Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Neurosurgery. Author manuscript; available in PMC 2011 April 1.
Published in final edited form as:
PMCID: PMC2847513

A Supplementary Grading Scale for Selecting Patients with Brain Arteriovenous Malformations for Surgery

Michael T. Lawton, MD,1,5 Helen Kim, PhD,2,3,5 Charles E. McCulloch, PhD,3 Bahar Mikhak, MS, MPH,2,5 and William L. Young, MD1,2,4,5, for the UCSF BAVM Study Project



Patient age, hemorrhagic presentation, nidal diffuseness, and deep perforating artery supply are important factors when selecting patients with brain arteriovenous malformations for surgery. We hypothesized that these factors outside of the Spetzler-Martin grading system could be combined into a simple, supplementary grading system that would accurately predict neurological outcome and refine patient selection.


A consecutive, single-surgeon series of 300 patients with AVMs treated microsurgically was analyzed in terms of change between preoperative and final postoperative Modified Rankin Scale scores. Three different multivariable logistic models (full, Spetzler-Martin, and supplementary models) were constructed to test the association of combined predictor variables with the change in MRS score. A simplified supplementary grading system was developed from the data which combined age, hemorrhagic presentation, and diffuseness in a manner analogous to the Spetzler-Martin grading system, with points assigned according to each variable and added together for a supplementary AVM grade.


Predictive accuracy was highest for the full multivariable model (receiver operating characteristic curve area, 0.78), followed by the supplementary model (0.73), and least for the Spetzler-Martin model (0.66). Predictive accuracy of the simplified supplementary grade was significantly better than that of the Spetzler-Martin grade (P=0.042), with ROC curve areas of 0.73 and 0.65, respectively. The predictive accuracy of the supplementary grade was only slightly less than a full point score with all 7 weighted variables (P=0.364), with areas under the ROC curve of 0.73 and 0.75, respectively.


This new AVM grading system supplements rather than replaces the well established Spetzler-Martin grading system, and is a better predictor of neurological outcomes after AVM surgery. The supplementary grading scale has high predictive accuracy on its own and stratifies surgical risk more evenly. Supplementary grades can confirm risk predicted by the Spetzler-Martin grade, or in cases of mismatched grades, may alter management decisions. The supplementary grading system is easily applicable at the bedside, where it is intended to improve preoperative risk prediction and patient selection for surgery.

Keywords: Arteriovenous malformation, Microsurgery, Patient selection, Prediction models, Spetzler-Martin grading system, Supplementary grading system


Judicious patient selection is essential to avoiding surgical complications and poor neurological outcomes with microsurgical resection of brain arteriovenous malformations (AVM). The combination of nidus size, deep venous drainage, and eloquence of adjacent brain that comprises the Spetzler-Martin grading scale provides a preliminary assessment of surgical risks 1, with low-grade AVMs (grade I – III) having acceptably low morbidity rates and high-grade AVMs (grade IV – V) having unacceptably high morbidity and mortality rates. As helpful as this simple grading scale is, it is crude at best and the recommendation to operate may be strengthened by considering additional risk factors.

Some of these factors are embedded within the Spetzler-Martin grading scale, like grade III subtype and functional eloquence. The dividing line between operability and non-operability does not run cleanly between grades III and IV, but rather between subtypes of the third grade. Our experience has demonstrated that medium-sized AVMs (3-6 cm diameter) in eloquent locations have morbidity rates that are higher than expected for grade III lesions (more like grade IV AVMs), while small-sized AVMs (< 3 cm diameter) with deep venous drainage in eloquent locations have morbidity rates that are lower than expected (more like grade II AVMs) 2. Similarly, structural anatomy defined by the Spetzler-Martin system as eloquent does not always equate with functional anatomy, because brain inhabited by an AVM often relocates functions that lie too close to the nidus. We have found that preoperative imaging with functional MRI (fMRI) more precisely localizes neurological function and allows us to individualize a patient's eloquence 3.

Other factors important to patient selection that are not part of the Spetzler-Martin grading scale include patient presentation, age, deep perforating artery supply, and diffuseness. Presentation with hemorrhage not only indicates AVMs with high risk of re-hemorrhage 4-6, but also facilitates surgery 7. Hematomas help separate AVMs from adjacent brain; evacuation of hematoma creates working space around the AVM that can minimize transgression of normal brain or access a deep nidus that might otherwise have been unreachable; and hemorrhage can obliterate some of the AVMs arterial supply to reduce its flow during resection. AVM hemorrhage and microsurgery can injure brain tissue, but young age and plasticity can enhance a patient's ability to recover neurological function 8. Compact AVMs with tightly woven arteries and veins often have distinct borders that separate cleanly from the adjacent brain, whereas diffuse AVMs with ragged borders and intermixed brain force the neurosurgeon to establish dissection planes that can extend into normal brain 9. Deep perforating arteries are thin, fragile, and difficult to occlude with cautery. Bleeding during surgery can escape into deep white matter tracts and cause significant deficits. All of these factors, hemorrhagic presentation, young age, compactness, and absence of deep perforator supply have been identified as predictors of good outcomes after microsurgical resection 7, 9.

We hypothesized that these factors outside the Spetzler-Martin grading system could be combined into a supplementary grading system that would accurately predict neurological outcome. We analyzed a consecutive surgical series of 300 patients to compare the predictive accuracy using the Spetzler-Martin scale and a new supplementary scale. We propose that this simple supplementary grading system can be used to improve and refine patient selection for AVM surgery.


Study Population

The study was approved by the University of California, San Francisco Committee on Human Research and conducted in compliance with Health Insurance Portability and Accountability Act (HIPAA) regulations. During an 11-year period, 392 patients underwent microsurgical resection of their AVMs by a single neurosurgeon (MTL). Since 2000, all patients with brain AVMs evaluated or treated at UCSF were enrolled prospectively in the UCSF BAVM registry. The study sample consisted of patients with AVMs treated with microsurgical resection from 2000 through 2007, of which 300 had complete demographic, anatomical, and clinical data.

Study Variables

Patient demographics included age (decade), gender, and race/ethnicity. Nidus size (diameter in cm), venous drainage (superficial, deep, or both), and eloquence were determined from preoperative angiograms, computed tomography (CT) scans, and magnetic resonance images (MRI) for each AVM, according to the Spetzler-Martin grading system. Presence of deep perforating arterial feeders was also determined from preoperative angiograms and included lateral lenticulostriate arteries from the M1 segment of middle cerebral artery, medial lenticulostriates from the A1 segment of anterior cerebral artery, anterior and posterior choroidal arteries, thalamoperforators from the posterior communicating artery and P1 segment of posterior cerebral artery, and brain stem perforators from the basilar trunk and vertebral arteries. Compact or diffuse AVM morphology was determined from preoperative angiograms, with computed tomography scans and magnetic resonance images used to identify intervening brain parenchyma within the nidus. Hemorrhagic presentation was defined as radiographic evidence of hemorrhage on CT or MRI, regardless of signs or symptoms.

The Modified Rankin Scale (MRS) was used to grade outcomes 10. A nurse clinician, under the supervision of a neurologist, performed MRS assessments at presentation, preoperatively, 3 months postoperatively, and up to 2 years postoperatively. Follow-up information was obtained during routine clinic visits or telephone interviews. Outcomes were analyzed in terms of change between preoperative and final postoperative MRS scores (MRS final – MRS preoperative). A final MRS score of 0 to 2 was considered a good outcome, and a final MRS > 2 was considered a poor outcome. Improvement was defined as a change in MRS score of less than or equal to 0 (improved or unchanged), and deterioration was defined as a change in MRS score of greater than 0 (worse or dead) 7.

Statistical Analysis

Data were analyzed using Intercooled STATA version 9 (STATA Corp, College Station, TX). Descriptive statistics included t tests and χ2 tests for comparison between change in MRS score and continuous and dichotomous variables, respectively. Univariable logistic regression analysis tested the association of each predictor with the change in MRS score dichotomized into deterioration (MRS change of 1-6) vs. no change or improvement (MRS change ≤0). Three different multivariable logistic models were constructed to test the association of combined predictor variables with the change in MRS score:

  • Full Model: logit (MRS change) = β0 + β1(unruptured presentation) + β2(age decade) + β3(eloquent location) + β4(size) + β5(diffuse) + β6(deep venous drainage) + β7(deep perforating artery) + β8(logtime).
  • Spetzler-Martin Score Model: logit (MRS change) = β0 + β1(eloquent location) + β2(size) + β3(deep venous drainage) + β4(logtime)
  • Supplementary Score Model: logit (MRS change) = β0 + β1(unruptured presentation) + β2(age decade) + β3(diffuse) + β4(deep perforating artery) + β5(logtime)

The full model included all pre-specified predictor variables. The Spetzler-Martin score model included the components of this grading system 1. The supplementary score model included nonhemorrhagic presentation, age, diffuseness, and deep perforating artery supply. All multivariable models included adjustment for the duration of follow-up (log transformed time), which influenced final MRS assessments. Statistical significance was set at P<0.05.

Predictive accuracy is the ability of a grading system to correctly classify patients into those who will be worse after surgery and those who will not. Receiver operating characteristic (ROC) analyses were performed after each multivariable logistic regression model, and the area under the ROC curves were compared for accuracy in predicting change in MRS score between these four models. An area under the ROC curve of 1.0 indicates perfect discrimination, whereas an area of 0.5 indicates no discrimination. Generally, an area under the ROC curve ≥ 0.70 is considered a clinically useful predictive model 11.

Supplementary AVM Grading System

A point scoring system was developed from the data that used the beta coefficients from the multivariable logistic regression models to weight the clinical and AVM characteristic values for each patient. These points were added together to get a total point score for each patient (Table 1). A simplified supplementary grading system was developed from the data, which included significant clinical and AVM characteristics not already expressed in the Spetzler-Martin scoring system. In a manner analogous to the Spetzler-Martin scoring system, points were assigned according to these variables and added together for a supplementary AVM grade (Table 1).

Table 1
Grading systems for brain AVMs. The full statistical model combines all 7 variables with points assigned, weighted, and added to generate a final score. For example, a 57 year-old man presents with an unruptured, 2 cm diameter AVM with superficial venous ...

The total point score, supplementary grade score, and Spetzler-Martin grade score were then analyzed as predictor variables in separate logistic regression models, adjusting again for the duration of follow-up, and ROC analyses were repeated. Areas under the ROC curves were tested for equality using a χ2 test.

Because we used all the data to build our prediction models, which could result in overly optimistic predictions, we performed a 10-fold cross validation 12. In this approach, the dataset is randomly split into 10 groups; the model is then constructed on the first 9 groups and applied to the remaining group. The model building and validation process is repeated 10 times with each sample used only once as the validation set, i.e., no patient is used both to develop and test the model. The area under the ROC curve is then estimated using data from the 10 validation sets.


Patient Demographics and AVM Characteristics

Patient demographics and AVM charateristics are summarized in Table 2. A total of 194 patients (65%) had embolization, radiosurgery, or both prior to microsurgical resection.

Table 2
Patient and brain AVM characteristics by change in preoperative and postoperative Modified Rankin Score (MRS).

Good outcomes after AVM resection were observed in 239 of the 300 patients (80%, MRS scores 0 – 2, Figure 1). Based on changes in MRS score, 227 patients (76%) were unchanged or improved, 55 patients (18%) were worse, and 18 patients (6%) died. Of the patients that were worse after treatment, the largest groups were those presenting with preoperative MRS scores of 0 (35 patients, 48%) or 1 (14 patients, 19%). Six of the patients that died presented in coma (preoperative MRS score 5, 8%) and failed to improve with aggressive management.

Figure 1
Neurological outcomes. Pre- and postoperative Modified Rankin Scale scores are shown.

Patients who did worse after treatment were older, male, and had a greater frequency of unruptured presentation compared to patients who improved or remained unchanged (Table 2). In addition, patients who did worse had AVMs that were larger, diffuse, in eloquent location, and had deep venous drainage. There was no difference in the proportion of patients with deep perforating artery supply (P=0.614).

Logistic Regression Analysis

Univariable logistic regression analysis identified age (P<0.001), AVM size (P=0.001), unruptured presentation (P=0.005), and diffuse nidus (P=0.016) as significant predictors of worsened MRS score (Table 3). Eloquence (P=0.058) and deep venous drainage (P=0.085) were borderline significant, whereas deep perforating artery supply was not associated with worsened MRS score (P=0.614).

Table 3
Univariable Analysis

Multivariable logistic regression analysis using the full model containing all variables identified unruptured presentation (OR=2.7), age (OR=1.4), and deep venous drainage (OR=2.0) as independent and significant predictors of worsened MRS score (P<0.05, Table 4). In the Spetzler-Martin score model, AVM size (OR=1.3) was the only significant predictor of worsened MRS score (Table 4). In the supplementary score model, age (OR=1.4), unruptured presentation (OR=2.3), and diffuseness (OR=2.3) were all independent predictors of worsened outcome (P<0.05, Table 4).

Table 4
Multivariable logistic regression analysis comparing three different models: full, Spetzler-Martin score, and supplementary score. OR = odds ratio; CI = confidence interval.

ROC Analysis

The area under the ROC curve, indicating the predictive accuracy of each model, was highest for the full multivariable model (0.78), followed by the supplementary score model (0.73, Table 4). The Spetzler-Martin score model had the lowest area under the ROC curve (0.66). The ROC curve areas were significantly different between the three models (P<0.001).

Patient outcome is predicted better using all 7 variables in the full multivariable model than with the 3 variables in the Spetzler-Martin scale, with increased specificity for the same sensitivity and increased sensitivity for the same specificity. For example, the sensitivity and specificity of predicting clinical deterioration is 52% and 63%, respectively, for an AVM with a Spetzler-Martin Grade III or higher. With the same sensitivity, the specificity increases to 81% when combining the Spetzler-Martin variables with other supplementary variables. Similarly, for the same specificity of the Spetzler-Martin score, the sensitivity increases to 75% when using all variables.

Supplementary AVM Grading Scale

Next, we constructed a supplementary grading system, which included only statistically significant variables from Model 3 (Table 4) not already expressed in the Spetzler-Martin scoring system. Therefore, points were assigned for patient age, presentation, and AVM diffuseness, analogous to the Spetzler-Martin scoring system (Table 1). These points were added together for a supplementary AVM grade that ranged from 1 to 5. Supplementary AVM grades were assigned immediately before surgical treatment. There was only one patient whose supplementary grade changed during treatment due to an intraprocedural hemorrhage during embolization of a previously unruptured AVM. Supplementary AVM grades were normally distributed, without the selection bias against higher grades seen in the distribution of Spetzler-Martin grades (Figure 2).

Figure 2
Distribution of Spetzler-Martin grades and supplementary grades, as applied to 300 patients.

Neurological outcomes by Spetzler-Martin grade and the new supplementary grade are shown in Table 5. Adding the Spetzler-Martin grade and the supplementary grade for each patient yielded a combined grade ranging from 1 to 10. A greater percentage of patients had improved neurological outcomes with decreasing combined grade, with stratification into low-risk (grades 1 – 3), moderate-risk (grades (4 – 6), and high-risk groups (grades 7 – 10) (Table 5).

Table 5
Patient outcomes according to Spetzler-Martin grade, supplementary grade scale, and combined grades.

The predictive accuracy of the new supplementary grade was significantly better than that of the Spetzler-Martin grade (P=0.042), with areas under the ROC curve of 0.73 (95% CI=0.67-0.79) vs. 0.65 (95% CI=0.58-0.72), respectively (Figure 3). The predictive accuracy of the supplementary grade was only slightly less than the combined score (P=0.364), with areas under the ROC curve of 0.73 (95% CI=0.67-0.79) and 0.75 (95% CI=0.69-0.81), respectively.

Figure 3
Receiver operating characteristic (ROC) analyses for the weighted point scores using all 7 variables (blue curve), the Spetzler-Martin grading system (green curve), and the supplementary grading system (red curve) (reference line shown in teal). The predictive ...

To evaluate whether the predictive accuracy of our supplementary grade was overly optimistic, we performed a 10-fold cross validation of the data so that no patient was used to both build and test the model. The 10-fold cross validation resulted in similar estimates with an area under the ROC curve of 0.72 compared to 0.73, suggesting that the model was not overly optimistic.


Supplementing the Spetzler-Martin Grading Scale

Our analysis of 300 patients undergoing microsurgical AVM resection demonstrated that the Spetzler-Martin grading system is a crude predictor of neurological outcomes (ROC area < 0.7), as defined by changes in MRS scores. This analysis also demonstrated that a supplementary grading system with three other variables (age, hemorrhagic presentation, and diffuseness) was a better predictor of neurological outcomes, and that the combination of all six variables was better still. This supplementary grading system is simple, analogous to the Spetzler-Martin grading system, and easily applicable at the bedside (Table 1).

Our previous research demonstrated that deep perforating artery supply was associated with increased surgical risk 9, but the current study did not confirm this association. These thin-walled, friable feeders are difficult to coagulate and reside along the deep margin of the nidus, where bleeding can result in intraparenchymal bleeding and dissection into eloquent white matter. Therefore, the lack of a significant association between perforator supply and outcome disagreed with our clinical experience. We attributed the lack of significance of deep perforating artery supply to the differences in patient population and methodology between studies. However, diffuseness was significantly associated with increased surgical risk, and deep perforators often contribute to the ragged borders of diffuse AVMs. Diffuseness may therefore indirectly address deep perforating artery supply in the supplementary grading scale.

Compact AVMs are easily recognized as a tight tangle of vessels on angiography, with little brain tissue within the nidus on MR images, and well defined margins between brain and AVM. However, diffuseness remains somewhat vague. Margins are ragged and irregular, as if the arteriovenous tangle was loosened and unraveled, with intervening brain parenchyma intermixed within the nidus. Diffuseness is easy to recognize on angiography (Figure 4), but difficult to quantify. We developed methodology to quantify diffuseness using computer-generated outlines of AVMs on angiograms, contour plots with varying image intensities, and calculations of nidus area-intensity profiles 9, but this methodology is far too complex for quick, bedside application. Therefore, the determination of diffuseness in the supplementary grading system is qualitative. Nonetheless, diffuseness belongs in the supplementary grading system because this angioarchitecture is critically important to surgical risk assessment. Compact AVMs have distinct dissection planes with clear separation between nidus and brain tissue, whereas diffuse AVMs have obscure planes that can draw the dissection too close to the nidus, resulting in hemorrhagic complications, or can force the dissection away from the nidus, compromising interspersed brain. Experienced neurosurgeons trained to analyze dissection planes on preoperative angiograms can identify diffuse AVMs reliably. Spears et al. examined interobserver variability in grading 233 brain AVMs and found substantial agreement when separate clinicians determined diffuseness (Kappa value = 0.67) 11.

Figure 4
Examples of compact and diffuse AVMs. Case 1 demonstrates a compact nidus in the right parieto-occipital region with distinct borders and a tight tangle of arteries and veins (internal carotid artery angiogram, (A) lateral and (B) anteroposterior views). ...

Age categories in the supplementary grading system are somewhat arbitrary. The 20-year cut-off was intended to capture pediatric patients who fare better than adults after AVM resection due to increased tolerance to surgery, increased neurological recovery, and/or neural plasticity 8. The 40-year cut-off was intended to separate adults into those with and without other medical co-morbidities. As with size categories in the Spetzler-Martin grading system, minor changes in the cut-offs in age categories did not diminish the predictive accuracy or utility of the supplementary scale.

Like the Spetzler-Martin grading system, the supplementary grading system is a tool to assess the risk of AVM resection. It is applied when analyzing a particular patient's AVM and formulating a management recommendation. However, unlike the Spetzler-Martin score, the supplementary score can change with other treatments or with time. After radiosurgery, an AVM can lose its diffuseness, it can rupture during the latency period, and a pediatric patient can transition to an adult patient. In this example, one would subtract 2 points and add 1 point to the supplementary score. Similarly, embolization can cause hemorrhage in a previously unruptured AVM and one would subtract 1 point (as in one of our patients). We did not encounter cases where embolization changed diffuseness in the supplementary score. In these examples, the supplementary score decreases to reflect the effects of previous treatment and thereby encourage surgical intervention. Therefore, the supplementary score is dynamic and must be re-evaluated as the patient's clinical circumstances change, whereas the Spetzler-Martin score is determined at initial diagnosis and carried throughout subsequent treatments or clinical events.

Other AVM Grading Systems

Other AVM grading systems have been proposed to improve surgical risk prediction and patient selection since the introduction of the Spetzler-Martin grading system in 1986 1. Tamaki et al. 13 assigned points for size (small or large), number of feeding artery systems (1-2 or ≥3), and location (superficial or deep), stratifying AVMs into five grades ranging from 0 to 4 that correlated with surgical difficulty, as measured by rate of total AVM excision and patients' Karnofsky scale score 14. These authors identified age as a significant predictor of outcome, but did not include it in the grading system. Their grading system was too similar to the Spetzler-Martin system and failed to gain acceptance.

After finding that neither AVM size nor venous drainage pattern influenced outcome in their experience, Hollerhage et al. proposed a grading system based on 5 territories of feeding artery supply (ACA, MCA, PCA, Rolandic branches, and ACoA shunt flow) 15. Their grading system was among the first to incorporate a clinical variable in addition to these anatomical variables, assigning 1 point for Hunt & Hess grades I – II and 2 points for Hunt & Hess grades III – V. The Hunt & Hess grade contained within this AVM grading system may be a surrogate for hemorrhagic presentation, but Hunt and Hess 16 designed their scale for aneurysm patients with subarachnoid hemorrhage and its application to AVM patients was awkward. Despite the possibility of an AVM grade as high as 7, grades ranged from 1 to 4 in this study and correlated with Glasgow Outcome Scale (GOS) scores 17. However, by measuring surgical results by final GOS rather than changes in GOS scores, this grading scale failed to recognize the surgical advantages associated with hemorrhagic presentation.

Perhaps the most comprehensive grading system for AVMs was proposed by Pertuiset et al. 18. In addition to angiographic factors like AVM location and feeding artery supply, this system analyzed number of AVM sectors, and caliber and straightening of feeding arteries. Hemodynamic factors included nidus volume, cerebral steal, and circulatory velocity of radiolabeled red blood cells. Notably, this system included age and previous hemorrhage in its clinical variables. Using elaborate tables, each variable was coded and these codes were added to generate operability scores ranging between 3 and 69, with AVMs scores less than 30 considered operable. The authors concluded that “the score system is a little too complicated …. It will not take more than 15 minutes to get the sum of the code numbers but it is absolutely necessary to choose a team of scrutators with more than one neurosurgeon; it is also necessary to add to the scrutators a biophysicist.” This system contained two of the variables in our supplementary grading system, but was too impractical for clinical use.

The University of Toronto Brain AVM Study Group developed a discriminative prediction model of neurological outcomes associated with AVM resection that recognized nidus diffuseness as a critical predictor variable, weighted predictor variables according to their statistical significance, and used MRS to measure outcomes 11. The Toronto model incorporated just 3 variables, weighted them with rounded odds ratios (eloquence = 4, diffuseness = 3, and deep venous drainage = 2), and added points to form a 9-point stratified risk score. Discrimination of this model for predicting permanent disabling neurological outcomes was high (area under the ROC curve, 0.79), and better than the Spetzler-Martin scale (area under the ROC curve, 0.69). Our full model point score was derived with a similar statistical approach, incorporating 7 variables weighted according to the beta coefficients from multivariable logistic regression models. The predictive accuracy of this grading system was high also (area under the ROC curve, 0.78), but such a grading system is too cumbersome to be practical. The weighted grading system of the Toronto group is much simpler, but it has not been widely applied in the years since its publication. In addition, it competes with the Spetzler-Martin grading system, re-affiliating eloquence and venous drainage with the newer scale. Our supplementary grading system with its own unique variables remains separate from the Spetzler-Martin scale and avoids this problem.

We envisioned a grading system that would supplement rather than replace the already entrenched Spetzler-Martin grading system. Simplicity is a critical aspect of a popular grading scale, and our supplementary grading scale is designed with this in mind. In addition, the two grading systems are analogous in their structure, which we hope will make the supplementary grading scale memorable. The supplementary grading scale has high predictive accuracy on its own (area under the ROC curve, 0.73 vs. 0.65 for the Spetzler-Martin grading system), and stratified surgical risk more evenly in our series (Table 5). Therefore, the supplementary grade can be considered separately, or it can be combined with the Spetzler-Martin grade. Patients with supplementary grades ≤ 3 or combined grades ≤ 6 stratify into low- or moderate-risk groups that predict acceptably low surgical morbidity.

Application of Supplementary AVM Grading System

Clinical decisions begin with an analysis of nidus size, venous drainage, and location. An analysis of supplementary factors can impact a management decision by confirming the risk predicted by the Spetzler-Martin grade. For example, an AVM with a low Spetzler-Martin grade (grade I – III) may be favorable for microsurgical resection, and a low supplementary grade (I – III) may strengthen the recommendation for surgery. In our experience, 186 patients (62%) had low-grade AVMs according to both grading systems, and 158 patients (85%) were improved or unchanged after surgery (Table 6). Conversely, an AVM with a high Spetzler-Martin grade (IV – V) may be unfavorable for microsurgical resection, and a high supplementary grade (IV – V) may strengthen the recommendation for nonoperative management. This experience included only surgical patients and therefore there were only 10 such patients (3%), of which half were worse after surgery. In these cases of matched Spetzler-Martin and supplementary grades, the supplementary grading system has a confirmatory role and may not alter management decisions. However, in cases of mismatched Spetzler-Martin and supplementary grades, the supplementary grading system may alter management decisions and therefore has a more important role.

Table 6
Impact of supplementary grading scale on surgical decision-making.

In our data, 83 patients (28%) had low Spetzler-Martin grades and high supplementary grades, and 34 of these patients (41%) were neurologically worse after surgery (Table 6), which is a higher morbidity than that of Spetzler-Martin grade IV AVMs (31%). Insight provided by the supplementary grade might have discouraged the recommendation for surgery in some of these patients (Figure 5). Similarly, 21 patients (7%) had high Spetzler-Martin grades and low supplementary grades, and 6 of these patients (29%) were neurologically worse after surgery (Table 6). This proportion of worsening was lower than the 35% morbidity for the overall group of Spetzler-Martin grade IV and V AVMs, and equivalent to the 30% morbidity seen for grade III AVMs. Again, insight provided by the supplementary grade might have encouraged the recommendation for surgery in some of these patients (Figure 6). Spetzler-Martin grade III AVMs have surgical risks that depend on the subtype, with small-sized/deep/eloquent AVMs (S1V1E1) associated with lower risk and medium-sized/eloquent AVMs (S2V0E1) associated with higher risk. In addition to considering the grade III subtype, considering the supplementary grade may influence surgical decisions for AVM patients at the borderline between high and low risk (Figure 7).

Figure 5
This 56 year-old woman presented with an incidental, unruptured AVM in the right medial parietal lobe, just posterior to the somatosensory strip on axial T2-weighted MR images (A, B). Angiography (right internal carotid artery injection, lateral (C) and ...
Figure 6
This 6 year-old boy presented with a cerebellar hemorrhage, as seen on the axial T2-weighted (A) and gadolinium-enhanced T1-weighted MR images (B). The associated AVM extended down to the deep cerebellar nuclei. Angiography (left vertebral artery injection, ...
Figure 7
This 54 year-old woman present with an intraventricular hemorrhage from a right anteromedial thalamic AVM, as seen on axial T2-weighted (A) and coronal T1-weighted MR images (B). Angiography (right internal carotid artery injection, lateral (C) and three-dimensional ...


These two grading systems analyze just 6 variables, and there may be other factors that influence surgical outcomes. For example, previous radiation has been shown to reduce surgical morbidity, in part because radiation changes AVM tissue to facilitate its resection 19. Although previous radiation is not assigned points from the supplementary grading scale, it may still influence the supplementary score if it has changed an AVM's diffuseness or bled during the latency period. We limited the supplementary grading scale to 3 variables, but there may be other variables of importance.

The supplementary grading system was derived from a surgical series that includes only operated AVMs, and therefore contains selection biases. However, we performed a 10-fold cross validation of our prediction model, and the results were similar, suggesting that the model was not overly optimistic. Nevertheless, we encourage the broader application of this grading system outside our institution to validate it on different cohorts of AVM patients.

The decision to resect a brain AVM is a complex art that requires a thorough appreciation of the lesion's anatomy, patient's history, neurosurgeon's skills, and family's preferences. No grading system, combination of grading systems, or simple algorithm can replace the discriminating process of patient selection. We offer this supplementary grading system as just another tool to guide the process of analyzing some of the critical factors that influence patient outcome, in order to make more rational choices when weighing known risk factors for spontaneous AVM rupture 4 against risk of intervention. The supplementary grading system is intended to improve preoperative risk prediction, and we expect that it will assist in patient selection for surgery. We anticipate that other grading systems will be developed to predict radiosurgical risks, embolization risks, and natural history risks. The clinician will be required to use these different scales and synthesize their insights to generate the best management plan for each individual patient.


This study was supported in part by NIH grants R01 NS034949 and P01NS44155 (WLY) and K23 NS058357 (HK), and by NIH/NCRR UCSF-CTSI Grant Number UL1 RR024131. The contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.


Financial Disclosure: The authors have no personal financial or institutional interest in any of the drugs, materials, or devices described in this article.

A listing of USCF Brain AVM Study Project members are found at The following members contributed to data registry efforts: Christopher F. Dowd, MD, Van V. Halbach, MD, Randall T. Higashida, MD, S. Claiborne Johnston, MD, Ph.D., Nerissa Ko, MD, Nancy Quinnine, RN, Brad Dispena, BS, Phillip Jolivalt, BS.

Authorship: There were many collaborators in this interdisciplinary project. In addition to reviewing the final manuscript, the authors' participation is noted below.

MT Lawton: study design, data collection

H Kim: study design, data analysis

C McCulloch: study design, data analysis

B Mikhak: data analysis

WL Young: study design, data collection, data analysis


1. Spetzler RF, Martin NA. A proposed grading system for arteriovenous malformations. Journal of neurosurgery. 1986 Oct;65:476–483. [PubMed]
2. Lawton MT. UCSF AVM Study Project: Spetzler-Martin Grade III arteriovenous malformations: surgical results and a modification of the grading scale. Neurosurgery. 2003 Apr;52(4):740–748. 748–749. discussion. [PubMed]
3. Vates GE, Lawton MT, Wilson CB, et al. Magnetic source imaging demonstrates altered cortical distribution of function in patients with arteriovenous malformations. Neurosurgery. 2002 Sep;51(3):614–623. 623–617. discussion. [PubMed]
4. Kim H, Sidney S, McCulloch CE, et al. Racial/ethnic differences in longitudinal risk of intracranial hemorrhage in brain arteriovenous malformation patients. Stroke; a journal of cerebral circulation. 2007 Sep;38(9):2430–2437. [PubMed]
5. Stapf C, Mast H, Sciacca RR, et al. Predictors of hemorrhage in patients with untreated brain arteriovenous malformation. Neurology. 2006 May 9;66(9):1350–1355. [PubMed]
6. Halim AX, Johnston SC, Singh V, et al. Longitudinal risk of intracranial hemorrhage in patients with arteriovenous malformation of the brain within a defined population. Stroke; a journal of cerebral circulation. 2004 Jul;35(7):1697–1702. [PubMed]
7. Lawton MT, Du R, Tran M, et al. Effect of presenting hemorrhage on outcome after microsurgical resection of brain arteriovenous malformations. Neurosurgery. 2005 Mar;56(3):485–493. [PubMed]
8. Sanchez-Mejia RO, Chenaputi S, Gupta N, Fullerton HJ, Young WL, Lawton MT. Superior outcomes in children versus adults after microsurgical resection of brain arteriovenous malformations. Journal of neurosurgery. 2006 Aug;105(2 Suppl):82–87. [PubMed]
9. Du R, Keyoung HM, Dowd CF, Young WL, Lawton MT. The effects of diffuseness and deep perforating artery supply on outcomes after microsurgical resection of brain arteriovenous malformations. Neurosurgery. 2007 Apr;60(4):638–646. 646–638. discussion. [PubMed]
10. van Swieten JC, Koudstaal PJ, Visser MC, Schouten HJ, van Gijn J. Interobserver agreement for the assessment of handicap in stroke patients. Stroke; a journal of cerebral circulation. 1988;19(5):604–607. [PubMed]
11. Spears J, Terbrugge KG, Moosavian M, et al. A discriminative prediction model of neurological outcome for patients undergoing surgery of brain arteriovenous malformations. Stroke; a journal of cerebral circulation. 2006 Jun;37(6):1457–1464. [PubMed]
12. Harrell FE., Jr . Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer-Verlag; 2001.
13. Tamaki N, Ehara K, Lin TK, et al. Cerebral arteriovenous malformations: factors influencing the surgical difficulty and outcome. Neurosurgery. 1991 Dec;29(6):856–861. 861–853. discussion. [PubMed]
14. Karnofsky DA, Burchenal JH. The clinical evaluation of chemotherapeutic agents in cancer. In: MacLeod CM, editor. Evaluation of Chemotherapeutic Agents. New York: Columbia University Press; 1949. p. 196.
15. Hollerhage HG, Dewenter KM, Dietz H. Grading of supratentorial arteriovenous malformations on the basis of multivariate analysis of prognostic factors. Acta neurochirurgica. 1992;117(34):129–134. [PubMed]
16. Hunt WE, Hess RM. Surgical risk as related to time of intervention in the repair of intracranial aneurysms. Journal of neurosurgery. 1968 Jan;28(1):14–20. [PubMed]
17. Jennett B, Bond M. Assessment of outcome after severe brain damage. The Lancet (Feb 13) 1975;1(7905):480–484. [PubMed]
18. Pertuiset B, Ancri D, Kinuta Y, et al. Classification of supratentorial arteriovenous malformations. A score system for evaluation of operability and surgical strategy based on an analysis of 66 cases. Acta neurochirurgica. 1991;110(12):6–16. [PubMed]
19. Sanchez-Mejia RO, McDermott MW, Tan J, Kim H, Young WL, Lawton MT. Radiosurgery facilitates resection of brain arteriovenous malformations and reduces surgical morbidity. Neurosurgery. 2009 Feb;64(2):231–238. 238–240. discussion. [PMC free article] [PubMed]