Gamma Knife radiosurgery for cerebral arteriovenous malformations: a systematic review and meta-analysis

Cerebral arteriovenous malformations (AVMs) are rare vascular lesions that rupture at an approximate annual rate of 2–4% [1]. Annual haemorrhage rates range from < 1%, for unruptured superficially located AVMs, up to 33% in ruptured AVMs with deep, location and venous drainage [12, 68]. Once haemorrhage occurs, the probability of experiencing new neurological deficit has been reported to be up to 50% and fatality ~ 10% [28, 66]. Gamma knife radiosurgery (GKRS) is an established intervention predominantly favoured for small (< 3.5 cm), surgically high-risk or complex AVMs. Successful AVM GKRS is contingent on abolishing the risk of intracranial haemorrhage (ICH) via complete nidal obliteration, whilst limiting the development of new neurological deficit from radiation-induced changes (RICs) [9, 26, 69]. The complications of AVM GKRS are mainly two-fold. Firstly, patients remain at risk of haemorrhage during the latency period between GKRS and nidus obliteration. The relative risk of AVM haemorrhage during this latency period versus the natural history of an untreated AVM remains a topic of debate [26, 28]. Secondly, RICs can occur following radiosurgery, influenced by AVM and treatment parameters including AVM volume, location and treatment dose [12]. Approximately 34% of patients develop RICs, ~ 8% of patients develop RIC-related neurological symptoms and ~ 3% will experience permanent neurologic deterioration. However, considerable variations in the reported rates, durations and definition of RICs still remain. Further, delayed adverse effects including radiation necrosis and cyst formation following GKRS have been reported, yet there is a paucity of literature conclusively describing their incidence [38]. A previous systematic review and meta-analysis quantified haemorrhage risk and obliteration rate following AVM stereotactic radiosurgery (SRS), but it had limitations [72]. The review included 69 observational studies, of which, however, approximately only a third solely treated AVMs with GKRS; the majority reporting on linear accelerator (LINAC) or other SRSmodalities. Additionally, the median duration of study followup was less than three years; an arguably limited time-frame to comprehensively analyse clinical outcomes of AVM GKRS, which evolve in a time-dependent manner over a latency period of 2–3 years following radiosurgical intervention [1, 12, 68]. Alongside imaging advancements and increasingly sophisticated radiation delivery systems since the inception of GKRS, there has also been an increasing knowledge base on AVM obliteration, RICs and post-GKRS ICH [12]. Whilst there are several studies that review or report on these outcomes, none are both systematic or comprehensive. We sought to systematically review the current literature and comprehensively quantify the efficacy: complication profile of GKRS for cerebral AVMs. Musa China and Amisha Vastani are joint first authors.


Introduction
Cerebral arteriovenous malformations (AVMs) are rare vascular lesions that rupture at an approximate annual rate of 2-4% [1]. Annual haemorrhage rates range from < 1%, for unruptured superficially located AVMs, up to 33% in ruptured AVMs with deep, location and venous drainage [12,68]. Once haemorrhage occurs, the probability of experiencing new neurological deficit has been reported to be up to 50% and fatality ~ 10% [28,66]. Gamma knife radiosurgery (GKRS) is an established intervention predominantly favoured for small (< 3.5 cm), surgically high-risk or complex AVMs. Successful AVM GKRS is contingent on abolishing the risk of intracranial haemorrhage (ICH) via complete nidal obliteration, whilst limiting the development of new neurological deficit from radiation-induced changes (RICs) [9,26,69].
The complications of AVM GKRS are mainly two-fold. Firstly, patients remain at risk of haemorrhage during the latency period between GKRS and nidus obliteration. The relative risk of AVM haemorrhage during this latency period versus the natural history of an untreated AVM remains a topic of debate [26,28]. Secondly, RICs can occur following radiosurgery, influenced by AVM and treatment parameters including AVM volume, location and treatment dose [12]. Approximately 34% of patients develop RICs, ~ 8% of patients develop RIC-related neurological symptoms and ~ 3% will experience permanent neurologic deterioration. However, considerable variations in the reported rates, durations and definition of RICs still remain. Further, delayed adverse effects including radiation necrosis and cyst formation following GKRS have been reported, yet there is a paucity of literature conclusively describing their incidence [38].
A previous systematic review and meta-analysis quantified haemorrhage risk and obliteration rate following AVM stereotactic radiosurgery (SRS), but it had limitations [72]. The review included 69 observational studies, of which, however, approximately only a third solely treated AVMs with GKRS; the majority reporting on linear accelerator (LINAC) or other SRSmodalities. Additionally, the median duration of study followup was less than three years; an arguably limited time-frame to comprehensively analyse clinical outcomes of AVM GKRS, which evolve in a time-dependent manner over a latency period of 2-3 years following radiosurgical intervention [1,12,68].
Alongside imaging advancements and increasingly sophisticated radiation delivery systems since the inception of GKRS, there has also been an increasing knowledge base on AVM obliteration, RICs and post-GKRS ICH [12]. Whilst there are several studies that review or report on these outcomes, none are both systematic or comprehensive. We sought to systematically review the current literature and comprehensively quantify the efficacy: complication profile of GKRS for cerebral AVMs.
Musa China and Amisha Vastani are joint first authors.
Musa China and Amisha Vastani contributed equally to this work.

Study selection
Two investigators (MC and AV) independently screened all titles and abstracts for eligibility. The full text of eligible studies was reviewed for inclusion. CSH, CT and PG acted as mediators in cases of disagreement. If multiple studies with overlapping data from a single institution were published, the study reporting the largest sample size and/or the most recent study were prioritised (Appendix 3 and 4).

Data extraction
Data extraction was performed independently by two authors (MC and AV) from each unique study cohort undergoing GKRS to ensure consistent extraction of patient, AVM and treatment characteristics. We extracted the following outcomes: complete nidus obliteration rate, ICH, radiologic evidence of RIC, symptomatic RIC events (transient or permanent), radionecrosis, cyst formation, radiation-induced neoplasm, seizures (new or worsening), deaths attributed to AVM/GKRS (case-related) and all-cause death rate. We used outcomes as described per patient. The proportion successfully obliterated was calculated as a proportion of all patients with available radiological follow-up and was defined as the presence of angiographically demonstrated complete obliteration and/or angiography-or-MRI confirmed obliteration. Post-GKRS ICH was defined as any AVMrelated haemorrhage detected through appropriate followup radiological imaging. RICs were classified as follows: (1) total radiologic RIC: any MRI evidence of peri-nidal T2-weighted hyperintensities after GKRS; (2) transient symptomatic RIC: radiologic RIC which correlated with new or worsening neurological symptoms, typically headache, seizure or focal neurological deficit, which resolved by the end of study follow-up; and (3) permanent RIC: symptomatic RIC without full recovery to pre-GKRS neurological baseline at the end of study follow-up.

Risk of bias
Two authors (MC and AV) assessed the risk of bias of each cohort as serious, moderate, low, or unclear risk according to the following domains of the Cochrane ROBINS-I tool [70]: (1) confounding bias, (2) selection bias, (3) bias in classification of interventions, (4) bias due to deviations from intended interventions, (5) attrition bias, (6) detection bias and (7) selective outcome reporting (Appendix 5).

Statistical analysis
We quantified the distribution of cohort-level characteristics with descriptive analyses. We calculated the incidence of clinical outcomes as a proportion of total patients treated with GKRS. We quantified the occurrence of haemorrhage and fatality at any time after GKRS either during the total person-years of follow-up stated or by multiplying the median (or mean) follow-up duration by total number of patients treated. We calculated annual incidence rates for haemorrhage and fatality, using Poisson distributions.
We aimed to perform the following subgroup analyses: (1) clinical and obliteration outcomes reported in studies stratifying by AVM Spetzler-Martin grades [67]: I-II, III and IV-V; (2) clinical and obliteration outcomes reported in studies stratifying by Radiosurgery-Based AVM Score [61,62].
A meta-analysis was performed for all primary outcomes. DerSimonian-Laird random effects models were used for all summary effect estimates with a Freeman-Tukey double arcsine transformation [11,25].
Heterogeneity was assessed using the quantity I 2 [34]. Heterogeneity was investigated using Baujat plots [3] and leaveone-out sensitivity analysis. Publication bias was assessed using Funnel plots and Egger's test was performed to assess Funnel plot symmetry.
Moderator analysis was performed using meta-regression techniques to determine association of patient, AVM and treatment characteristics with clinical and radiological outcomes. Patients were stratified by the following prespecified variables: age, sex, study mid-year, median margin treatment dose (Gy), median AVM volume (cm 3 ), deep location (%), eloquent location (%), deep venous drainage (%). R version 4.0.4 (The R Foundation for Statistical Computing, Vienna, Austria) was used for all statistical analyses. For all statistical tests, P < 0.05 was considered statistically significant.

Study selection
After screening 3402 publications, 166 full-text studies were reviewed with 34 studies included in final analysis (Fig. 1, Appendix 3,4). A detailed summary of included studies is presented in Appendix 6.

Study characteristics
In the 35 cohorts (34 studies) receiving GKRS, including a total of 8673 patients with 41,544 patient-years of follow-up, the median cohort-level patient characteristics were: population size 182 patients, age at initial treatment 35 years, 53% male, follow-up duration 60 months, 49% presented with ICH, 71% AVMs were lobar, 20% deep and 72% in eloquent locations (Table 1 and Appendix 6).  Thirty-three (97%) studies were from single centres and one (3%) was multi-centre (Appendix 7). Thirteen (39%) studies were from North America [4,[14][15][16]18 Overall, included studies were at moderate risk of bias (Appendix 5). All studies were at moderate risk of bias due Table 2 Outcomes/incidence rate following single-stage gamma knife radiosurgery for brain arteriovenous malformations

Table 6
Overview of AVM obliteration rates reported in studies stratified by Spetzler-Martin grade IV-V Arteriovenous malformation (AVM); follow-up (FU); gamma-knife (GK); radiosurgery-based AVM score (RBAS); Spetzler-Martin (SM) a Statistics generated based on whole-study parameters (e.g., not necessarily just grade IV-V patients) Spetzler to confounding as none of the studies were randomised and none concealed treatment allocation. Fifteen studies (44%) were at moderate risk of selection bias; most reasoned to the retrospective selection of patients with a minimum of 2 years of follow-up. No studies were at risk of bias due to classification of intervention or from performance bias from deviation of intended intervention. Four studies (12%) were at moderate risk of attrition bias for obliteration outcomes from loss of patients to follow-up [6,8,54,57]. All four studies had > 80% of total patients treated available for analysis at the end of follow-up. All studies were retrospective, except one which identified and followed consecutive patients prospectively (Appendix 7) [65]. Three studies (9%) were deemed moderate risk of reporting bias [6,8,56]. Overall, no studies were deemed to be at serious risk of bias for any of the seven domains of the ROBINS-I tool [70].
Cohorts with higher proportions of AVMs with strictly deep venous drainage (p = 0.005) and eloquent location (p = 0.026) were associated with higher haemorrhage rates. The chance of complete AVM obliteration was higher in more recent cohorts (p = 0.021) with younger patients (p = 0.018) or AVMs with deep venous drainage (p = 0.035). A higher risk of both transient and permanent symptomatic RIC was associated with cohorts with higher proportions of AVMs in deep locations (p = 0.002 and 0.005). Appendix 8 summarises moderator analysis performed for primary outcomes.

Discussion
We have systematically reviewed the three primary outcomes defining successful AVM GKRS: rate of AVM obliteration, chance of post-GKRS ICH and risk of radiationinduced complications, to establish a comprehensive and contemporary risk: benefit profile for AVM GKRS.
Starke et al. previously reported on 2236 AVMs treated with GKRS between 1988 and 2013, with mean volume 4.3 cm 3 , 20.5 Gy margin dose and follow-up of 7 years. This multi-centre study reported a comparable overall nidus obliteration rate of 64.7%, haemorrhage rate of 7.4% with annual incidence of 1.1%, whilst symptomatic RICs were transient in 6.7% and permanent in 2.7% of patients at the end of follow-up [69]. The most contemporary systematic review of SRS for brain AVMs was performed by van Beijnum et al. in 2013, investigating the outcomes following treatment of 9436 AVMs across multiple interventions (surgery, SRS or embolisation). They reported a median post-SRS haemorrhage rate of 5.8% and an annual ICH and case-related fatality rate of 1.7 and 0.5 per 100 person-years [72].
Generally, nidus obliteration can be achieved in 65-85% of patients after a 3-5 year latency period [20,23,60,61,69]. Of note, van Beijnum et al. reported a significantly lower median angiography-confirmed obliteration rate of 38% following AVM SRS [72]. This can partly be reasoned to their strict calculation of obliteration rate, calculated as a proportion of patients confirmed obliterated as a proportion of all SRS-treated patients, which did not account for patients without complete angiography radiological followup to confirm obliteration or those patients lost to followup. Also, obliteration was calculated from 69 SRS studies (GKRS-modality: 22/69) with a shorter median follow-up of 35 months. In our pooled analysis, 60.5% of patients achieved angiography-confirmed obliteration and overall nidus obliteration rate was 69.7% with a median time to obliteration of 35.4 months after initial radiosurgical intervention. Whilst realisation of nidus obliteration on neuroimaging occurs at relatively varying intervals due to differences in the intervals of radiological follow-up, our median study follow-up (60 months) is a sufficient duration, we believe, to accommodate this post-GKRS latency period whilst supporting the validity of our findings.
On behalf of the International Stereotactic Radiosurgery Society (ISRS), a systematic review was performed to establish SRS practice guidelines for SM I/II AVMs [30]. Of 1102 SM I/II AVMs (78% SM II, median margin dose 23 Gy, volume 2.4 cm 3 ), overall reported obliteration rate was 80%. Of 888 SM I/II AVMs (72% SM II) in our study, 74% achieved obliteration. Both findings highlight that appropriately selected, low-grade AVMs can expect a significant chance of successful outcome.
Spetzler-Martin grade III AVMs are angioarchitecturally heterogeneous lesions, straddling the boundary between low grade (I-II) AVMs, which favour intervention with microsurgical resection, and high grade (IV-V) AVMs, which favour conservative management. Our findings suggest grade III AVMs experience a reasonable risk-to-benefit profile following GKRS; of 505 AVMs, 69% achieved obliteration.
Whilst our pooled and SM-stratified obliteration values entice direct comparisons with obliteration outcomes following surgical resection, caution is warranted as patient selection for GKRS versus surgical resection is undoubtedly to have been influenced by individual patient and AVM features which were not available for further detailed analysis.
Previous studies have reported on patient, AVM and treatment factors associated with the development of RICs, of which, increasing radiation dose and irradiated AVM volume, have been well established [5,13,17,40]. Kano et al. investigated the incidence of and treatment parameters that contributed to the development of RICs following AVM GKRS. Further, 775 patients (median volume 3.6 cm 3 and dose 20 Gy) underwent single-stage GKRS with ≥ 2 years of follow-up, with reported symptomatic and permanent RIC rates of 7.1% and 3%, respectively. Increased AVM volume, 12-Gy volume, higher margin dose and deep location were associated with a higher rate of developing symptomatic RICs [40]. In our meta-analysis, we found no significant association between either median study margin dose or AVM volume and the development of transient symptomatic (p = 0.45 and 0.43) or permanent RICs (p = 0.17 and 0.19). We believe that this may be due to lack of individual patient data, differences in patient selection and variations in GKRS treatment protocols among included studies; all of which precluded a sufficiently rigorous analysis of the relationship among study nidus volume, margin dose, and symptomatic RIC occurrence. Furthermore, this dose-volume relationship has been further refined to specify the '12-Gray volume', the volume of brain tissue receiving radiation dose of 12 Gy or more, which is strongly correlated with the risk of RICs [20,55]. Interestingly, only one study in our review [64] referred to the 12-Gy volume in their analysis of RIC outcomes. Studies suggest these dose-volume relationships are more strongly associated with the development of post-SRS RICs and less influential for the occurrence of symptomatic RICs [29,55].
One AVM angioarchitectural feature consistently reported to be associated with the development of symptomatic RICs is deep location, more specifically, the brainstem [19,21,22,61]. This was confirmed in our findings for both transient (p = 0.002) and permanent symptomatic RICs (p = 0.005) and is reflected in the RBAS grading system where deep location confers a higher risk feature associated with poorer outcome post-GKRS [61,62].
The relative risk of AVM haemorrhage during the latency period between radiosurgery and obliteration versus an untreated AVM's natural history remains a topic of debate [12]. In comparison to the generally accepted 2-4% natural haemorrhage rate of untreated AVMs [1], decreased, unchanged or increased rates of haemorrhage after radiosurgery have been reported [27,50,59]. In our pooled analysis, the annual haemorrhage rate following GKRS was 1.4%. This suggests radiosurgical intervention may afford partial protection from AVM rupture during the latency interval before nidus obliteration; however, we cannot definitively exclude that the decline in haemorrhage rate following GKRS is not part of the natural course of the disease or due to selection bias of included studies.
Further, in our analysis, we were unable to determine whether pre-GKRS AVM rupture status, yielded a significantly lower or higher post-GKRS haemorrhage risk as (1) the majority of studies did not stratify haemorrhage rate by rupture status, and (2) we did not have access to individual patient data to perform rigorous sub-group analysis within individual cohorts. Yen et al. [73] evaluated the rates of preand post-GKRS haemorrhage in a cohort of 1204 AVMs. The annual AVM haemorrhage risk from birth to radiosurgery, assuming patients are at risk of haemorrhage from birth, was 2.0% for the entire cohort and 3.7% for AVMs with prior haemorrhage. Post-GKRS, the annual haemorrhage risk until obliteration was 2.5%, stratified as 2.8% and 2.2% for AVMs with and without prior haemorrhage. Kano et al. [42] reported on a cohort of 407 ruptured AVMs. The annual haemorrhage rate between birth and radiosurgery was 3.4%, reducing to 1.3% following radiosurgical intervention.
Overall, it appears GKRS reduces the haemorrhage risk of AVMs before obliteration, an attribute that may be more pronounced in patients who present with haemorrhage.
Cyst formation (CF) is an uncommon complication following SRS. In our review, 15 studies totalling 3446 patients reported a pooled incidence of 2%. This is approximated by the only comprehensive analysis of CF following GKRS by Ilyas et al., reporting a pooled CF rate of 2.9% in 2562 patients following GKRS, with a mean latency period of 6.5 years from time of GKRS to cyst detection [37]. A shorter median study follow-up (5 years) in our study, may partly explain a slightly reduced CF incidence. Reassuringly, in most cases, cysts are managed conservatively with relative success, especially if they are asymptomatic, radiologically stable or do not exert significant local mass effect [12].

Limitations
The strengths of this review include adherence to PRISMA guidelines, its pre-specified protocol and formal validated risk-of-bias assessment. Studies included in this review generally followed a similar AVM GKRS treatment and  follow-up protocol, of which the median length of followup across all studies was approximately 5 years.
In terms of limitations, the included studies were all nonrandomised, mostly retrospective and at moderate risk of bias. In most analyses, heterogeneity was present and substantial, with the highest I 2 values found in cohorts reporting obliteration outcomes. We are aware that certain studies may have been excluded whilst strictly adhering to our inclusion criteria which required outcomes data on obliteration, haemorrhage and RICs cumulatively or excluded solely due to English language restriction (Appendix 3, 4).
Most studies (29) in this review reported an aggregated angiography or MRI confirmed obliteration rate, some in addition or as an alternative to an angiography-only confirmed obliteration rate. Whilst MR-imaging has been shown to exhibit 77-85% sensitivity and 89-95% specificity for AVM obliteration detection [48,63], there is a risk of incorrectly assuming AVMs to be obliterated by MRI at the time of assessment with subsequent over-estimation of the true nidus obliteration rate, especially when obliteration status is confirmed solely using MRI. False-positive nidal obliteration according to MRI can be detrimental to the patient because any residual arteriovenous shunting represents a persistent haemorrhage risk with the subsequent possibility of adverse outcome. In our pooled analysis, 60.5% of patients achieved angiography-confirmed obliteration whilst overall nidus obliteration rate, confirmed with either angiography or MRI, was 69.7%. Our findings for SM-stratified obliteration rates also illustrated a consistent and clinically significant incongruence between rates of obliteration confirmed by angiography or MRI and obliteration confirmed by angiography alone. Whilst it can be appreciated why some patients may choose to forgo DSA, for example due to its invasiveness, risk of procedural complications or false reassurance from the absence of AVM or GKRS-related symptoms, ultimately it remains the gold standard of accurately assessing AVM obliteration status and should remain the critical and definitive determinant of nidal obliteration in any post-GKRS imaging protocol [10,33].
In this review, we obtained aggregate patient, AVM and outcome data with a tendency for studies to report on the whole patient cohort with limited further sub-group analysis. Very few studies stratified outcomes by individual patient and AVM features, e.g. by location, AVM volume, margin dose or SM grade, which have been wellestablished to affect radiosurgical outcomes. Further, the reporting of a clinically heterogeneous cohort of patients with AVMs is likely to contribute to heterogenous pooled estimates of outcomes, as illustrated in our findings. Interpreting these summary outcome estimates, without robust sub-group analysis can mitigate the generalisability of our findings for individual patients and future AVM treatment decision-making.
Standardised prospective multi-centre recording of patient, AVM and treatment characteristics and reporting of outcomes is needed to be certain of individual AVM risks and benefit prediction. Stratifying clinical outcomes further by AVM and pre-specified treatment subgroups will contribute to the selection of a more homogenous set of AVMs from which meaningful comparisons of outcomes can be made and valid conclusions can be drawn.

Conclusion
Gamma knife radiosurgery is a safe and effective treatment option for cerebral AVMs. Appropriately selected patients can expect a significant chance of successful obliteration coupled with minimal risks of haemorrhage and radiation-induced complications. Future studies would be strengthened by attempting to report on a homogenous set of study participants, in terms of pre-specified patient, AVM angioarchitectural or treatment parameters, which would allow for a more conclusive risk: benefit profile of AVM GKRS to be established. provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.