Construction and evaluation of different glioblastoma prognosis scores based on gene expression databases

Introduction: Glioblastoma (GBM) is the most common malignant brain tumor, and the prognosis of GBM pa-tients is unfavorable.More studies are needed to develop new prognostic tools for predicting GBM patients' prognosis.This study aims to construct generisk score (GRS) models based on gene expression databases.Methods: Genomic data of GBM were downloaded from the CGGA, TCGA, MYO, and CPTAC.Patients were divided into two groups with overall survival (OS) of more or less than 15 months.Top 31 genes from our previous study and clinical data such as age, gender, and IDH wildtype/mutant were used to develop two GRS models.Cox methods in SPSS v26 were applied in this study.Results: A total of 551 (334 male, mean age 55.5 13.3 years) cases were used.Fourgene (TGFB1, CCL2, CD274, and TNFRSF1A; from the combination of four databases) and eight-gene (EGFR, TGFB1, SPP1, AGT, TNFRSF1A, CDK1, FOXO3, and CEP55; from CGGA) risk scores were developed.Two models could separate OS samples into high and low-risk groups, and AUCs of 0.984 and, 0.998 were achieved that showed excellent discriminating power at the training set (all: p < 0.0001).For the 8-GRS model, the OS of cases in the high-risk group was poorer than that in the low-risk group when used on another's datasets at the validation set, however, it was not significant.Conclusion: Four-and eight-gene prognostic signatures were identified and constructed to predict OS in GBM patients.This study may provide innovative insights into the treatment of GBM.