Recursive partitioning analysis for the prediction of stereotactic radiosurgery brain metastases lesion control.

ABSTRACT

PURPOSE: The objective of this investigation was to identify independent pretreatment factors that predict for control of local brain metastases (BM) in a large single-institution series of patients receiving stereotactic radiosurgery (SRS). Recursive partitioning analysis was used to potentially identify a class of patients with durable lesion control characteristics. METHODS: A retrospective SRS database containing baseline characteristics, treatment details, and follow-up data of newly diagnosed patients with 1-3 BM (on magnetic resonance imaging) treated with linear accelerator-based SRS was created. Three study endpoints were used time to progression (primary endpoint, individual lesion progression; n = 536), time to first progression (secondary endpoint, first lesion progression on an individual patient basis; n = 380), and overall survival (secondary endpoint; n = 380). Recursive partitioning analysis (RPA) was performed to identify predictors of time to progression. RESULTS: Multivariable analysis demonstrated that lesion aspect/phenotype and radiotherapy schedule were independent factors associated with both progression outcomes. Presence of tumor necrosis was found to be associated with a significant hazard of progression (hazard ratio >3), whereas use of the most intense radiotherapy fractionation schedule (21 Gy in one fraction) was associated with significant reductions in progression (hazard ratio <0.3). RPA using SRS dose and lesion aspect/phenotype was created and described three distinct prognostic groups. CONCLUSIONS: RPA of a large retrospective database of patients receiving SRS confirmed previous observations regarding the importance of SRS dose and lesion aspect/phenotype in lesion control and overall survival. The SRS lesion analysis may help to stratify future clinical trials and better define patient care options and prognosis.