MethodsStudy population

SELECT2 is a prospective, randomized, open-label, controlled trial with blinded outcome assessment, conducted at 31 centers across the USA, Canada, Europe, Australia, and New Zealand with patient enrollment occurring between October 2019 and September 2022. Detailed inclusion and exclusion criteria have been published elsewhere.5 7 Adults (≥18 years of age) with no baseline disability (modified Rankin Scale (mRS) 0 or 1) and a diagnosis of acute ischemic stroke due to occlusion of the internal carotid artery (ICA) or middle cerebral artery (MCA) M1 segment presenting with National Institutes of Health Stroke Scale (NIHSS) score >6 and presence of a large stroke, defined as (a) non-contrast CT (Alberta Stroke Program Early CT Score (ASPECTS) 3–5), (b) CT perfusion ischemic core (relative cerebral blood flow <30%) volume ≥50 mL) or (c) MR diffusion (apparent diffusion coefficient (ADC) <620×10-6 mm2/s) volume ≥50 mL, were randomized within 24 hours of when they were last known to be well. All participating centers obtained requisite institutional review board (IRB) or equivalent ethics committee approvals before beginning enrollment. Further details regarding IRB or ethics committee approval are provided in the ethics approval statement at the end of the article. Informed consent was obtained from all enrolled patients or legal authorized representatives. Consented and enrolled patients were randomized to receive EVT versus standard medical care using a centralized, web-based system in a 1:1 ratio. Stent retrievers, aspiration catheters, or a combination of devices were used for thrombectomy, at the discretion of the neurointerventionalist. While general anesthesia was discouraged in the study protocol, the anesthesia approach was at the discretion of the local neurointerventional and anesthesia team, without any specific protocol recommendations. Standard medical care, including IV thrombolytics if eligible, based on local institutional protocols and regional guidelines, was also provided. To reduce the risk of hemorrhagic transformation in cases with extracranial ICA occlusion, immediate stenting of the extracranial lesion was strongly discouraged and stenting beyond 72 hours of the procedure was encouraged. Use of dual antiplatelet therapy and heparin was also discouraged.

All patients with available follow-up imaging to determine presence, type, and grade of hemorrhage were included in the analyses.

Imaging evaluation

All patients received standardized neuroimaging evaluation, including non-contrast CT, CT/MR angiography, and CT perfusion (CTP)/MR diffusion-perfusion imaging at the time of presentation, with automated processing of CT/MR perfusion imaging using iSchemaView RAPID perfusion software at local sites in real time to obtain estimates of ischemic core and penumbra volumes. A follow-up MRI (preferred) or non-contrast CT (if MRI not available) at 24 hours to 7 days after randomization was recommended for all patients. Additional follow-up images to evaluate and monitor patients’ clinical status were obtained at the discretion of local investigators. All neurological images acquired during the index hospitalization, including baseline imaging (non-contrast CT, CT/MR angiography, and CT/MR perfusion), procedure angiograms, and follow-up neuroimaging were collected for central core lab review. Ischemic core volume was defined as the larger of the CTP (relative cerebral blood flow <30%)/MRI (ADC <620×10-6 mm/s2) core volume and non-contrast CT hypodensity volume (manually delineated) and was used as a covariate to adjust for the extent of ischemic injury.

Hemorrhage evaluation

Intracranial hemorrhage, defined as hemorrhage in any intracranial compartment (parenchymal, intraventricular, subarachnoid, subdural, epidural), and intracerebral hemorrhage, defined as parenchymal hemorrhage only, were included in the analysis. Follow-up image(s) acquired between 24 hours and 7 days were independently reviewed by core labs in Houston and Melbourne to determine the presence, type, and grade of hemorrhage. Neuroimaging and clinical symptoms of all patients with symptomatic hemorrhages were also reviewed by the medical monitor. Disagreements were resolved by mutual decision between the treatment team and medical monitor. sICH was defined using the SITS MOST definition8 (parenchymal hematoma type 2 (PH2) or remote PH hemorrhage with deterioration of ≥4 points on NIHSS score up to 24 hours (±6 hours) follow-up window).

A secondary post-hoc evaluation was completed using the Heidelberg Bleeding Classification9 (HBC) algorithm to differentiate hemorrhage grades.

Outcomes evaluation

Trained evaluators, blinded to treatment assignment, performed an assessment of the patient’s functional status 90 days (±30 days) after randomization. The primary outcome was the distribution of the mRS scores at 90 days. Scores of 5 (severe disability) and 6 (death) were merged to avoid considering a shift of mRS scores from 6 to 5 as an improvement. Secondary outcomes included (a) functional independence (mRS 0–2), (b) independent ambulation (mRS 0–3), and (c) severe disability or death (mRS 5–6).

Statistical analyses

Hemorrhage types and grades were described using the HBC and stratified by the type of treatment received by the patient (EVT vs medical management (MM)). Furthermore, baseline clinical and imaging characteristics were described and compared between patients with and without any intracranial hemorrhage. Continuous variables were described using median (IQR), whereas categorical variables were described using counts and proportions.

Probabilistic Index Modelling,10 11 adjusted for received treatment (EVT vs MM), age, NIHSS at presentation, time from last known well to randomization, core volume estimates, and ASPECTS on baseline CT, was conducted to evaluate the difference in the distribution of mRS scores between patients who did and did not have intracranial hemorrhage. The effect of any intracranial hemorrhage on the distribution of functional outcomes was reported using adjusted generalized odds ratios (aGenOR) with 95% confidence intervals (95% CI), where ties were split equally between the groups. For binary secondary outcomes, modified Poisson regression models with robust standard errors were used to characterize the effect of the presence of any intracranial hemorrhage. Models were adjusted for received treatment (EVT vs MM), age, NIHSS at presentation, time from last known well to randomization, ischemic core volume, and ASPECTS on baseline CT. Additionally, sensitivity analysis was also performed using models adjusted for history of diabetes, oral anticoagulant use, and IV thrombolytics in addition to the aforementioned covariates.

All analyses were performed using STATA 1712 and R version 4.2.2.13 All hypotheses were evaluated using two-sided tests. Analyses were considered exploratory and no adjustments for multiple comparisons were performed. Missing data were not imputed.

ResultsBaseline characteristics, hemorrhage rates, and distribution

Between September 2019 and September 2022, 352 subjects were enrolled; 178 patients were randomized to receive EVT, and two patients crossed over from MM and received EVT. One patient did not have follow-up neuroimaging completed and was excluded from the analysis, thus 351 patients were included in the final analysis population.

Any intracranial hemorrhage was observed in 194 (55%) patients, while intracerebral hemorrhage was observed in 189 (54%) patients (table 1). Any intracranial hemorrhages were significantly more frequent in EVT treated patients (134 (75%)) compared with those who received MM only (60/172 (35%)) (P<0.001). Similarly, any intracerebral hemorrhage was also more frequently observed after EVT (EVT 130 (73%) vs MM 59 (34%), P<0.001). Hemorrhagic transformation type 1 (HI1, HBC 1a) or type 2 (HI2, HBC 1b) (online supplemental figure S1) accounted for 93% of all intracranial hemorrhages in both groups. In the EVT group, parenchymal hematoma type 1 (PH1 <30% of infarcted tissue, HBC 1c) was observed in four (2.2%) and PH2 (≥30% of infarcted tissue with substantial mass effect, HBC 2) in one (0.6%) of the participants, while the MM group had no patients with PH1 (0%), two (1.2%) with PH2, and one (0.6%) patient with remote PH. The proportion of patients experiencing remote PH (HBC 3a), intraventricular hemorrhage (HBC 3b), subarachnoid hemorrhage (HBC 3c), and subdural hemorrhage (HBC 3d) were 0%, 0%, 2.2%, and 0%, respectively, in the EVT group. The primary safety outcome, sICH, was observed in 0.6% of EVT patients and 1.2% of MM patients using the SITS-MOST criteria. Of note, seven patients with tandem lesions received intraprocedure stenting, with none experiencing sICH, but six had any intracranial hemorrhage. Use of antiplatelet and anticoagulant agents before stroke was documented in 29% (EVT 26%, MM 30%) and 14% (EVT 16%, MM 11%) of individuals, respectively.

Table 1

Hemorrhage rates in the as-treated patients by treatment received using the Heidelberg Bleeding Classification

Baseline characteristics among EVT patients with and without intracerebral hemorrhage are shown in table 2. The proportion of patients with IV thrombolysis (20.0% vs 22.4%), occlusion location (ICA 46% vs 39%, MCA 54% vs 61%), tandem occlusion (30.8% vs 32.7%), core volume estimates (82.5 mL vs 81 mL), history of atrial fibrillation (26.9% vs 22.4%), and general anesthesia (57% vs 61.2%) were similar between those with and without intracerebral hemorrhage, respectively.

Table 2

Baseline characteristics by any intracerebral hemorrhage among patients receiving EVT

Association of baseline images and procedural parameters with hemorrhagic transformation

There was no association between increasing rates of intracerebral hemorrhage with higher core infarct volumes (<50 mL: 53%, 50–100 mL: 74%, 100–150 mL: 71%, 150 mL or larger: 83%, P for trend 0.10) or ASPECTS (ASPECTS 3: 74%, ASPECTS 4: 77%, ASPECTS 5: 71%, P for trend 0.74). The proportion of successful reperfusion (modified Thrombolysis In Cerebral Infarction (mTICI) 2b-3) between those with and without any intracerebral hemorrhage was 80.8% versus 77.6%, respectively.

Hemorrhagic transformation association with clinical outcomes and thrombectomy treatment benefit

Functional outcomes including the distribution of the 90-day mRS score (no intracerebral hemorrhage: 4 (3–6) vs any intracerebral hemorrhage: 4 (3–6); aGenOR 1.00, 95% CI 0.68 to 1.47, P>0.99) (table 3, figure 1), functional independence (mRS 0–2) (no intracerebral hemorrhage: 20.4% vs any intracerebral hemorrhage: 20.2%; aRR 1.38, 95% CI 0.76 to 2.52, P=0.29), and independent ambulation (no intracerebral hemorrhage: 44.9% vs any intracerebral hemorrhage: 35.7%; aRR 1.00, 95% CI 0.70 to 1.44, P=0.99) did not differ between those with and without intracerebral hemorrhage. In patients receiving EVT, the presence of any intracerebral hemorrhage demonstrated no difference in early neurological worsening (no intracerebral hemorrhage: 12.2% vs any intracerebral hemorrhage: 29.2%; aRR 1.92, 95% CI 0.89 to 4.14) or increase in mean (SD) infarct growth (no intracerebral hemorrhage: 69.4 (71.4) mL vs any intracerebral hemorrhage: 100 (87.4) mL; adjusted coefficient 25.04, 95% CI −1.80 to 51.89).

Figure 1

Distribution of modified Rankin Scale (mRS) scores for patients with and without any intracerebral hemorrhage (ICH), stratified based on treatment received. 90-day functional outcomes were missing for one patient receiving endovascular thrombectomy (EVT) and three patients receiving medical management (MM).

Table 3

Outcomes by any intracerebral hemorrhage among patients receiving EVT

Similar results were observed after adjusting for history of diabetes, oral anticoagulant use, and IV thrombolytics in addition to the other pre-specified covariates (online supplemental table 1). No significant interaction between hemorrhagic transformation status and treatment received (EVT vs MM) for 90-day mRS distribution (Pinteraction 0.77), mRS 0–2 (Pinteraction 0.50) or mRS 0–3 (Pinteraction 0.69) was observed.

Discussion

In this secondary analysis of the SELECT2 randomized trial, the proportion of patients experiencing any hemorrhagic event was higher than those reported in prior EVT trials of patients with smaller core infarcts. Any hemorrhagic transformation was more frequent in the EVT group compared with those receiving best medical care. However, hemorrhagic transformation was largely asymptomatic and was not associated with worse clinical outcome after adjusting for other clinical covariates as well as the type of treatment. Most importantly, parenchymal hemorrhage or sICH were infrequent and not higher with EVT.

This 0.6% sICH rate in SELECT2 EVT patients is markedly lower than the 6.9%, 4.4%, and 5.1% rates reported in the RESCUE-LIMIT,3 ANGEL-ASPECT,4 and TENSION6 studies, respectively (table 4). There are several potential explanations for this observed difference. We reported sICH using a central core laboratory blinded to treatment assignment as well as an independent medical monitor to assess hemorrhage grades and clinical deterioration. Other differences that might explain lower sICH rates in SELECT2 include a younger median age, lower proportion of Asian patients, and lower rates of IV thrombolysis usage in the SELECT2 EVT group (20.8%) compared with RESCUE-Japan LIMIT (26.7%), ANGEL-ASPECT (28.7%), and TENSION (39%).

Table 4

Reported incidence of any intracranial hemorrhage and sICH in various trials evaluating thrombectomy in patients with large ischemic stroke

While pragmatic, the SELECT27 protocol put in place certain restrictions such as excluding patients with established hypodensity and signs of hemorrhage on baseline imaging, limiting blood pressure (BP) fluctuations during the EVT procedure by discouraging general anesthesia, and reducing post-procedure risk of bleeding by discouraging use of dual antiplatelet agents until more than 24 hours after the procedure. In patients treated outside the study protocol these restrictions may not be applied stringently. While the overall risk of sICH and parenchymal hemorrhage was low in SELECT2, the hemorrhage risk in excluded patients has not been evaluated systematically and may show an increase in comparison to what was observed in SELECT2.

A low rate of sICH is not surprising in this population. Among patients with large infarcts who already have a severe neurological deficit, a hemorrhage must be very severe or occur in a part of the brain unaffected by infarction to be symptomatic, either by NIHSS criteria or the more sensitive criterion of any neurological worsening. Finally, often neglected in thinking about sICH is the temporality of outcomes; the final outcome at 90 days includes accounts for any early worsening due to hemorrhage. Given that the primary outcome at 90 days is substantially better overall in the EVT group, any early difference in adverse events would not be a major concern.

Intracranial hemorrhage after ischemic stroke can be associated with both treatment or even spontaneous. Notably, petechial hemorrhage is part of the natural history of ischemic stroke, and is common both without treatment and in association with thrombolysis and/or EVT. It is usually asymptomatic, therefore not considered an adverse event. Some types of intracranial hemorrhage, in particular subarachnoid hemorrhage, usually do not occur spontaneously or in association with thrombolysis, but are associated with procedural complications such as wire perforation or tearing of perforators during thrombectomy. However, most of these hemorrhagic events do not influence prognosis significantly. The HBC includes outcome measures that capture events relevant to thrombectomy such as subarachnoid, subdural, intracerebral, and intraventricular hemorrhage.9 We reclassified hemorrhagic occurrence by HBC with low prevalence of these events, especially in patients receiving EVT.

In this study, nearly three quarters of patients with large core infarct developed some intracranial hemorrhage after EVT. Despite EVT patients having twice the rate of any intracerebral hemorrhage compared with MM patients, their overall clinical outcomes were still more favorable. Since intracranial hemorrhage is often assessed as a quality measure among stroke centers, expectations about its incidence should be considered in relation to initial core size. The rate of sICH among EVT patients using the SITS-MOST8 definition was only 0.6%. These findings suggest the following conclusions. First, with the likelihood of poor clinical outcome being high even without hemorrhagic transformation in those receiving best medical care, the risks of hemorrhagic transformation in those receiving EVT appears to be well-balanced and plausibly acceptable. Second, the stroke severity in these patients is expected to be significantly high to begin with, and thus a 4-point increase in NIHSS would require the hemorrhage to be much worse than those presenting with smaller stroke. This is especially true for a punctate hemorrhagic transformation in an already large infarcted area, which is likely to be clinically irrelevant. Third, the disparity between any intracranial hemorrhage (75%) and sICH (0.6%) raises questions as to what type of bleeding classification system may be most appropriate for evaluating safety outcomes in patients with large core, although the proportion of patients experiencing any parenchymal hemorrhage was also low (2%). The overall better clinical outcomes with EVT despite high rates of hemorrhage, and the non-significant effect of hemorrhagic transformation on overall functional outcome after adjusting for treatment, provide reassurances about the relative safety of EVT.

Our findings suggest that while not reaching the statistical significance, early neurological worsening and an increase in infarct growth were numerically more frequent after intracerebral hemorrhage. However, we did not observe an association between any intracerebral hemorrhage and 90-day functional outcomes. It is difficult to ascertain causation between these events considering their temporality is not always clear, as infarct growth may precede hemorrhagic transformation and vice versa. Furthermore, any associations observed between increased hemorrhagic transformation and larger infarct growth may simply be driven by an underlying confounding effect of large core stroke.

Compared with thrombectomy patients with small core, there appears to be an inverse relationship between the size of the pre-existing infarct before thrombectomy and the adverse effects of intracranial hemorrhage. A pooled individual patient-level data analysis of 389 ‘small core’ patients from three prospective thrombectomy trials provides a point of comparison summarizing the adverse effect of intracranial hemorrhage.14 Median ASPECTS in this study was 9 and IV thrombolysis was given in 66% of patients. Any ICH was observed in 21.6% of patients undergoing EVT. They found that patients with any intracranial hemorrhage (P<0.001), hemorrhagic transformation (P<0.001), hemorrhagic infarct (P=0.002), and parenchymal hematoma (P=0.002) were much less likely to have functional independence at 90 days than those without hemorrhage. On the other hand, in the SELECT2 cohort with a median ASPECT of 4, there was no difference in mortality and improved 90-day median mRS, mRS 0–2, and mRS 0–3 despite twice the incidence of any intracranial hemorrhage in the EVT group (74.4%).

The etiology of hemorrhagic transformation, especially parenchymal hemorrhage, is considered multifactorial, with ischemic core size at baseline, administration of IV thrombolysis, procedure parameters (reperfusion success, number of attempted device passes, procedure length) and pre-, peri- and post-procedure BP control potentially contributing to the occurrence. Even after accounting for these factors, the variance in symptomatic hemorrhage occurrence remains largely unexplained; much of symptomatic hemorrhage is simply not predictable.

We could not perform analyses to evaluate procedure parameters such as procedure success or number of attempted device passes on parenchymal or symptomatic hemorrhage because of the limited incidence of these outcomes in our trial, especially within the thrombectomy arm. Furthermore, while our analyses demonstrated limited association of hemorrhagic transformation and worsening functional outcomes, opportunities remain to optimize clinical outcomes in large core patients receiving EVT, by decreasing the risk and grade of hemorrhagic transformation and/or reducing the effect of hemorrhagic transformation. Potential avenues include neuroprotective agents to reduce hemorrhagic transformation such as 3K3A-APC15 (being evaluated in RHAPSODY-II), active intensive BP management, reversal of thrombolytics if administered, and the use of procoagulant agents such as rFVIIa.16 Additionally, identification and management of modifiable risk factors associated with hemorrhagic transformation such as thrombolytic agents, anesthesia type, and brain imaging biomarkers17 should also be pursued actively as a research avenue.

Limitations to this study include the smaller than planned sample size in the SELECT2 trial because of early termination. This may have confounded our safety results due to insufficient power and the overall low occurrence of the primary safety event. However, when considering the sample size differences, the SELECT2 intracranial hemorrhage data are largely supportive of other published large core intracranial hemorrhage results. Details regarding post-procedure antiplatelet/anticoagulation regimen, including agent used, initiation time and duration, were not available and precluded detailed analysis.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

This study involves human participants and was approved by Contractual Agreements which have been executed with the following sites with complete IRB approval. List of the IRB addresses are below. 1. University Hospitals Cleveland Medical Center, Cleveland, OH. 2. University of Texas Health Science Center at Houston, Houston, TX. 3. The University of Kansas Medical Center, Kansas City, KS. 4. Baptist Health Jacksonville, FL. 5. Cleveland Clinic Foundation, Cleveland, OH. 6. Rush University Medical Center, Chicago, IL. 7. The University of Texas at Austin, Austin, TX. 8. Thomas Jefferson University Hospital, Philadelphia, PA. 9. Abington Memorial Hospital, Abington, PA. 10. OhioHealth – Riverside Methodist Hospital, Columbus OH. 11. Valley Baptist Medical Center, Harlingen, TX. 12. Ascension Columbia St. Mary’s Hospital, Milwaukee, WI. 13. The University of Iowa Hospitals and Clinics, Iowa City, IA. 14. Kaiser Permanente Southern California Group, Los Angeles, CA. 15. Ascension St. Vincent Indiana, Indianapolis, IN. 16. University Tennessee Health Science Center, Memphis, TN. 17. Hospital of the University of Pennsylvania, Philadelphia, PA. 18. Spectrum Health Hospital, Grand Rapid, MI. 19. Westchester Medical Center, Valhalla, NY. The study is also approved by equivalent ethical committees in international sites: 1. Royal Melbourne Hospital, Victoria, Australia. 2. Royal Adelaide Hospitals, Adelaide, Australia. 3. Liverpool Hospitals, Sydney, Australia. 4. Christchurch Hospital - Canterbury District Health Board, New Zealand. 5. University Health Network/University of Toronto, Ontario, Canada. 6. University of Alberta, Calgary, Alberta, Canada. 7. Val d’Hebron University Hospital, Barcelona, Spain. 8. Germans Trias Hospital, Barcelona, Spain. IRB numbers where available: University Hospitals Cleveland Medical Center Advarra IRB# Pro00056862 University of Texas Health Science Center at Houston IRB# HSC-MS-18-0997 University of Kansas Medical Center IRB# STUDY00144001 Baptist Health Medical Center IRB# 19-51 Cleveland Clinic Foundation IRB# 19-943 Rush University Medical Center IRB# 18111201-IRB02 University of Texas at Austin IRB# 2019050084 Riverside Methodist Hospital – OhioHealth Research Institute WCG WIRB IRB# 20192455 Kaiser Permanente – Los Angeles IRB# 12305 Abington Jefferson Health IRB# 19-046 Valley Baptist Medical Center IRB# 2019-161 Ascension Wisconsin IRB# Protocol 1330 University of Iowa Hospital and Clinics IRB# 201909801 Semmes Murphey Clinic IRB# 20-20 Thomas Jefferson University Hospitals IRB# 19-046 The Hospital of the University of Pennsylvania IRB# 834809 Corewell Health (previously as: Spectrum Health) IRB# 2019-620 Ascension St Vincent IRB# R20210068. Westchester Medical Center WCG WIRB IRB# 20192455 Royal Melbourne Hospital, Victoria, Australia Research Governance Officer IRB# 2019.337 Royal Adelaide Hospitals, Adelaide, Australia IRB# HREC/58741/MH-2019 Liverpool Hospitals, Sydney, Australia Research Governance Officer IRB# 2019.337 Christchurch Hospital - Canterbury District Health Board, New Zealand IRB# 20/STH/19 University Health Network/University of Toronto, Ontario, Canada IRB# 19-5052 Val d’Hebron University Hospital, Barcelona, Spain IRB# PR(AG)263/2019 Germans Trias Hospital, Barcelona, Spain IRB# PI-20-030 Clinical and Provincial Hospital of Barcelona, Barcelona, Spain IRB# HCB/2019/0732 Hospital Universitari de Bellvitge, Barcelona, Spain IRB# PR170/21 University Clinical Hospital of Valladolid, Valladolid, Spain IRB# CASVE-NM-19-385 Participants gave informed consent to participate in the study before taking part.