J Neurosurg 114:842–849, 2011

A 3-tier classification of cerebral arteriovenous

malformations

Clinical article
Robert F. Spetzler, M.D., and Francisco A. Ponce, M.D.
Division of Neurological Surgery, Barrow Neurological Institute, St. Joseph’s Hospital and Medical Center,
Phoenix, Arizona

Object. The authors propose a 3-tier classification for cerebral arteriovenous malformations (AVMs). The clas-
sification is based on the original 5-tier Spetzler-Martin grading system, and reflects the treatment paradigm for these
lesions. The implications of this modification in the literature are explored.
Methods. Class A combines Grades I and II AVMs, Class B are Grade III AVMs, and Class C combines Grades
IV and V AVMs. Recommended management is surgery for Class A AVMs, multimodality treatment for Class B, and
observation for Class C, with exceptions to the latter including recurrent hemorrhages and progressive neurological
deficits. To evaluate whether combining grades is warranted from the perspective of surgical outcomes, the 3-tier
system was applied to 1476 patients from 7 surgical series in which results were stratified according to Spetzler-
Martin grades.
Results. Pairwise comparisons of individual Spetzler-Martin grades in the series analyzed showed the fewest
significant differences (p < 0.05) in outcomes between Grades I and II AVMs and between Grades IV and V AVMs.
In the pooled data analysis, significant differences in outcomes were found between all grades except IV and V (p =
0.38), and the lowest relative risks were found between Grades I and II (1.066) and between Grades IV and V (1.095).
Using the pooled data, the predictive accuracies for surgical outcomes of the 5-tier and 3-tier systems were equivalent
(receiver operating characteristic curve area 0.711 and 0.713, respectively).
Conclusions. Combining Grades I and II AVMs and combining Grades IV and V AVMs is justified in part be-
cause the differences in surgical results between these respective pairs are small. The proposed 3-tier classification of
AVMs offers simplification of the Spetzler-Martin system, provides a guide to treatment, and is predictive of outcome.
The revised classification not only simplifies treatment recommendations; by placing patients into 3 as opposed to 5
groups, statistical power is markedly increased for series comparisons. (DOI: 10.3171/2010.8.JNS10663)

Key Words      •      Spetzler-Martin grading system      •

arteriovenous malformation      •      disease classification

T Methods
he Spetzler-Martin grading system31 was designed
as a tool to assist with the complexity surround-
ing surgical decision making for cerebral AVMs.7
Since its introduction in 1986, this grading system has
been frequently cited in the medical literature,25,26 and
other large surgical series have validated its use.2,7,9,11 Its
popularity reflects both its simplicity and its predictive
capabilities with respect to postoperative deficits.21
The management of AVMs at the Barrow Neurologi-
cal Institute follows a treatment paradigm whereby Grades
I and II lesions are managed similarly, as are Grades IV
and V.30 This study reflects our management strategy by
evaluating a 3-tier AVM classification and comparing
this modification to the standard 5-tier Spetzler-Martin
grading system.
Abbreviations used in this paper: AVM = arteriovenous malfor-
mation; mRS = modified Rankin Scale; ROC = receiver operating
characteristic; RR = relative risk.
Proposed 3-Tier Model
The proposed system consists of 3 classes of AVM
and is derived by combining Spetzler-Martin Grades I
and II AVMs into Class A and Grades IV and V lesions
into Class C; Grade III AVMs become Class B (Fig. 1).
Each class of AVM corresponds to a separate treatment
paradigm (Table 1).
Identification of Articles
We examined whether combining the grades as de-
scribed is justified from the perspective of surgical out-
comes reported in the literature. The database Thomson’s
ISI Web of Science (accessed April 2010) was used to
identify 672 published works that have cited the Spet-
zler-Martin grading system.31 The types of documents
included 490 articles, 85 proceedings papers, 46 reviews,
20 editorial materials, 20 letters, 7 reprints, 2 notes, and

842 J Neurosurg / Volume 114 / March 2011

Three-tier classification of arteriovenous malformations

Fig. 1.  Diagrammatic representation of the combinations of graded variables (size, eloquence, and venous drainage) for each
class of AVM. The Spetzler-Martin system assigns a score of 1 for small AVMs (< 3 cm), 2 for medium (3–6 cm), and 3 for large
(> 6 cm). The eloquence of adjacent brain is scored as either noneloquent (0) or eloquent (1). The venous drainage is scored
as superficial only (0) or including drainage to the deep cerebral veins (1). Scores for each feature are totaled to determine the
grade. In the system described in this article, Class A includes Spetzler-Martin Grades I and II; Class B includes Grade III; and
Class C includes Grades IV and V. Modified from Spetzler and Martin. (Modified with permission from Spetzler RF, Martin NA: A
proposed grading system for arteriovenous malformations. J Neurosurg 65:476–483, 1986.)

J Neurosurg / Volume 114 / March 2011 843


TABLE 1: Proposed 3-tier classification of cerebral AVMs with
treatment paradigm
Class Spetzler-Martin Grade Management
A I & II resection
B III multimodality treatment
C IV & V no treatment*
*  Exceptions for treatment of Class C AVMs include recurrent hemor-
rhages, progressive neurological deficits, steal-related symptoms, and
AVM-related aneurysms.
1 meeting abstract. The abstracts of articles and proceed-
ings papers were reviewed. Articles were characterized
based on their primary emphasis, including surgical,
endovascular, or radiosurgical management. An overlap-
ping subset focused on pediatrics. Other topics included
validation of novel imaging techniques and assessment of
interobserver variability with the grading system.
We restricted our analysis to large surgical series that
stratified outcomes based on the 5-tier Spetzler-Martin
system. When articles originated from the same institu-
tion and had overlapping patient cohorts, only the most
recent article was included. Seven series were identified,
including that by Spetzler and Martin31 (Table 2). We
applied the proposed 3-tier classification system to the
outcomes of the 1476 cases reported in these series and
compared the predictive accuracy with regard to clinical
outcomes to that of the 5-tier system.
Statistical Analysis
Statistical analyses include the Fisher exact prob-
ability test to compare outcomes of individual grades and
classes of AVMs, the fitted ROC curve area to evaluate
the predictive accuracy of the 2 systems with regard to
outcomes, 5,29 linear regression to evaluate chronological
trends in management and impact of study size, and the
modified Wald method to calculate 95% CIs for propor-
tions.22 The significance level was set at p < 0.05. An ROC
area of at least 0.70 indicates that a predictive model is
clinically useful.18,29,32
Pooling of Data From Various Series
The 7 series reported a total of 11 outcome sets: 4 se-
R. F. Spetzler and F. A. Ponce
ries reported both early and late outcome, and 3 reported
a single outcome set. A pooled analysis of these outcomes
was performed in which the single outcome sets were
combined with the late outcome sets. Each series defined
outcomes in terms of postoperative deficits, although the
definition of deficit varied (see Appendix). All series ulti-
mately grouped outcomes in a binary fashion (for exam-
ple, deficit versus no deficit,7,31 better/same versus worse/
dead,18 excellent/good versus fair/poor/dead11). In pooling
the data, we combined the binary (that is, “positive” and
“negative”) deficit-based outcomes that the authors had
adopted in their respective analyses.
Results
Statistical Assessment of the 5-Tier System
Table 3 shows the rates of negative outcomes from
the 7 series that were evaluated. Whether a significant
difference in outcome existed between individual grades
in an outcome set was determined using the Fisher exact
test. Pairwise comparisons were performed between each
of the 5 grades (10 pairwise comparisons per outcome
set) for the 11 outcome sets (Table 4). Overall, 53 (48%)
of the 110 pairwise comparisons of grades demonstrated
significant differences in outcome (95% CI 39%–57%).
Among the 11 outcome sets (columns in Table 4),
the highest proportion of significant pairwise differences
was 8 of 10 tests, seen in Davidson and Morgan’s series.2
In that study, the 2 comparisons that lacked significance
were between Grades I and II AVMs and between Grades
IV and V lesions. The lowest proportion of significantly
different combinations was 2 of 10 tests, seen in Spetzler
and Martin31 and the late results in Hamilton and Spet-
zler.7 The size of a series correlated with the number of
significant pairwise differences between grades in the
late or single outcome sets (p = 0.03).
The frequency with which each pair of grades was
significantly different in the 7 outcome sets was then
evaluated (rows in Table 4). The lowest frequency was
between Grades I and II (1 of 11 tests [9%]) and between
Grades IV and V (2 of 11 tests [19%]). The highest fre-
quencies were seen between Grades I and IV (9 of 11 tests
[82%]) followed by Grades II and IV and Grades I and V
(8 of 11 tests [73%]).

TABLE 2: Summary of 7 studies on surgical management of AVMs

No. in Spetzler-Martin Grade (%)

Authors & Year No. of Cases I II III IV V
Spetzler & Martin, 1986 100 23 (23) 21 (21) 25 (25) 15 (15) 16 (16)
Heros et al., 1990 153 12 (8) 35 (23) 44 (29) 41 (27) 21 (14)
Hamilton & Spetzler, 1994 120 16 (13) 24 (20) 36 (30) 32 (27) 12 (10)
Schaller et al., 1998 150 33 (22) 48 (32) 44 (29) 21 (14) 4 (3)
Hartmann et al., 2000 124 12 (10) 36 (29) 47 (38) 26 (21) 3 (2)
Davidson & Morgan, 2010 529* 98 (19) 198 (37) 169 (32) 54 (10) 10 (2)
Lawton et al., 2010 300 56 (19) 123 (41) 90 (30) 29 (10) 2 (1)
total 1476 250 (17) 485 (33) 455 (31) 218 (15) 68 (5)

*  Surgically treated cases.

844 J Neurosurg / Volume 114 / March 2011

Three-tier classification of arteriovenous malformations

When the positive and negative outcomes of each

series were pooled (Table 5), 9 of the 10 pairwise com-

37 (26–49)
31 (25–37)
Combined

18 (15–22)

18 (15–22)
32 (27–38)
8 (6–10)
10 (7–13)
4 (2–7)
parisons were significantly different, the exception being
between Grades IV and V AVMs (p = 0.38). The lowest
RRs were between Grades I and II (RR 1.066, 95% CI
1.027–1.107) and between Grades IV and V (RR 1.095,
Lawton et al.

100 (37–100)
14 (10–20) 30 (21–40)

14 (10–20) 30 (21–40)
34 (24–47) 35 (21–53)
24 (18–33)

20 (14–26)
17 (21–45) 31 (17–49)
95% CI 0.895–1.340; not significant).
9 (3–20)

Statistical Assessment of the 3-Tier System

 
Stratification of outcomes reported in each series
 

based on the 3-tier system is shown in Table 3. Three pair-

Davidson &

5 (7–12)
Morgan

wise tests were performed for each of the 11 outcome sets

1 (0–6)
0.5 (0–3)

0.7 (0–3)
to determine whether differences were significant (Table
4). In the series reported by Davidson and Morgan,2 all
 
 

pairwise comparisons were significant. No pairwise test

32 (20–46)

32 (20–46)

was significant in the series published by Lawton et al.,18

62 (44–77)
29 (18–43)
36 (22–52)

65 (46–81)
33 (6–80)
8 (0–38)
Late

although the differences between Classes A and B and

Hartmann et al.

between Classes A and C approached significance (p

= 0.066 and p = 0.060, respectively). The Spetzler and
Martin31 series showed significance only between Classes
100 (47–100)
73 (54–87)

76 (58–88)
32 (20–46)

32 (20–46)
29 (18–43)
36 (22–52)

A and C, but the difference between Classes A and B

8 (0–38)
Early

approached significance (p = 0.054). The remaining 8

TABLE 3: Rate of negative outcomes in 11 outcome sets obtained from 7 studies, stratified by 5-tier and 3-tier systems*

outcome sets showed significance in 2 of the 3 pairwise

tests. Overall, 20 (61%) of the 33 pairwise comparisons
Rate of Negative Outcomes Reported†

demonstrated significant differences in outcome (95% CI

44%–75%). Of the 11 tests performed on each pair, sig-
40 (23–59)
50 (15–85)
23 (13–37)

23 (13–37)
38 (21–29)
3 (0–17)
0 (0–6)

1 (0–7)

nificant differences in outcomes between Classes A and

Late

B were defined in 3 (27%), between Classes B and C in 7

Schaller et al.

(64%), and between Classes A and C in 10 (91%). In the

 

pooled analysis, all pairwise comparisons showed signifi-

100 (54–100)

64 (44–80)
25 (15–39)
57 (42–70)

57 (42–70)
21 (13–31)
57 (37–76)

cant differences in outcome (Table 5).

15 (6–31)
Early

Analysis of ROC Area

 

Of the 11 outcome sets tested, the 5-tier system

 

showed the ROC area was < 0.7 in 2, 0.7–0.79 in 5, and

20 (11–35)
22 (11–39)

0.8–0.89 in 4 (Table 6). With the 3-tier system, an ROC

12 (3–46)
0 (0–15)
3 (0–15)

3 (0–15)
0 (0–17)

0 (0–9)
Hamilton & Spetzler
Late‡

area < 0.7 was seen in 3 sets, 0.7–0.79 in 3, 0.8–0.89 in 4,

and ≥ 0.9 in 1. The mean ROC area was 0.770 for both the
5-tier and the 3-tier system, and the difference was not
*  See Appendix for details on outcomes measured in the respective series.
31 (18–49)
50 (25–75)

significant (p = 0.93). When the outcomes were pooled,

36 (24–51)

†  Outcomes presented as percentages, with the 95% CI in parentheses.

4 (0–22)
3 (0–15)

3 (0–15)
0 (0–17)

2.5 (0–14)
Early

the ROC area was 0.711 for the 5-tier system and 0.713
for the 3-tier system, suggesting that the predictive ac-
curacies of both systems were clinically useful18,29,32 and
equivalent.
38 (21–59)

21 (13–33)
8 (0–38)

12 (5–26)
3 (0–16)

4 (4–15)
0 (0–7)

0 (0–7)
Late

Discussion
Heros et al.

‡  Late outcome data not presented for 5 patients.

 

The proposed 3-tier classification reflects a treatment

 

paradigm for AVMs based on the Spetzler-Martin grade

11 (4.5–24)

11 (4.5–24)
71 (50–86)

50 (38–62)
39 (26–54)

that is practiced at our institution. In terms of manage-

0 (0–22)

4 (4–15)
6 (1–20)
Early

ment strategies, we find that the practical distinctions

between Grades I and II AVMs and between Grades IV
 

and V lesions are limited. In the 7 surgical series consid-

 

ered in this study, we found little difference between the

Spetzler & Martin

outcomes of Grades I and II or Grades IV and V AVMs.

27 (10–55)

29 (16–47)
31 (14–56)

16 (6–35)
16 (5–36)
5 (0–24)
0 (0–13)

2 (0–13)

In condensing the breakdown of AVMs from 5 tiers to 3

tiers, we demonstrate that the 3-tier system retains a ca-
 

pability of predicting poor neurological outcomes that is

equivalent to the 5-tier system.
 
System

Class A: Grades I and II AVMs

5-tier

3-tier
  IV
  III

  C
  V

  A
  B
  II
  I

The differences in outcomes between Grades I and II

J Neurosurg / Volume 114 / March 2011 845

 R. F. Spetzler and F. A. Ponce

TABLE 4: Summary of results from pairwise comparisons of individual tiers within the 5-tier and 3-tier systems from 7 surgical series*

Heros et al. Hamilton & Spetzler Schaller et al. Hartmann et al. Davidson
System Spetzler & Martin Early Late Early Late Early Late Early Late & Morgan Lawton et al. Total (+)
5-tier
  I, II − − − − − − − − − − +  1
  II, III − − − − − + + − − + −  3
  III, IV − + + + + − − + + + −  7
  IV, V − + + − − − − − − − −  2
  I, III − − − − − + + − − + +  4
  II, IV − + − + + + + + + + −  8
  III, V − + + + − − − + − + −  5
  I, IV + + − + − + + + + + +  9
  II, V − + + + − + + − − + −  6
  I, V + + − + − + + + − + +  8
3-tier
  A, B − − − − − + + − − + −  3
  B, C − + + + + − − + + + −  7
  A, C + + + + + + + + + + − 10

*  Pairwise comparisons were done using the Fisher exact test. A “+” indicates that differences in outcomes were significant (p < 0.05), and a “−” indicates
that differences were not significant (p ≥ 0.05). The column labeled “Total (+)” indicates the number of outcome sets (of 11 total) with significant differ-
ences for each pairwise comparison.

AVMs appear to be minimal. In pairwise analysis, only 1 differences within Grade III AVMs. The modification by
of 11 outcome sets demonstrated a significant difference de Oliveira et al.3 divides the grade into 2 subgroups: IIIA
between these 2 grades. Although the pooled analysis (large) and IIIB (small, in eloquent areas) and proposes
was significant, the RR associated with surgery in these embolization plus surgery for the former and radiosur-
grades was the smallest of the 10 pairwise comparisons. gery for the latter. The modification by Lawton17 suggests
In the 3-tier classification, Grades I and II AVMs are breaking this grade down into all 4 subgroups, and de-
combined as Class A.
The recommended management of Class A AVMs
is microsurgical resection. In the initial report from our TABLE 5: Analysis of pooled data*
institution, the incidence of deficits in this cohort was
2%.31 The subsequent prospective application of the sys- System p Value RR 95% CI
tem showed a 2.5% incidence of early deficits, and no late
5-tier  
deficits.7 The recent study by Davidson and Morgan2 re-
ported a 0.7% risk of adverse surgery-related outcomes   I, II 0.003 1.066 1.027–1.107
for Class A AVMs, further reinforcing the role of surgery   II, III 0.0003 1.100 1.045–1.158
as first-line therapy.   III, IV 0.0002 1.187 1.076–1.309
While the benefit of including endovascular embo-   IV, V 0.38† 1.095 0.895–1.340
lization preoperatively in the management of Class A
  I, III <0.0001 1.173 1.117–1.232
AVMs is under evaluation, procedural risk of this mo-
dality should be considered. A recent series of 47 AVMs   II, IV <0.0001 1.305 1.189–1.433
treated preoperatively with Onyx included 25 patients   III, V 0.0006 1.300 1.079–1.566
with Class A AVMs, of which 4 (16%) showed a decline   I, IV <0.0001 1.392 1.270–1.525
in their mRS score after embolization.35 The potential   II, V <0.0001 1.430 1.190–1.718
role of embolization as a curative single modality has also
  I, V <0.0001 1.524 1.270–1.830
been suggested, although the risk profile appears to be
higher than with surgery.16,36 Although we do not believe 3-tier  
that radiosurgery is indicated for this class of AVM, a re-   A, B <0.0001 1.125 1.073–1.180
cent study reported an increased rate of hemorrhage af-   B, C <0.0001 1.208 1.103–1.322
ter radiosurgery in low-grade compared with high-grade   A, C <0.0001 1.358 1.251–1.475
AVMs.15
*  Data consist of p values (according to the Fisher exact test) and RR,
Class B: Grade III AVMs with 95% CIs for pairwise comparisons (also according to the Fisher
A number of authors have suggested that the Spet- exact test) of each tier for both the 5-tier and 3-tier systems.
zler-Martin grading system be modified to emphasize the †  Not significant.

846 J Neurosurg / Volume 114 / March 2011

Three-tier classification of arteriovenous malformations
TABLE 6: Fitted ROC area values for 11 outcome sets using 5-tier recommended for only 5% of Class C AVMs evaluated by
and 3-tier systems the senior author (R.F.S.),8 these numbers must be viewed
in the context of a strong selection bias. In the 7 series
ROC Area reported here, we detected a significant chronological re-
Authors & Year 5-Tier System 3-Tier System duction, based on year of publication, in the proportion
of AVMs being treated in each cohort that are Grades IV
Spetzler & Martin, 1986 0.780 0.774
and V (p = 0.04 and p = 0.03, respectively). This finding
Heros et al., 1990 may reflect an overall trend toward conservative manage-
  early 0.845 0.861 ment of this class of AVM.
  late* 0.783 0.735
Hamilton & Spetzler, 1994 Important Variables and Alternative Grading Systems
  early 0.851 0.913 Examples of widely adopted classification systems
  late 0.790 0.800 in neurosurgery26 include the Glasgow Outcome Scale,14
Schaller et al., 1998 Glasgow Coma Scale,34 Fisher grade,6 Hunt and Hess
  early 0.742 0.767
grade,13 and the Simpson grade.28 Like the Spetzler-Mar-
tin grade, these systems provide useful, clinically predic-
  late 0.829 0.850 tive models and are simple to use.
Hartmann et al., 2000 The Spetzler-Martin grading system divides AVMs
  early 0.727 0.685 into 5 grades based on their size, location, and venous
  late 0.656 0.639 drainage. Additional variables that affect surgical risk
Davidson & Morgan, 2010 0.856 0.868 associated with AVMs have been identified, including
Lawton et al., 2010 0.637 0.613 age,1 clinical status,19 perforating vessel supply,1,20 and
compactness of nidus,29 and alternative grading systems
pooled data† 0.711 0.713
have incorporated these variables.12,18,19,23,24,29,33 However,
*  Empirical value due to degenerate fitting. increasing the number of categories with the introduction
†  Pooled data exclude early results. of new variables results in fewer patients being assigned
to each category, which in turn weakens the statistical
scribes different management strategies for the 3 of these power of the results from a particular study. Given the
that were encountered. infrequency with which AVMs are encountered, a more
In the proposed 3-tier system, Grade III AVMs re- extensive grading system for these lesions is impracti-
main separate as Class B. The management of Class B cal.18 Conversely, by reducing the number of categories,
AVMs is more individualized and typically requires a the proposed 3-tier system makes significance easier to
multimodality approach drawing upon microsurgical, en- achieve with smaller numbers. Indeed, some earlier stud-
dovascular, and radiosurgical techniques. These nuances ies have combined grades in this manner to improve sta-
reflect the complexity and heterogeneity of this class, as tistical power.4,21
implied in the earlier modifications.3,17
Limitations of This Study
Class C: Grades IV and V AVMs First, our analysis was restricted to only 7 studies.
Among the 11 outcome sets analyzed, lack of signifi- However, these were large series, providing data on 1476
cance between Grades IV and V AVMs was seen in all patients. Two of these studies originated from our insti-
but one of the series. In the pooled analysis, the differ- tution, but their contribution reflected only 15% of pa-
ence in outcomes between Grades IV and V AVMs was tients.
the only comparison that was not significant. In the 3-tier Second, the studies included in the analysis did not
classification, these 2 grades are combined as Class C. use a uniform definition of positive and negative out-
Unless a patient has repeated hemorrhages or pro- comes. Our justification for pooling the data is that the
gressive neurological deficits, conservative management outcomes are all defined in relation to the occurrence of
is recommended for Class C AVMs. Additional caveats new treatment-related neurological deficits. To address
include the presence of flow-related aneurysms, which this limitation, we considered lowering the p value for
are treated by microsurgical or endovascular obliteration, significance for the pooled analysis (that is, to p < 0.001).
and steal-related deficits amenable to endovascular em- Doing so, however, would only have rendered insignifi-
bolization.8 cant the difference between Grades I and II AVMs (p =
The recommendation of no treatment is based on the 0.003), which would further support combining these 2
high reported rates of morbidity associated with surgery, grades.
coupled with the lack of immediate protection and possi- Third, when the proportion of patients evaluated who
ble elevated risk of hemorrhage associated with radiation are not offered surgery varies between grades, a selec-
or partial embolization. At our institution, the surgical tion bias is introduced and may confound the analysis.
results for these lesions show a rate of new neurologi- Consequently, differences in outcomes between tiers may
cal deficits of 29% in the series reported by Spetzler and be rendered statistically insignificant even when a true
Martin,31 and of 20% in that reported by Hamilton and difference in risk exists. The influence of selection bias
Spetzler.7 However, considering that complete resection is on outcomes is addressed in the report by Davidson and

J Neurosurg / Volume 114 / March 2011 847


Morgan,2 who caution that their surgical results for elo-
quent Grades III, IV, and V AVMs are not generalizable
because 14% of the patients evaluated who had similar
lesions had been refused surgery due to its perceived dif-
ficulty. Considering the fact that we recommend complete
treatment of Class C AVMs for only 5% of our patients,8
selection bias may be partially responsible for the rela-
tively few significant differences in surgical outcomes
seen between grades in the 2 studies from our institu-
tion.7,31 The decision-making process profoundly affects
the results of a surgical series and can account for the
differences in rates of morbidity and mortality among
different studies.8 As treatment patterns trend toward a
more conservative approach for Class C AVMs, subse-
quent improvements in surgical outcomes may render dif-
ferences more difficult to detect, making the combining
of categories desirable.
Conclusions
We propose that the original 5-tier Spetzler-Martin
grading system be condensed to a 3-tier classification that
reflects a treatment paradigm for these lesions. This sys-
tem serves as a guide for treatment and is not intended
to replace individual analysis of AVMs. The modifica-
tion is easily applied to earlier studies that have used the
Spetzler-Martin grading system because the criteria are
the same. It retains the predictive accuracy of outcomes,
as shown in both cohorts from our institution as well as
in other large AVM series. In eliminating what are argu-
ably 2 redundant tiers from the 5-tier system, the 3-tier
system offers the advantage of simplification.30 The re-
vised classification not only simplifies treatment recom-
mendations, but by placing patients into 3 as opposed to 5
groups, statistical power is markedly increased for series
comparisons.
Appendix
Description of Outcomes in Each Study
Spetzler and Martin. 31 “Complications were broken down
into the categories of minor deficit, major deficit, and mortality.”
Heros et al.11 Early: “Excellent, no neurological deficit, other
than a partial field defect; good, minimal neurological deficit (pa-
tient fully independent); fair, moderate neurological deficit (patient
fully independent); poor, moderate to severe neurological deficit
(patient dependent).” Late: “Excellent, no neurological deficit
other than a partial visual field defect, no seizures or only well-
controlled seizures (patient able to work full time); good, minimal
neurological deficit, no seizures or only well-controlled seizures
(patient able to work full time); fair, mild to moderate neurologi-
cal deficit and/or difficulty in controlling seizures (patient unable
to work full time but fully independent); poor, moderate to severe
neurological deficit (patient incapacitated and dependent).”
Hamilton and Spetzler.7 Table 4 data comparing “unchanged
or improved” versus “treatment-associated new neurological defi-
cit.”
Schaller et al. 27 Table 2 data showing incidence of “early new
postoperative deficit”; Table 3 data showing “permanent neuro-
logical deficits (major plus minor).”
Hartmann et al.9 “Any worsening of patients’ preoperative
848
R. F. Spetzler and F. A. Ponce
Rankin scale score was coded as ‘new neurological deficit.’ New
neurological deficits were classified as ‘disabling’ when Rankin
scores were 3, 4, or 5 (for patients with preoperative Rankin scores
of 2 or worse, any score increase was classified as disabling.)”
Davidson and Morgan. 2 “Outcomes assessment was per-
formed using the modified Rankin Scale (mRS) score, which was
allocated preoperatively at 6 weeks and 12 months of follow-up. A
poor outcome was considered to be any patient with a 12-month
mRS score greater than 1. To differentiate between the neurologic
effects of AVM presentation (hemorrhage, focal neurologic defi-
cit) and the effects of treatment, adverse outcomes were attributed
to one of the following factors at the 6-week clinical assessment:
natural history, surgery, embolization, or focused irradiation. For
the purposes of this article, an adverse outcome due to surgery also
included patients who had an adverse outcome due to planned pre-
operative embolization… To account for the gradual improvement
in neurologic deficits, only those patients whose surgery-related
deficits persisted at the 12-month assessment were declared to
have had an adverse outcome due to surgery.”
Lawton et al.18 “Outcomes were analyzed in terms of change
between preoperative and final postoperative mRS scores (mRS
final − mRS preoperative)… 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).”
Disclosure
The authors report no conflict of interest concerning the mate-
rials or methods used in this study or the findings specified in this
paper.
Author contributions to the study and manuscript preparation
include the following. Acquisition of data: Ponce. Analysis and
interpretation of data: Ponce. Drafting the article: Ponce. Critically
revising the article: Ponce. Reviewed final version of the manuscript
and approved it for submission: both authors. Administrative/techni-
cal/material support: Ponce. Study supervision: Spetzler.
Acknowledgment
Kristina Kupanoff, Ph.D., provided assistance with statistical
analysis.
References
  1.  Batjer HH, Devous MD Sr, Seibert GB, Purdy PD, Bonte FJ:
Intracranial arteriovenous malformation: relationship be-
tween clinical factors and surgical complications. Neurosur-
gery 24:75–79, 1989
  2.  Davidson AS, Morgan MK: How safe is arteriovenous malfor-
mation surgery? A prospective, observational study of surgery
as first-line treatment for brain arteriovenous malformations.
Neurosurgery 66:498–505, 2010
  3.  de Oliveira E, Tedeschi H, Raso J: Comprehensive manage-
ment of arteriovenous malformations. Neurol Res 20:673–
683, 1998
  4.  Deruty R, Pelissou-Guyotat I, Mottolese C, Amat D, Bascoul-
ergue Y: Prognostic value of the Spetzler’s grading system in a
series of cerebral AVMs treated by a combined management.
Acta Neurochir (Wien) 131:169–175, 1994
  5.  Eng J: ROC Analysis: Web-based Calculator for ROC
Curves. (http://www.jrocfit.org) [Accessed September 7,
2010]
  6.  Fisher CM, Kistler JP, Davis JM: Relation of cerebral vasos-
pasm to subarachnoid hemorrhage visualized by computer-
ized tomographic scanning. Neurosurgery 6:1–9, 1980
  7.  Hamilton MG, Spetzler RF: The prospective application of a
J Neurosurg / Volume 114 / March 2011
Three-tier classification of arteriovenous malformations
grading system for arteriovenous malformations. Neurosur-
gery 34:2–7, 1994
  8.  Han PP, Ponce FA, Spetzler RF: Intention-to-treat analysis of
Spetzler–Martin Grades IV and V arteriovenous malforma-
tions: natural history and treatment paradigm. J Neurosurg
98:3–7, 2003
  9.  Hartmann A, Stapf C, Hofmeister C, Mohr JP, Sciacca RR,
Stein BM, et al: Determinants of neurological outcome after
surgery for brain arteriovenous malformation. Stroke 31:
2361–2364, 2000
10.  Heros RC: Spetzler–Martin Grades IV and V arteriovenous
malformations. J Neurosurg 98:1–2, 2003
11.  Heros RC, Korosue K, Diebold PM: Surgical excision of cere-
bral arteriovenous malformations: late results. Neurosurgery
26:570–578, 1990
12.  Höllerhage HG, Dewenter KM, Dietz H: Grading of supraten-
torial arteriovenous malformations on the basis of multivari-
ate analysis of prognostic factors. Acta Neurochir (Wien)
117:129–134, 1992
13.  Hunt WE, Hess RM: Surgical risk as related to time of inter-
vention in the repair of intracranial aneurysms. J Neurosurg
28:14–20, 1968
14.  Jennett B, Bond M: Assessment of outcome after severe brain
damage. Lancet 1:480–484, 1975
15.  Kasliwal MK, Kale SS, Gupta A, Kiran NAS, Sharma MS,
Sharma BS, et al: Outcome after hemorrhage following Gam-
ma Knife surgery for cerebral arteriovenous malformations.
Clinical article. J Neurosurg 110:1003–1009, 2009
16.  Katsaridis V, Papagiannaki C, Aimar E: Curative emboliza-
tion of cerebral arteriovenous malformations (AVMs) with
Onyx in 101 patients. Neuroradiology 50:589–597, 2008
17.  Lawton MT: Spetzler-Martin Grade III arteriovenous malfor-
mations: surgical results and a modification of the grading
scale. Neurosurgery 52:740–749, 2003
18.  Lawton MT, Kim H, McCulloch CE, Mikhak B, Young WL:
A supplementary grading scale for selecting patients with
brain arteriovenous malformations for surgery. Neurosur-
gery 66:702–713, 2010
19.  Luessenhop AJ, Gennarelli TA: Anatomical grading of su-
pratentorial arteriovenous malformations for determining op-
erability. Neurosurgery 1:30–35, 1977
20.  Morgan MK, Drummond KJ, Grinnell V, Sorby W: Surgery
for cerebral arteriovenous malformation: risks related to len-
ticulostriate arterial supply. J Neurosurg 86:801–805, 1997
21.  Morgan MK, Sekhon LH, Finfer S, Grinnell V: Delayed neu-
rological deterioration following resection of arteriovenous
malformations of the brain. J Neurosurg 90:695–701, 1999
22.  Newcombe RG: Two-sided confidence intervals for the single
proportion: comparison of seven methods. Stat Med 17:857–
872, 1998
23.  Pertuiset B, Ancri D, Kinuta Y, Haisa T, Bordi L, Lin C, et al:
Classification of supratentorial arteriovenous malformations.
A score system for evaluation of operability and surgical strat-
J Neurosurg / Volume 114 / March 2011
egy based on an analysis of 66 cases. Acta Neurochir (Wien)
110:6–16, 1991
24.  Pollock BE, Flickinger JC: A proposed radiosurgery-based
grading system for arteriovenous malformations. J Neuro-
surg 96:79–85, 2002
25.  Ponce FA, Lozano AM: Highly cited works in neurosurgery.
Part I: the 100 top-cited papers in neurosurgical journals. A
review. J Neurosurg 112:223–232, 2010
26.  Ponce FA, Lozano AM: Highly cited works in neurosurgery.
Part II: the citation classics. A review. J Neurosurg 112:233–
246, 2010
27.  Schaller C, Schramm J, Haun D: Significance of factors con-
tributing to surgical complications and to late outcome after
elective surgery of cerebral arteriovenous malformations. J
Neurol Neurosurg Psychiatry 65:547–554, 1998
28.  Simpson D: The recurrence of intracranial meningiomas after
surgical treatment. J Neurol Neurosurg Psychiatry 20:22–
39, 1957
29.  Spears J, Terbrugge KG, Moosavian M, Montanera W, Wil-
linsky RA, Wallace MC, et al: A discriminative prediction
model of neurological outcome for patients undergoing sur-
gery of brain arteriovenous malformations. Stroke 37:1457–
1464, 2006
30.  Spetzler RF: Introduction. Arteriovenous malformation grad-
ing system. J Neurosurg 108:185, 2008
31.  Spetzler RF, Martin NA: A proposed grading system for arte-
riovenous malformations. J Neurosurg 65:476–483, 1986
32.  Swets JA: Measuring the accuracy of diagnostic systems. Sci-
ence 240:1285–1293, 1988
33.  Tamaki N, Ehara K, Lin TK, Kuwamura K, Obora Y, Kanaza-
wa Y, et al: Cerebral arteriovenous malformations: factors in-
fluencing the surgical difficulty and outcome. Neurosurgery
29:856–863, 1991
34.  Teasdale G, Jennett B: Assessment of coma and impaired con-
sciousness. A practical scale. Lancet 2:81–84, 1974
35.  Weber W, Kis B, Siekmann R, Jans P, Laumer R, Kühne D:
Preoperative embolization of intracranial arteriovenous mal-
formations with Onyx. Neurosurgery 61:244–254, 2007
36.  Yu SC, Chan MS, Lam JM, Tam PH, Poon WS: Complete
obliteration of intracranial arteriovenous malformation with
endovascular cyanoacrylate embolization: initial success and
rate of permanent cure. AJNR Am J Neuroradiol 25:1139–
1143, 2004
Manuscript submitted April 30, 2010.
Accepted August 31, 2010.
Please include this information when citing this paper: published
online October 8, 2010; DOI: 10.3171/2010.8.JNS10663.
Address correspondence to: Robert F. Spetzler, M.D., c/o
Neuroscience Publications, Barrow Neurological Institute, 350
West Thomas Road, Phoenix, Arizona 85013. email: neuropub@
chw.edu.
849