Long‐Term Outcomes and Natural Course of Giant Cerebral Arteriovenous Malformations (> 6 cm): Insights From a National Multicenter Propensity Score‐Matched Cohort

ABSTRACT

Objectives

This study evaluated the natural history of giant cerebral arteriovenous malformations (AVMs) over 6 cm and compared long‐term outcomes of interventional treatment versus conservative management.

Materials and Methods

Patients with AVMs > 6 cm were identified from a national multicenter prospective registry (MATCH). Rupture risk factors were analyzed with uni‐ and multivariate models. Propensity score matching balanced intervention and conservative groups. The primary outcome was long‐term hemorrhagic stroke or death; secondary outcomes included obliteration rates and neurological status. Subgroup and sensitivity analyses assessed robustness.

Results

From August 2011 to December 2021, 380 patients with giant AVMs were enrolled. Annual rupture risk was 2.4% for unruptured lesions, 9.4% for previously ruptured, and 3.6% overall. Ventricular involvement (OR = 3.61) and draining vein stenosis (OR = 2.61) were independent hemorrhage risk factors. Over a mean follow‐up of 7.3 years, intervention did not significantly reduce hemorrhagic stroke or death compared to conservative management but was associated with higher hemorrhagic stroke risk (HR = 2.04) and neurological deterioration, despite higher obliteration rates (39.1%). Stratified analysis suggested microsurgery and embolization alone were less favorable, while embolization plus radiosurgery more effectively reduced hemorrhagic stroke or death. Subgroup analysis indicated conservative management was preferable for higher S‐M grades, unruptured AVMs, and eloquent brain regions.

Conclusion

The annual rupture rate of giant AVMs (> 6 cm) is approximately 3.6%. Interventional treatment for giant AVMs is not superior to conservative management, and the risks of hemorrhagic stroke and neurological deterioration remain substantial—particularly for unruptured, high‐grade, or eloquently located AVMs.

This multicenter prospective study of 380 patients with giant cerebral arteriovenous malformations (> 6 cm) found an overall annual rupture rate of 3.6%. Interventional treatment did not decrease long‐term hemorrhagic stroke or death compared with conservative management and carried higher risks of hemorrhage and neurological deterioration. Combined embolization and radiosurgery showed relatively better outcomes, while conservative management was favored for unruptured, high‐grade, or eloquent AVMs.

1. Introduction

Cerebral arteriovenous malformations (AVMs) are neurological disorders characterized by abnormal vascular structures in the brain, primarily featuring aberrant connections between arteries and veins. These malformed vessels intertwine, allowing blood to bypass the normal capillary bed to form lesions [1]. These abnormal vascular structures frequently cause severe neurological complications, with hemorrhagic stroke being the most critical, making AVMs one of the leading causes of spontaneous cerebral hemorrhage in young adults [2]. AVMs demonstrate considerable heterogeneity with numerous subtypes, complicating treatment decisions [3]. The most conventional classification system is the Spetzler‐Martin grading, which categorizes lesions based on size: less than 3 cm, 3–6 cm, and greater than 6 cm [4]. Lesions exceeding 6 cm are classified as giant AVMs [5]. Giant AVMs have a low incidence, constituting less than 10% of all AVMs [6]. These lesions are characterized by their enormous size and, due to their extensive involvement, frequently affect eloquent areas, deep locations, or other critical structures, increasing surgical complexity and potential neurological damage [7]. Treatment approaches including microsurgical resection (MS), endovascular embolization (EM), and stereotactic radiosurgery (SRS) present distinct risks and challenges. Therefore, determining the safest and most effective long‐term therapeutic strategy for this high‐risk, technically challenging AVM subtype is crucial.

MS is generally recognized as the standard curative treatment for AVMs [8]. However, the extensive involvement, structural complexity, and technical challenges associated with giant AVMs cannot be overlooked [9, 10]. EM therapy may provide long‐term benefits for some specially selected AVMs, but giant AVMs are typically more difficult to successfully embolize due to their substantial volume and complex vascular architecture [11]. Recent guidelines suggest that embolization‐assisted stereotactic radiosurgery may be advantageous for high‐grade large AVMs; however, some reports indicate that SRS efficacy may be compromised by artifacts from embolic agents in SRS planning images and radiation absorption [12]. Additionally, multimodal, multistage treatment strategies are recommended for large AVMs [8, 13], though the primary limitations are the lack of individualization and evidence‐based protocols for multimodal and staged planning [14]. Notably, previous studies focusing on giant AVMs have predominantly been case reports, single‐arm, single‐center investigations, or analyses of individual treatment modalities, lacking adequate sample sizes and clearly defined control comparisons. To date, there are no definitive treatment recommendations or large‐scale multicenter studies evaluating multiple treatment modalities to address this issue.

Given these uncertainties, large‐scale, meticulously designed research is needed to elucidate the natural history and relative efficacy of interventions for giant AVMs exceeding 6 cm. This prospective multicenter study aims to provide reliable, comprehensive long‐term data clarifying the natural history of giant AVMs and comparing risk–benefit outcomes in terms of hemorrhagic stroke or mortality rates, obliteration rates, and neurological outcomes across different treatment modalities. These findings will help guide clinicians in selecting the most appropriate therapeutic approach for giant AVMs.

2. Method

2.1. Data Sources and Study Design

The data for this study were obtained from a retrospective multicenter registry, the MATCH study (the registry of multimodality treatment for brain AVMs in China, NCT 04572568), from August 2011 to December 2021 [15]. The primary objective of the MATCH study is to explore the natural history of AVM in the Asian population and to identify the most effective and personalized treatment strategies. The reliability of the MATCH registry has been validated through established protocols and peer‐reviewed publications [16, 17, 18]. A detailed procedure of data quality management is shown in Data S1. This study obtained the approval of the institutional ethics committee (IRB approval number: KY 2020‐003‐01) in compliance with the 1964 Helsinki Declaration. All patients who enrolled in the registry gave their written informed consent at admission. This study followed the STROCSS guidelines for observational cohort studies [19].

Giant AVMs (> 6 cm) were defined as lesions with a maximum diameter exceeding 6 cm, in accordance with the Spetzler‐Martin grading system. Inclusion criteria comprised AVMs measuring greater than 6 cm in diameter. Exclusion criteria encompassed patients with insufficient baseline clinical parameters, absence of pre‐intervention radiological imaging, or unavailable post‐treatment radiological and clinical outcome data. All eligible patients were incorporated into the natural history analysis, whereas patients with continued follow‐up were included in the analysis of long‐term treatment outcomes.

2.2. Cohort Definition

For long‐term outcome analysis, patients were divided into conservative and interventional treatment groups. Conservative treatment included pharmacological management or emergency procedures (e.g., EVD or decompressive craniectomy) without direct AVM‐targeted intervention. Interventional treatment comprised all modalities aimed at lesion obliteration and was further categorized into microsurgery, embolization, and embolization plus stereotactic radiosurgery (EM + SRS). Microsurgical cases required preoperative catheter angiography and MRI, with surgeries performed by neurosurgeons experienced in AVM resection. Embolization was performed via arterial access using microcatheter‐based techniques, primarily utilizing Onyx as the embolic agent. SRS was conducted with Gamma Knife, using 3D stereotactic MRI for target delineation and treatment planning by senior neuroradiologists. Dosage and planning were tailored to AVM location and volume, following established protocols. This stratification enabled comprehensive comparison of treatment modalities and outcomes.

2.3. Baseline Characteristics

The study collected data on demographic variables (including age at diagnosis and sex), clinical presentations (seizures, neurological deficits, life‐saving craniotomies for hematoma evacuation or ventricular catheter drainage, and modified Rankin Scale [mRS] score at admission), as well as radiological characteristics of AVMs (feeding arteries, location of lesion and draining veins, size, and angiographic features). Radiological characteristics were defined according to the reporting terminology guidelines [20], and assessed by digital subtraction angiography (DSA) and MRI. The angioarchitectural parameters were verified by neurosurgery residents who received training from qualified senior neuroradiologists.

2.4. Outcomes and Follow‐Up

To assess long‐term efficacy, the primary outcome was a composite of hemorrhagic stroke or death during follow‐up. Hemorrhagic stroke was defined as a symptomatic event (new neurological deficit, seizure, or severe headache) with AVM‐related intracranial or subarachnoid hemorrhage confirmed by CT or MRI. It is generally recognized that acute hemorrhage occurring within 2 weeks postoperatively is more likely attributable to surgical technical factors rather than the intrinsic hemorrhagic risk of the AVM itself. Inclusion of such acute postoperative cases could confound the assessment of the intervention's effect on the lesion's natural hemorrhage risk. Therefore, cases with acute postoperative hemorrhage within 2 weeks were excluded from our analysis. Secondary outcomes included AVM occlusion (absence of abnormal flow on MRI or DSA) and neurological outcomes, assessed by the modified Rankin Scale (mRS). Disability was defined as mRS > 2, and neurological deterioration as worsening mRS at final follow‐up. Clinical follow‐up was conducted at 3 months, annually up to 3 years, and then every 5 years, using telephone or record review (Data S2). Strategies were employed to minimize follow‐up bias. Radiological assessments were independently performed by two senior neuroradiologists.

2.5. Controlling for Confounding

To reduce the effects of potential confounding bias, we used propensity scores to control for pretreatment imbalances on baseline characteristics. These factors that could potentially affect both the selection of therapeutic strategies and outcomes were specifically referred to as “confounding by indication” [21]. Propensity score matching (PSM) methods were recommended to address this issue in clinical research [22]. All available baseline characteristics, including demographic factors, clinical presentation at admission, morphological features, and angioarchitectural parameters, were matched between the two groups separately. In this study, we used the nearest‐neighbor method with the caliper radius of 0.2 without replacement to do 1:1 patient matching. Covariate balance was assessed with standardized mean differences (SMD) with an acceptable balance set at values less than 0.1. Even with PSM, residual confounding from unmeasured factors could remain. Thus, we used E‐values to assess how strong such factors would have to be to nullify the observed association [23]. The matched baseline variables include: age at diagnosis, mRS score at admission, gender, rupture status, Spetzler–Martin grade; lesion‐related characteristics (ventricular involvement, frontal lobe, temporal lobe, parietal lobe, occipital lobe, cerebellum, brain stem, basal ganglia, deep lesions, subtentorial lesions, and involvement of functional areas); blood supply artery characteristics (multiple blood supply, deep perfusion, flow‐related aneurysms); diffuse lesions; and venous drainage characteristics (deep drainage, single drainage, drainage stenosis, and venous aneurysms).

2.6. Statistical Analyses

All statistical analyses were performed using R (version 4.2.2), with significance set at p < 0.05. Continuous variables were reported as mean ± SD or median (IQR), as appropriate. Binary logistic regression identified factors associated with AVM rupture. PSM was applied to compare interventional and conservative treatment groups, with baseline characteristics evaluated before and after matching. Categorical variables were presented as percentages. Analyses focused on the post‐PSM cohort. Kaplan–Meier curves illustrated cumulative incidence of primary outcomes, and hazard rates were compared across treatment strategies. Attributable risk (AR), incidence rates, and hazard ratios (HR) with 95% confidence intervals were calculated for primary outcomes using Cox models. For secondary outcomes, AR and odds ratios (OR) were determined via logistic regression. Subgroup and sensitivity analyses were conducted based on AVM grade, demographics, and treatment stages to assess robustness.

3. Result

3.1. Study Population, Natural History of Rupture, and Baseline Characteristics

From August 2011 to December 2021, the MATCH registry enrolled 4286 patients diagnosed with AVMs. Following comprehensive screening, 380 patients (8.9%) with giant AVMs were identified. Baseline characteristics stratified by rupture status are presented in Table 1. For patients with unruptured AVMs at initial diagnosis, 31 hemorrhagic events occurred during 1294 patient‐years of conservative observation, yielding an annual rupture rate of 2.40%. In contrast, patients with ruptured AVMs at diagnosis experienced 26 hemorrhagic events during 277 patient‐years of conservative observation, corresponding to an annual rupture rate of 9.39%. Collectively, 57 hemorrhagic events were documented during 1571 patient‐years of conservative management, resulting in an overall annual rupture rate of 3.63%. At initial presentation, 27.6% of patients exhibited ruptured AVMs, while 72.3% presented with unruptured lesions. Multivariate logistic regression analysis identified ventricular involvement (adjusted OR: 3.61, 95% CI [2.02–6.45], p < 0.001) and venous drainage stenosis (adjusted OR: 2.61, 95% CI [1.42–4.81], p = 0.002) as independent risk factors for AVM rupture (Table 2).

TABLE 1.

Risk factors related to rupture.

Variables	Total (n = 331)	Unrupture (n = 244)	Rupture (n = 87)	p
Age at diagnosis, Mean ± SD	23.64 ± 12.62	23.89 ± 11.60	22.93 ± 15.17	0.594
mRS at admission, M (Q1, Q3)	1.00 (1.00, 2.00)	1.00 (1.00, 1.00)	1.00 (1.00, 2.00)	0.758
Female, n (%)	119 (35.95)	84 (34.43)	35 (40.23)	0.333
S‐M grade, n (%)				0.904
III, n (%)	62 (18.73)	47 (19.26)	15 (17.24)
IV, n (%)	148 (44.71)	109 (44.67)	39 (44.83)
V, n (%)	121 (36.56)	88 (36.07)	33 (37.93)
Location
Involved ventricles, n (%)	206 (62.24)	135 (55.33)	71 (81.61)	< 0.001
Frontal lobe, n (%)	112 (33.84)	88 (36.07)	24 (27.59)	0.151
Temporal lobe, n (%)	137 (41.39)	99 (40.57)	38 (43.68)	0.614
Parietal lobe, n (%)	143 (43.20)	112 (45.90)	31 (35.63)	0.097
Occipital lobe, n (%)	81 (24.47)	64 (26.23)	17 (19.54)	0.213
Cerebellum, n (%)	23 (6.95)	16 (6.56)	7 (8.05)	0.639
Brain stem, n (%)	7 (2.11)	6 (2.46)	1 (1.15)	0.768
Basal ganglia, n (%)	44 (13.29)	27 (11.07)	17 (19.54)	0.046
Any deep‐seated, n (%)	98 (29.61)	63 (25.82)	35 (40.23)	0.011
Infratentorial, n (%)	26 (7.85)	19 (7.79)	7 (8.05)	0.939
Eloquent, n (%)	223 (67.37)	167 (68.44)	56 (64.37)	0.486
Feeder artery
Multi‐source, n (%)	213 (64.35)	166 (68.03)	47 (54.02)	0.019
Deep perfusion, n (%)	196 (59.21)	145 (59.43)	51 (58.62)	0.896
Flow‐related aneurysm, n (%)	72 (21.75)	48 (19.67)	24 (27.59)	0.124
Diffuse nidus, n (%)	143 (43.20)	106 (43.44)	37 (42.53)	0.883
Drainage
Any deep drainage, n (%)	167 (50.45)	118 (48.36)	49 (56.32)	0.202
Single drainage, n (%)	61 (18.43)	44 (18.03)	17 (19.54)	0.756
Drainage stenosis, n (%)	45 (13.60)	27 (11.07)	18 (20.69)	0.025
Venous aneurysm, n (%)	111 (33.53)	93 (38.11)	18 (20.69)	0.003

Abbreviations: χ², Chi‐square test; M, median; mRS, modified Rankin Scale; Q₁, 1st Quartile; Q₃, 3st Quartile; S‐M, Spetzler‐Martin; SD, standard deviation; t, t‐test; Z, Mann–Whitney test.

TABLE 2.

Risk factors related to rupture uni‐ and multi‐variable analysis.

Variables	Univariable					Multivariable
	β	S.E	Z	p	OR (95% CI)	β	S.E	Z	p	OR (95% CI)
Female	0.22	0.24	0.92	0.357	1.24 (0.78 ~ 1.97)
S‐M grade
III					1.00 (Reference)
IV	0.24	0.32	0.73	0.462	1.27 (0.68 ~ 2.37)
V	0.18	0.33	0.56	0.578	1.20 (0.63 ~ 2.29)
Location
Involved ventricles	1.36	0.28	4.83	< 0.001	3.88 (2.24 ~ 6.74)	1.28	0.3	4.32	< 0.001	3.61 (2.02 ~ 6.45)
Frontal lobe	−0.4	0.25	−1.59	0.112	0.67 (0.41 ~ 1.10)
Temporal lobe	0.11	0.23	0.47	0.64	1.12 (0.71 ~ 1.76)
Parietal lobe	−0.44	0.24	−1.84	0.066	0.64 (0.40 ~ 1.03)
Occipital lobe	−0.31	0.28	−1.12	0.261	0.73 (0.43 ~ 1.26)
Cerebellum	0.35	0.41	0.85	0.393	1.42 (0.64 ~ 3.16)
Brain stem	−0.3	0.81	−0.37	0.714	0.74 (0.15 ~ 3.64)
Basal ganglia	0.65	0.31	2.07	0.038	1.92 (1.04 ~ 3.56)	0.04	0.4	0.11	0.916	1.04 (0.48 ~ 2.28)
Any deep‐seated	0.69	0.24	2.88	0.004	2.00 (1.25 ~ 3.20)	0.26	0.31	0.83	0.404	1.29 (0.71 ~ 2.37)
Subtentorial	0.14	0.4	0.36	0.72	1.15 (0.53 ~ 2.51)
Eloquent	−0.1	0.24	−0.42	0.671	0.90 (0.56 ~ 1.45)
Feeding artery
Deep perfusion	0.09	0.23	0.39	0.7	1.09 (0.69 ~ 1.73)
Flow‐related aneurysm	0.4	0.26	1.51	0.13	1.49 (0.89 ~ 2.51)
Diffuse nidus	−0.05	0.23	−0.22	0.828	0.95 (0.60 ~ 1.50)
Drainage
Any deep drainage	0.24	0.23	1.03	0.302	1.27 (0.81 ~ 1.99)
Single drainage	0.06	0.29	0.19	0.846	1.06 (0.60 ~ 1.88)
Drainage stenosis	0.95	0.3	3.22	0.001	2.59 (1.45 ~ 4.63)	0.96	0.31	3.08	0.002	2.61 (1.42 ~ 4.81)

Note: *Bold values: statistical significances.

Abbreviations: CI, confidence interval; OR, odds ratio; S‐M, Speztler‐Martin.

Following comprehensive screening, 331 patients who underwent either conservative or interventional treatment with long‐term clinical follow‐up were included in the analysis (215 received interventional treatment, 116 received conservative management). Detailed baseline characteristics are presented in Tables S1 and S2. After PSM, 198 patients (99 in each cohort) were included in the comparative analysis between conservative and interventional treatment groups, with all recorded baseline characteristics demonstrating comparable distributions as illustrated in Table S3 and Figure S1. The median follow‐up duration for the entire cohort was 6.84 years (IQR: 3.03–9.83), with similar follow‐up periods between groups (conservative versus interventional treatment: 7.00 [3.04, 10.29] versus 6.33 [2.99, 8.79] years, respectively). Detailed follow‐up protocols and reassessment criteria are documented in Table S4. Patient selection methodology is illustrated in Figure 1.

FIGURE 1.

Flowchart of patient selection.

3.2. Long‐Term Outcomes of Conservative Treatment vs. Interventional Treatment in Comparison

After PSM, 22.7% were ruptured AVMs. The risk of hemorrhagic stroke or death did not demonstrate a significant difference between the interventional and conservative treatment groups (conservative versus interventional: 1.89 versus 3.33 per 100 person‐years; absolute risk [AR], 1.44 per 100 person‐years (95% CI: −0.22 to 3.10), p = 0.089; hazard ratio [HR], 1.88 (95% CI: 0.96 to 3.69), p = 0.061). When analyzing primary outcomes individually, the incidence of hemorrhagic stroke (AR, 1.57 (95% CI: −0.06 to 3.21) per 100 person‐years, p = 0.060; HR, 2.03 (95% CI: 1.02 to 4.06), p = 0.039) was significantly higher in the interventional group compared to the conservative group, while AVM‐related mortality (AR, 0.42 (95% CI: −0.29 to 1.14) per 100 person‐years, p = 0.245; HR, 2.83 (95% CI: 0.55 to 14.63), p = 0.194) showed no significant between‐group difference (Table 3). Kaplan–Meier curves illustrating the cumulative incidence of hemorrhagic stroke or death are presented in Figure 2. Regarding secondary outcomes, the lesion obliteration rate was significantly higher in the interventional treatment group (AR, 39.08 (95% CI: 25.94 to 52.22), p < 0.001). The rate of neurological deterioration was significantly higher in the interventional group compared to the conservative group (OR, 3.15 (95% CI: 1.47 to 6.78), p = 0.003); however, no significant difference was observed in the rate of neurological disability (OR, 1.72 (95% CI: 0.79 to 3.77), p = 0.174).

TABLE 3.

Outcomes of conservative treatment vs. interventional treatment.

	Conservativetreatment (%/per 100 person‐year)	Interventional treatment (%/per 100 person‐year)	Attributable risk a (95% CI)	p	HR (95% CI)/OR (95% CI) a	p	E‐value (95% CI) b
Primary outcomes
Hemorrhage stroke or death	14 (1.89)	24 (3.33)	1.44 (−0.22 ~ 3.10)	0.089	1.88 (0.96 ~ 3.69)	0.061	—
Symptomatic hemorrhagic stroke	13 (1.75)	24 (3.33)	1.57 (−0.06 ~ 3.21)	0.060	2.04 (1.02 ~ 4.06)	0.039	3.49 (1.16 ~ 7.58)
AVM related death	2 (0.27)	5 (0.69)	0.42 (−0.29 ~ 1.14)	0.245	2.83 (0.55 ~ 14.63)	0.194	—
Secondary outcomes
Obliteration	0 (0.00)	34 (39.08)	39.08 (25.94 ~ 52.22)	< 0.001	—	—	—
Disability	12 (12.12)	19 (19.19)	7.07 (−3.00 ~ 17.15)	0.169	1.72 (0.79 ~ 3.77)	0.174	—
Neurofunctional decline	11 (11.11)	28 (28.28)	17.17 (6.35 ~ 27.99)	0.002	3.15 (1.47 ~ 6.78)	0.003	5.75 (2.30 ~ 13.04)

Note: The matched baseline variables include: age at diagnosis, mRS score at admission, gender, rupture status, Spetzler–Martin grade; lesion‐related characteristics (ventricular involvement, frontal lobe, temporal lobe, parietal lobe, occipital lobe, cerebellum, brain stem, basal ganglia, deep lesions, subtentorial lesions, and involvement of functional areas); blood supply artery characteristics (multiple blood supply, deep perfusion, flow‐related aneurysms); diffuse lesions; and venous drainage characteristics (deep drainage, single drainage, drainage stenosis, and venous aneurysms).

Abbreviations: AVM, arteriovenous malformation; CI, confidence interval; HR, hazard ratio; OR, odds ratio.

^a The results were calculated with the conservative treatment group as the reference. The metrics of the primary outcomes were expressed as rate per 100 patient years and hazard ratios, and the secondary outcomes were expressed as proportion and odds ratios.

^b E‐values were calculated for HRs or ORs with statistical significance.

FIGURE 2.

Cumulative incidence of primary outcomes in conservative treatment vs. interventional treatment. ((A) Hemorrhagic stroke or death; (B) Hemorrhagic stroke; (C) Deathṣ).

3.3. Long‐Term Outcomes of Conservative Treatment vs. Different Interventional Strategies

In the comparison of different treatment strategies, we conducted propensity score matching analyses for various interventional approaches compared to conservative management. Results demonstrated that compared to conservative treatment, MS, standalone EM, and EM + SRS showed no significant differences in long‐term risk of hemorrhagic stroke or death (MS: AR, 0.82 (95% CI: −1.26 to 2.91), p = 0.439; EM: AR, 1.11 (95% CI: −0.80 to 3.02), p = 0.253; EM + SRS: AR, −0.21 (95% CI: −3.14 to 2.71), p = 0.888). Interestingly, despite the absence of statistical significance, the hazard ratios across these three interventional approaches demonstrated distinct trends (MS: HR, 1.60 (95% CI: 0.66 to 3.91); EM: HR, 1.50 (95% CI: 0.59 to 3.86); EM + SRS: HR, 0.72 (95% CI: 0.26 to 2.01)), ultimately suggesting that compared to conservative management, MS and EM tended to disfavor intervention, while EM + SRS tended to favor interventional treatment (Table 4). Kaplan–Meier curves illustrating the cumulative incidence of hemorrhagic stroke or death are presented in Figure 3. Among patients who underwent EM, 18% received curative embolization, 50% underwent palliative embolization, and 19% received targeted embolization. Immediate post‐procedural angiographic assessment demonstrated complete obliteration in 1% of patients, near‐total or substantial obliteration in 11%, and partial obliteration in 84%. Regarding secondary outcomes, compared to conservative treatment, MS achieved a significantly higher obliteration rate compared with conservative management (93.33% vs. 0.00%, p < 0.001). In contrast, no significant differences in obliteration rates were observed for EM or EM + SRS compared with conservative treatment (EM vs. conservative, 3.51% vs. 0.00%, p = 0.150; EM + SRS vs. conservative, 9.53% vs. 0.00%, p = 0.137). Regarding neurological outcomes, MS was associated with a significantly higher incidence of neurological deterioration than conservative management (25.00% vs. 6.67%, p = 0.010), whereas EM and EM + SRS did not differ significantly from conservative treatment in this respect (EM vs. conservative, 28.07% vs. 14.04%, p = 0.071; EM + SRS vs. conservative, 24.32% vs. 16.22%, p = 0.388).

TABLE 4.

Outcomes different strategies vs. conservative treatment.

	Conservative treatment (%/per 100 person‐year)	Microsurgery (%/per 100 person‐year)	Attributable risk a (95% CI)	p	HR (95% CI)/OR (95% CI) a	p	E‐value (95% CI) b
Primary outcomes
Hemorrhage stroke or death	9 (2.03)	12 (2.85)	0.82 (−1.26 ~ 2.91)	0.439	1.60 (0.66 ~ 3.91)	0.296	—
Symptomatic hemorrhagic stroke	9 (2.03)	11 (2.61)	0.59 (−1.45 ~ 2.62)	0.572	1.45 (0.58 ~ 3.61)	0.420	—
AVM related death	1 (0.23)	2 (0.48)	0.25 (−0.54 ~ 1.04)	0.537	2.35 (0.21 ~ 26.01)	0.473	—
Secondary outcomes
Obliteration	0 (0.00)	56 (93.33)	93.33 (87.02 ~ 99.65)	< 0.001	—	—	—
Disability	7 (11.67)	12 (20.00)	8.33 (−4.64 ~ 21.31)	0.208	1.89 (0.69 ~ 5.20)	0.216	—
Neurofunctional decline	4 (6.67)	15 (25.00)	18.33 (5.69 ~ 30.98)	0.004	4.67 (1.45 ~ 15.05)	0.010	8.81 (2.26 ~ 29.59)

	Conservative treatment (%/per 100 person‐year)	Embolization (%/per 100 person‐year)	Attributable risk a (95% CI)	p	HR (95% CI)/OR (95% CI) a	p	E‐value (95% CI) b
Primary outcomes
Hemorrhage stroke or death	8 (1.62)	12 (2.73)	1.11 (−0.8 ~ 3.02)	0.253	1.50 (0.59 ~ 3.86)	0.394	—
Symptomatic hemorrhagic stroke	8 (1.62)	10 (2.28)	0.66 (−1.15 ~ 2.46)	0.474	1.28 (0.48 ~ 3.39)	0.624	—
AVM related death	1 (0.20)	5 (1.14)	0.94 (−0.14 ~ 2.01)	0.088	4.76 (0.55 ~ 41.04)	0.118	—
Secondary outcomes
Obliteration	0 (0.00)	2 (3.51)	3.51 (−1.27 ~ 8.29)	0.150	—	—	—
Disability	4 (7.02)	9 (15.79)	8.77 (−2.79 ~ 20.33)	0.137	2.48 (0.72 ~ 8.59)	0.151	—
Neurofunctional decline	8 (14.04)	16 (28.07)	14.04 (−0.71 ~ 28.78)	0.062	2.39 (0.93 ~ 6.15)	0.071	—

	Conservative treatment (%/per 100 person‐year)	Embolization + stereotactic radiosurgery (%/per 100 person‐year)	Attributable risk a (95% CI)	p	HR (95% CI)/OR (95% CI) a	p	E‐value (95% CI) b
Primary outcomes
Hemorrhage stroke or death	8 (3.09)	8 (2.88)	−0.21 (−3.14 ~ 2.71)	0.888	0.72 (0.26 ~ 2.01)	0.531	—
Symptomatic hemorrhagic stroke	8 (3.09)	8 (2.88)	−0.21 (−3.14 ~ 2.71)	0.888	0.72 (0.26 ~ 2.01)	0.531	—
AVM related death	1 (0.39)	0 (0.00)	−0.39 (−1.14 ~ 0.37)	0.317	—	—	—
Secondary outcomes
Obliteration	0 (0.00)	2 (9.52)	9.52 (−3.03 ~ 22.08)	0.137	—	—	—
Disability	4 (10.81)	3 (8.11)	−2.7 (−16.02 ~ 10.62)	0.691	0.73 (0.15 ~ 3.50)	0.692	—
Neurofunctional decline	6 (16.22)	9 (24.32)	8.11 (−10.12 ~ 26.33)	0.383	1.66 (0.52 ~ 5.26)	0.388	—

Abbreviations: AVM, arteriovenous malformation; CI, confidence interval; HR, hazard ratio; OR, odds ratio.

^a The results were calculated with the conservative treatment group as the reference. The metrics of the primary outcomes were expressed as rate per 100 patient years and hazard ratios, and the secondary outcomes were expressed as proportion and odds ratios.

^b E‐values were calculated for HRs or ORs with statistical significance.

FIGURE 3.

Cumulative incidence of primary outcomes in conservative treatment vs. different strategies. (MS vs. conservative treatment [(A) Hemorrhagic stroke or death; (B) Hemorrhagic stroke; (C) Death]; EM vs. conservative treatment [(D) Hemorrhagic stroke or death; (E) Hemorrhagic stroke; (F) Death]; EM + SRS vs. conservative treatment [(G) Hemorrhagic stroke or death; (H) Hemorrhagic stroke; (I) Death]).

3.4. Subgroup Analyses and Sensitivity Analyses

Subgroup analysis by Spetzler‐Martin grade (Figure 4) demonstrated that no positive events were observed in the conservative treatment group among S‐M III patients. However, across S‐M IV to S‐M V subtypes (S3E0V1, S3E1V0, S3E1V1), outcomes increasingly favored conservative management as grade increased, although these differences did not reach statistical significance. In subgroup analyses of different AVM types (Figure 5), significantly elevated risks of hemorrhagic stroke or death following interventional treatment were observed in unruptured AVMs and AVMs located in eloquent regions, although no interaction effects were detected between subgroups. Figure 6 presents subgroup analyses across the three interventional strategies, with results consistent with the primary analysis. Sensitivity analyses (Table S5) evaluated the relative efficacy of treatments across different study design cohorts. The analysis results of the multimodal multistage treatment, long‐term follow‐up cohort, and unmatched cohort are consistent with the preliminary analysis.

FIGURE 4.

Subgroup analysis for primary outcomes in S‐M grades.

FIGURE 5.

Subgroup analysis for primary outcomes in conservative treatment vs. interventional treatment.

FIGURE 6.

Subgroup analysis for primary outcomes in conservative treatment vs different strategies ((A) MS; (B) EM; (C) EM + SRS).

4. Discussion

As a rare subtype of AVMs, giant AVMs over 6 cm in diameter have been infrequently reported in the literature, making it essential to understand their natural history and optimal management strategies. In this study, we aimed to evaluate the natural rupture risk in patients with giant AVMs and compare the long‐term risk differentials between interventional and conservative treatment approaches. Our findings revealed an overall annual hemorrhage rate of 3.63% for giant AVMs, with rates of 2.40% for unruptured lesions and 9.39% for previously ruptured AVMs. These overall results align with the commonly reported annual hemorrhage rates of 2%–4% [24, 25], suggesting that despite their extensive involvement and higher likelihood of incorporating other high‐risk structural factors, giant AVMs do not demonstrate significantly increased rupture risks as might be expected. Independent risk factors for hemorrhage in giant AVMs included intraventricular extension and draining vein stenosis, both physiologically plausible due to increased drainage pressures and impaired outflow. Our analysis showed that interventional therapy raised long‐term hemorrhage risk without affecting mortality and was linked to higher neurological deterioration, though not permanent disability. While interventions achieved lesion obliteration, MS, EM, and EM + SRS did not significantly change long‐term hemorrhagic stroke or death risk compared to conservative management. Trend analysis favored conservative management for MS and EM, with EM + SRS showing a slight benefit for intervention. MS was clearly associated with increased neurological deterioration. Subgroup analyses highlighted greater risks with intervention in unruptured AVMs and eloquent regions. Sensitivity analyses consistently supported conservative management for giant AVMs.

This national multi‐center prospective cohort study focused on giant AVMs, a rare subtype comprising less than 10% of all AVMs. Due to their rarity, few studies have addressed their natural history or long‐term hemorrhage risk. Limited patient numbers and treatment selection bias hinder robust, comparable research [26, 27, 28]. Single‐arm studies may overestimate treatment safety and efficacy, limiting reliability. Our prospective multi‐center study overcomes previous limitations of small samples, retrospective designs, and single‐arm approaches, providing stronger evidence for managing giant AVMs. Prior reports estimate the annual hemorrhage risk for giant AVMs at 2.2%–4.5% [6], which is consistent with our finding of approximately 3.63%. Interestingly, despite potentially harboring more risk factors, giant AVMs do not demonstrate significantly elevated hemorrhage risk. This phenomenon may be attributed to the larger vascular capacity within extensive AVMs, conferring greater tolerance to hemodynamic fluctuations [29]. Additionally, some studies suggest that smaller AVMs actually carry a higher rupture risk, which aligns with our observations. Moreover, for previously ruptured giant AVMs, the annual hemorrhage rate is nearly four times that of unruptured lesions, consistent with existing evidence. Regarding factors promoting AVM rupture, we identified intraventricular extension and draining vein stenosis as playing crucial roles. For these massive, high‐volume, and high‐flow AVMs, stenosis and deep ventricular involvement signify compromised venous outflow. This promotes lesional congestion and subsequently leads to AVM rupture [30, 31, 32, 33, 34, 35].

Treatment approaches for giant AVMs typically adopt strategies designed for large AVMs (defined as those exceeding 3 cm in diameter or 10 mL in volume). Current guidelines recommend multimodal, multi‐stage interventional approaches for large AVMs, employing volume‐staged or dose‐staged methods to progressively achieve lesion obliteration [3, 8]. This represents a highly personalized treatment paradigm with considerable heterogeneity in implementation. Several neurosurgical centers have extensively utilized staged SRS as their primary treatment modality, reporting obliteration rates ranging from 33% to 53%, with complication rates varying dramatically from 4% to 67%. Significant methodological variations exist between centers regarding staging approaches; while some centers employ volume‐staged or dose‐staged SRS monotherapy, others supplement SRS with adjunctive embolization or microsurgical resection [25, 26, 27, 36, 37]. Consequently, despite consensus on the fundamental principle of multimodal, multi‐stage treatment, optimal methodological specifics remain unestablished. Previous literature on embolization therapy has demonstrated that in cohorts with an average lesion diameter of 5 cm, embolization achieved only a 26.9% obliteration rate while carrying a 36% rate of serious complications [38, 39, 40]. In our current investigation, all interventional treatments showed a trend toward increased risk of hemorrhagic stroke or mortality compared to conservative management, with significantly higher long‐term hemorrhage rates in the intervention group. Regarding obliteration outcomes, our study demonstrated a 39% overall obliteration rate across all treatment modalities, consistent with previous research findings. However, substantial heterogeneity persists across different studies and treatment selection strategies. This methodological heterogeneity was similarly evident in the ARUBA trial [41], which partially explains why its findings have not gained widespread acceptance in clinical practice [42, 43].

Similar outcomes were observed across treatment strategies: neither MS, EM, nor EM + SRS significantly reduced long‐term risk compared to conservative management. Intervention and surgical groups showed higher rates of neurological deficits, suggesting that interventional treatments may increase neurological morbidity without lowering risk. Thus, interventional treatments cannot be routinely recommended over conservative management for giant AVMs. In the comparison between interventional and conservative management, an apparent difference between the two groups was observed; however, statistical testing using both p‐values and effect size did not demonstrate significance, most likely due to insufficient sample size and consequent lack of statistical power. Although the calculated AR values yielded relatively small p‐values, these still failed to reach statistical significance. Similarly, sensitivity analyses indicated that the outcomes between interventional and conservative approaches did not differ significantly. Despite the potential limitation of inadequate statistical power, the consistent findings across different analytical methods suggest that interventional treatment is not superior to conservative management and may even be disadvantageous with regard to hemorrhagic stroke outcomes. Notably, EM + SRS showed a trend toward reduced hemorrhagic stroke or mortality, partially supporting current guidelines, though this was not statistically significant. Study limitations include heterogeneity in multimodal protocols and the high surgical risk associated with these patients. Careful selection of treatment strategies is essential, considering the limitations of each modality, such as resection margins in MS, vascular access in EM, and volume or latency issues in SRS. Future management should focus on more sophisticated, individualized combined approaches for this complex patient population [12, 44, 45, 46, 47, 48]. Protective patient factors, especially in unruptured cases, are important to consider. Subgroup analysis showed a higher risk of hemorrhagic stroke or mortality with interventional treatment versus conservative management in these patients. Achieving complete obliteration in giant AVMs remains challenging; the overall obliteration rate with intervention was 39%, and even surgical resection reached only 93%, both lower than rates for other AVM subtypes. Excessive pursuit of complete obliteration may increase neurological risk, as shown by worse outcomes after intervention and surgery. The technical complexity of giant AVMs likely limits the potential for favorable long‐term results. In our multicenter registry cohort, only a small proportion of patients with giant AVMs larger than 6 cm underwent combined embolization and microsurgical resection. After propensity score matching, the sample size of this subgroup remained insufficient to provide adequate statistical power for reliable comparative analysis. Therefore, this treatment strategy was not included in the final analysis of therapeutic outcomes. Ischemic stroke represents an important adverse outcome following AVM treatment. Both previous reports and our observations suggest that such events are more commonly encountered after endovascular therapy, potentially related to the accuracy of embolic material deployment and treatment‐related complications. However, the focus of the present study was primarily on overall outcome differences between conservative management and interventional approaches. Given the relatively low incidence of ischemic stroke in conservatively managed AVMs within our cohort, the comparison between these strategies with respect to ischemic events may have limited clinical interpretability. Further investigations specifically designed to evaluate post‐treatment ischemic stroke risk are warranted to better clarify this issue.

This study demonstrates several methodological strengths. First, we utilized a large dataset derived from a prospective multicenter registry, ensuring a balanced data source and enhancing statistical power for primary outcome assessment. Additionally, our comparative analysis of multiple treatment modalities generated results that could be cross‐validated across different interventional approaches. Third, we implemented PSM to control for confounding variables associated with indication bias, effectively mitigating potential biases inherent in non‐randomized treatment selection. To address potential unmeasured confounding factors, E‐values were calculated to estimate the strength of unmeasured factors required to explain observed associations, thereby enhancing the robustness and reliability of our findings. Finally, multiple sensitivity analyses employing different study designs were conducted to evaluate whether methodological variations would affect the comparative effectiveness between the two treatment strategies.

However, several limitations warrant consideration. Firstly, due to the low incidence rate of giant AVMs, the cohort size and scope of intervention strategies may be limited, and the sample size after matching may be small, potentially leading to insufficient statistical power. Nevertheless, the consistency of the sensitivity analysis results with the main analysis demonstrates the high robustness of the conclusions. Second, the median follow‐up period of 6 years may insufficiently reflect the complete natural history of AVMs. Nevertheless, given the high‐risk profile of interventions for patients with giant AVMs, this follow‐up duration still provides considerable reference value. Third, despite adherence to standardized treatment protocols in the registry, intervention methodologies across centers may have varied due to clinical judgment and the absence of definitive guidelines in existing literature. Our study may provide important clinical reference points for future research, potentially contributing to reduced variability in treatment approaches. Fourthly, in our multicenter registry cohort, few patients with giant AVMs larger than 6 cm chose embolization plus microsurgery as their treatment option, and the sample size after PSM matching was insufficient to support adequate statistical power. Fifthly, as the cohort in this study is mostly from Chinese centers and the majority of patients are Chinese, the conclusions may be limited by ethnicity and affect extrapolation. In the future, more international multicenter cohort studies are expected.

5. Conclusions

The annual rupture rate of giant AVMs (> 6 cm) is approximately 3.6%. Interventional treatment for giant AVMs is not superior to conservative management, and the risks of hemorrhagic stroke and neurological deterioration remain substantial—particularly for unruptured, high‐grade, or eloquently located AVMs.

Author Contributions

Nan Li: methodology, software, data curation, investigation, validation, formal analysis, visualization, writing – original draft. Yukun Zhang: methodology, software, data curation, investigation, validation, formal analysis, writing – original draft. Chengzhuo Wang: methodology, software, data curation, investigation, validation, formal analysis, visualization, writing – original draft. Heze Han: methodology, investigation, validation. Li Ma: methodology, investigation, validation. Ruinan Li: investigation, resources. Zhipeng Li: investigation, resources. Haibin Zhang: resources. Kexin Yuan: resources. Anqi Li: resources. Qinghui Zhu: resources. Yongenbo Su: resources. Dezhi Gao: investigation. Hengwei Jin: investigation. Youxiang Li: investigation. Shibin Sun: investigation. Yuanli Zhao: conceptualization, supervision, funding acquisition. Shuo Wang: conceptualization, supervision, funding acquisition. Yu Chen: conceptualization, supervision, funding acquisition, writing – review and editing. Hao Wang: conceptualization, supervision, funding acquisition, writing – review and editing. Xiaolin Chen: conceptualization, supervision, funding acquisition, writing – review and editing. Jizong Zhao: conceptualization, supervision, writing – review and editing. All authors read and approved the final version of the manuscript.

Funding

This work was supported by the Natural Science Foundation of China (grant no. 82202244 to Yu Chen, and 81,771,234 and 82,071,302 to Yuanli Zhao), and the National Key Research and Development Program of China (Grant No. 2022YFB4702800 to Yuanli Zhao, and No. 2021YFC2501101 and 2020YFC2004701 to Xiaolin Chen).

Ethics Statement

This study obtained the approval of the institutional ethics committee of Beijing Tiantan Hospital (IRB approval number: KY 2020‐003‐01).

Consent

All patients who enrolled in the registry gave their written informed consent at admission.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: ene70485‐sup‐0001‐supinfo.docx.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.