The initial treatment for large arteriovenous malformations (AVMs) is currently driven by institutional preference without a clear ideal treatment option. Although the short-term outcomes from A Randomized Trial of Unruptured Brain Arteriovenous Malformations (ARUBA) questioned the value of upfront therapy in low-risk, small AVMs, large AVMs have a much different natural history.¹ Treatment is generally favored even in the absence of hemorrhage because these entities exhibit a higher propensity to bleed.²

Primary treatment modalities for AVMs include stereotactic radiosurgery (SRS), embolization, and surgery. Surgery is the gold standard and remains optimal for smaller lesions in more accessible locations of the brain, especially for patients with prior hemorrhage with low risk of complications.^3,4 Single-stage radiosurgery is an alternative option better suited for small lesions in higher-risk locations with surgery. The major drawback with SRS remains associated with a latent period before cure, which is associated with a hemorrhage risk. During the latent period after SRS this risk may be slightly lower than prior to treatment, but it remains a risk factor for subsequent hemorrhage and its associated morbidity and mortality.^5,6 For single-stage SRS, the larger the lesion or deeper the location, the lower the dose that is prescribed to the margin in order to minimize the risk of an adverse radiation effect (ARE). However, lowering the dose prescribed lessens the likelihood of obliteration and thereby increases the risk of subsequent rupture.^6–8

Volume-staged SRS (VS-SRS) is an alternative approach to single-stage SRS. With a VS-SRS approach the nidus is divided into separate target volumes that are irradiated in different sessions, with a 3- to 6-month interval between treatment sessions. The interface between volumes allows more follow-up, and this is not dissimilar to dosimetric differences with use of smaller adjacent isocenters, in that the cumulative volume receiving 12 Gy (V12 Gy) to the normal brain decreases and theoretically reduces toxicity while maintaining reasonable prescription doses.^9,10 Attempting to partially dissociate the inherent relationship between dose and lesion size is the goal. In addition, dose-response within the volume-staged setting appears evident, suggesting that maintaining adequate prescription doses has significant value in optimizing treatment response.^11,12

Moving beyond rates of cure and rupture, the incidents of symptomatic AREs following radiosurgery have not been analyzed in detail for very large malformations within any approach. Deeper areas have also been estimated to have a higher risk of AREs for a given dose, suggesting a much higher risk of AREs in the single-stage setting.⁷ Many of these large AVMs that require a volume-staged approach inevitably cross multiple regions, involve eloquent structures, and incorporate these high-risk deep structures, and therefore would place patients at a theoretically very high risk of AREs. Reports on any of the approaches for treating large AVMs have been limited due to the infrequency of these lesions, yet VS-SRS has shown increasing promise of greater rates of obliteration and lower rates of toxicity. Reports of any in-depth evaluation of toxicity in the volume-staged setting to further optimize outcomes within this treatment paradigm are absent from the medical literature.

In this multiinstitutional retrospective review of prospective volume staging through the International Gamma Knife Research Foundation (IGKRF) we sought to evaluate for predictors of symptomatic AREs following VS-SRS. This previously evaluated cohort demonstrated that treatment response and obliteration improved with dose escalation.¹³ Doses of at least 17.5 Gy were strongly associated with improved rates of partial response, complete obliteration, and cure. The 5- and 10-year cure rates were 33.7% and 76.8% in evaluable patients compared to 6.4% and 20.6% with less than 17 Gy per volume stage, yet it is unclear if this also resulted in increased toxicity. That is the emphasis of this report.

Methods

Treatment

Patients were all treated with a framed approach on the Gamma Knife radiosurgical system including models U, B, C, 4C, and Perfexion based on the technology available at the time of VS-SRS. Multidisciplinary radiosurgery dose planning was then performed by the radiation oncologist in conjunction with a neurosurgeon, neuroradiologist, and medical physicist.

As previously reported:

Two general methods were used to plan VS-SRS. In one approach, a plan was created at the time of the first stage of treatment covering the entire nidus, and then the plan was divided into 2–4 volume stages, functioning as first volume stage and pre-plan(s) for the subsequent stage(s). This approach was utilized in 52.9% of patients with minimal adjustments to the pre-plan on each successive volume stage. The alternative approach, planning each volume stage only on the date of each treatment, was used in 46.3% of patients. This approach allowed for any anatomical changes with associated treatment response and potentially minimized under-dosing of the junction zones between nidus sub-volumes. Regardless of planning method, the interval between successive volume stages was 3–6 months.¹³

Dose summation is not feasible in Leksell GammaPlan software (Elekta AB) and would not allow for changes in architecture that were accommodated for between volume stages. An example of treatment planning is in Fig. 1.

Example of the volume-staged AVM approach, showing a coronal view of a 35.4 cm³ (3.5 × 5.1 × 5.0 cm) nidus (solid red line) treated in 3 stages with 17 Gy per stage. The prescription dose lines for the first and second stages are both in blue, and for the third and final stage the prescription dose line is in yellow, with the 12-Gy isodose line in green. Figure is available in color online only.

In the event of incomplete obliteration, salvage therapy was often considered for patients 3 years after completion of VS-SRS if complete obliteration had not occurred and/or if the patient experienced hemorrhage within the latent period.

Nidal architecture was scored as “diffuse” or “compact” based on the physician’s judgment on pretreatment angiography. No objective or quantifiable metrics of diffuse and compact exist but were previously noted to be significant in predicting response and cure.¹³

Categorization of a clinical worsening attributable to radiotherapy, and therefore an ARE, versus attributable to hemorrhage, and therefore excluded from analysis of AREs, was based on the treating physicians’ and institutions’ review and the timeline of events in the advent of both an ARE and hemorrhage. Radiographic imaging features at the time of an ARE were noted, such as the presence of FLAIR abnormality.

Results

AREs in VS-AVM

A total of 64 patients (25%) experienced a total of 82 ARE events, of which 19 were permanent. This includes incidents of hemiparesis (n = 22), headache (n = 20), seizure onset/exacerbation (n = 15), cognitive decline (n = 8), visual field cut (n = 8), sensory loss (n = 5), and aphasia (n = 4). The onset of the ARE was at a median of 1.28 years (range 0.1–7.32 years).

Location appeared to be associated with AREs. Parietal lobe involvement was associated with AREs (p = 0.005, HR 2.37, 95% CI 1.30–4.31), whereas deep locations were not associated with increased risk of an ARE. Interestingly, size was also associated with increased risk of AREs, especially in the Z (craniocaudal) dimension (p < 0.001, HR 1.90, 95% CI 1.34–2.69).

With ROC analysis, maximal linear dimension was the only appreciable factor associated with predicting AREs, especially within the Z-axis, with an area under the curve > 0.6 (0.638 with standard error of 0.42, and 0.716 with standard error of 0.43; Fig. 2). Of patients experiencing an ARE, a maximal linear dimension in the Z-dimension ≥ 3.28 cm was strongly associated with the ARE, with a 72.5% sensitivity and 58.3% specificity. Only 14 of 116 patients with a Z maximal dimension < 3.28 had AREs. This length was then used in multivariable analysis models involving the entire cohort. Limiting analysis to lesions above this threshold did not alter this analysis—larger AVMs with maximal linear dimensions ≥ 3.28 had a continued increased risk of an ARE with increasing linear dimension in the Z-dimension (p = 0.001).

ROC analysis for AREs. Please note that diagonal segments are produced by ties. GKRS = Gamma Knife radiosurgery. Figure is available in color online only.

Dosimetric and timing factors, including time interval between volume stages, prescription dose, maximum dose, V12 Gy for each stage, and V12 Gy summated for all stages, did not appear significant on univariate analysis (Table 2). Only a trend toward higher V12 Gy in stage 2 and prescription isodose contour per stage for both stage 1 and stage 2 appeared. Also, no accepted models of outcome in AVM were associated with toxicity in evaluating volume-staged patients.

TABLE 2.

Univariate analyses of parameters for association with AREs in patients treated with VS-SRS

Variable	Symptomatic AREs for All AVMs		Symptomatic AREs w/ Deep Involvement Only		Symptomatic & Permanent AREs for All AVMs
	p Value	HR (95% CI)	p Value	HR (95% CI)	p Value	HR (95% CI)
Age; continuous	NS		NS		NS
Sex; male = 0 vs female = 1	NS		0.072	0.397 (0.145–1.087)	NS
History of embo; yes = 1 vs no = 0	NS		NS		0.034	0.561 (0.294–1.073)
Type of AVM; diffuse = 1 vs compact = 0	NS		NS		NS
Locations; yes = 1 vs no = 0
Deep	NS		NS		NS
Temporal	NS		0.014	3.850 (1.321–11.224)	NS
Parietal	0.005	2.370 (1.303–4.309)	NS		NS
Occipital	NS		NS		0.050	4.615 (0.998–21.238)
Total AVM vol, cm3	NS		NS		NS
Maximal dimension
Any, cm; continuous	0.096	1.250 (0.961–1.626)	0.037	1.573 (1.028–2.405)	NS
Z, cm; continuous	<0.001	1.898 (1.339–2.690)	0.005	2.536 (1.303–4.938)	NS
Z; ≥3.28 cm = 1 vs <3.28 cm = 0	<0.001	3.693 (1.862–7.322)	0.004	7.188 (1.871–27.616)	NS
Y, cm; continuous	NS		NS		NS
X, cm; continuous	NS		0.055	1.559 (0.972–2.499)	NS
Time interval btwn vol stages; continuous	NS		NS		NS
No. of vol stages; continuous	NS		NS		0.046	3.931 (1.022–15.114)
Max dose at stage 1, Gy	NS		NS		NS
Margin dose at stage 1, Gy	NS		NS		NS
Prescribed isodose contour at stage 1, %	0.091	0.886 (0.769–1.020)	NS		NS
No. of isocenters at stage 1; continuous	NS		NS		NS
Target vol stage 1, cm3	NS		NS		NS
V12 Gy stage 1, cm3	NS		NS		NS
Max dose at stage 2, Gy	NS		NS		NS
Margin dose at stage 2, Gy	NS		NS		NS
Prescribed isodose contour at stage 2, %	0.080	0.856 (0.720–1.019)	NS		NS
No. of isocenters at stage 2; continuous	NS		NS		NS
Vol-treated stage 2, cm3	0.086	1.032 (0.996–1.070)	0.055	1.068 (0.999–1.142)	NS
V12 Gy stage 2, cm3	NS		NS		NS
V12 Gy summated all stages, cm3	NS		NS		0.037	1.040 (1.002–1.079)
mRBAS; continuous	NS		NS		NS
VRAS; continuous	NS		NS		NS
SM grade; continuous	NS		NS		NS

Max = maximum; NS = not significant.

On multivariable analysis of the entire VS-SRS cohort, parietal lobe and maximal linear dimension in the Z-dimension remained the only important variables in predicting AREs (Table 3).

TABLE 3.

Multivariable analyses of parameters for association with AREs in patients treated with VS-SRS

Variable	Symptomatic AREs for All AVMs		Symptomatic AREs w/ Deep Involvement Only		Symptomatic & Permanent AREs for All AVMs
	p Value	HR (95% CI)	p Value	HR (95% CI)	p Value	HR (95% CI)
Sex	NS		0.060	(0.081–1.052)	NS
Locations; yes = 1 vs no = 0
Parietal	0.015	2.299 (1.179–4.485)	NS		NS
Temporal	NS		0.033	4.173 (1.119–15.565)	NS
Occipital	NS		NS		0.030	0.269 (0.082–0.880)
Maximal dimension
Z, cm; continuous	Excluded		0.010	2.663 (1.265–5.609)	NS
Z; ≥3.28 cm = 1 vs <3.28 cm = 0	<0.001	4.116 (2.033–8.333)	Excluded		Excluded
Isodose stage 1, %	0.076	0.877 (0.760–1.014)	NS		NS
No. of vol stages	NS		NS		0.057	4.464 (0.957–20.827)

Variables that were not significant were removed from the multivariable model by backward stepwise selection.

Discussion

Large AVMs are technically challenging lesions to treat. After SRS, the risk of morbidity related to subsequent bleeding persists within the latent period, and a more arduous natural history remains with or without intervention.⁶ There have been analyses assessing various approaches for these initially nonsurgically treated lesions, such as a volume-staged approach in which the nidus is treated piecemeal, and a dose-staged approach in which the entire nidus is encompassed with each treatment, either with traditional fractionation schedules or repeat treatments spaced months to years apart. However, there have been no previous analyses of the predictors of AREs associated with a volume-staged approach for very large AVMs. Given that there is no established threshold for what constitutes large, we would argue that any lesion of a size that would require a volume-staged approach to maintain adequate marginal dose would constitute a large AVM for radiosurgical purposes. The threshold for increased risk of adverse events from a single session at doses of 17–18 Gy is approximately 7 cm³. However, from the lesions in this report, this risk appears to be mostly driven by maximal linear Z-dimension.

In our analysis previously reported from this same cohort, we suggested that clear responses were found with continued dose escalation with a volume-staged approach.¹³ This appeared especially true for large AVMs with a diffuse nidus architecture. In contrast, patients with diffuse nidus architecture did not exhibit different rates of AREs. Dose and volume were not predictors of AREs within this report. This strongly suggests that a volume-staged approach does successfully dissociate dose and volume to a degree, and that diffuse lesions will disproportionately benefit from moderate dose escalation because they will be more likely to have large volume reductions enabling salvage therapy or obliteration without increased risk of an ARE.

Even more surprising, patients with deep involvement did not have an increased risk of an adverse event. Based on the previously reported estimates of neurological deficit from AVMs treated with single-stage SRS, a 10-cm³ malformation involving the basal ganglia, brainstem, corpus callosum, or thalamus alone would have a > 50% estimated risk of a permanent symptomatic ARE.⁷ Here there was a < 30% risk of symptomatic, transient, or permanent AREs in patients with and without deep involvement, and the overall risk of a permanent ARE was < 10%.

The role of patient selection in this cohort did probably limit the view that all patients with any deep invasion are not at an increased risk of an ARE. For instance, only 3 patients had brainstem involvement and therefore extrapolation to all deep locations may not be possible, but the rates of AREs in patients with thalamus, corpus callosum, and/or basal ganglia involvement appear to be low with a volume-staged approach. Lesions treated with more volume stages were maintained in the model for permanent AREs, and there was a trend toward increased risk of an ARE with larger number of treatment stages; perhaps progressive volume staging with smaller volume per stage may have diminishing returns with regard to minimizing the risk of persistent toxicity.

The emphasis on parietal or temporal involvement and Z-dimension length may suggest that other factors may be at play. The finding of a similar cutoff point for an ARE—3.28 cm compared to 3 cm for SM grading, with maximal linear dimension—suggests that some of these toxicities may mimic surgery-related phenomena rather than classically accepted radiobiological phenomena. These large lesions must either displace or contain eloquent cortex given their sheer extent in almost all cases. Reorganization of eloquent cortex, especially for language and motor functions, is a phenomenon that has been described within AVMs.^14,15 Hypothetically, limitations in reorganization in the parietal lobe for more superficial lesions and in the temporal lobe for deeper lesions may have resulted in an increased risk of overall and permanent AREs in this cohort. An increased involvement in the craniocaudal dimension may further limit that reorganization. There may be further architectural and predisposing features that accentuate or augment this reorganization, including but not limited to a threshold of size that may have interplay with or total independence from normal tissue factors in a response to SRS that is beyond theoretical genomic and dosimetric factors. Furthermore, the lack of a radiographic correlate in many instances of clinically apparent AREs again emphasizes that standard and accepted rules of thumb may not extrapolate to these extreme scenarios.

Although there are limitations in this analysis with regard to evaluation of complicated dosimetric features and it is complicated to evaluate any volume-staged approach given the competing risks, first and foremost, the results were consistent across subsets of patients and stages evaluated. The redundancy of the results is reassuring in that within this cohort, the predictors of AREs were location and craniocaudal extent of the lesion. Further limitations include the study’s retrospective nature and lack of central review of degrees of obliteration or established threshold for diffuse and compact architecture. This report exceeds the scale and breadth of evaluable follow-up from any previous report, but some reports have noted a continued rise in incidence of radiographic changes over time even in the presence of a low incidence of symptomatic AREs. This suggests that perhaps more radiographic changes and possibly symptomatic events will occur with further follow-up.¹⁶

There was no clear modifiable risk feature to optimize radiotherapy, yet further advancements in radiation optimization through differing dosing schemes are interesting. Notably, there appear to be no associations between AREs and a “shot-packing” approach, which one would associate with an increased number of isocenters used per volume stage.¹⁷ Although comparisons between volume-staged and dose-staged approaches appeared to suggest that VS-SRS may have a lower risk of symptomatic AREs and higher rates of obliteration, there has yet to be an approach combining these two methods in which different portions of the tumor or lesion receive hypofractionated^3–5 treatments.¹⁸ There are potential phenomena that are not well understood regarding incidence of AREs and hemorrhage within this cohort. Unquantifiable overlap of radiosurgical dose may be causing functional damage that is not appreciated on imaging and possible vascular damage causing secondary deficits, possibly including but not limited to venous occlusion, which may be different in different eras of treatment. Combining the two approaches may minimize the biological effects of theoretical overlap. Extrapolating this volume-staged technique to other large benign tumors may also further reduce the risk of AREs compared to a dose-staged fractionated approach if dissociation of dose and volume truly has occurred and should be investigated.

Conclusions

The incidence of symptomatic AREs following a VS-SRS approach for a very large AVM is associated with parietal involvement and increased size in the craniocaudal dimension.

Disclosures

Dr. Liščák is a consultant for Elekta AB. Dr. Lunsford is a consultant for the Insightec DSMB, and has direct stock ownership in Elekta AB.

Author Contributions

Conception and design: Seymour. Acquisition of data: Seymour, Chan, Kano, Lehocky, Jacobs, Chytka, Liščák, CC Lee, Yang, Ding, Sheehan, Feliciano, Rodriquez-Mercado, Chiang, Hess, McShane, Vasas, Sneed. Analysis and interpretation of data: Seymour, Ye, Ding. Drafting the article: Seymour. Critically revising the article: Seymour, Chan, Grills, Lunsford, Chytka, Liščák, CC Lee, Yang, Sheehan, Feliciano, Rodriquez-Mercado, Chiang, McShane, JYK Lee, Kaufmann, Sneed. Reviewed submitted version of manuscript: Seymour, Chan, McDermott, Grills, Kano, Lunsford, Chytka, Liščák, CC Lee, Sheehan, Feliciano, Rodriquez-Mercado, Chiang, Sommaruga, McShane, JYK Lee, Kaufmann, Sneed. Approved the final version of the manuscript on behalf of all authors: Seymour. Statistical analysis: McDermott, Ye, Kano, Sommaruga. Study supervision: Seymour.

Supplemental Information