Int. J. Radiation Oncology Biol. Phys., Vol. 46, No. 5, pp.

1143–1148, 2000
Copyright © 2000 Elsevier Science Inc.
Printed in the USA. All rights reserved
0360-3016/00/$–see front matter

PII S0360-3016(99)00513-1

CLINICAL INVESTIGATION Central Nervous System

DEVELOPMENT OF A MODEL TO PREDICT PERMANENT SYMPTOMATIC

POSTRADIOSURGERY INJURY FOR ARTERIOVENOUS
MALFORMATION PATIENTS

JOHN C. FLICKINGER, M.D.,*† DOUGLAS KONDZIOLKA, M.D.,*†

L. DADE LUNSFORD, M.D.,*†‡ AMIN KASSAM, M.D.,†§ LOI K. PHUONG,M.D.,㛳 ROMAN LISCAK, M.D.,¶
㛳
AND BRUCE POLLOCK, M.D. FOR THE ARTERIOVENOUS MALFORMATION RADIOSURGERY STUDY GROUP

Departments of *Radiation Oncology, †Neurological Surgery, ‡Radiology, and §Biostatistics, University of Pittsburgh School of
Medicine, Pittsburgh, PA; 㛳Department of Neurosurgery, Mayo Clinic, Rochester, MN; and ¶Department of Neurosurgery, Hospital Na
Homolce, Prague, Czech Republic

Purpose: To better predict permanent complications from arteriovenous malformation (AVM) radiosurgery.
Methods and Materials: Data from 85 AVM patients who developed symptomatic complications following
gamma knife radiosurgery and 337 control patients with no complications were evaluated as part of a multi-
institutional study. Of the 85 patients with complications, 38 patients were classified as having permanent
symptomatic sequelae (necrosis). AVM marginal doses varied from 10 –35 Gy and treatment volumes from
0.26 – 47.9 cc. Median follow-up for patients without complications was 45 months (range: 24 –92).
Results: Multivariate analysis of the effects of AVM location and the volume of tissue receiving 12 Gy or more
(12-Gy-Volume) allowed construction of a significant postradiosurgery injury expression (SPIE) score. AVM
locations in order of increasing risk and SPIE score (from 0 –10) were: frontal, temporal, intraventricular,
parietal, cerebellar, corpus callosum, occipital, medulla, thalamus, basal ganglia, and pons/midbrain. The final
statistical model predicts risks of permanent symptomatic sequelae from SPIE scores and 12-Gy-Volumes. Prior
hemorrhage, marginal dose, and Marginal-12-Gy-Volume (target volume excluded) did not significantly improve
the risk-prediction model for permanent sequelae (p > 0.39).
Conclusion: The risks of developing permanent symptomatic sequelae from AVM radiosurgery vary dramati-
cally with location and, to a lesser extent, volume. These risks can be predicted according to the SPIE
location-risk score and the 12-Gy-Volume. © 2000 Elsevier Science Inc.

Radiosurgery, Stereotactic surgery, Arteriovenous malformation, Complications, Radiation injury.

INTRODUCTION affected patients, so that symptomatic postradiosurgery se-

quelae develop only in approximately 9% of patients (4 –7).
Radiosurgery is a highly effective way of reducing the risk
Delayed effects of radiosurgery are difficult to study
of hemorrhage in properly selected patients with cerebral
because of the time needed for their development, the ne-
arteriovenous malformations (AVM) (1–3). Radiosurgery
cessity to distinguish temporary from permanent injury with
induces an injury response in the AVM nidus leading to
additional follow-up, and the variability introduced by dif-
eventual complete obliteration in 60 to 90% of cases, de-
pending upon AVM size, configuration, as well as the ferences in location, treatment volume, and radiation dose
radiosurgery targeting and treatment techniques used (1–3). distributions. Prior studies focused first on the endpoint of
AVM radiosurgery can sometimes unfortunately induce un- combined symptomatic and asymptomatic postradiosurgery
wanted radiation injury in surrounding brain tissue (4 –10). imaging changes without regard to permanence because of
Magnetic resonance (MR) scans show postradiosurgery im- limited data and the greater number of events to study (4, 7).
aging changes in the brain surrounding AVM in approxi- The total volume of tissue receiving 12 Gy or more (includ-
mately 30% of patients, depending on the treatment volume ing the target), termed the “12-Gy-Volume” was found to
and, to a lesser extent, the dose administered (4 –7). Fortu- accurately reflect the risk of developing postradiosurgery
nately, these effects are asymptomatic in two-thirds of the imaging changes (4). The 12-Gy-Volume, which depends

Reprint requests to: John C. Flickinger, M.D., Joint Radiation Sneed, D. Larson, V. Smith, M. W. McDermott, L. Miyawaki
Oncology Center, 200 Lothrop Street, Pittsburgh, PA 15213. E- (U. C. San Francisco), J. Yamamoto (Tokyo), J. Chilton (Midwest
mail: jflickin⫹@[Link] Gamma Knife Center, Kansas City), R. A. Morantz, B. Young (U.
Acknowledgments—The following other members of the Arterio- of Kentucky), H. Jokura (Tohoku U., Sendai, Japan).
venous Malformation Radiosurgery Study Group contributed to Accepted for publication 15 November 1999.
this study: D. A. Gorman, P. J. Schomberg (Mayo Clinic), P.

1143
1144 I. J. Radiation Oncology ● Biology ● Physics Volume 46, Number 5, 2000

on treatment volume and dose, is a parameter that reflected 1994 were included as part of the multicenter AVM Radio-
radiation injury risk only for similar distributions of radia- surgery Study Group analysis of complications outcomes in
tion within a limited range of doses, and was reported with 102 patients (5, 11). That study also included 55 other
a warning not to regard 12 Gy as a true “threshold dose” (4). patients with symptomatic postradiosurgery sequelae from
Analyses of the endpoint of symptomatic postradiosurgery 850 patients who had gamma knife radiosurgery at other
sequelae found that AVM location dramatically affected institutions (11). The previous study also included 14 pa-
whether the postradiosurgery imaging changes were symp- tients who received gamma knife radiosurgery and three
tomatic or not (4, 5, 7). gamma knife treated patients with asymptomatic sequelae
A postradiosurgery imaging expression (PIE) score was (two small cysts and one case of middle cerebral artery
constructed from analysis of the Pittsburgh AVM radiosur- stenosis) who were excluded from this total of 85 patients
gery experience to help predict the development of symp- with complications included in the present study. Data on
tomatic postradiosurgery injury from location and dose/ patients without complications were not submitted as part of
volume which appears best represented by the 12-Gy- the AVM Radiosurgery Study Group study on the outcome
Volume (5). There are shortcomings in using this to predict of radiosurgery complications (11). Complete data on the
radiosurgery complications and to guide therapy. The most complementary set of 302 patients without symptomatic
serious shortcoming is the lack of any differentiation be- postradiosurgery sequelae were available for the University
tween temporary and/or minor sequelae (examples: head- of Pittsburgh patients who were part of that study. These
ache, easily controlled seizures) compared to more serious data (n ⫽ 302) were combined with data on 35 AVM
permanent radiation injuries because of the limited amount radiosurgery patients from Prague (n ⫽ 21) and Mayo
of data in the series. Although avoiding temporary postra- Clinic (n ⫽ 14) who did not develop any complications with
diosurgery sequelae is desirable, unlike permanent symp- ⬎2-yr follow-up. The combined data (n ⫽ 337) served as a
tomatic radiation injury, it would not be worth reducing the representative control group for this study to compare to
chance of obliteration (and taking a greater chance of hem- cases with complications from all institutions (n ⫽ 85). The
orrhage and/or death) by using lower doses to avoid only median follow-up for patients without complications was 45
temporary, minor side effects. Another problem is that months (range: 24 –92). Among the 85 patients who devel-
because of limited data, the analysis of location effects did oped complications, the median follow-up from radiosur-
not simultaneously assess all of the various locations in a gery was 34 months (range: 9 –140). The median follow-up
multivariate analysis. Instead the effects of each location after the onset of postradiosurgery neurological sequelae
were assessed in two-variable models including 12-Gy- was 21 months (range: 2–95). Of the 85 patients with
volume, leading to a univariate assessment of each location complications, 38 were classified as having permanent post-
with respect to other locations. The limitations of the data radiosurgery sequelae (symptomatic radiation necrosis) un-
(only 30 complications with 11 location categories to assess resolved after 2 years of subsequent follow-up.
in different combinations) and the consequent lack of com-
plete multivariate modeling of all locations raises serious
doubts about the accuracy of the PIE model. Despite these Treatment parameters
limitations, the risks of symptomatic sequelae correlated Radiosurgery was performed with a Gamma Knife (Ele-
more closely with the PIE score than any other variable kta Instruments, Atlanta, GA) in all patients. The years of
including the 12-Gy-Volume (which still remained highly treatment were from 1985 to 1996 (median ⫽ 1992). The
predictive). median minimum target dose (Dmin) was 20 Gy (range:
Because permanent radiosurgery complications are infre- 10 –35). The median value for the maximum dose (Dmax)
quent, as well as multifactorial, there are no adequate stud- was 36 Gy (range: 20 – 60). The median treatment volume
ies of what factors determine whether a postradiosurgery was 3.5 cc (range: 0.13– 47.9). The median 12-Gy-Volume,
complication will be temporary or permanent. The partici- which is the total volume of tissue (including the target)
pants of this study pooled data on AVM postradiosurgery receiving 12 Gy or more, was 7.39 cc (range: 0.64 –55.3).
complications to better define their character, and to define The median number of isocenters irradiated was two (range:
factors affecting their resolution. 1–21). Treatment volume inhomogeneity expressed as the
This paper seeks to refine the prediction of AVM radio- maximum-to-minimum-dose ratio varied from 1.11 to 3.33
surgery complications by constructing a mathematical risk (median ⫽ 2.0) (12). The treatment isodose volume (nor-
model with permanent symptomatic radiation injury as the malized to maximum dose) varied from 30 to 90% (medi-
endpoint and a more detailed representation of the effects of an ⫽ 50%). Another factor assessed was the Marginal-12-
location. Gy-Volume, which is the volume of tissue outside the
treatment volume that receives 12 Gy or more (the Margin-
al-12-Gy-Volume equals the 12-Gy-Volume minus the
METHODS AND MATERIALS
treatment volume). The Marginal-12-Gy-Volume should
Data from symptomatic postradiosurgery complications theoretically correlate with complications better than the
in 30/332 AVM patients who received gamma knife radio- 12-Gy-Volume if the reaction of the target volume does not
surgery at the University of Pittsburgh between 1987 and contribute to complications in surrounding normal tissue.
 Prediction model for symptomatic necrosis from AVM radiosurgery ● J. C. FLICKINGER et al. 1145

The Marginal-12-Gy-Volume varied from 0 to 27.4 cc (me- Table 1. Derivation of the significant postradiosurgery injury
dian ⫽ 3.64 cc). expression (SPIE) risk-location score from the results of
multivariate logistic regression analysis*

Statistical methods Regression SPIE PIE

Variable coefficient score score
Location was classified into 11 categories in the database:
frontal lobe, cerebellum, temporal lobe, parietal lobe, oc- Frontal 2.35 0.00 1
cipital lobe, basal ganglia, medulla, thalamus, intraventric- Temporal 3.48 1.89 2
ular, pons, or corpus callosum. A value of 0 was entered for Intraventricular 4.57 3.72 4
Parietal 5.24 4.83 2
each location not involved by the AVM and 1 if that was the Cerebellar 5.26 4.87 2
only site involved. When multiple sites were involved, Corpus callosum 5.93 5.99 4
fractional values were entered into the database for each site Occipital 5.96 6.04 3
involved according to the number of sites (0.5 for each of Medulla 6.51 6.96 4
two sites, 0.333 for three sites, etc.) Thalamus 6.96 7.71 4
Basal ganglia 7.14 8.01 3
Multivariate logistic regression analysis of the relation- Pons/midbrain 8.33 10.00 4
ship of different variables to permanent symptomatic post- 12 Gy Volume 0.0747
radiosurgery injury was performed using SPSS software. To
maintain the same 9% proportion of symptomatic post- * The results are compared to the previously described postra-
radiosurgery sequelae as in the Pittsburgh series, the number diosurgery injury expression (PIE) score.
of asymptomatic control cases was duplicated proportion-
ally to balance the number of symptomatic non-Pittsburgh
Final complication prediction model
cases without controls. The resulting database therefore
SPIE scores for individual patient’s AVM location were
included 944 cases: 85 with symptomatic sequelae (38
assessed in stepwise (forward conditional) multivariate lo-
permanent, 47 temporary), 302 asymptomatic controls, and
gistic regression (Table 2). Also included in the analysis
557 duplicate control cases.
were 12-Gy-Volume, and two other variables (history of
An initial logistic regression model of permanent symp-
prior hemorrhage and marginal dose or Dmin) that appeared
tomatic postradiosurgery sequelae was constructed that in-
to affect resolution of symptomatic sequelae in prior studies
cluded the 12-Gy-Volume (previously shown to correlate
and the previously described PIE score (5, 10). The final
with complications) and scores for all of the different loca-
regression model included only SPIE score and 12-Gy-
tions. The regression coefficients from this initial model
Volume. History of prior hemorrhage, marginal dose
were used to construct a location score. The effects of other
(Dmin), Marginal-12-Gy-Volume, and PIE score did not
treatment parameters were then examined in combination
significantly add to the predictive value of the model (p ⱖ
with the location score and 12-Gy-Volume to see if any of
0.39). Figures 1 and 2 illustrate the probability of permanent
these other factors affected the risks of developing perma-
symptomatic postradiosurgery injury for different single-
nent symptomatic postradiosurgery sequelae.
location sites according to 12-Gy-Volume predicted from
the final logistic regression model.
The 12-Gy-Volume cannot be calculated prior to com-
RESULTS
pletion of a radiosurgery treatment plan. It may be useful to
Modeling of location effects estimate this parameter ahead of time to predict radiosur-
Table 1 shows the results of multivariate logistic regres- gery risks prior to the day of the procedure. Figure 3 shows
sion analysis of the effects of location on distinguishing the relationship between the equivalent average diameter for
between patients with and without persistent symptomatic the target volume to 12-Gy-Volume in this data, with dose
postradiosurgery sequelae. A single model including all prescriptions roughly following the similar guidelines of the
different locations and the 12-Gy-Volume was constructed. integrated logistic formula and Kjellberg’s 1% isoeffect line
Although 12-Gy-Volume was significantly associated with (2, 12).
permanent postradiosurgery sequelae (p ⫽ 0.0001), none of
the individual locations had a significant effect by them-
DISCUSSION
selves (0.73 ⬍ p ⬍ 0.92). The significant postradiosurgery
injury expression (SPIE) score was constructed from the This study found that permanent symptomatic postradio-
regression coefficients for each location by normalizing surgery sequelae were significantly correlated with location,
them to a scale of 0 –10. The lowest risk region with a SPIE as assessed by SPIE score, and with 12-Gy-Volume. We
score of 0 was the frontal lobe, while the highest risk region were able to develop a more refined score for accounting for
with a SPIE score of 10 was the pons/midbrain. The previ- volume effects (the SPIE score) by simultaneous multivar-
ously described PIE score is included for comparison. In- iate assessment of all locations using a larger database than
traventricular locations changed in assessed relative risk that used for the PIE score (which did not differentiate
between the previous PIE and the newer SPIE scores the between temporary and permanent injuries). We were able
most of all locations (from PIE ⫽ 4/4 to SPIE ⫽ 2.9/10). to construct logistic risk prediction curves for permanent
1146 I. J. Radiation Oncology ● Biology ● Physics Volume 46, Number 5, 2000

Table 2. Results of stepwise multivariate logistic regression modeling of the risks of permanent symptomatic postradiosurgery injury
using the SPIE location-risk score and 12-Gy-Volume with assessment of other variables*

Regression Risk ratio (95%

Variable p Value coefficient ⫾ SE confidence interval)

SPIE score ⬍0.00001 0.7506 ⫾ 0.1243 2.12 (1.67–2.70)

12-Gy-Volume (V12) 0.00001 0.0734 ⫾ 0.0191 1.08 (1.04–1.12)/cc
Constant ⬍0.00001 ⫺7.8713 ⫾ 0.8570
Prior hemorrhage 0.3879 Not entered in model
Marginal-12-Gy-Volume 0.4691 Not entered in model
Maximum dose (Dmax) 0.5871 Not entered in model
Marginal dose (Dmin) 0.6150 Not entered in model
Number of isocenters 0.6614 Not entered in model
Treatment volume 0.6795 Not entered in model
Max. to min. dose ratio 0.8722 Not entered in model
PIE score 0.8780 Not entered in model

* The estimated probability (P) of developing a permanent symptomatic postradiosurgery injury (necrosis) from the final logistic
regression model is: P(necrosis) ⫽ e B /(1 ⫹ e B ), where B ⫽ constant (⫺7.8713) ⫹ 0.7506*(SPIE) ⫹ 0.0734*(V12).

symptomatic postradiosurgery injury from this analysis, assessment. Differences between using any symptomatic
with dose and volume parameters for the treatment repre- postradiosurgery injury (including temporary sequelae, such
sented by the single parameter, 12-Gy-Volume. as headache or seizure alone) and the endpoint of permanent
There are several findings in this study that differed from symptomatic postradiosurgery injury may also have led to
previous complication studies and attempts to model or differences. The inclusion of more data from the multi-
predict complications. Several AVM locations, such as in- institutional Arteriovenous Malformation Radiosurgery
traventricular, were found to have a different relative risk in Study Group also may have contributed to these differences.
the SPIE score than in the previous PIE score. This is most There are a large number of AVM locations to study (11
likely from the simultaneous multivariate modeling of all categories were used in this study). This forced us to split up
locations with 12 Gy volume. Because all of the intraven- the limited amount of data on which we devised the SPIE
tricular locations were also scored as being located in the score. Because of these limitations, the SPIE score formula
adjacent brain (including some high-risk locations such as should be regarded as an initial risk estimate that needs to be
thalamus), the multivariate assessment found that the risk verified in other data, perhaps with recalculation of SPIE
was not as great as indicated by univariate location risk scores for different locations.

Fig. 1. Predicted risks of permanent symptomatic postradiosurgery injury according to location and 12-Gy-Volume for
arteriovenous malformations in frontal, intraventricular, cerebellar, occipital, thalamic, and pons/midbrain locations.
 Prediction model for symptomatic necrosis from AVM radiosurgery ● J. C. FLICKINGER et al. 1147

Fig. 2. Predicted risks of permanent symptomatic postradiosurgery injury according to location and 12-Gy-Volume for
arteriovenous malformations in temporal, parietal, cerebellar, corpus callosum, medulla, and basal ganglia locations.

Prior hemorrhage and marginal dose were not found to be sion analysis. The multi-institutional Arteriovenous Malfor-
significant once the 12-Gy-Volume and SPIE score were mation Radiosurgery Study Group analysis found that a
accounted for in the stepwise multivariate logistic regres- history of prior hemorrhage significantly lessened recovery

Fig. 3. Correlation of equivalent average diameter (diameter for a sphere of the same treatment volume as the AVM
treatment volume) with 12-Gy-Volume (total volume receiving 12 Gy or more including the target) in 422 AVM patients
who underwent gamma knife radiosurgery. Although 12-Gy-Volume depends on the treatment volume and the marginal
dose prescribed, most dose prescriptions followed the guidelines of the 3% risk curve for the integrated logistic formula
or the similar guidelines of Kjellberg’s 1% dose–volume isoeffect line for radiation necrosis. This allows the
12-Gy-Volume to be estimated from the AVM treatment volume (or average equivalent diameter) alone if similar
dose–volume prescription guidelines are used. The correlation coefficient, r2, is the proportion (91.59 %) of the total
variation in the 12 Gy volume that is explained by the equivalent average diameter.
1148 I. J. Radiation Oncology ● Biology ● Physics Volume 46, Number 5, 2000

from symptomatic postradiosurgery injury (p ⫽ 0.01, with for AVM located in the pons/midbrain, basal ganglia, and
41% complete recovery in patients with prior hemorrhage medulla approach zero. Logic and past experience indicate
vs. 66% without). It was therefore somewhat surprising that that the risks should approach zero as 12-Gy-Volume ap-
a history of prior hemorrhage did not predict permanent proaches zero. This appears to be a limitation from the
symptomatic postradiosurgery imaging in the same data- necessity of fitting one relatively simple model to limited
base. A previous analysis of the University of Pittsburgh data for a complex problem. Ideally, separate risk models
data alone found that hemorrhage did not significantly affect should be constructed from large separate data bases for
the development of symptomatic postradiosurgery injury. A AVMs limited to individual locations. In this series, the
slight trend for less symptomatic complications occurring in number of AVM located exclusively in the pons/midbrain,
patients with prior hemorrhage could cancel out the effect of basal ganglia, and thalamus (whose treatment volumes var-
poorer complication recovery in these patients. An expla- ied from 0.51 to 36 cc) were 3, 3, and 6 respectively,
nation for these opposing trends is that prior hemorrhage although 26, 22, and 35 other AVM partially involved these
could conceal the effects of a mild postradiosurgery injury, respective sites. Another limitation of this study is that there
but not injuries that are more severe (and less likely to
is no way to account for individual differences in radiation
resolve).
sensitivity. Hopefully in the near future simple tests will be
Marginal dose (Dmin) was assessed in the initial multi-
available to identify patients with increased radiation sen-
variate analysis of permanent symptomatic postradiosurgery
sitivity of their AVM and normal brain tissue that will allow
injury to test if 12-Gy-Volume alone adequately represented
these patients to be treated safely and effectively with lower
dose–volume effects and because an initial analysis of re-
doses of radiation. Despite these limitations, Figs. 1 and 2
covery from symptomatic injury indicated it might be im-
portant. The multi-institutional Arteriovenous Malforma- provide reasonable predictions of the risks of radiosurgery
tion Radiosurgery Study Group analysis found that marginal for most AVM patients according to their 12-Gy-Volume
dose did not affect recovery from symptomatic postradio- and SPIE location score.
surgery injury. The analysis in this paper (which included In summary, the risks of developing permanent symp-
all of the data in that study plus control cases from the tomatic sequelae from AVM radiosurgery seem to be well
University of Pittsburgh, Mayo Clinic, and Prague with no predicted according to the SPIE location-risk score devel-
complications) confirmed that marginal dose was not sig- oped in this study and the 12-Gy-Volume. The equations
nificantly predictive of the overall endpoint of symptomatic and figures in this study should provide reasonable guides
postradiosurgery injury. for estimating risks from AVM radiosurgery using doses
The model of permanent symptomatic postradiosurgery and techniques similar to those in this study. Extrapolating
sequelae based on 12-Gy-Volume and SPIE score from this these findings to treatment of other targets and/or with other
study has several limitations. Figures 1 and 2 show signif- techniques (especially more homogeneous dose distribu-
icant risks for radiation necrosis remaining as the volumes tions with doses near 12 Gy) is likely to be unreliable.

REFERENCES
1. Flickinger JC, Pollock BE, Kondziolka D, Lunsford LD. A stein A, Simons S, Wu A. Radiosurgery and brain tolerance:
dose–response analysis of arteriovenous malformation oblit- An analysis of neurodiagnostic imaging changes following
eration after radiosurgery. Int J Radiat Oncol Biol Phys 1996; gamma knife radiosurgery for arteriovenous malformations.
36:873– 879. Int J Radiat Oncol Biol Phys 1992;23:19 –26.
2. Kjellberg R, Hanamura T, Davis K, Lyons S, Butler W, 8. Voges J, Treuer H, Lehrke R, Kocher M, Staar S, Muller RP,
Adams R. Bragg-peak proton-beam therapy for arteriovenous Sturm V. Risk analysis of linear accelerator radiosurgery. Int
malformations of the brain. N Engl J Med 1983;309:269 –273. J Radiat Oncol Biol Phys 1996;138:1055–1063.
3. Loeffler JS, Alexander E, Siddon R, Saunders W, Coleman N, 9. Lax I, Karlsson B. Prediction of complications in gamma knife
Winston K. Stereotactic radiosurgery for intracranial arterio- radiosurgery of arteriovenous malformations. Acta Onco-
venous malformations using a standard linear accelerator. Int logica 1996;35:49 –55.
J Radiat Oncol Biol Phys 1989;17:673– 677. 10. Karlsson B, Lax I, Soderman M. Factors influencing the risk
4. Flickinger JC, Kondziolka D, Kalend AM, Maitz AH, for complications following gamma knife radiosurgery for
Lunsford LD. Radiosurgery-related imaging changes in sur- cerebral malformations. In: Karlsson B. Gamma knife radio-
rounding brain: Multivariate analysis and model evaluation. surgery of cerebral arteriovenous malformations. Stockholm:
Radiosurgery 1996;1:229 –236. Repro Print AB; 1996. p. 34 – 42.
5. Flickinger JC, Kondziolka D, Maitz AH, Lunsford LD. Anal- 11. Flickinger JC, Kondziolka D, Lunsford LD, Pollock BE,
ysis of neurological sequelae from radiosurgery of arterio- Yamamoto M, Gorman DA, Schomberg PA, Sneed P, Larson
venous malformations: How location affects outcome. Int J D, Smith V, McDermott MW, Miyawaki L, Chilton J, Mo-
Radiat Oncol Biol Phys 1997;40:273–278. rantz RA, Young B, Jokura H, Liscak R. A multi-institutional
6. Flickinger JC, Kondziolka D, Pollock BE, Maitz AH, analysis of complication outcomes after arteriovenous malfor-
Lunsford LD. Complications from arteriovenous malforma- mation radiosurgery. Int J Radiat Oncol Biol Phys. In press.
tion radiosurgery: Multivariate analysis and risk modeling. Int 12. Flickinger JC. The integrated logistic formula and prediction
J Radiat Oncol Biol Phys 1997;38:485– 490. of complications from radiosurgery. Int J Radiat Oncol Biol
7. Flickinger JC, Lunsford LD, Kondziolka D, Maitz AM, Ep- Phys 1989;17:879 – 885.