Cerebral arteriovenous malformations (AVMs) are

Transcription

1 CLINICAL ARTICLE J Neurosurg 126: , 2017 Comparative analysis of arteriovenous malformation grading scales in predicting outcomes after stereotactic radiosurgery Bruce E. Pollock, MD, 1,2 Curtis B. Storlie, PhD, 3 Michael J. Link, MD, 1,4 Scott L. Stafford, MD, 2 Yolanda I. Garces, MD, 2 and Robert L. Foote, MD 2 Departments of 1 Neurological Surgery, 2 Radiation Oncology, 3 Biomedical Statistics and Informatics, and 4 Otorhinolaryngology, Mayo Clinic College of Medicine, Rochester, Minnesota OBJECTIVE Successful stereotactic radiosurgery (SRS) for the treatment of arteriovenous malformations (AVMs) results in nidus obliteration without new neurological deficits related to either intracranial hemorrhage (ICH) or radiationinduced complications (RICs). In this study the authors compared 5 AVM grading scales (Spetzler-Martin grading scale, radiosurgery-based AVM score [RBAS], Heidelberg score, Virginia Radiosurgery AVM Scale [VRAS], and proton radiosurgery AVM scale [PRAS]) at predicting outcomes after SRS. METHODS The study group consisted of 381 patients with sporadic AVMs who underwent Gamma Knife SRS between January 1990 and December 2009; none of the patients underwent prior radiation therapy. The primary end point was AVM obliteration without a decline in modified Rankin Scale (mrs) score (excellent outcome). Comparison of the area under the receiver operating characteristic curve (AUC) and accuracy was performed between the AVM grading scales and the best linear regression model (generalized linear model, elastic net [GLMnet]). RESULTS The median radiological follow-up after initial SRS was 77 months; the median clinical follow-up was 93 months. AVM obliteration was documented in 297 patients (78.0%). Obliteration was 59% at 4 years and 85% at 8 years. Fifty-five patients (14.4%) had a decline in mrs score secondary to RICs (n = 29, 7.6%) or ICH (n = 26, 6.8%). The mrs score declined by 10% at 4 years and 15% at 8 years. Overall, 274 patients (71.9%) had excellent outcomes. There was no difference between the AUC for the GLMnet (0.69 [95% CI ]), RBAS (0.68 [95% CI ]), or PRAS (0.69 [95% CI ]). Pairwise comparison for accuracy showed no difference between the GLMnet and the RBAS (p = 0.08) or PRAS (p = 0.16), but it did show a significant difference between the GLMnet and the Spetzler-Martin grading system (p < 0.001), Heidelberg score (p < 0.001), and the VRAS (p < 0.001). The RBAS and the PRAS were more accurate when compared with the Spetzler-Martin grading scale (p = 0.03 and p = 0.01), Heidelberg score (p = 0.02 and p = 0.02), and VRAS (p = 0.03 and p = 0.02). CONCLUSIONS SRS provides AVM obliteration without functional decline in the majority of treated patients. AVM grading scales having continuous scores (RBAS and PRAS) outperformed integer-based grading systems in the prediction of AVM obliteration without mrs score decline after SRS. KEY WORDS arteriovenous malformation; complication; grading scale; stereotactic radiosurgery; vascular disorders Cerebral arteriovenous malformations (AVMs) are congenital lesions with an incidence of 1.12 to 1.34 per 100,000 person-years. 2,28 Most patients become symptomatic in the 2nd through 4th decades of life with intracranial hemorrhage (ICH), seizures, or headaches. However, the number of incidentally discovered cerebral AVMs continues to rise as more patients undergo MRI. The management of patients diagnosed with cerebral AVMs is based on a number of factors, including presentation, patient age, AVM size, and AVM location. Resection is the preferred treatment for patients with an accessible AVM that has bled, whereas stereotactic radiosurgery (SRS) is generally performed in patients with deeply located AVMs. 11,12,17 The benefit of resection compared with SRS is the immediate elimination of future ICH risk. Although there is little debate about the treatment of patients ABBREVIATIONS AUC = area under the receiver operating characteristic curve; AVM arteriovenous malformation; GLMnet = generalized linear model, elastic net; ICH = intracranial hemorrhage; KPS = Karnofsky Performance Status; mrs = modified Rankin Scale; RBAS = radiosurgery-based AVM score; RIC = radiation-induced complication; SRS = stereotactic radiosurgery; VRAS = Virginia Radiosurgery AVM Scale. SUBMITTED June 5, ACCEPTED November 13, INCLUDE WHEN CITING Published online April 8, 2016; DOI: / JNS AANS, 2017

2 AVM grading scale comparison with ruptured AVMs, the publication of A Randomized trial of Unruptured Brain Arteriovenous Malformations (ARUBA) has questioned whether physicians have been overtreating patients with unruptured AVMs. 16 Consequently, methods that permit the prediction of outcomes for different AVM management techniques are desirable to compare against the natural history of untreated cerebral AVMs. The Spetzler-Martin grading system is the most frequently used scale to predict outcomes after the excision of cerebral AVMs. 27 This grading scale is based on AVM size, AVM location, and pattern of venous drainage. However, a small AVM in the Spetzler-Martin grading scale (< 3 cm) could have a volume anywhere from < 1 ml to approximately 14 ml. In addition, AVMs located in the basal ganglia, thalamus, or brainstem are considered at equal risk for radiation-induced complications (RICs) as AVMs located in or near critical cortical locations, such as the sensorimotor cortex. As a result, Pollock and Flickinger in 2002 proposed a radiosurgery-based AVM score (RBAS) to predict the chance of AVM obliteration without new deficits after SRS. 22 Based on AVM volume, patient age, and AVM location, the RBAS was later modified 21 and has been proven to be a valid instrument to predict outcomes not only after Gamma Knife SRS, but also linear-accelerator SRS, CyberKnife SRS, and proton-based SRS. 3,6,8,9,24,31 Over the past several years, 3 additional AVM grading scales (Heidelberg score, Virginia Radiosurgery AVM Score [VRAS], and proton radiosurgery AVM scale [PRAS]) have been developed to predict outcomes after SRS. 10,15,29 In this report, we compare these 5 AVM grading systems with regard to their ability to predict obliteration without functional decline after radiosurgical management. Methods Patients All aspects of this retrospective study were approved by the institutional review board of the Mayo Clinic, Rochester, Minnesota. A total of 471 AVM patients were identified from our prospective registry as having undergone SRS between 1990 and Patients with hereditary hemorrhagic telangiectasia (n = 11), those with prior radiotherapy or SRS (n = 10), those with Wyburn-Mason syndrome (n = 2), and those with partial AVM coverage (n = 2) were excluded from the study. Ten patients (2.6%) refused research authorization. Patients with less than 2 years of clinical and radiological follow-up were also excluded (n = 55, 11.7%). The characteristics of the 381 patients (169 men and 212 women) in this study are detailed in Table 1. The Spetzler-Martin grading system, modified RBAS, Heidelberg score, VRAS, and PRAS were calculated for each patient (Tables 2 and 3). Radiosurgery Radiosurgical procedures were performed using various versions of the Leksell Gamma Knife (Elekta Instruments). Dose planning was performed using a combination of stereotactic biplanar angiography and either contrast-enhanced CT or MRI. Three hundred fifty-eight TABLE 1. Patient characteristics Factor Value* Age, yrs Median 39.3 Range 3 82 Prior hemorrhage 118 (31.0) Prior resection 30 (7.9) Prior embolization 20 (5.3) Location Hemispheric 297 (78.0) Cerebellar 22 (5.8) Deep 62 (16.3) Eloquent 293 (76.9) Diameter, mm Median 26 Range 6 62 * Values are presented as the number of patients (%) unless indicated otherwise. Deep location is defined as basal ganglia, thalamus, or brainstem. Eloquent location is defined as sensorimotor, language, or visual cortex, hypothalamus, thalamus, brainstem, cerebellar nuclei, or regions directly adjacent to these structures. patients (94.0%) had complete nidus coverage in a single SRS procedure, whereas 23 patients (6.0%) underwent staged-volume SRS. A median of 5 isocenters (range 1 26) were used to cover a prescription isodose volume of 4.8 ml (range ml). The median AVM margin dose was 18 Gy (range Gy); the median maximum dose was 36 Gy (range Gy). In neurologically stable patients, follow-up consisted of MRI and clinical examination at 1, 2, and 3 years after SRS. If follow-up MRI findings were consistent with obliteration, then angiography was requested 2 or more years after SRS to confirm obliteration. Patients with MRI results showing persistent nidus and patients with residual AVM on follow-up angiography 3 or more years after SRS were evaluated for repeat SRS or resection based on their age, clinical condition, and the AVM response from the first SRS procedure. Additional Procedures Sixty-four patients (16.8%) underwent repeat SRS at a median of 42 months (range months) after their initial procedure. A median of 5 isocenters (range 1 12) were used to cover a median prescription isodose volume of 2.4 ml (range ml). The median margin dose was 18.0 Gy (range Gy); the median maximum dose was 36 Gy (range Gy). Twenty-three patients (6.0%) underwent AVM resection after SRS. Ten patients (2.6%) had AVM resection performed at a median of 24 months (range 7 70 months) after ICH. Four patients (1.0%) had AVM resection performed at a median of 48 months (range months) due to residual nidus. One patient (0.3%) had AVM resection performed at 29 months due to symptomatic radiation necrosis. Eight patients (2.1%) had AVM resection performed at a median of 110 months (range months) 853

3 B. E. Pollock et al. TABLE 2. Arteriovenous malformation grading scales Grading Scale & Year SM, <3 cm = I; 3 6 cm = II; >6 cm = III Modified RBAS, Variables Size Vol Patient Age Location Venous Drainage Presentation NA NA Noneloquent = 0; eloquent = 1* Type of Scale (range) No = 0; yes = 1 NA Integer-based (1 5) in yrs deep = 1) NA 0.1 vol in ml 0.02 age 0.5 (not deep = 0, NA NA Continuous HS, <3 cm or 3 cm NA 50 yrs or >50 yrs NA NA NA Integer-based (1 3) VRAS, NA <2 cm 3 = 0; 2 4 cm 3 = 1; >4 cm 3 = 2 NA Noneloquent = 0; eloquent = 1* NA No bleed = 0; bleed = 1 Integer-based (0 4) PRAS, NA 0.26 vol in ml NA 0.7 (not deep = 0, deep = 1) NA NA Continuous HS = Heidelberg score; NA = not applicable; SM = Spetzler-Martin. * Eloquent location is defined as sensorimotor, language, or visual cortex, hypothalamus, thalamus, brain stem, cerebellar nuclei, or regions directly adjacent to these structures. Deep location is defined as basal ganglia, thalamus, or brainstem. Heidelberg score: 1 = (< 3 cm and 50 years); 2 = (either < 3 cm or 50 years); 3 = ( 3 cm and > 50 years). due to symptomatic cyst formation or edema. One patient (0.3%) underwent placement of a ventriculoperitoneal shunt following ICH 50 months after staged-volume SRS. Outcomes The primary outcome was AVM obliteration without a decline in the patient s modified Rankin Scale (mrs) score after SRS. AVMs that exhibited subtotal obliteration on follow-up angiography, defined as persistent arteriovenous shunting without visible nidus, were classified as obliterated. 1,32 Obliteration was defined on MRI as an absence of flow voids on T1- and T2-weighted images. 18,23 Patients undergoing surgery due to ICH or residual AVM (n = 14) and those undergoing repeat SRS at other centers (n = 3) were defined as having incomplete obliteration. Patients functional status before and after SRS from 1990 until 2005 were based on the Karnofsky Performance Status (KPS). 20 The mrs has been used as a measure of patients functional status before and after SRS since The final mrs score for patients undergoing surgery due to ICH or residual AVM (n = 14) and patients undergoing repeat SRS at other centers (n = 3) was based on their preoperative status. The final mrs score of patients having surgery secondary to RICs (n = 9) was based on their postoperative status at last clinical follow-up. Statistical Analysis Data collection for this study was completed in December The median radiological follow-up after patients first SRS was 77 months (range months). The median clinical follow-up was 93 months (range months); 108 patients (28.3%) had more than 10 years of clinical follow-up. Kaplan-Meier analysis was performed to determine the rates of obliteration, ICH, RICs, and decline in mrs score. The best linear regression model (generalized linear model, elastic net [GLMnet]) 34 to predict AVM obliteration without mrs score decline was developed based on the factors used to create the different AVM grading scales (age, prior bleed, deep location, AVM diameter, eloquent location, deep venous drainage, and AVM volume). The area under the receiver operating characteristic curve (AUC) and 95% confidence intervals were calculated for the GLMnet, RBAS, and PRAS. The grading scale AUC estimates are based on validation data, whereas the GLMnet was estimated using these data, and its AUC was calculated using a 10-fold cross-validation to make it compa- TABLE 3. Patient distribution based on the different AVM grading scales Grading Scale Value* SM grade I 44 (11.6) II 115 (30.2) III 158 (41.4) IV V 64 (16.8) Modified radiosurgery-based AVM score Median 1.37 Range HS (49.9) (40.9) 3 35 (9.2) VRAS score 0 27 (7.1) 1 41 (10.8) (32.3) (39.1) 4 41 (10.8) PRAS Median 1.33 Range * Values are presented as the number of patients (%) unless otherwise specified. 854

4 AVM grading scale comparison rable to that of the AVM grading scales. The 95% confidence intervals for AUC were obtained via a nonparametric bootstrap of the validation data, using 1000 bootstrap samples. The AUC could not be calculated accurately for the discrete AVM grading scales (Spetzler-Martin grading scale, Heidelberg score, and VRAS) because of the small number of points available (3 5 points) for AUC estimation. The best attainable overall accuracy (the proportion of cases classified correctly by the rule) for each of the 5 different AVM grading scales was compared with that for the GLMnet using the nonparametric bootstrap with 1000 samples. Statistical significance was defined as p < Analyses were conducted using the open source R software ( Results AVM Obliteration Two hundred forty-eight patients (65.1%) had obliteration confirmed by angiography (n = 169) or MRI (n = 79) after initial SRS. Forty-nine of 64 patients (76.6%) undergoing repeat SRS achieved obliteration (angiography, n = 35; MRI, n = 14) for an overall obliteration rate of 78.0%. The rate of obliteration was 59% at 4 years and 85% at 8 years. Complications Thirty-seven patients (9.7%) had an ICH after SRS. Eleven patients (2.9%) had no deficit, 13 patients (3.4%) developed new deficits (hemiparesis, n = 7; ataxia, n = 2; vegetative state, n = 2; diplopia, n = 1; visual field loss, n = 1), and 13 patients (3.4%) died. The rate of ICH was 8% at 4 years and 11% at 8 years. Twenty-nine patients (7.6%) developed a permanent RIC after SRS. The deficits included hemiparesis (n = 11), visual field loss (n = 5), new seizures (n = 4), aphasia (n = 2), sensory loss (n = 2), ataxia (n = 2), and diplopia (n = 2). One patient (0.3%) died of complications related to the treatment of radiation necrosis. The rate of RIC was 4% at 4 years and 9% at 8 years. Functional Status The patients mrs scores before SRS were 0 (n = 119, 31.2%), 1 2 (n = 252, 66.1%), and 3 (n = 10, 2.6%). After SRS, 20 patients (5.2%) showed improvement in their mrs score, primarily from 1 to 0 as they had resolution of headaches (n = 11), seizures (n = 8), or trigeminal neuralgia (n = 1). The majority of patients scores (n = 306, 80.3%) were unchanged. Fifty-five patients (14.4%) had a decline (median -2) in their mrs score at a median of 29 months (range months) after SRS. The rate of mrs score decline was 10% at 4 years and 15% at 8 years. Analysis of AVM Grading Scales Two hundred seventy-four patients (71.9%) had excellent outcomes after 1 or more SRS procedures. The rate of obliteration without mrs score decline was 57% at 4 years and 82% at 8 years. Figure 1 shows the AUC for the GLMnet relative to the different AVM grading scales (Table 4). There was no difference between the AUC for the GLMnet and the RBAS (p = 0.42) or the PRAS (p = 0.60). There was no difference between the AUC for the RBAS and the PRAS (p = 0.70). Pairwise comparison for accuracy showed no difference between the GLMnet and the RBAS (p = 0.08) or PRAS (p = 0.16), but a significant difference between the GLMnet and the Spetzler-Martin grading system (p < 0.001), Heidelberg score (p < 0.001), and the VRAS (p < 0.001) (Table 5). The RBAS and the PRAS were more accurate when compared with the Spetz ler-martin grading system (p = 0.03 and p = 0.01), Heidelberg score (p = 0.02 and p = 0.02), and VRAS (p = 0.03 and p = 0.02), respectively. Discussion The management of cerebral AVM is of great interest to neurosurgeons despite their relative rarity. Grading scales are commonly used by clinicians to not only predict outcomes for individual patients but also to permit comparative analysis between different series. Established grading scales are based on factors relevant to the treatment method that can be determined accurately and reliably by independent observers. For example, patient age, history of ICH, AVM size, and AVM location will have low interobserver variability and be useful in AVM grading systems, while other potentially important factors such as nidus morphology (diffuse vs compact) and angioarchitecture are more subjective, making their incorporation into any grading system difficult. 19,30,33 Equally important, grading scales cannot be so complicated as to preclude their application on a day-to-day basis. Although integer-based grading scales such as the Spetzler-Martin grading system, Heidelberg score, and VRAS are practical, this approach assumes that the chosen factors have an equal effect on outcomes. Therefore, if one factor has a greater influence on outcomes than the other variables in a grading scale, giving them equal weighting would result in a simple, but possibly inaccurate, scale. Conversely, the RBAS and PRAS do incorporate some basic algebra, but this level of mathematics is typically taught in middle school, and therefore it is not unreasonable to think that physicians with many years of postgraduate education can perform such functions without difficulty. Consequently, grading scales need to represent a balance between accuracy and practicality. Finally, because potentially confounding variables such as smoking, hypertension, immune status, and medication usage may contribute to a patient s outcome after AVM SRS, it is clear that no predictive method can be perfect due to the uncertainty that is inherent to humans and their behaviors. Testing of any clinical grading system requires that the study group does not include patients who were used to develop the grading scale. This approach provides external validity of the grading scale and confirms that its predictions can be generalized to other patient populations. In the current study we found that AVM grading scales developed using regression analysis and having continuous scores (RBAS and PRAS) outperformed integer-based AVM grading scales at predicting obliteration without functional decline after SRS. The RBAS was developed in 2002 as a collaborative effort between the Mayo Clinic and the University of Pittsburgh to predict AVM obliteration without neurological decline after single-session 855

5 B. E. Pollock et al. FIG. 1. Comparison of the GLMnet and the AVM grading scales. Left: Comparison of the GLMnet (solid line, AUC = 0.69) with the RBAS (dotted line, AUC = 0.68) and PRAS (dashed line, AUC = 0.68). Right: Comparison of the GLMnet (solid line, AUC = 0.69) with the Spetzler-Martin grading system (triangles), Heidelberg score (pluses), and VRAS (circles). The AUC could not be calculated accurately for the Spetzler-Martin grading system, Heidelberg score, and VRAS secondary to the small number of points available (3 5 points) for AUC estimation. SRS. 22 The RBAS was simplified in 2008 using location as a 2-tiered variable (basal ganglia, thalamus, or brainstem vs other) rather than the original 3-tiered classification. 21 There was no difference between the original or modified RBAS with regard to outcomes after SRS, and this grading scale has been widely used since its publication. In a separate report, we recently analyzed the factors related to obliteration, post-srs hemorrhage, and RIC using the same patient cohort as the current study. 24 Obliteration was more common in patients with hemispheric or cerebellar AVM and smaller treatment volumes. Bleeding after SRS was more common in older patients, patients with deep AVMs, and larger treatment volumes. RICs were more common with larger treatment volumes. Therefore, all factors used in the RBAS were found to correlate with obliteration (AVM volume and deep location), post- SRS ICH (age, deep location, and AVM volume), or RICs (AVM volume). In 2014, Hattangadi-Gluth et al. published the results of single-fraction proton beam SRS for 248 AVM patients (254 AVMs) undergoing treatment between 1991 and Multivariate analysis found that increasing target volume and deep AVM location were associated with a lower chance of obliteration. This resulted in the creation of the PRAS, which, like the RBAS, is based on a continuous scale rather than an integer-based system. Factors used in the PRAS were predictors of obliteration (AVM volume and deep location), post-srs ICH (AVM volume, deep location), and RIC (AVM volume). It is notable that the PRAS was developed to predict obliteration alone without regard to neurological condition, whereas the RBAS and the current analysis used obliteration without functional decline as the primary end point. Pairwise comparison showed that the integer-based AVM grading scales (Spetzler-Martin grading system, Heidelberg score, and VRAS) were significantly less accurate than the GLMnet, the RBAS, and the PRAS. The Spetzler-Martin grading system has been shown to be predictive of outcomes after AVM resection, but was not designed as a prognostic method for SRS. Only one factor from the Spetzler-Martin grading system, AVM diameter (which directly correlates with AVM volume), is a significant predictor of obliteration, post-srs ICH, or RICs. Similarly, only one variable used in the VRAS (AVM volume) correlated with obliteration, post-srs ICH, and RICs, whereas prior ICH and eloquent location were not predictive factors. The effect of prior ICH on SRS outcomes has been mixed, with some studies showing higher obliteration rates for ruptured AVMs, 14,26 while other studies have found no relationship between prior ICH and obliteration, post-srs ICH, or RICs. 4,5,10 13,15 In addition, patients with critically located hemispheric AVMs seem to have outcomes comparable to patients with noncritically located hemispheric or cerebellar AVMs. Recently, a group at the University of Virginia performed a matched cohort study comparing 134 patients with AVM located in the primary sensorimotor cortex to patients with noneloquent lobar AVM. 7 No difference was noted in the rate of AVM obliteration or deficits between the 2 groups. The authors concluded that eloquent location does not confer the same risk for SRS as it does for resection. Likewise, Bowden et al. reviewed 171 patients undergoing SRS for AVM in the postgeniculate visual pathways. 4 The rate of obliteration was 67% at 4 years; the risk of a new visual deficit was only 5% at 5 years. The Heidelberg score, developed in 2011 by Milker-Zabel and colleagues from the University of Heidelberg, is a prognostic score to predict AVM oblit- TABLE 4. AUC comparison of the continuous AVM grading scales Grading Scale AUC 95% CI Accuracy GLMnet RBAS PRAS SM NA NA 0.72 HS NA NA 0.71 VRAS NA NA

6 AVM grading scale comparison TABLE 5. Pairwise accuracy comparison of the AVM grading scales* GLMnet SM RBAS HS VRAS PRAS GLMnet < <0.001 < SM < RBAS HS < VRAS < PRAS * Values are p values. eration after linear accelerator based SRS. 15 Based on the multivariate analysis of 293 patients undergoing SRS from 1996 to 2006, the Heidelberg score comprises 2 dichotomized variables (patient age and AVM diameter) to create 3 grades or classes: 1) age 50 years and AVM diameter < 3 cm, 2) either age > 50 years or AVM diameter 3 cm, and 3) age > 50 years and AVM diameter 3 cm. An increase of 1 point resulted in a 44% decrease in the chance of obliteration. Although both variables in this score relate to the results of AVM SRS, the discrete nature of this scale provides fewer ROC points and thus less chance to find optimal cutoffs to increase the sensitivity and specificity of this model. Several limitations to our study should be recognized. First, as a retrospective study the number of complications could have been underestimated. However, the clinical and radiological data in our registry were acquired in a prospective fashion over the past 25 years, and the majority of patients underwent follow-up performed at our institution. Second, in our early SRS practice we used the KPS to measure activities of daily living, so it is possible that the conversion of KPS score into mrs score introduced some errors on patients functional capacity, especially for patients with KPS scores of 60 to 80. Nonetheless, we felt that conversion of our data into a form that was more compatible with the cerebrovascular literature was important and worthwhile to facilitate comparisons between different AVM management approaches. Third, despite a follow-up interval longer than than 10 years for more than 25% of the study population, patients remain at risk for mrs score decline due to either ICH from nonobliterated AVM or late RICs. Conclusions SRS provides AVM obliteration without functional decline in the majority of treated patients. AVM grading scales having continuous scores (RBAS and PRAS) outperformed integer-based grading systems in predicting AVM obliteration without mrs score decline after SRS. References 1. Abu-Salma Z, Nataf F, Ghossoub M, Schlienger M, Meder JF, Houdart E, et al: The protective status of subtotal obliteration of arteriovenous malformations after radiosurgery: significance and risk of hemorrhage. Neurosurgery 65: , Al-Shahi R, Bhattacharya JJ, Currie DG, Papanastassiou V, Ritchie V, Roberts RC, et al: Prospective, population-based detection of intracranial vascular malformations in adults: the Scottish Intracranial Vascular Malformation Study (SIVMS). Stroke 34: , Andrade-Souza YM, Zadeh G, Ramani M, Scora D, Tsao MN, Schwartz ML: Testing the radiosurgery-based arteriovenous malformation score and the modified Spetzler-Martin grading system to predict radiosurgical outcome. J Neurosurg 103: , Bowden G, Kano H, Caparosa E, Tonetti D, Niranjan A, Monaco EA III, et al: Stereotactic radiosurgery for arteriovenous malformations of the postgeniculate visual pathway. J Neurosurg 122: , Bowden G, Kano H, Tonetti D, Niranjan A, Flickinger J, Lunsford LD: Stereotactic radiosurgery for arteriovenous malformations of the cerebellum. J Neurosurg 120: , Colombo F, Cavedon C, Casentini L, Francescon P, Causin F, Pinna V: Early results of CyberKnife radiosurgery for arteriovenous malformations. J Neurosurg 111: , Ding D, Yen CP, Xu Z, Starke RM, Sheehan JP: Radiosurgery for primary motor and sensory cortex arteriovenous malformations: outcomes and the effect of eloquent location. Neurosurgery 73: , 824, Franzin A, Snider S, Boari N, Scomazzoni F, Picozzi P, Spatola G, et al: Evaluation of prognostic factors as predictor of AVMS obliteration after Gamma Knife radiosurgery. Acta Neurochir (Wien) 155: , Hattangadi JA, Chapman PH, Bussière MR, Niemierko A, Ogilvy CS, Rowell A, et al: Planned two-fraction proton beam stereotactic radiosurgery for high-risk inoperable cerebral arteriovenous malformations. Int J Radiat Oncol Biol Phys 83: , Hattangadi-Gluth JA, Chapman PH, Kim D, Niemierko A, Bussière MR, Stringham A, et al: Single-fraction proton beam stereotactic radiosurgery for cerebral arteriovenous malformations. Int J Radiat Oncol Biol Phys 89: , Kano H, Kondziolka D, Flickinger JC, Yang HC, Flannery TJ, Niranjan A, et al: Stereotactic radiosurgery for arteriovenous malformations, Part 4: management of basal ganglia and thalamus arteriovenous malformations. J Neurosurg 116:33 43, Kano H, Kondziolka D, Flickinger JC, Yang HC, Flannery TJ, Niranjan A, et al: Stereotactic radiosurgery for arteriovenous malformations, Part 5: management of brainstem arteriovenous malformations. J Neurosurg 116:44 53, Kano H, Lunsford LD, Flickinger JC, Yang HC, Flannery TJ, Awan NR, et al: Stereotactic radiosurgery for arteriovenous malformations, Part 1: management of Spetzler-Martin Grade I and II arteriovenous malformations. J Neurosurg 116:11 20, Liscák R, Vladyka V, Simonová G, Urgosík D, Novotný J Jr, Janousková L, et al: Arteriovenous malformations after Leksell gamma knife radiosurgery: rate of obliteration and complications. Neurosurgery 60: , Milker-Zabel S, Kopp-Schneider A, Wiesbauer H, Schlegel W, Huber P, Debus J, et al: Proposal for a new prognostic score for linac-based radiosurgery in cerebral arteriovenous malformations. Int J Radiat Oncol Biol Phys 83: , Mohr JP, Parides MK, Stapf C, Moquete E, Moy CS, Overbey JR, et al: Medical management with or without interventional therapy for unruptured brain arteriovenous malformations (ARUBA): a multicentre, non-blinded, randomised trial. Lancet 383: , Nagy G, Major O, Rowe JG, Radatz MWR, Hodgson TJ, Coley SC, et al: Stereotactic radiosurgery for arteriovenous malformations located in deep critical regions. Neurosurgery 70: , O Connor TE, Friedman WA: Magnetic resonance imaging assessment of cerebral arteriovenous malformation oblitera- 857

7 B. E. Pollock et al. tion after stereotactic radiosurgery. Neurosurgery 73: , Paúl L, Casasco A, Kusak ME, Martínez N, Rey G, Martínez R: Results for a series of 697 arteriovenous malformations treated by Gamma Knife: influence of angiographic features on the obliteration rate. Neurosurgery 75: , , Pollock BE, Brown RD Jr: Use of the Modified Rankin Scale to assess outcome after arteriovenous malformation radiosurgery. Neurology 67: , Pollock BE, Flickinger JC: Modification of the radiosurgerybased arteriovenous malformation grading system. Neurosurgery 63: , Pollock BE, Flickinger JC: A proposed radiosurgery-based grading system for arteriovenous malformations. J Neurosurg 96:79 85, Pollock BE, Kondziolka D, Flickinger JC, Patel AK, Bissonette DJ, Lunsford LD: Magnetic resonance imaging: an accurate method to evaluate arteriovenous malformations after stereotactic radiosurgery. J Neurosurg 85: , Pollock BE, Link MJ, Stafford SL, Garces YI, Foote RL: Stereotactic radiosurgery for arteriovenous malformations: the effect of treatment period on patient outcomes. Neurosurgery, in press 25. Raffa SJ, Chi YY, Bova FJ, Friedman WA: Validation of the radiosurgery-based arteriovenous malformation score in a large linear accelerator radiosurgery experience. J Neurosurg 111: , Shin M, Maruyama K, Kurita H, Kawamoto S, Tago M, Terahara A, et al: Analysis of nidus obliteration rates after gamma knife surgery for arteriovenous malformations based on long-term follow-up data: the University of Tokyo experience. J Neurosurg 101:18 24, Spetzler RF, Martin NA: A proposed grading system for arteriovenous malformations. J Neurosurg 65: , Stapf C, Mast H, Sciacca RR, Berenstein A, Nelson PK, Gobin YP, et al: The New York Islands AVM Study: design, study progress, and initial results. Stroke 34:e29 e33, Starke RM, Yen CP, Ding D, Sheehan JP: A practical grading scale for predicting outcome after radiosurgery for arteriovenous malformations: analysis of 1012 treated patients. J Neurosurg 119: , Taeshineetanakul P, Krings T, Geibprasert S, Menezes R, Agid R, terbrugge KG, et al: Angioarchitecture determines obliteration rate after radiosurgery in brain arteriovenous malformations. Neurosurgery 71: , Wong GK, Kam MK, Chiu SK, Lam JMK, Leung CHS, Ng DWK, et al: Validation of the modified radiosurgery-based arteriovenous malformation score in a linear accelerator radiosurgery experience in Hong Kong. J Clin Neurosci 19: , Yen CP, Varady P, Sheehan J, Steiner M, Steiner L: Subtotal obliteration of cerebral arteriovenous malformations after Gamma Knife surgery. J Neurosurg 106: , Zipfel GJ, Bradshaw P, Bova FJ, Friedman WA: Do the morphological characteristics of arteriovenous malformations affect the results of radiosurgery? J Neurosurg 101: , Zou H, Hastie T: Regularization and variable selection via the elastic net. J R Stat Soc, B 67: , 2005 Disclosures The authors report no conflict of interest concerning the materials or methods used in this study or the findings specified in this paper. Author Contributions Conception and design: Pollock. Acquisition of data: Pollock, Link, Stafford, Garces, Foote. Analysis and interpretation of data: all authors. Drafting the article: Pollock, Storlie. Critically revising the article: Pollock, Link, Stafford, Garces, Foote. Reviewed submitted version of manuscript: Pollock, Link, Stafford, Garces, Foote. Approved the final version of the manuscript on behalf of all authors: Pollock. Statistical analysis: Pollock. Administrative/ technical/material support: Pollock. Correspondence Bruce E. Pollock, Department of Neurological Surgery, Mayo Clinic, 200 First Street SW, Rochester, MN pollock.bruce@mayo.edu. 858