Patient selection

All patients who underwent EVT with the Solitaire device in the STAR (n=202), SWIFT (n=89), and SWIFT PRIME (n=98) trials were included in this pooled analysis. All trials were large prospective multicenter studies involving patients with AIS with anterior circulation emergent large vessel occlusion (ELVO). The detailed designs, methods, and results of each individual trial have been previously published.7–10 All three study protocols were approved by the local ethics committee at each of the participating sites. All patients were enrolled after an informed consent form had been signed or country-specific requirements had been met for enrollment without explicit informed consent in emergency circumstances.

Data collection

Outcome data were adjudicated by an independent imaging/angiography core laboratory and a Clinical Events Committee (CEC). The core laboratory assessors were blinded to clinical outcome. Variables scored by the core laboratory and CEC were: location of occlusion, final revascularization grades, collaterals grading, Alberta Stroke Program Early CT score (ASPECTS), presence of ICH with corresponding subtypes, and other adverse events. In SWIFT and SWIFT PRIME, the assessors were blinded to study group assignments. The data were collected and provided by Medtronic Neurovascular. The company had no role in the design, conduct, analysis, or reporting of the present study.

ICH categorization

Different ICH subtypes were categorized by the core laboratory and CEC according to the European Cooperative Acute Stroke Study (ECASS) III definition as: (a) hemorrhagic transformation within the ischemic territory (HT), a category which included both hemorrhagic infarct (HI) and parenchymal hematoma (PH); and (b) symptomatic intracranial hemorrhage (sICH), defined as any PH, subarachnoid hemorrhage (SAH), or intraventricular hemorrhage (IVH) associated with death, or worsening of National Institutes of Health Stroke Scale score (NIHSS) by ≥4 within 24 hours.11 Other imaging core laboratory-adjudicated ICH categories included: SAH, remote intracranial hemorrhage, and IVH. All the above referenced ICH subtypes were already categorized by the CEC and core laboratory for each individual trial and provided to us for the present analysis. From the pathophysiologic perspective, we analyzed HT (including all HI and PH and excluding SAH) as a separate category, which most likely represents post-reperfusion hemorrhage and less likely involves other potential mechanisms such as vessel dissection and wire perforation. The presence of any ICH was also categorized and analyzed separately. In the present study, all six separate hemorrhage subtypes (any ICH, HI, HT, PH, SAH, and sICH) were correlated with 15 candidate baseline clinical, imaging, and procedural variables (tables 1 and 2). 

Statistical analysis

Standard descriptive statistics were employed, including means, SD and medians for continuous variables and frequency distributions for categorical variables. For unadjusted between-group comparisons, t-tests were used for continuous variables and Fisher’s exact test for categorical variables. Multivariate stepwise logistic regression was used to identify predictors of any ICH and individual hemorrhage subtypes using the list of candidate predictors cited in table 1. Univariate analyses were run for each individual variable. Predictors with univariate p values <0.20 were included in the multivariable analyses. Stepwise selection with entry criteria of p<0.2 and retention criteria of p<0.05 were used to obtain the final regression model. Odds ratios were derived from these models and presented along with their 95% CIs and corresponding p values. Fisher’s exact test was used to compare the frequency of functional independence, defined as modified Rankin Scale (mRS) score of 0–2 at 90 days, among subgroups of patients with and without ICH and subtypes of ICH. Additional logistic regression, adjusted for baseline ASPECTS, was performed to compare ICH subtypes with 90-day functional independence. For all statistical analyses, two-tailed p values are presented, with values <0.05 considered statistically significant. Analyses were conducted in SAS Version 9.4 (SAS Institute, Cary, North Carolina, USA) and R Version 3.2 (R Foundation for Statistical Computing, Vienna, Austria). Table 2 represents the distribution of key baseline patient characteristics of the studied cohort. 

Clinical and procedural predictors of ICH risk

Any ICH was observed in 84/389 (21.6%), HT in 75/389 (19.3%), HI in 54/389 (14%), PH in 21/389 (5.4%), SAH in 9/389 (2.3%), and sICH only in 4/389 (1%) of the analyzed patients. The most significant predictors of separate hemorrhage subtypes based on both univariate and multivariate analyses are shown in table 3. Both ASPECTS and general anesthesia (GA) had a strong inverse association with any ICH (ASPECTS: OR 0.80, 95% CI 0.66 to 0.98, p=0.032; GA: OR 0.36, 95% CI 0.18 to 0.71, p=0.003) and HT (ASPECTS: OR 0.78, 95% CI 0.68 to 0.91, p=0.001; GA: OR 0.54, 95% CI 0.31 to 0.92, p=0.023). Collateral grade was also inversely associated with any ICH, but this association did not reach statistical significance (p=0.057). Time to treatment was a significant hemorrhage predictor, with longer times increasing the risk of HT (OR 1.08, 95% CI 1.03 to 1.12, p=0.001) and PH (OR 1.11, 95% CI 1.02 to 1.20, p=0.015). IV-tPA use prior to thrombectomy was also a significant predictor of PH (OR 7.63, 95% CI 1.52 to 17.35, p=0.013). No significant predictors of SAH and sICH were identified, although a non-significant inverse association between sICH and platelet count was noted (OR 0.98, 95% CI 0.96 to 1.0, p=0.084).

Analysis of clinical outcome based on ICH subtype

Among all patients treated with Solitaire in this dataset, 210/378 (55.6%) were functionally independent at 90 days. The frequency of different hemorrhage subtypes seen within this group is shown in table 4. Additional adjusted analysis was performed to eliminate potential confounding between large ischemic core (low ASPECTS) and the occurrence of various ICH subtypes. Overall, patients with any ICH (p<0.001), HT (p<0.001), HI (p=0.002), and PH (p=0.002) were much less likely to have functional independence at 90 days than those who did not have hemorrhage. No significant difference in terms of functional independence was observed in patients with SAH (p=0.46). None of the patients with sICH were independent at 90 days table 5.

Impact of general anesthesia

Our study is the first to report an inverse association between GA and the occurrence of ICH in the setting of neurothrombectomy. Although GA did not impact the most severe hemorrhage subtypes (sICH and PH), these findings provide signals about its protective effect against hemorrhage that warrant further exploration. While no definitive explanation regarding the relationship between GA and ICH can be drawn from our data, there are several theoretical possibilities: (1) less patient motion in the setting of GA with more precise intracranial navigation and lower chance of microvascular avulsion; (2) protective effect of anesthetic medications against cerebral ischemic reperfusion injury12–15; (3) reduced pain and agitation with lower chance of blood pressure surges and uncontrolled hypertension, which is an independent risk of HT.16 In addition, hypotension that frequently occurs with GA during thrombectomy may also protect the ischemic tissue against HT after reperfusion. However, it is also important to emphasize that hypotension can be deleterious to the cerebral collateral circulation prior to revascularization. Profound hemodynamic and ventilation changes in the setting of GA have been associated with worse outcome after thrombectomy,17 and are likely key factors leading to an overall worse outcome seen in multiple retrospective case series and post hoc analyses correlating GA with worse clinical outcome.18–23 In contrast, three single-center prospective randomized studies exploring the difference between GA and conscious sedation (CS) demonstrated equivalence in their primary outcomes.24–26 Furthermore, two of these three recent trials showed superiority in terms of functional independence of GA over CS as a secondary outcome.24 26 All three trials had specific GA protocols with well-defined ventilation and blood pressure targets designed to minimize the risk of collateral failure prior to endovascular revascularization. Accordingly, two recent meta-analyses involving newer thrombectomy devices demonstrated inconclusive results regarding the clinical benefit of either procedural method of sedation.27 28 Given the prior confounding data, the most recent guidelines for early management of AIS specifically outlined the need for further randomized data to identify the best anesthetic technique.6 Our findings of the protective effect of GA against ICH further contribute to the existing clinical equipoise regarding the optimal method of sedation during EVT.

Impact of baseline ASPECTS

The size of the baseline ischemic core is a well-established predictor of hemorrhagic risk with intravenous and endovascular reperfusion.29–33 The strong association of baseline ASPECTS with any ICH and HT in our study reaffirms these previous data and further establishes the ischemic volume as one of the most important factors influencing hemorrhage after reperfusion. Cerebral collateral circulation, which is inversely proportional to the ischemic core volume, is another important and well-known variable associated with an increased risk of HT after intra-arterial therapies for AIS33–35 and also was linked to ICH in our study. Both variables demonstrate the increased susceptibility of the ischemic core tissue to HT after thrombectomy.

Impact of time to treatment

Prolonged time from onset to treatment is one of the main determinants of clinical outcome in all AIS therapies.36–38 The well-established pathophysiologic paradigm is that progression of the ischemic core lesion in the setting of LVO is highly dependent on the temporal dynamic evolution of collateral circulation, which fails over time.39 As discussed above, ischemic core volume and collaterals are potent predictors of HT. Ischemia-induced blood–brain barrier disruption,40 activation of metalloproteinases,41 and formation of free radicals42 rendering the ischemic brain susceptible to hemorrhage are proportionally related to the degree and size of the ischemic changes and occur at maximum intensity within 6–24 hours after onset.43 44 Thus, the association between time to treatment and hemorrhagic risk in our study is not surprising as it likely represents the increased vulnerability of the ischemic brain tissue to reperfusion hemorrhage in the setting of prolonged ischemia. Similar findings have been reported in previous studies: prolonged time from onset to EVT has been associated with an increased risk of basal ganglionic hemorrhage and sICH.32 33 Experience with faster intravenous thrombolysis further supports this notion: in a study involving over 70 000 patients, reduction of door-to-needle time for IV-tPA administration after implementation of national quality initiatives resulted in a significant decrease in the incidence of ICH.45

Impact of IV-tPA

Similar to GA, the data regarding the benefit of bridging IV-tPA therapy in the setting of EVT for ELVO is a topic of ongoing controversy and debate.46 The main reason supporting its use is improved recanalization and potential avoidance of thrombectomy, particularly in patients with distal and smaller clots.47 However, the small chance of recanalization after early administration of IV-tPA for LVO in comprehensive or thrombectomy-ready centers may be obviated by rapid endovascular revascularization. A small single-center study demonstrated equivalent outcomes with EVT alone versus IV-tPA/EVT bridging therapy.48 Another consideration against bridging therapy in ELVO patients is the increased cost of IV-tPA.49 In addition, the coagulopathy induced by the tPA clearly accentuates the hemorrhagic risk after reperfusion.50 51 These post-thrombolytic hemostasis changes have been correlated with cardioembolic infarcts,52 which comprise most of the LVO etiology and further raise the concern of an augmented risk of hemorrhage after EVT. Unlike the other predictors, IV-tPA was uniquely associated with PH in our study. PH has been considered as a separate entity due to its highest impact on clinical outcome among all hemorrhagic subtypes,53 which further emphasizes the potentially hazardous effect of bridging IV-tPA therapy in conjunction with EVT. A similar finding of a higher PH risk with IV-tPA from ‘real-world’ non-randomized retrospective data involving older generation devices was reported by Nogueira et al.54 As such, our findings contribute to the existing data of the increased hemorrhagic risk of IV-tPA, reaffirm the controversy regarding the utility of IV-tPA in the setting of ELVO, and support the need for a well-powered randomized multicenter trial comparing bridging therapy versus EVT alone.

Impact of ICH subtypes on outcome

We explored the incidence of functional independence (mRS 0–2) with ICH versus the incidence of functional independence without ICH for each hemorrhagic subtype. Although this analysis involved simple univariate comparisons, it provides important information regarding the potential impact of each type of hemorrhage on clinical outcome. All hemorrhage subtypes were associated with a lower likelihood of independent clinical outcome, except for SAH. This finding replicates prior data, emphasizing that the imaging finding of SAH alone without parenchymal involvement and/or clinical deterioration may not have a significant long-term clinical impact.55–57

None of the currently used definitions of sICH optimally predict functional outcome.58 The data regarding the long-term clinical effect of different radiographic subtypes are also unclear. While PH has been cited as the most potentially hazardous to clinical outcome,53 HI has been considered a relatively ‘benign’ entity and even correlated with improved outcome as a marker of early reperfusion after IV-tPA in some studies.59 Conversely, more detailed analyses have identified asymptomatic HI-2 as an independent predictor of poor outcome after IV-tPA and EVT.54 60 Our data support the notion that any ICH, HI, and HT can have a potential impact on functional outcome, even if considered asymptomatic by trial definitions. However, our findings also highlight that the stringent definition of sICH used in this dataset (any PH, SAH, or IVH associated with death or worsening of NIHSS score by ≥4 within 24 hours) and recent thrombectomy trials has the strongest impact on functional outcome as none of the patients with sICH achieved mRS 0–2 at 90 days.

Limitations

Our study has several limitations. The main limitation is the relatively small number of symptomatic hemorrhages in the dataset as the patient population included in the three trials is highly selected and may not represent ‘real-world’ experience. Another important drawback is the difference in the design of the three studies, conducted in three difference continents over a time span of 4 years. Although all patients were treated with the same device, these factors introduce potential heterogeneity in the analyzed data due to differences in patient population and neurointerventional practices. In addition, our clinical outcome data were derived on the basis of analyses adjusted only for baseline ASPECTS, instead of multivariate regression, focusing only on individual hemorrhage effect and excluding other important clinical and procedural variables. However, the goal of the outcome analysis was to analyze the clinical effect of each ICH subtype compared with all other hemorrhages rather than identifying overall predictors of clinical outcome.