Identification and validation of a gene expression signature that predicts outcome in malignant glioma patients

Abstract

Better understanding of the underlying biology of malignant gliomas is critical for the development of early detection strategies and new therapeutics. This study aimed to define genes associated with survival. We investigated whether genes selected using random survival forests model could be used to define subgroups of gliomas objectively. RNAs from 50 non-treated gliomas were analyzed using the GeneChip Human Genome U133 Plus 2.0 Expression array. We identified 82 genes whose expression was strongly and consistently related to patient survival. For practical purposes, a 15-gene set was also selected. Both the complete 82 gene signature and the 15 gene set subgroup indicated their significant predictivity in the 3 out of 4 independent external dataset. Our method was effective for objectively classifying gliomas, and provided a more accurate predictor of prognosis. We assessed the relationship between gene expressions and survival time by using the random survival forests model and this performance was a better classifier compared to significance analysis of microarrays.