Introduction

Endovascular treatment for cerebral aneurysms has significantly improved over the past few decades [1]. Traditional methods such as unassisted and assisted coil embolization techniques have been thoroughly investigated, with many studies supporting their efficacy and safety [2,3,4,5]. However, the emergence of flow diverters (FDs) has introduced a paradigm shift in treating cerebral aneurysms, offering novel approaches that differ fundamentally from coiling techniques [6].

Despite the approval and rapid adoption of competing FDs from different manufacturers in various regions, their optimal usage and device selection still need to be explored [7]. First generation FDs receiving regulatory approval included: Pipeline Embolization Device (PED) (Medtronic, Irvine, California), [8, 9] Silk (Balt Extrusion, Montmorency, France), [8, 9] Flow Re-direction Endoluminal Device (FRED) (Microvention, Tustin, California), [10, 11] and Surpass Streamline (Stryker Neurovascular, Fremont, California), [7] among others. While many studies have reported experiences with PED and Silk, there needs to be more published data on other devices, such as FRED and Surpass Streamline [7]. Comparison between all these devices in a single study has yet to be attempted, and never within a matched consortium.

The Pipeline for Uncoilable or Failed Aneurysms (PUFS) trial and subsequent studies have highlighted the potential of PED, and by extension FDs in general, in treating complex aneurysms [12]. These investigations show promising results in terms of aneurysm occlusion rates and low adverse events. Still, they also underscore the risk of thromboembolic complications, which was felt initially to be a frequent cause of peri-procedural morbidity [12].

Therefore, as experience with FDs continues to grow, it is imperative to evaluate their relative and absolute efficacy and safety critically. This study aims to provide a retrospective, multicenter analysis comparing the effectiveness of different flow diverter devices in treating internal carotid artery (ICA) sidewall cerebral aneurysms.

Methods

Study Design

This retrospective review of prospectively maintained databases from eighteen academic institutions across North America and Europe. The study spanned from 2009 to 2016 and included patients treated with one of four flow diverter devices: PED, Silk, FRED, and Surpass Streamline for ICA sidewall cerebral aneurysms. All FDs included were non-surface modified, as the study period spanned only the initial FD epoch for purity of comparison. A global cohort was maintained to achieve generalizability. Institutional Review Board approval was obtained locally at all participating centers, and informed consent from patients was waived due to retrospective study design.

Aneurysm locations were categorized into three primary groups: cavernous, paraclinoid, and supraclinoid. Additionally, aneurysms that did not fit into these categories were grouped under ‘other ICA-related aneurysms’, which mainly include locations such as Siphon, cervical, petrous, Lacerum, and ICA Transitional locations. Patients included in this study were those treated for sidewall ICA aneurysms with one of the four flow diverter devices: PED, Silk, FRED, or Surpass Streamline. The procedures were performed between 2009 and 2016, and only those with available angiographic and clinical follow-up data were included. Patients were excluded if their aneurysms were located outside the ICA sidewall, if they were treated with flow diverter devices not included in this study, or if there was insufficient follow-up data or incomplete medical records. In addition, aneurysm-branches and ruptured aneurysms were excluded from the study.

Data Collection

Data collected included patient demographics, aneurysm characteristics, platelet function test results, antiplatelet regimen, procedural details, complications, angiographic, and functional outcomes. The primary endpoints of this study were the adequate occlusion rate and the retreatment rate. Adequate occlusion was defined as either complete occlusion (Raymond-Roy class 1) or neck remnant (Raymond-Roy class 2) at the last follow-up. The retreatment rate was defined as the necessity for any additional procedures following the initial treatment. The secondary endpoints included functional outcomes and complication rates. Functional outcomes were measured by the modified Rankin Scale (mRS) score, with a favorable outcome defined as mRS 0–1. Complication rates encompassed both thromboembolic and hemorrhagic complications observed during and after the procedure.

Procedure Details

Prior to the procedure, patients were administered aspirin 325 mg daily and clopidogrel 75 mg daily for a duration ranging from 3 to 14 days. Antiplatelet resistance testing was discretional. All procedures were performed under general anesthesia. Dual antiplatelet therapy was continued for at least three months post-procedure, with aspirin maintained indefinitely.

Angiographic and Clinical Outcomes

Angiographic outcomes were evaluated using digital subtraction angiography (DSA), computed tomographic angiography (CTA), or magnetic resonance angiography (MRA). Occlusion was categorized using the three-point Raymond-Roy (RR) occlusion scale: complete occlusion (RR1), neck remnant (RR2), and aneurysm remnant (RR3). Adequate occlusion was defined as either complete occlusion or neck remnant with lack of an aneurysm remnant (RR1 + RR2).

Complications and Outcomes

Both thromboembolic and hemorrhagic complications were accounted for in the study, from the date of the procedure up to the last follow-up. Thromboembolic complications during the procedure were identified on digital subtraction angiography (DSA) and could include thrombus formation, slow filling of a previously normally filling vessel, or vessel dropout. Treatment for intraprocedural thromboembolism was at the discretion of the interventionalist and could include additional antiplatelet medication, anticoagulation, thrombolytics, mechanical thrombectomy, or observation. Postprocedural thromboembolic complications were identified through a combination of clinical and radiographic findings. Imaging obtained to detect an ischemic stroke could include any combination of a non-contrast CT, CTA, or MRI. An ischemic complication was considered symptomatic if the patient reported symptoms or demonstrated signs attributable to thromboembolism. Hemorrhagic complications were identified intraoperatively on DSA, or on post-procedure imaging obtained guided by clinical concerns. Intracranial hemorrhages were considered symptomatic if the patient reported symptoms or demonstrated signs attributable to the hemorrhage.

Supplementary Figures were added, showcasing one of the sidewall aneurysms treated by one of the centers in this study (Supplementary Fig. 1).

Statistical Analysis

Statistical analyses were performed using R Studio Version 4.2.2. The estimation of PS weights was conducted using generalized boosted modeling (GBM) methodology. In this study, the mean Kolmogorov-Smirnov (KS.mean) statistic was employed as a stopping criterion for assessing and summarizing balance across pretreatment variables. A 10,000-tree GBM model was utilized, with an interaction depth of 3, a shrinkage value of 0.01, and a bag fraction of 1. The end propensity scores for numerous treatments using observable covariates was estimated using the “The Toolkit for Weighting and Analysis of Nonequivalent Groups (twang)” package [13].

The analyses used balance diagnostics for propensity score weights, as per known best practices [14]. The diagnostic analysis primarily examined the absolute standardized mean difference (ASMD) between weighted and unweighted variables. In our analysis, we deemed any ASMD greater than 0.10 to indicate covariate imbalance. The literature has proposed that any value equal to or below 0.10 can be considered as a trivial difference in terms of relative balance [15, 16]. A variant of doubly robust estimation is employed, wherein additional adjustments are made for any covariates that exhibit imbalances. This approach aims to enhance the accuracy of the estimation and minimize the mean square error [17, 18].

The missing data were handled according to twang recommendations by assigning NA values to the missing entries and including them in the propensity score estimation. The weighting algorithm treats these NAs as a separate group, ensuring the missing data is accounted for during the balancing process [13]. Missing data is documented in Supplementary Fig. 2.

Lastly, descriptive statistics were used to provide a concise summary of patient demographics, aneurysm characteristics, and treatment outcomes within the four device groups, namely Silk, PED, FRED, and Surpass. The median and Interquartile range (IQR) were used to represent continuous variables, whilst frequencies and percentages were employed to show categorical information. Moreover, we conducted Chi-square tests for categorical data, while Kruskal-Wallis rank sum test for continuous variables. These tests were utilized to compare the baseline characteristics and treatment outcomes among the four device groups. A p-value below the threshold of 0.05 was deemed to be statistically significant.

The utilization of Kaplan-Meier curves was employed to examine the likelihood of achieving sufficient occlusion with various devices and sites of aneurysms. The log-rank tests were employed to assess and compare the survival curves across the various groups. The study utilized a Cox proportional hazards model to ascertain the characteristics that are correlated with a heightened chance of inadequate occlusion. A multivariable logistic regression was done to investigate which devices affect the functional outcomes, retreatment rates, and thromboembolic complications. Variables adjusted in both Cox and logistic regression models were chosen from the literature and based on senior authors experience.

Results

Baseline Characteristics

The study encompassed a total of 444 patients undergoing aneurysm treatment with four flow-diverting devices: Silk (n = 60), PED (n = 117), FRED (n = 177), and Surpass (n = 90). The cohort predominantly comprised females (n = 378, 85.1%). The mean age across all patients was 55.4 ± 11.9 years, displaying no statistically significant variation across devices (p = 0.397). Aneurysm location was significantly different in one instance (p < 0.001), being that Other ICA-related aneurysms were most prevalent in the FRED group (53.7%). This was addressed with PS for comparison (Table 1).

Table 1 Baseline Characteristics of Patients
Full size table

Treatment Outcomes

Most interventions employed a single flow-diverting device (n = 358, 80.6%), while only a few (n = 67, 15.1%) used coiling as an adjunctive to flow-diverting device with no significant difference across the groups (p = 0.444). Single-device deployment was most prevalent in the FRED group (98.9%, p < 0.001). Complete aneurysm occlusion at last follow-up was achieved in 79.3% of cases, with the PED group presenting the highest rate (83.8%, p = 0.036). The necessity for retreatment was infrequent across the cohort (n = 18, 4.3%) and did not vary significantly among the device groups (p = 0.692). Favorable functional outcomes, denoted by modified Rankin Scale (mRS) score of 0–1, were observed in 92.7% of patients. The PED group exhibited the highest proportion of favorable functional outcomes (98.1%, p = 0.066), albeit not a statistically significant difference. Hemorrhagic complications were exceedingly rare, occurring in only 1.0% of cases, but were exclusively observed in the Surpass group (p = 0.001). (Table 2), which was primarily involving two instances of intraparenchymal hemorrhage and two instances of subarachnoid hemorrhage.

Table 2 Treatment Outcomes of Patients Included in the Study
Full size table

In our Kaplan-Meier analysis, evaluating the adequate occlusion probability of different flow diverters over an 8‑year follow-up period, distinct patterns were observed, as shown in Fig. 1 (p = 0.0047). Regarding aneurysm location, other locations showed certain patterns of occlusion. However, no significant differences were noticed (p = 0.1) (Fig. 1).

Fig. 1
figure 1

Kaplan-Meier Curve for FDs and different carotid sidewall aneurysm locations

Full size image

Balance Diagnostics for the Propensity Score (PS) Estimation

Initially, the covariates in the study showed significant imbalances. For example, the maximum standardized effect size for “Other ICA-related location of aneurysms” was 0.82, and for “neck,” it was 0.807. However, after applying the Kolmogorov-Smirnov mean stopping method, these imbalances were substantially reduced. The maximum standardized effect size for “Other ICA-related location of aneurysms” improved to 0.468, and for “Pretreatment mRS: 0–1” it decreased to 0.253, indicating an effective partial adjustment for most covariates (Fig. 2) (Supplementary Table 1). All imbalance variables (ASMD > 0.1) were included in a doubly robust estimation, wherein additional adjustments to these variables were made by including them in the Cox and logistic regression models.

Fig. 2
figure 2

Comparison of the absolute standardized mean difference (ASMD) between treatment groups on pretreatment covariates before and after weighting

Full size image

PS estimation requires sufficient “common support” between treatment groups to ensure adequate covariate balance. In our analysis, there was some visual evidence of a lack of common support across treatment methods, but the lack of overlap was not extreme. In addition, the absence of weights with a value of zero and weights with extreme values suggests that common support was not an issue in achieving covariate balance (Fig. 3).

Fig. 3
figure 3

Overlap of the empirical propensity score distributions—showing the spread of the estimated propensity scores in the treatment groups

Full size image

Multivariable Logistic Regression and Cox Proportional-Hazards Regression

After adjusting the model to sex, age, location of aneurysms, the maximum diameter of aneurysms, pretreatment mRS, the presence of daughter sac, morphology of the aneurysm (saccular or not), the presence of multiple aneurysms, aneurysm dimensions (depth, height, and neck) and previous treatment, none of the devices included in this study were found to significantly affect the functional outcomes (mRS) (PED (OR:0.15; CI: 0.02–1.37, p = 0.093), FRED (OR:0.1; CI: 0.01–1.06, p = 0.055), Surpass (OR:0.35; CI: 0.02–5.24, p = 0.4)), retreatment rates (PED (OR: 0.76; CI: 0.07–8.07, p = 0.8), FRED (OR:3.15; CI: 0.36–28, p = 0.3), Surpass (OR:2.2; CI: 0.31–15.7, p = 0.4)), and thromboembolic complications (PED (OR:0.24; CI: 0.05–1.1, p = 0.066), FRED (OR:0.37; CI: 0.09–1.56, p = 0.2), Surpass (OR:0.61; CI: 0.12–3.07, p = 0.5)) (Table 3).

Table 3 Cox Proportional-Hazards Regression and Multivariable Logistic Regression Models
Full size table

However, when Cox proportional-hazards regression was adjusted for the above variables, Surpass showed the highest probability of adequate occlusion (HR: 4.59; CI:2.75–7.66, p < 0.001), followed by FRED (HR: 2.23; CI: 1.44–3.46, p < 0.001) and PED (HR: 1.72; CI: 1.10–2.70, p = 0.018). Similarly, Surpass also showed the highest probability of complete occlusion (HR: 5.21; CI: 3.09–8.80, p < 0.001), followed by FRED (HR: 2.28; CI: 1.43–3.63, p < 0.001), and PED (HR: 1.79; CI: 1.09–2.93, p = 0.021) (Table 3).

Discussion

This retrospective study aimed to evaluate the comparative efficacy of four first-generation flow diverters, namely PED, Silk, FRED, and Surpass Streamline, in managing sidewall cerebral aneurysms. While the PED exhibited the highest complete occlusion rate at the last follow-up, the crucial insights from our Cox proportional-hazards analysis indicate that Surpass has the highest probability of achieving adequate occlusion, followed by FRED, PED and Silk.

The PUFS trial was the first prospective, international multicenter series focused on treatment of complex aneurysms using PED [12]. The findings from the trial concluded that PED is a safe and effective treatment of large and giant intracranial internal carotid artery aneurysms of the segments aforementioned, as evidenced by the elevated rates of complete aneurysm occlusion coupled with low rates of adverse neurologic events. Subsequent research has reaffirmed the safety and efficacy of PED in managing aneurysms of diverse morphologies and located in various on- and off-label anatomical sites, [1, 12, 19,20,21,22,23] including those in the posterior circulation [24]. A study by Adeeb et al., [25] encompassing 465 aneurysms treated with PED, reported that complete occlusion (100%) was achieved in 78.2% of aneurysms, while near complete occlusion (90–99%) was observed in 7.6%, and partial occlusion (< 90%) in 14.2% of cases. The rate of aneurysm occlusion continued to increase with time following flow diversion, reaching 83.9% in aneurysms that were followed for > 12 months. The rate of retreatment was 6.3%, with no incidence of recanalization following complete occlusion. Our study aligns with these findings where PED was observed to achieve a complete occlusion in 83.8% aneurysms at last follow-up. In addition, the overall retreatment rate of FDs was 4.3% in our study.

In five studies that compared PED and FRED, [26,27,28] and PED and SURPASS, [29] PED, FRED and Surpass, [30] there was no significant difference between any in terms of aneurysmal occlusion at last follow-up. In contrast, our study showed that Surpass has the highest probability of achieving adequate occlusion at the last follow-up, followed by FRED, PED, and Silk. It is understood that Surpass is occasionally challenging to position compared to its contemporaries, and while we enrolled only consecutive cases, there could still be an element of selection bias for those which were successfully deployed. It may also explain hemorrhage exclusive to the device, albeit rare.

The results of our study confirm that FDs are viable options, depending on the aneurysm location, size, morphology, previous treatments, and type of device used. Our Kaplan-Meier analysis showed that ICA aneurysms sustained efficacy over the observed 8 years with a probability of adequate occlusion more than 80%. This is consistent with published studies. For instance, two meta-analyses by Brinjikji et al. [9] and Briganti et al. [7] reported a complete occlusion rate of 76 and 81.5%, respectively. ICA aneurysms were prevalent in up to 90% of cases in both studies [7, 9].

Despite this, the study encompasses several limitations warranting consideration. Firstly, its retrospective design inherently possesses the risk of selection bias and data incompleteness, which may potentially influence the outcomes. Secondly, the multicenter design, spanning eighteen academic institutions, introduces the possibility of variations and biases in procedural techniques, device selection, and clinical decision-making, which may affect the results; the heterogeneity across centers could potentially confound interpretations. We ensured that only the original epoch of flow diverters was analyzed, with the understanding however that Pipeline Flex so quickly supplanted the original as to consider it in the same group. The lack of randomization further adds to the potential for confounding, as the assignment of FD type may correlate with specific patient or aneurysm characteristics. Statistical limitations are also present; the study acknowledges a slight imbalance in the propensity score estimation which may affect the robustness of the statistical comparisons.

Conclusion

This retrospective, multicenter study sheds light on the relative effectiveness of different FDs in the treatment of ICA sidewall cerebral aneurysms. While no significant differences between these devices were observed in terms of retreatment rates, thromboembolic complications, and functional outcomes, Surpass showed higher probability rates of achieving adequate occlusion at last follow-up, followed by FRED, PED, and Silk. Future prospective studies are needed to validate these results.