion from conscious sedation to GA in comparison to patients who received GA before or at the beginning of mechanical thrombectomy. Our diverse findings strengthen the assumption that periprocedural ICH and symptomatic ICH are multifactorially influenced events, which cannot be easily predicted by familiar clinical and procedural parameters or might need different, new strategies in acquiring predictive factors, such as the integration of radiomic features [31]. The complex manner of ICH after MT entails the need for further, prospectively conducted studies with larger cohorts.

Conclusion

Wake-up strokes and patients who are treated within a drip-and-ship concept are especially vulnerable for symptomatic ICH, supporting the notion that especially in those vulnerable patients endovascular recanalization performed as fast as possible is a major goal in stroke triage. Our findings also underline that complete recanalization in contrast to subtotal recanalization can be considered as a protective factor against ICH. Further research in larger patient populations is necessary to increase our understanding about the complex nature of these intracranial hemorrhages.