External Validity of a Risk Stratification Score Predicting Early Distant Brain Failure and Salvage Whole-Brain Radiation Therapy After Stereotactic Radiosurgery for Brain Metastases

Abstract

Early distant brain failure (DBF) after stereotactic radiosurgery (SRS) and need for early salvage whole brain radiation therapy (WBRT) for patients (pts) with brain metastases (BM) remains a substantial clinical problem. A risk stratification score (Emory score) was recently published for predicting risk of early DBF and salvage WBRT (“early” defined as ≤6 months from SRS) using pre-treatment factors (Press et al., Cancer 2015). The purpose of this study was to determine the external validity of this scoring system using an independent validation cohort. The records of consecutive pts with BM treated with SRS between 2010-2013 from Levine Cancer Institute (LCI, Charlotte, NC) were reviewed. The Emory score uses number of BM, tumor histology, total BM volume, and prior WBRT as factors. This score was applied to the LCI validation cohort to determine low (LR), intermediate (IR), and high-risk (HR) groups. DBF and salvage WBRT rates were estimated using cumulative incidence with competing risk analysis. Pts with previous WBRT were excluded from the salvage WBRT analyses. Multivariate analysis (MVA) used Fine and Gray method. Overall, 247 consecutive pts underwent SRS for BM, of which 41 had previous WBRT. Early DBF rates (n = 247) were 22.6%, 42.2%, and 44% for LR, IR, and HR groups, respectively. Relative to the HR group (reference), the LR group had significantly lower risk of early DBF (hazard ratio [HR] 0.52, P = 0.004), but there was no difference for the IR group (HR 0.95, P = 0.79). Early salvage WBRT rates (n = 206) were 8.9%, 19.3%, and 19.4%, respectively. Similarly, relative to the HR group (reference), the LR group had lower risk of early salvage WBRT (HR 0.53, P = 0.09), but there was no difference for the IR group (HR 1.19, P = 0.53). Applying the Emory MVA model for early DBF to the LCI cohort confirmed the prognostic significance for higher risk of early DBF for higher number of BM (≥ 3, P<0.001), melanoma histology (P = 0.007), and smaller total BM volume (<0.3cc, P = 0.004), however neither previous WBRT (P = 0.69) nor breast histology (P = 0.8) retained prognostic significance. The AUC of the Emory MVA model applied to the LCI cohort was 0.622. The Emory risk score was able to stratify risk of early (≤ 6 months) DBF and salvage WBRT between 2 levels (LR and IR/HR groups), but not between IR and HR in the independent LCI validation cohort. Additionally, the prognostic value of number of BM, melanoma histology, and total BM volume for early DBF was confirmed, but was not for prior WBRT and breast histology. These results serve to refine the predictive use of the Emory score for early DBF after SRS and reinforce the importance of validating predictive models in independent cohorts.