Endovascular Intervention Versus Surgery in Ruptured Intracranial Aneurysms in Equipoise: A Systematic Review

All data are available within the article and Data Supplement. A protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO; CRD42018090396) on March 28, 2018 to guide the process of the study.¹³ A logic framework¹² was established to delineate the chain of logic clarifying the mechanism of interventions that relate to outcomes (Figure I in the Data Supplement).

Based on the framework, the population, intervention, comparator and outcome of interest was determined. Full details in terms of definition, time of assessment, and outcome measurement were published.¹³ The population was ruptured aneurysms with defined characteristics that were equally treatable by both methods (objective equipoise). To achieve objective equipoise, we included aneurysms of the anterior circulation only. Aneurysms of posterior circulation and those with nonsaccular morphologies were excluded, that is, dissecting, blood-blister, fusiform, or aneurysms that were giant (>25 mm) or tiny (<3 mm). For intervention and comparator, we defined endovascular intervention as any method performed with a catheter to prevent rebleeding and surgery as any method performed by craniotomy. Finally, for outcomes, we determined both rebleeding and adverse events as primary outcomes. Rebleeding was recorded as defined in the particular study and included rebleedings occurring during the hospitalization period for subarachnoid hemorrhage (SAH) and not longer than 3 months after treatment. Adverse events were defined as procedural complications with evidence of neurological deterioration during or within 7 days after the procedure.² Functional outcomes, delayed cerebral ischemia and seizure were the secondary outcomes. The assignment of dependency as the secondary outcome was based on the indirect relationship between the outcome and the interventions (Figure I in the Data Supplement). In brief, the neurological grade on admission, reflecting early brain injury, is the most important prognostic factor for outcome after SAH and therefore contributes in large part to the risk of subsequent dependency.¹⁴ Aneurysm repair procedures prevent rebleeding (repeated injury to the brain) and do not directly modify the initial early brain injury. We defined dependency as a modified Rankin Scale (mRS) score of >2 or a Glasgow Outcome Scale score of severe disability, persistent vegetative state, or death. The timing was at 3 to 6 and at 12 months after treatment. For delayed cerebral ischemia, we used the definition of Vergouwen et al.^15,16 Good neurological grade SAH was defined as World Federation of Neurosurgical Societies or Hunt and Hess grades of I to III. We included randomized and NRS and excluded single-arm studies (see protocol for detail).

To explore sources of heterogeneity, study level and patient characteristics that were potential effect modifiers were defined a priori in the protocol¹³ and were to be investigated in subgroup analyses. These were studies that initiated treatment in <48 hours, mentioned a preference of craniotomy for intraparenchymal hemorrhage (mostly for middle cerebral artery aneurysms), included a small portion of posterior circulation (not greater than 15%), and were from high-volume centers (>35 SAH cases per year).

Subsequently, the components of population, intervention, comparator, and outcome were used to formulate the search strategy. Figure II in the Data Supplement indicates the relationship between 3 concepts: SAH, endovascular intervention, and surgery. The search strategy was designed to include all 3 concepts while limiting the target to studies that involved both the treatments. We searched Medline, Embase, the Cochrane Central Register of Controlled Trials (CENTRAL), the Institute of Medical Information & Library (IMI&L, China), and other gray publications illustrated in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines flowchart (Figure 1). The search strategy for each aforementioned database and the articles yielded are available (Methods I in the Data Supplement).

The initial search was performed in April 2018, followed by an updated search in June 2019 (Methods I in the Data Supplement). Studies published in English or Chinese were included. The latter was included because the first and third authors were proficient at the language. The date of intervention was limited to after January 1, 2005 to understand the outcome in the era after ISAT.^1,9 Study titles, abstracts, and full text were independently screened for inclusion by 2 authors and discrepancies were decided by consensus discussion with a third author. Prepiloted forms (Methods II and III in the Data Supplement) were used in the process of study inclusion and data extraction to enhance transparency and reproducibility.

The effect of bias in individual studies was assessed with Risk of Bias in Nonrandomized Studies of Interventions (ROBINS-I).¹⁷ The overall judgment for a specific outcome was derived according to the table provided by ROBINS-I (Data Supplement).¹⁷ The bias of the body of evidence was assessed with Grading of Recommendations, Assessment, Development and Evaluation (GRADE).¹⁸ The bias judgment of ROBINS-I was integrated into the risk of bias domain of GRADE according to an established guideline.¹⁹ Details of GRADE and the integration are described in Methods IV in the Data Supplement.

Meta-analysis was planned if the data were appropriate for pooling.¹³ Summary of estimates was presented as Peto odds ratio for rare events (rebleeding) and risk ratio for other outcomes.¹³ Consequently, both fixed-effects and random-effects meta-analyses were analyzed, although the latter was prespecified in the protocol.¹³ If pooling was inappropriate, a narrative synthesis was planned.¹³ The definitions of CI, heterogeneity, and P value are conventional (CI 95%; I² <40%: unimportant; 30%–60%: moderate; 50%–90%: substantial, 75%–100%: considerable; P value <0.05: significant; P value for interaction <0.1: significant).²⁰ Review Manager 5.3.5 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014)²¹ was used for meta-analysis and web-based GRADEpro GDT for GRADE.²²

Meta-analysis of 15 studies (Table 2; Figure 2A) showed a higher incidence of rebleeding in the endovascular intervention than the surgery group (Peto OR, 2.18 [95% CI, 1.29–3.70]; 3104 participants; 15 studies; I²=0%). The GRADE certainty of evidence was very low (see Methods IV in the Data Supplement for definition, downgrade criteria, and the implication of GRADE). As the assessment of ROBINS-I for rebleeding was moderate (Table 1), the grading process started as moderate. One level was downgraded for the domain of imprecision because the rebleeding event rates (2.6% for endovascular; 1.3% for surgery) did not meet the optimal information size (event rates <100).³⁴ Another one level was downgraded in indirectness³⁵ because most studies did not indicate the time of rebleeding after the treatment of the aneurysm. The funnel plot of rebleeding and all other outcomes did not suggest publication bias or systemic heterogeneity (Figure III in the Data Supplement).

Meta-analysis of 11 studies (Figure 2B) showed fewer patients with adverse events in the endovascular than in the surgery arm (RR, 0.71 [95% CI, 0.53–0.95]; 1661 participants; 11 studies; I²=14%). The GRADE certainty of evidence was low. As the assessment of ROBINS-I was moderate (Table 1), the grading process started as moderate. One level was downgraded for indirectness³⁵ because the causal relationship between neurological deterioration and the aneurysm repair procedure was not robustly described in all included studies.

Meta-analysis of 18 studies (Figure 3A) showed fewer patients with endovascular intervention that were dependent compared with those undergoing surgery (RR, 0.82 [95% CI, 0.73–0.93]; 4081 participants; 18 studies; I²=15%). For dependency at 12 months, the meta-analysis of 10 studies (Figure 3B) also found less dependency after endovascular compared with surgical repair (RR, 0.76 [95% CI, 0.66–0.86]; 1981 participants; 10 studies; I²=0%). As the assessment of ROBINS-I was serious (Table 1), the grading process started as low. In the absence of downgrading, the GRADE certainty of evidence remained low for dependency at both times.

Our analysis (Figure 3C) showed no difference in delayed cerebral ischemia between endovascular intervention and surgery (RR, 0.89 [95% CI, 0.71–1.12]; 2608 participants; 12 studies; I²=25%). The meta-analysis also showed no difference in seizure frequency between groups (Figure 3D; RR, 0.59 [95% CI, 0.30–1.17]; 1683 participants; 4 studies; I²=0%).

In the stratified subgroup analysis (Figures IV through VIII in the Data Supplement), no significant subgroup effect was found when the outcomes were investigated against characteristics determined a priori. During study selection, we discovered some institutes had a policy to treat poor-grade SAH with endovascular intervention. Therefore, the characteristic was investigated post hoc. The P value for interaction ranged within 0.19 to 0.79 for all analyses that were of even covariate distribution (Figure V through VII in the Data Supplement), including the characteristics added post hoc.

After the publication of ISAT¹ in 2002, numerous studies compared the 2 treatments and concluded that there was no statistically difference between functional outcome after endovascular intervention and surgery. This is evident in the forest plot of this study (Figure 3A and 3B). The 95% CI of 18 studies in the forest plots in Figure 3A crossed 1, whereas only 3 did not. Overall, the individual studies suggest that there was no statistical difference in terms of functional outcome. Conversely, the 95% CI for many of the studies include the possibility of favoring endovascular or surgical management. By pooling the effect estimates, this study demonstrated better functional outcome after endovascular intervention. Therefore, this meta-analysis of 29 studies (27 NRS) is consistent with those of ISAT¹ and of other systematic reviews.^2,36 This study differs from prior meta-analyses in that we only analyzed studies performed after 2005. We found low certainty of evidence in the estimate of the functional outcome and only randomized trials can more accurately answer the question whether current endovascular aneurysm repair has better outcomes than surgical repair.⁹

The ISAT included a low proportion of poor-grade patients. The real-world population has a higher proportion of poor-grade patients. This was emphasized by 2 Cochrane reviews that stated the evidence supporting endovascular intervention was more applicable to good-grade patients.^2,36 In the current study, we found a total of 29% (n=1640 out of 5759) of patients were poor grade. In contrast, the proportion in ISAT was 6% (Figure 4). Within the limits and biases of the included studies, even with a several times higher proportion of poor-grade patients, endovascular intervention still was associated with better functional outcome.

This study gathered one of the largest datasets on rebleeding after endovascular treatment and surgery outside ISAT¹ (Table 2). Interestingly, the findings are similar to those of a meta-analysis published in 2013 in which the majority of patients were treated before 2005 (OR, 2.33 [95% CI, 1.52–3.57], data inversed for comparison; 8 studies; 5282 participants; I²=0%).³⁷ This implies that ISAT rebleeding rates are relatively unchanged even with newer endovascular technologies.

This study highlights the importance of the balance of rebleeding rate and procedural complications in contributing to functional outcome (see Logic framework, Figure I in the Data Supplement). Although the precise contributions of rebleeding and repair complications to poor outcome cannot be determined from the published data, an explanation in absolute numbers can facilitate understanding. The pooled risk difference of rebleeding between endovascular intervention and surgery is 0.02 ([95% CI, 0.00–0.03], Mantel-Haenszel). The inverse of that risk difference indicates that we need to treat 50 patients in the surgical arm (instead of endovascular intervention) to prevent a single rebleeding event (associated with endovascular intervention). On the contrary, with a similar calculation, we need to treat 17 patients with endovascular intervention to reduce the risk of a poor functional outcome (associated with surgery). This demonstrates that the low incidence of rebleeding has a minimal influence on the statistics of functional outcome and that despite this, endovascular intervention remained the better option although the predicted occurrence of rebleeding was relatively higher.

This analysis had narrow inclusion criteria such as the objective equipoise and the time limit of 2005. The results also do not apply to nonsaccular or posterior circulation aneurysms. Most importantly, confounding factors and biases are inherent in NRS. These were objectively assessed using ROBINS-I and systematically reflected in GRADE. It needs to be remembered that the findings are classified as low certainty of evidence.