Predicting Stereotactic Radiosurgery Outcomes for Brain Metastases
Abstract
Many brain metastases (BMs) patients are subjected to disease retreatment due to stereotactic radiosurgery (SRS) failure (progression rate of 4%-30%). This led us to attempt to find predictors of SRS BM progression using statistical techniques and quantitatively demonstrate the need for a uniform definition of progression to validate machine learning models going forward. We analyzed how BM characteristics could predict progression on a multi-institutional dataset and found that the distance between BMs was a retrospective predictor of progression (median distance of 5.5cm versus 3.4cm for non-progressing and progressing BMs). We also found that performance of a random decision forest model designed to predict BM progression was affected by varying the definition of progression (AUC varied by 0.08), demonstrating the need for a more uniform definition of progression. The testing and validation of new predictors and models could help implement these techniques clinically to improve treatment decision making for BM patients.