You are on page 1of 9



Department of Neurosurgery,
Helsinki University
Central Hospital,
Helsinki, Finland
OBJECTIVE: Long-term follow-up studies in patients with brain arteriovenous malfor-
Reza Dashti, M.D. mations (AVM) have yielded contradictory results regarding both risk factors for rup-
Department of Neurosurgery, ture and annual rupture rate. We performed a long-term follow-up study in an unse-
Helsinki University lected, consecutive patient population with AVMs admitted to the Department of
Central Hospital,
Neurosurgery at Helsinki University Central Hospital between 1942 and 2005.
Helsinki, Finland
METHODS: Patients with untreated AVMs were followed from admission until death,
Seppo Juvela, M.D., Ph.D. occurrence of AVM rupture, initiation of treatment, or until the end of 2005. Patients with
Department of Neurosurgery, at least 1 month of follow-up were included in further analysis. Annual and cumulative
Helsinki University incidence rates of AVM rupture as well as several potential risk factors for rupture were
Central Hospital,
Helsinki, Finland,
analyzed using Kaplan-Meier life table analyses and Cox proportional hazards regres-
Department of Neurosurgery, sion models.
Turku University Central Hospital,
RESULTS: We identified 238 patients with a mean follow-up period of 13.5 years (range,
Turku, Finland
1 month–53.1 years). The average annual risk of hemorrhage from AVMs was 2.4%.
Kristjan Väärt, M.D. The risk was highest during the first 5 years after diagnosis, decreasing thereafter. Risk
Department of Neurosurgery, factors predicting subsequent AVM hemorrhage in univariate analysis were young age,
Helsinki University previous rupture, deep and infratentorial locations, and exclusively deep venous drainage.
Central Hospital, Previous rupture, large AVM size, and infratentorial and deep locations were independ-
Helsinki, Finland
ent risk factors according to multivariate models.
Mika Niemelä, M.D., Ph.D. CONCLUSION: According to this long-term follow-up study, AVMs with previous rup-
Department of Neurosurgery, ture and large size, as well as with infratentorial and deep locations have the highest risk
Helsinki University of subsequent hemorrhage. This risk is highest during the first few years after diagnosis
Central Hospital,
but remains significant for decades.
Helsinki, Finland
KEY WORDS: Arteriovenous malformation, Cerebrovascular, Intracerebral hemorrhage, Stroke, Subarachnoid
Aki Laakso, M.D., Ph.D. hemorrhage
Department of Neurosurgery,
Helsinki University Neurosurgery 63:823–831, 2008 DOI: 10.1227/01.NEU.0000330401.82582.5E
Central Hospital,
Helsinki, Finland

rteriovenous malformations (AVM) of The complex cerebrovascular anatomy of
Reprint requests:
Aki Laakso, M.D., Ph.D., the brain are congenital vascular lesions AVMs makes them a challenge to treat, and the
Department of Neurosurgery, that account for approximately 2% of all treatment itself carries significant risks. To eval-
Helsinki University hemorrhagic strokes (2, 5). Despite the relative uate the possible benefit of a risky treatment,
Central Hospital, rarity of the disease (with an estimated current one needs to understand the natural history
Topeliuksenkatu 5,
P.O. Box 266,
detection rate of approximately 1/100 000 and prognosis of the disease. The complexity of
Helsinki, FIN-00029-HUS, Finland. person-years) (2), AVMs pose a significant neu- AVMs makes them a rather heterogeneous
Email: rological problem because patients are mostly group of lesions in terms of various factors pos-
young and otherwise healthy. Moreover, the sibly affecting the risk of rupture and subse-
Received, February 6, 2008. availability of noninvasive imaging has rapidly quent hemorrhagic stroke. Over the past sev-
Accepted, June 26, 2008. increased the detection of incidental AVMs. eral decades, several groups have contributed
to our knowledge of the hemorrhage risk asso-
ciated with untreated AVMs (1, 3, 4, 6–9, 11,
ABBREVIATION: AVM, arteriovenous malforma-
13–15, 17, 19, 20, 22). The existing literature,
however, is inconclusive about both the annual



rupture rate and the factors affecting this risk. Some of the patients were followed until death, the first new occurrence of hemor-
inconsistencies between older and recent reports are explained rhagic stroke, initiation of any treatment of the AVM (neurosurgical,
by the use of more sophisticated statistical methods in the lat- endovascular, or radiosurgical), or last contact (at the end of 2005). Only
ter. At the same time, however, the development of therapeu- patients with at least 1 month of hemorrhage- and treatment-free sur-
tic strategies has made it increasingly rare that a large propor- vival time were included for further analysis. The characteristics of
patients experiencing an AVM rupture during the follow-up period
tion of patients would be followed conservatively; therefore,
were compared with rupture-free patients using Pearson’s χ2 test for cat-
historical patient cohorts have significant inherent value egorical variables and the Mann-Whitney U test for age and follow-up
because of the lack of selection bias. time. The annual risk of AVM rupture was calculated as the number of
Finland is well suited for population-based epidemiological patients with new hemorrhage during follow-up divided by person-
studies because of a high-quality public health care system in years of follow-up. Cumulative rates of AVM rupture were estimated
which AVMs are treated in public university hospitals with using the Kaplan-Meier product-limit method, and the resulting curves
population-wide responsibility, comprehensive and accessible were compared using the log-rank test. The Cox proportional hazards
hospital records and vital statistics, and a relatively stable and model with a forward stepwise regression procedure was used to deter-
homogeneous population. One of the most cited studies on the mine the significance of several variables in predicting the relative risk
natural history of patients with AVMs is based on a series of (hazard ratio) of subsequent AVM rupture. These variables were age;
sex; AVM rupture before admission; AVM size (categorized as small
160 patients admitted to the Department of Neurosurgery at
[⬍2.5 cm in diameter], medium [2.5–5.0 cm], or large [⬎5 cm]); infra- or
Helsinki University Central Hospital between 1942 and 1975 supratentorial location and deep or superficial location; pattern of
(15). We have now extended that series to include 238 patients venous drainage (categorized as superficial, deep, or both superficial
admitted between 1942 and 2005 and applied Kaplan-Meier and deep); and interactions between rupture status, size, location, and
life table analyses and Cox proportional hazards regression venous drainage pattern. Because of the observation that annual rupture
models to define the annual risk of hemorrhage and risk factors rates were highest during the first few years after diagnosis, log-rank
for hemorrhage. tests and Cox regression analyses were performed not only for the
whole follow-up period but also for only the first 5 years. A 2-tailed
P value of less than 0.05 was considered statistically significant.
Patients RESULTS
All 631 consecutive patients with AVMs who were admitted to the All Patients
Department of Neurosurgery at Helsinki University Central Hospital
(current catchment population of almost 2 million people) from 1942 to Of the 631 AVM patients admitted between 1942 and 2005, 393
2005 were identified. This department was the only neurosurgical clinic patients had less than 1 month of hemorrhage- and treatment-
in Finland until the late 1960s. A diagnosis of AVM was based on angiog- free follow-up time: 7 had died, 1 had recurrent bleeding, 2 were
raphy or histology. Patients with spinal AVMs, dural arteriovenous fis- discharged abroad, and treatment was initiated in 383 patients
tulae, caroticocavernous fistulae, and vein of Galen aneurysms were not within the first month after admission. Thus, 238 AVM patients
included in the study. Size, location, and angioarchitecture of AVMs were included in further analyses (see Table 1 for details). More
were evaluated from angiography, computed tomography, or magnetic than three-fourths (182 patients, 77%) of patients had been
resonance imaging studies. Because a detailed characterization of various admitted before the end of 1980s. During the same period, how-
types of associated aneurysms (incidental, flow related, intranidal, or ever, only 54% (199 of 368) of all patients admitted before the end
venous) was not readily available for all older cases, we decided not to
of 1980s had received any treatment for their AVM, demonstrat-
include the existence of aneurysms in the statistical analysis. Anterior
and posterior paracallosal, basal ganglia, intraventricular, trigonal, tem-
ing the rather conservative treatment policy of the period from
poromesial, pontomesencephalic, and deep cerebellar AVMs were con- which most of the follow-up data have been gathered. The mean
sidered deep and all other locations superficial. Pontomesencephalic and follow-up period was 13.5 years (range, 1 month–53.1 years).
cerebellar AVMs were considered infratentorial and all other locations Complete follow-up was obtained for all but 3 (1.3%) patients.
supratentorial. An AVM was considered ruptured before admission if AVMs had ruptured in 139 patients before admission. The diag-
there were signs of bleeding on a computed tomographic scan or lumbar nosis of previous hemorrhage was based on history alone in 19
puncture or if a patient had a history of severe, sudden headache and patients and was verified by computed tomography, magnetic
bleeding was not ruled out. The severity of an AVM rupture was esti- resonance imaging, lumbar puncture, or operative findings in
mated from hospital records when possible and scored according to the the remainder of the patients. During the total follow-up period
classification of Hunt and Hess (10), in which increasing severity corre-
of 3222 person-years, 77 patients experienced a hemorrhage from
sponds to increasing grade (Grades 1–5). All available follow-up data
were collected starting from the admission to a neurosurgical referral
AVM, yielding an overall annual rupture rate of 2.4%. The sever-
center until death or the end of the year 2005. The study was approved ity of hemorrhage on the Hunt and Hess scale was Grade 1 in 2
by the ethical committee of Helsinki University Central Hospital. patients (3%), Grade 2 in 23 (30%), Grade 3 in 13 (17%), Grade 4
in 8 (10%), Grade 5 in 24 (31%), and could not been reliably esti-
Statistical Methods mated retrospectively in 7 (9%) patients. The risk of new hemor-
Statistical analysis was performed by two authors (AL, SJ) using SPSS rhagic event was highest during the first few years after diagno-
software (version 13.0; SPSS Inc., Chicago, IL). For the purpose of life sis. The annual rupture rate was almost 3 times higher during
table analyses and the Cox proportional hazards regression model, the first 5 years (4.6%) than thereafter (1.6%) (Fig. 1A; Table 2).

824 | VOLUME 63 | NUMBER 5 | NOVEMBER 2008


TABLE 1. Characteristics of patients according to the occurrence TABLE 1. Continued

of arteriovenous malformation rupture during follow-upa No rupture Rupture All patients
No rupture Rupture All patients (n ⴝ 161) (n ⴝ 77) (n ⴝ 238)
(n ⴝ 161) (n ⴝ 77) (n ⴝ 238) Deep cerebellar 1 (0.6) 1 (1.3) 2 (0.8)
Age at admission (years), mean ⫾ SDb hemispheric
34.4 ⫾ 15.1 29.3 ⫾ 14.4 32.8 ⫾ 15.0 Multiple 1 (0.6) 1 (1.3) 2 (0.8)
Sex, no. (%) Supra- or infratentorial AVM, no. (%)
Female 62 (39) 35 (46) 97 (41) Supratentorial 151 (94) 67 (87) 218 (92)
Male 99 (62) 42 (55) 141 (59) Infratentorial 9 (6) 9 (12) 18 (8)
AVM rupture before admission, no. (%)c Not classifiable 1 (1) 1 (1) 2 (1)
Yes 79 (49) 60 (78) 139 (58) Superficial or deep AVM, no. (%)d
No 82 (51) 17 (22) 99 (42) Superficial 122 (76) 48 (62) 170 (71)
AVM location, no. (%) Deep 38 (24) 28 (36) 66 (28)
Frontal 35 (21.7) 8 (10.4) 43 (18.1) Not classifiable 1 (1) 1 (1) 2 (1)
Parasylvian 12 (7.5) 3 (3.9) 15 (6.3) AVM size, no. (%)
Temporal 18 (11.2) 6 (7.8) 24 (10.1) Small (⬍25 mm) 62 (39) 31 (33) 88 (37)
Rolandic 20 (12.4) 7 (9.1) 27 (11.3) Medium (25–50 mm) 70 (44) 26 (34) 96 (40)
Parietal 21 (13.0) 12 (15.6) 33 (13.9) Large (⬎50 mm) 26 (16) 21 (27) 47 (20)
Occipital 10 (6.2) 9 (11.7) 19 (8.0) Not known 3 (2) 4 (5) 7 (3)
Anterior paracallosal 3 (1.9) 4 (5.2) 7 (2.9) Venous drainage, no. (%)
Posterior paracallosal 7 (4.3) 2 (2.6) 9 (3.8) Cortical and deep 33 (21) 9 (12) 42 (18)
Basal ganglia 16 (9.9) 10 (13.0) 26 (10.9) Cortical 84 (52) 38 (49) 122 (51)
Intraventricular 1 (0.6) 4 (5.2) 5 (2.1) Deep 40 (25) 24 (31) 64 (27)
Trigonal 3 (1.9) 1 (1.3) 4 (1.7) Not known 4 (3) 6 (8) 10 (4)
Temporomesial 4 (2.5) 1 (1.3) 5 (2.1) Follow-up
Whole hemispheric 1 (0.6) 0 (0) 1 (0.4) Total person-years 2489 734 3222
Pontomesencephalic 3 (1.9) 5 (6.5) 8 (3.4) Median (range), years 9.3 (0.1–53.1) 5.0 (0.1–41.8) 7.4 (0.1–53.1)
Cerebellar vermis 2 (1.2) 1 (1.3) 3 (1.3) a
SD, standard deviation; AVM, arteriovenous malformation.
Superficial cerebellar 3 (1.9) 2 (2.6) 5 (2.1) b
P ⫽ 0.010, Mann-Whitney U test.
hemispheric c
P ⬍ 0.001, Pearson’s χ2 test.
P ⬍ 0.05, Pearson’s χ2 test.

Univariate Analyses of Risk Factors for Rupture almost 12% during the first 5 years in infratentorial AVMs
The characteristics of patients who experienced an AVM rup- (Table 2). Similarly, cumulative rupture rates were very vari-
ture during the follow-up period are compared with those who able, ranging from 18% in 20 years in AVMs with both cortical
remained rupture-free in Table 1. Patients with an AVM rupture and deep venous drainage to 76% in 20 years in infratentorial
during the follow-up period were significantly younger at the AVMs (Table 2).
time of admission and were more likely to have a previously Univariate Cox regression analysis (Table 3) revealed that pre-
ruptured AVM and a deeply located AVM (Table 1). vious rupture, infratentorial location, and deep location were
Previous rupture, infratentorial location, and deep location associated with a significantly increased relative risk (approxi-
were associated with significantly higher AVM rupture rates mately 2- to 2.5-fold) for AVM rupture during the entire fol-
during the entire follow-up period than previously unruptured, low-up period. Exclusively deep venous drainage increased the
supratentorial, or superficially located AVMs, respectively (Fig. relative risk significantly more than sixfold for the first 5 years
1, B, D, and E; Table 2). Exclusively deep venous drainage but not for the entire follow-up period. Age, sex, and AVM size
increased the AVM rupture rate significantly only during the did not affect rupture risk in univariate analysis.
first 5 years, but not during the entire follow-up period (Fig. 1F;
Table 2). Sex and AVM size did not significantly affect the rup- Multivariate Analyses of Risk Factors for Rupture
ture rate (Fig. 1C; Table 2). Annual rupture rates varied sub- We performed multivariate Cox regression analyses to define
stantially, depending on stratifying factors and time point, from independent risk factors for AVM rupture using 3 different mod-
1.0% later than 5 years from admission in small AVMs to els (Table 3). First, we used a forward stepwise procedure to test all



ture, infratentorial location, deep

A B location, and large size when the
entire follow-up period was
In the second model, we
entered all the same variables
and calculated adjusted rela-
tive risks for all of them with-
out removing any variables by
the stepwise procedure. In this
model, only large size was
identified as a significant in-
dependent risk factor during
the first 5 years and previous
rupture, infratentorial location,
and large size during the entire
C D follow-up period.
In the third model, we again
used a forward stepwise proce-
dure but also introduced inter-
actions between all other vari-
ables except sex and age into the
model. This model did not pro-
duce any significant independ-
ent risk factors when only the
first 5 years were analyzed, but
previous rupture, infratentorial
location, large size, and interac-
tion between deep location and
previous rupture were all signif-
icant independent risk factors
during the entire follow-up
period. Interaction between
deep location and previous rup-
ture increased the relative risk
of rupture more than fourfold in
this model, suggesting that this
combination makes the AVM
especially hazardous.

Our natural history study
has the longest follow-up per
AVM patient ever published
and only modest selection
FIGURE 1. Kaplan-Meier curves demonstrating cumulative rates of arteriovenous malformation rupture as the func-
bias. Although patient selec-
tion of follow-up time in years: all patients together (A) and patients stratified by previous rupture (B), AVM size tion to active treatment may
(C), superficial or deep location (D), supra- or infratentorial location (E), and pattern of venous drainage (F). See introduce some bias influenc-
Table 2 for the number of patients in each stratum. BAVM, brain arteriovenous malformation; vs., versus. ing the results (i.e., the least
complex lesions are most
likely to be treated early, lead-
variables listed in Table 3, excluding interactions between vari- ing to shorter natural history follow-up time), there are 2 fac-
ables. According to this model, previous rupture and deep location tors in the current study that will perhaps decrease the impact
remained in the model as statistically significant independent risk of this bias. First, most of the follow-up time is from the period
factors for AVM rupture during the first 5 years and previous rup- before the 1990s when the treatment policy was rather conser-

826 | VOLUME 63 | NUMBER 5 | NOVEMBER 2008


TABLE 2. Annual and cumulative rupture rates in relation to previous rupture, supra- or infratentorial location, superficial or deep location,
arteriovenous malformation size, and pattern of venous drainagea
Cumulative rupture rates,
Annual rupture rates (%) Log-rank P values
% (95% CI)
No. of
Characteristic 0–5 years ⬎ 5 years Whole 5 years 20 years First 5 years Entire
after after follow-up after after after follow-up
admission admission period admission admission admission period
All patients 238 4.7 1.6 2.4 21 (15–27) 39 (32–47)
Sex 0.265 0.250
Male 141 4.0 1.5 2.1 18 (11–25) 37 (27–47)
Female 97 5.8 1.7 2.8 25 (15–35) 43 (31–66)
Previous rupture 0.011 0.016
Ruptured 139 6.2 1.7 2.8 26 (19–34) 45 (27–63)
Unruptured 99 2.3 1.3 1.6 10 (3–17) 29 (16–42)
Supra- or infratentorial AVM 0.023 0.008
Supratentorial 218 4.3 1.5 2.2 19 (13–25) 37 (29–45)
Infratentorial 18 11.6 3.6 6.7 45 (18–72) 76 (51–100)
Superficial or deep AVM 0.003 0.003
Superficial 170 3.5 1.4 1.9 16 (10–22) 35 (27–44)
Deep 66 8.9 2.2 4.1 35 (22–49) 53 (38–67)
AVM size 0.807 0.220
Small 88 5.0 1.0 1.9 22 (12–32) 33 (21–45)
Medium 96 4.2 1.6 2.3 17 (9–26) 38 (25–51)
Large 47 5.5 2.7 3.5 24 (11–36) 52 (35–69)
Venous drainage 0.013 0.111
Cortical and deep 42 1.2 1.9 1.7 5 (0–13) 18 (3–33)
Cortical 122 4.5 1.4 2.1 20 (12–28) 38 (29–47)
Deep 64 8.1 1.6 3.4 34 (20–48) 52 (37–68)

AVM, arteriovenous malformation; CI, confidence interval.

vative compared with today’s standards, and, therefore, study by Ondra et al. (15) had a longer follow-up period, but
patients with a wide variety of lesions were followed untreated the authors did not stop the follow-up at the first bleeding
for a rather long period (often years). Second, even if the distri- event as should have been done for life table and annual rup-
bution of various demographic and anatomic factors in the ture rate analyses.
study cohort is influenced by the exclusion of certain patients We identified several risk factors predicting AVM rupture,
because of treatment, multivariate models were used to dissect including young age, previous rupture, large size, deep and
the independent effects of these factors. Its results indicate that infratentorial location, and exclusively deep venous drainage. The
previously ruptured, large, and infratentorially and deeply most consistent risk factor was previous rupture because it pre-
located AVMs have the highest risk of subsequent hemorrhage. dicted future rupture in practically all models. The highest rupture
This risk is highest during the first few years after diagnosis rate, almost 12% per year, was observed in infratentorial AVMs
and decreases thereafter. during the first 5 years after diagnosis, whereas the highest rela-
Management of patients with AVMs requires reliable knowl- tive risk, more than sixfold, was associated with AVMs with exclu-
edge of the natural course of the disease. Only with sufficient sively deep venous drainage during the same period. According
understanding of these complex lesions can treatment-associ- to multivariate analyses, previous rupture, large size, and infraten-
ated risks be weighed against the prognosis, leading to deci- torial and deep locations were independent risk factors. Moreover,
sions that will be beneficial for the patient. With a total follow- interaction between deep location and previous rupture was an
up time of 3222 person-years (mean, 13.5 person-years per independent risk factor, with a more than fourfold relative risk.
patient), our study is the most extensive natural history series The comparison of our results with those published by
of AVM patients analyzed using Kaplan-Meier life table analy- Ondra et al. (15) is inevitable because their patients are also
sis and Cox multivariate modeling published to date. The included in our series. Two important differences are readily



notable. They reported a higher overall annual

rupture rate (4% compared with 2.4%), and they
TABLE 3. Relative risk and 95% confidence interval associated with age, sex, previous rupture, supra- or infratentorial location, deep or superficial location, arteriove-

2.02 (1.08–3.78)b 2.63 (1.43–4.84)d

3.51 (1.73–7.15)d 3.30 (1.63–6.66)d

3.07 (1.37–6.87)c 2.89 (1.43–4.84)c

4.21 (1.19–14.9)b
1.65 (0.90–3.05)
Model 3 did not observe a significant difference in rup-
ture rate between patients with previously rup-


tured and unruptured AVMs as we did. The first
discrepancy is easily explained by different sta-
tistical methods. Instead of using Kaplan-Meier
Relative risk during the whole period (95% CI)

life table analyses and following the patients

0.99 (0.97–1.01)
0.76 (0.46–1.26)

2.10 (0.90–4.91)

1.74 (0.93–3.27)

2.04 (0.82–5.13)
1.67 (0.67–4.15)
only until the first bleeding, they simply divided

Not tested
Model 2

all observed AVM ruptures (often multiple, as


many as 12 in one patient) by the total observa-
tion time. The second discrepancy is much more
difficult to explain, but is also likely related to
methodological differences. In our series, previ-
ous rupture predicted bleeding consistently and

3.13 (1.55–6.30)d
1.92 (1.12–3.29)b 2.23 (1.23–4.05)c

2.48 (1.23–5.01)b 2.89 (1.30–6.43)c

1.83 (1.08–3.10)

1.80 (0.98–3.32)

proved to be a significant risk factor without

Not tested
Model 1

doubt. Still, even though previous rupture has


also been recognized as a risk factor by several

other groups (3, 4, 9, 13, 19, 22), this finding is not
shared by every available study (7, 14, 20). One
methodological limitation of the current study is
0.99 (0.97–1.01)
0.77 (0.49–1.21)

1.99 (1.24–3.17)

1.14 (0.66–1.97)
1.65 (0.92–2.94)

1.38 (0.67–2.87)
2.08 (0.97–4.49)

that in some historical patients (19 of 139), the

nous malformation size, and venous drainage pattern in predicting arteriovenous malformation rupturea


Not tested

diagnosis of AVM rupture before admission was



based on history alone and was not verified by

P ⬍ 0.005. imaging or lumbar puncture.
P ⬍ 0.05.
P ⬍ 0.01.

Some AVMs are known to increase in size over

the years, whereas, on rare occasions, some oth-


ers may thrombose and disappear spontaneously

3.20 (1.26–8.14)b

(16, 21). Size assessment in our series was based

0.99 (0.97–1.02)
0.66 (0.34–1.29)
2.09 (0.86–5.05)

2.68 (0.98–7.32)
2.57 (0.76–0.86)

1.40 (0.60–3.27)

4.92 (0.98–24.7)
3.83 (0.78–18.8)

CI, confidence interval; Model 1, Cox proportional hazards model with a forward stepwise regression
procedure; Model 2, adjusted model with main effects; Model 3, Model 1 including interactions;

on the first angiography done after admission.

Not tested
Model 2

Large AVM size was a very consistent independ-


Relative risk during the first 5 years (95% CI)

ent risk factor for future bleeding in all multi-

variate analysis models. According to various
Cox regression models, AVMs larger than 5 cm in
diameter had approximately 3- to 3.5-fold rela-
2.44 (1.06–5.62)b

2.04 (1.04–3.99)

tive risk compared with AVMs smaller than 2.5

Not tested

cm. However, size did not affect rupture risk sig-

Model 1

nificantly in univariate analyses. This probably

reflects the fact that small AVMs, conversely, are

more likely to present with hemorrhage (12, 18,
22), probably because they are less likely than
large ones to become symptomatic without
2.76 (1.22–6.26)b

2.65 (1.11–6.32)b

6.42 (1.48–27.9)b
0.98 (0.96–1.01)
0.70 (0.37–1.31)

2.50 (1.32–4.71)

0.85 (0.40–1.78)
1.10 (0.50–2.46)

3.59 (0.84–15.4)

bleeding. Patients with small AVMs may, there-


Not tested

fore, tend to experience proportionally more



AVM ruptures, not because small size would be

a risk factor in itself, but through the association
with previous rupture. This phenomenon may
AVM, arteriovenous malformation.

prevent the effect of large AVM size on the rup-

ture rate to be visible unless multivariate models
Cortical and deep

are used. This observation also elegantly demon-

Venous drainage
Previous rupture

previous rupture

Deep location*

strates the difference between factors associated

Age (per year)

with hemorrhagic presentation and true risk fac-


AVM size
Male sex





tors predicting future hemorrhage. The fact that

small AVM size predicts hemorrhagic presenta-
tion but, in fact, significantly decreases the risk of
bleeding also suggests that a significant propor-

828 | VOLUME 63 | NUMBER 5 | NOVEMBER 2008


tion of small unruptured AVMs remain undetected and undiag- REFERENCES

nosed for a lifetime. Other groups (8, 14, 20) have also found
1. Brown RD Jr, Wiebers DO, Forbes G, O’Fallon WM, Piepgras DG, Marsh
large AVM size to be a risk factor for rupture, whereas others WR, Maciunas RJ: The natural history of unruptured intracranial arteriove-
(1, 3, 9, 13, 19, 22) have not found size to have a predictive nous malformations. J Neurosurg 68:352–357, 1988.
value. It is noteworthy, however, that none of the AVM studies 2. Choi JH, Mohr JP: Brain arteriovenous malformations in adults. Lancet
using Kaplan-Meier analysis and Cox models have suggested Neurol 4:299–308, 2005.
3. Crawford PM, West CR, Chadwick DW, Shaw MD: Arteriovenous malforma-
that small AVM size would predict AVM rupture.
tions of the brain: Natural history in unoperated patients. J Neurol
Deep and infratentorial locations were independent predic- Neurosurg Psychiatry 49:1–10, 1986.
tors of future bleeding as well. These findings have also been 4. Forster DM, Steiner L, Håkanson S: Arteriovenous malformations of the
reported by others (6, 14, 19, 20, 22). The explanation for this brain. A long-term clinical study. J Neurosurg 37:562–570, 1972.
phenomenon may be related to perforator involvement in the 5. Friedlander RM: Arteriovenous malformations of the brain. N Engl J Med
356:2704–2712, 2007.
angioarchitecture of AVMs in these locations, rendering the 6. Fults D, Kelly DL Jr: Natural history of arteriovenous malformations of the
feeding vessels less tolerant to high blood flow. Likewise, exclu- brain: A clinical study. Neurosurgery 15:658–662, 1984.
sively deep venous drainage may increase the rupture rate by 7. Graf CJ, Perret GE, Torner JC: Bleeding from cerebral arteriovenous malfor-
increasing the pressure gradient in the AVM nidus compared mations as part of their natural history. J Neurosurg 58:331–337, 1983.
8. Guidetti B, Delitala A: Intracranial arteriovenous malformations. Conservative
with AVMs with more extensive venous drainage systems.
and surgical treatment. J Neurosurg 53:149–152, 1980.
However, deep venous drainage was not an independent risk 9. Halim AX, Johnston SC, Singh V, McCulloch CE, Bennett JP, Achrol AS,
factor in our series (although it has been in some others [13, Sidney S, Young WL: Longitudinal risk of intracranial hemorrhage in patients
19]), and it is possible that it increases the risk only through its with arteriovenous malformation of the brain within a defined population.
association with the deep AVM location. Stroke 35:1697–1702, 2004.
10. Hunt WE, Hess RM: Surgical risk as related to time of intervention in the
In our series, patients who experienced an AVM rupture dur- repair of intracranial aneurysms. J Neurosurg 28:14–20, 1968.
ing the follow-up period were significantly younger on admis- 11. Itoyama Y, Uemura S, Ushio Y, Kuratsu J, Nonaka N, Wada H, Sano Y,
sion (5 years on average) than those who did not. However, Fukumura A, Yoshida K, Yano T: Natural course of unoperated intracranial
age did not have explanatory power in univariate and multi- arteriovenous malformations: Study of 50 cases. J Neurosurg 71:805–809,
variate Cox models. In contrast to our results, Stapf et al. (19)
12. Langer DJ, Lasner TM, Hurst RW, Flamm ES, Zager EL, King JT Jr:
reported that increasing age was associated with the risk of rup- Hypertension, small size, and deep venous drainage are associated with risk
ture, whereas Yamada et al. (22) found children to be at of hemorrhagic presentation of cerebral arteriovenous malformations.
increased risk. Sex did not affect risk at all in our study, whereas Neurosurgery 42:481–489, 1998.
both male and female sex have been reported by others (13, 22) 13. Mast H, Young WL, Koennecke HC, Sciacca RR, Osipov A, Pile-Spellman J,
Hacein-Bey L, Duong H, Stein BM, Mohr JP: Risk of spontaneous haemor-
to increase the risk of rupture. At present, we do not have an rhage after diagnosis of cerebral arteriovenous malformation. Lancet
explanation for these discrepancies. 350:1065–1068, 1997.
The rupture rates seemed to be consistently higher during 14. Mine S, Hirai S, Ono J, Yamaura A: Risk factors for poor outcome of untreated
the first few years after diagnosis, regardless of stratifying fac- arteriovenous malformation. J Clin Neurosci 7:503–506, 2000.
15. Ondra SL, Troupp H, George ED, Schwab K: The natural history of sympto-
tors. A similar pattern has been observed by some other groups
matic arteriovenous malformations of the brain: A 24-year follow-up assess-
with sufficiently long follow-up periods (6, 11, 22), but even ment. J Neurosurg 73:387–391, 1990.
opposite findings have been reported (1). The reason for this 16. Pasqualin A, Vivenza C, Rosta L, Scienza R, Da Pian R, Colangeli M:
phenomenon is unknown, but it is interesting to note that only Spontaneous disappearance of intracranial arteriovenous malformations.
10 AVMs in our series were asymptomatic and incidental, and Acta Neurochir (Wien) 76:50–57, 1985.
17. Pellettieri L, Carlsson CA, Grevsten S, Norlén G, Uhlemann C: Surgical ver-
the appearance of symptoms ultimately leading to diagnosis sus conservative treatment of intracranial arteriovenous malformations: A
may have coincided with some hemodynamic changes within study in surgical decision-making. Acta Neurochir Suppl (Wien) 29:1–86,
these lesions, making them more rupture prone during that 1979.
period in the patients’ lives. Nevertheless, this observation has 18. Spetzler RF, Hargraves RW, McCormick PW, Zabramski JM, Flom RA,
Zimmerman RS: Relationship of perfusion pressure and size to risk of hem-
important implications for the management of AVMs that are
orrhage from arteriovenous malformations. J Neurosurg 76:918–923, 1992.
known to have been rupture-free for a long time. 19. Stapf C, Mast H, Sciacca RR, Choi JH, Khaw AV, Connolly ES, Pile-Spellman
In conclusion, results from this extensive natural history fol- J, Mohr JP: Predictors of hemorrhage in patients with untreated brain arteri-
low-up study in AVM patients to date indicate that previously ovenous malformation. Neurology 66:1350–1355, 2006.
ruptured, large, and infratentorially and deeply located AVMs 20. Stefani MA, Porter PJ, terBrugge K, Montanera W, Willinsky RA, Wallace C:
Large and deep brain arteriovenous malformations are associated with risk
have the highest risk of future hemorrhage. This risk is highest of future hemorrhage. Stroke 33:1220–1224, 2002.
during the first few years after diagnosis and decreases there- 21. Waltimo O: The change in size of intracranial arteriovenous malformations.
after, but remains significant for decades. Our results will be of J Neurol Sci 19:21–27, 1973.
significant value for management decisions concerning patients 22. Yamada S, Takagi Y, Nozaki K, Kikuta KI, Hashimoto N: Risk factors for sub-
sequent hemorrhage in patients with cerebral arteriovenous malformations.
with these highly complex lesions.
J Neurosurg 107:965–972, 2007.

Disclosures Acknowledgment
This work was supported by the Maire Taponen Foundation, the We thank Annika Kytölä, R.N., for assistance in gathering follow-up data
Paavo Nurmi Foundation, and the Neurological Foundation of Finland. from the patients.



COMMENTS Undoubtedly, even complete understanding of these issues will still

fall short of an ideal treatment algorithm, as advanced neuroimaging
H ernesniemi et al. evaluated 631 patients between 1942 and 2005.
They excluded patients with less than 1 month of hemorrhage-
and treatment-free follow-up, resulting in a group of 238 patients. The
studies continue to identify more and more intranidal aneurysms and
subtle venous outflow obstruction, which undoubtedly contribute to
natural history risks and certainly could not be identified by imaging
mean follow-up duration was 13.5 years (range, 1 month–53.1 years).
studies from the mid-20th century. Follow-up studies have contributed
They report an annual rupture rate of 2.4% (77 patients during a total
to our understanding of the natural history of these complex lesions. To
of 3222 person-years). Univariate analyses of risk factors for rupture
revealed that patients who experienced rupture during follow-up were make matters even more complex, neurosurgeons are faced with an
younger, were more likely to have previously ruptured arteriovenous expanding armamentarium of weapons to treat these lesions. At the
malformations (AVM), and were more likely to have a deeply located very least, the new knowledge communicated in this report will assist
AVM. Additionally, deep venous drainage increased the AVM rupture the neurosurgeon and patient regarding the most important and diffi-
rate during the first 5 years only. An increased relative risk (2- to 2.5- cult decision they face: to treat or not to treat.
fold increase) was seen for previously ruptured AVMs and AVMs in Daniel Surdell
infratentorial locations and deep locations, with a 6-fold increased rel- Stephen L. Ondra
ative risk for exclusively deep venous drainage. Multivariate analyses H. Hunt Batjer
found that, when the follow-up period was analyzed as a whole, pre- Chicago, Illinois
vious rupture, infratentorial location, deep location, and large size were
independent risk factors. Additionally, multivariate analysis revealed
that the interaction between deep location and previous rupture was an 1. Ondra SL, Troupp H, George ED, Schwab K: The natural history of sympto-
independent risk factor associated with a 4-fold relative risk of rupture. matic arteriovenous malformations of the brain: A 24-year follow-up assess-
ment. J Neurosurg 73:387–391, 1990.
As the authors point out, we now have several long-term natural his-
2. Stapf C, Mast H, Sciacca RR, Choi JH, Khaw AV, Connolly ES, Pile-Spellman
tory studies evaluating the behavior of intracranial AVM. There are
J, Mohr JP: Predictors of hemorrhage in patients with untreated brain arteri-
some common themes but also conflicting data, as is carefully dis- ovenous malformation. Neurology 66:1350–1355, 2006.
cussed in this article. Although statistical methodologies are obviously
important, it is possible that the original series of Ondra et al. (1) had
a higher annual bleeding rate because only symptomatic patients were
included in the analysis. We would presume that the more contempo-
H ernesniemi et al. report the result of a retrospective survival study
in 631 consecutive patients with AVM admitted to the Department
of Neurosurgery in Helsinki University Hospital between 1942 and 2005.
rary patients are a mixture of symptomatic and asymptomatic patients.
We published a study on the natural history of AVMs in 2007 (1), and the
In the series of Ondra et al. (1), patients were enrolled and had the
results of the 2 studies can be compared. The study by Hernesniemi et al.
majority of follow-up in the pre-computed tomography era. The abil-
found a 2.4% annual risk of hemorrhage from AVMs. In our recent study,
ity to distinguish seizure from hemorrhage was based on angiography,
performed at Kyoto University, we reported that the annual risk of hem-
if any study was performed. This difference in imaging capability and
orrhage was 6.84% in the hemorrhagic group and 3.12% in nonhemor-
the need to make assumptions as to the cause of symptoms is likely the
rhagic group (1). One explanation for the lower percentage found by
reason for any discrepancy in the rate of bleeding between this study
Hernesniemi et al. may be the different statistical methods used, as
and the study by Ondra et al. (1). The discrepancy between the current
described in their Discussion. Another reason is the difference in the fol-
report and the data of Stapf et al. (2) regarding age differences is more
low-up period. The mean follow-up period in their study was longer
difficult to understand.
than in our study, and this characteristic may affect the results.
This report contains the most extensive natural history data ever
Risk factors predicting subsequent hemorrhage were reported to be
assembled for brain AVMs and represents a major contribution to the lit-
previous rupture, large size, infratentorial location, and deep location
erature. The variances between these data and prior reports, as refer-
in the study by Hernesniemi et al. We reported that risk is higher for
enced above, are critical to dissect so that we will ultimately be able to
children, females, and patients with deep-seated lesions (1). Like the
advise our patients with more precision. The actual rate of hemorrhage
authors, we also do not have a definite explanation for these differ-
likely lies between the results of the different published studies.
ences. In the 2 studies, deep-seated AVMs have an equal tendency
Differences in lesions and selection bias no doubt account for some of
toward subsequent hemorrhage. Six of 9 previous studies on the natu-
the study variability. This study is the largest and statistically most
ral history of hemorrhagic AVMs reported that deep location increased
sophisticated yet published. When the article by Ondra et al. (1) was
the risk of hemorrhage, as discussed in our article (1).
published, the authors were well aware of the limits of the statistical
The important characteristic of this report was that the mean follow-
methods that were available. Of even more concern were the limits of
up period was extremely long (13.5 yr). As the authors indicate,
imaging in the era of study enrollment. This made assessment of hem-
Finland is known to be well suited for population-based epidemiolog-
orrhage much more difficult. The current study is limited by a small, but
ical studies. Thus, the authors could investigate the patients for a long
real, selection bias and very short follow-up in some patients. The fact
follow-up period. In sum, this is a very valuable report for understand-
that treatment intervention often occurred after the first repeat hemor-
ing the natural history of cerebral AVMs.
rhage has a major effect on the ability to assess the risk of repeat hem-
orrhage. This approach necessitates statistical methods and assump- Yasushi Takagi
tions necessary for establishing the rate of subsequent hemorrhage as Kyoto, Japan
opposed to following the actual rate. Having said this, the current study Nobuo Hashimoto
does provide the data that are most appropriate for decisions on treat- Osaka, Japan
ment today, given the technology available. It is difficult to imagine
any center or physician following patients through multiple repeated
hemorrhages and neurological decline. This was an appropriate strategy 1. Yamada S, Takagi Y, Nozaki K, Kikuta K, Hashimoto N: Risk factors for sub-
during the original enrollment for the study by Ondra et al., (1) given sequent hemorrhage in patients with cerebral arteriovenous malformations.
the technology available at the time. It would not be appropriate today. J Neurosurg 107:965–972, 2007.

830 | VOLUME 63 | NUMBER 5 | NOVEMBER 2008


T his study is an important contribution to the brain AVM literature.

Notably, it is an expansion of the patient cohort, both in size and
follow-up period, of the often cited study by Ondra et al. (1), pub-
For the first time, in a prospective fashion, the authors have shed
light on the issues of deep versus superficial location and rupture,
brain AVM size and rupture, and type of venous drainage and rupture,
lished in 1990, on the natural history of symptomatic AVMs. The as well as infratentorial versus supratentorial location and associated
authors acknowledge that comparison of the 2 studies is inevitable, and rupture rates. This study will certainly be of benefit when patients are
their comments on the differences are interesting. The study confirms encountered in the office, because, although this type of stratification
many of the known risk factors that predict rehemorrhage of AVMs, has long been suspected as affecting rates of hemorrhage, the current
such as previous rupture, infratentorial and deep location, large AVM study documents the degree of stratification.
size (all significant in the multivariate analysis), and deep venous
Christopher S. Ogilvy
drainage (significant in the univariate analysis). It is important to note
Boston, Massachusetts
that other studies support lower rates of rebleeding in large and deep
AVMs, but the cohort size and length of follow-up in the Helsinki AVM
database adds strength to the contrary.
Notably, the annual hemorrhage risk from AVM was 2.4%, with the
I n an effort to further define the risk of hemorrhage related to AVMs,
the authors present a comprehensive longitudinal study of patients
diagnosed with brain AVMs over a period of some 60 years. Taking
highest risk during the first 5 years after diagnosis (4.6%), compared
advantage of a relatively unique sociological opportunity, they accumu-
with more than 5 years after diagnosis (1.6%). A 12%/y rupture rate
lated more than 3000 patient-years of follow-up data on almost 300
was found for infratentorial AVMs during the first 5 years after diag-
patients and then examined the data as rigorously as is consistent with
nosis (by univariate analysis) and a 6-fold risk of bleeding was associ-
the vagaries of such patient populations. The study is burdened with
ated with exclusively deep venous drainage over the same time period
numerous selection and statistical caveats (most of which the authors
(by univariate analysis). In 1 of the 3 statistical multivariate models
point out themselves), and none of the findings are earth-shattering: the
used, the combination of deep location and previous rupture increased
truest predictors of hemorrhage were previous hemorrhage, young age
the relative rupture rate by 4-fold, suggesting that patients with these
at presentation, deep/infratentorial location, isolated deep venous
AVM characteristics are particularly prone to bleeding.
drainage, etc., and the highest risk of bleeding occurred in the initial 5
As appropriately mentioned by the authors, a limitation of the
years after diagnosis. Nonetheless, in this work, the Helsinki group has
study is that in 19 of the 139 patients whose AVMs subsequently bled,
made a major contribution toward consolidating our understanding of
diagnosis of AVM rupture was based only on history, without imag-
AVM natural history and thereby clarifying appropriate therapeutic
ing or lumbar puncture confirmation. Nonetheless, this study repre-
risks and indications.
sents a major contribution to our understanding of the natural history
of AVMs. Duke S. Samson
Dallas, Texas
Marco Lee
Gary K. Steinberg
Stanford, California T he authors present a study regarding the natural history of cerebral
AVMs that is commendable for both its large size and the length of
follow-up. They followed 238 patients with untreated AVMs for a
mean of 13.5 years to determine rates of AVM rupture and risk factors
1. Ondra SL, Troupp H, George ED, Schwab K: The natural history of sympto-
matic arteriovenous malformations of the brain: A 24-year follow-up assess- for hemorrhage in untreated patients. The annual rupture rate was
ment. J Neurosurg 73:387–391, 1990. 2.4%. Previous rupture, large AVM size, and infratentorial and/or
deep locations were identified as independent risk factors for hemor-
rhage in a multivariate analysis that accounted for potential confound-
T he authors performed a long-term follow-up in patients with brain
AVMs. They had a unique opportunity to follow patients over an
extended interval, given their geographic location and health care sys-
ing variables.
This study represents one of the largest studies of AVM natural his-
tory published to date. In terms of total patient-years of follow-up, it is
tem. The authors studied a large number of patients with a significant
the most extensive study. Despite its few limitations, the wealth of data
interval of follow-up. They identify the annual risk for all AVMs in the
offers insight into the rupture rate of untreated AVMs and has the
study as a hemorrhage rate of 2.4%/y. They performed univariate and
potential to impact clinical decisions, particularly the optimal timing
multivariate analyses and found young age, previous rupture, deep
and modality of treatment.
and infratentorial location, and exclusively deep venous drainage to be
significant in a univariate analysis of subsequent rupture. Previous David Wilson
rupture, large AVM size, and infratentorial and deep locations were Robert F. Spetzler
independent risk factors in the multivariate analysis. Phoenix, Arizona