Development of a recursive partitioning analysis for prediction of radiation necrosis following single-fraction stereotactic radiosurgery for intact brain metastases

Abstract Purpose/objective Radiation necrosis (RN) is a potential complication after stereotactic radiosurgery (SRS) for brain metastases. This study develops a recursive partitioning analysis (RPA) to identify patients at risk for RN following SRS. Methods Patients who underwent single-fraction SRS for intact brain metastases at a single institution from 2017 to 2021 were identified. Cox regression identified factors associated with RN, and variables with p < 0.1 were included in the RPA. Patients with staged SRS, incomplete records, or less than 3 months of follow-up were excluded. Results The study included 170 patients with 919 lesions, with median follow-up of 9 months. Primary disease sites were non-small cell lung cancer (NSCLC, 49%), breast cancer (12%), melanoma (11%), renal cancer (6%), and others (22%). Median prescription dose was 24 Gy, and median maximum lesion dimension (MLD) was 0.7 cm. RN occurred in 110 (12.2%) lesions, of which 32 (3.5%) were symptomatic, at median of 4.9 months after SRS. Variables for RPA included primary disease site, tumor location, MLD, prior SRS, number of SRS targets, dosimetry, prior hemorrhage, and concurrent systemic therapy. RPA identified four groups: Group 1 (MLD ≤ 0.8 cm, non-breast/NSCLC/renal), Group 2 (MLD ≤ 0.8 cm, breast/NSCLC/renal), Group 3 (MLD > 0.8 cm, no post-SRS hemorrhage), and Group 4 (MLD > 0.8 cm, post-SRS hemorrhage). Two-year RN free survival was 99% (Group 1), 89% (Group 2), 70% (Group 3), and 52% (Group 4). Conclusion This is the first RPA model for RN after single-fraction SRS, which may aid in risk assessment and distinguishing RN from tumor progression.