Dose—volume prediction of radiation-related complications after proton beam radiosurgery for cerebral arteriovenous malformations

Abstract

Object. The use of radiosurgery for the treatment of cerebral arteriovenous malformations (AVMs) and other lesions demands an accurate understanding of the risk of radiation-related complications. Some commonly used formulas for predicting risk are based on extrapolation from small numbers of animal experiments, pilot human treatment series, and theoretical radiobiological considerations. The authors studied the incidence of complications after AVM radiosurgery in relation to dose, volume, and other factors in a large patient series. Methods. A retrospective review was conducted in 1329 patients with AVM treated by Dr. Raymond Kjellberg at the Harvard Cyclotron Laboratory (HCL) between 1965 and 1993. Dose and volume were obtained from HCL records, and information about patient follow up was derived from concurrent clinical records, questionnaires, and contact with referring physicians. Multivariate logistic regression with bootstrapped confidence intervals was used. Follow up was available in 1250 patients (94%); the median follow-up duration was 6.5 years. The median radiation dose was 10.5 Gy and the median treatment volume was 33.7 cm3. Twenty-three percent of treated lesions were smaller than 10 cm3. Fifty-one permanent radiation-related deficits occurred (4.1%). Of 1043 patients treated with a dose predicted by the Kjellberg isoeffective centile curve to have a less than 1% complication risk, 1.8% suffered radiation-related complications. Actual complication rates were 4.7% for 128 patients treated at Kjellberg risk centile doses of 1 to 1.8%, and 34% for 61 patients treated at risk centile doses of 2 to 2.5%. The fitted logistic model showed that complication risk was related to treatment dose and volume, thalamic or brainstem location, and patient age. Conclusions. The Kjellberg isoeffective risk centile curve significantly underpredicted actual risks of permanent complications after proton beam radiosurgery for AVMs. Actual risks were best predicted using a model that accounted for treatment dose and volume, lesion location, and patient age.