Recursive Partitioning Analysis of Patients with Spine Metastases Treated with Stereotactic Body Radiotherapy

Abstract

Uncertainty exits regarding the outcomes of patients receiving stereotactic body radiotherapy (SBRT) for spinal metastases. We sought to create a prognostic index using recursive partitioning analysis (RPA) for overall survival (OS). After institutional review board (IRB) approval, we performed a retrospective review on 334 patients treated with SBRT between 2005 and 2017. The cohort was split 2:1 into modelling and validation datasets. RPA was performed in the modeling cohort of patients to identify any association of the following variables with OS after spine SBRT: tumor histology, gender, age, race, performance status (KPS), controlled primary, extraosseous metastases, previous surgery, previous radiation therapy, time from primary diagnosis to treatment with SBRT, extent of spinal disease (single or multiple), spine disease type (bone only, epidural only, or bone plus epidural), paraspinal extension, equivalent dose in 2-Gy fractions (EQD2) for a/b ratio of 2, and EQD2 a/b ratio of 10. Median OS for the entire cohort was 9.8 months (95% CI 7.8-11.8). RPA analysis identified two prognostic groups (p<0.001). RPA1 included patients with bone only disease or extraosseous metastases and controlled primary from breast/prostate primary; RPA2 included extraosseous metastases from non-breast/prostate primary or uncontrolled breast/prostate primary. Median OS was 18.0 months (IQR: 6.9-29.0) and 5.4 months (IQR: 4.1-6.8) for RPA1 and RPA2, respectively. Multivariate Cox regression including the RPA and all other factors found to be significant initially in univariate analysis confirmed the RPA as the most significant variable predicting survival after spine SBRT in the validation data set. Patients treated for spine metastases with SBRT have variability in OS. An RPA classification system was developed with plans to further refine OS outcomes with additional variables. The RPA1 classification can successfully identify patients with sufficiently long expected OS that prospective analysis for late effects could be feasible in this cohort.