Object: Radiosurgery is an effective treatment strategy for properly selected patients harboring arteriovenous malformations (AVMs). Grading scales that are currently used to predict patient outcomes after AVM resection are unreliable tools for the prediction of the results of AVM radiosurgery.
Methods: A grading system was developed to predict outcomes following AVM radiosurgery, based on the multivariate analysis of data obtained in 220 patients treated between 1987 and 1991 (Group 1). The dependent variable in all analyses was excellent patient outcome (complete AVM obliteration without any new neurological deficit). The grading scale was tested on a separate set of 136 patients with AVMs treated between 1990 and 1996 at a different center (Group 2). One hundred twenty-one (55%) of 220 Group 1 patients had excellent outcomes. Multivariate analysis identified five variables related to excellent patient outcomes: AVM volume (p = 0.001), patient age (p < 0.001), AVM location (p < 0.001), previous embolization (p = 0.02), and number of draining veins (p < 0.001). Regression analysis modeling permitted removal of two significant variables (previous embolization and number of draining veins) and resulted in the following equation to predict patient outcomes after AVM radiosurgery: AVM score = (0.1)(AVM volume in cm3) + (0.02)(patient age in years) + (0.3)(location of lesion: frontal or temporal) = 0; parietal, occipital, intraventricular, corpus callosum, cerebellar = 1; or basal ganglia, thalamic, or brainstem = 2). Seventy-nine (58%) of 136 Group 2 patients had excellent outcomes. All variables in the model remained significant for the Group 2 patients: AVM volume (p = 0.01), patient age (p = 0.01), and AVM location (p < 0.001). Testing of the entire model on the Group 2 patients demonstrated that the AVM score could be used to predict patient outcomes after radiosurgery (p < 0.0001). All patients with an AVM score of 1 or lower had an excellent outcome compared with only 39% of patients with an AVM score higher than 2. The Spetzler-Martin grade (p = 0.13), the K index (p = 0.26), and the obliteration prediction index (p = 0.21) did not correlate with excellent patient outcomes.
Conclusions: Despite significant differences in preoperative patient characteristics and dose prescription guidelines at the two centers, the proposed AVM grading system strongly correlated with patient outcomes after single-session radiosurgery for both patient groups. Although further testing of this model by independent centers using prospective methodology is still required, this system allows a more accurate prediction of outcomes from radiosurgery to guide choices between surgical and radiosurgical management for individual patients with AVMs.