Survival Rate Prediction in Glioblastoma Patients Using Machine Learning

Abstract

Glioblastoma disease (GBM) has been widely recognized as the most deadly type of brain tumor. Extensive efforts of the medical and research community have been made to better understand this tumor, but still the survival probability of patients remains extremely low. Existing research works are mainly based on the analysis of non-omic patient’s data (MRI) for predicting their overall survival. However, omic data such as gene expression subtype, methylation, and IDH1 Mutation provide rich information about the tumor and their correlation with the survival probability. This paper proposes a novel approach to predicting the overall survival (OS) rate of Glioblastoma (GBM) patients based on omic data. We apply different machine learning algorithms to data from GBM patients (n= 577) from The Cancer Genome Atlas (TCGA) to evaluate their ability in predicting the survival probability and to highlight the most correlated features. Our simulation results show that support vector machine (SVM) outperforms random survival forest (RSF) and linear regression models in terms of concordance index.Additionally, a significant correlation (r = 0.74) is observed between G-CIMP methylation status and IDH1 mutation. The study underscores the potential of omic data and machine learning in predicting overall survival in glioblastoma patients.