Survival Risk Prediction Models of Glioma Based on IDH and 1 p/19q

To explore the prognostic factors of 3 kinds of gliomas based on molecular features (IDH and 1p/19q), we performed a univariate survival analysis on the RNA-seq data of 653 patients in The Cancer Genome Atlas. Separately, 12205 (20.18%), 6125 (10.13%) and 5206 (8.61%) genes were associated with the overall survival (OS) of the IDH-wildtype, IDH-mutation 1p/19q intact and IDH-mutation 1p/19q codeletion. With these OS related genes, the pathway enrichment revealed that alcoholism, systemic lupus erythematosus, hematopoietic cell lineage and diabetes might affect the OS of glioma patients. Besides, the more OS related genes a glioma had, the poorer prognosis it would be. With the selected genes, we used Lasso regression to set up three models that can effectively predict the survival risk of each type. Totally, the three models contain 76 genes expression level. Moreover, none of the 76 genes was repeated in another model, which suggested the enormous difference among the three subtypes. SERPINA5, RP11.229A12.2 and RP11.62F24.2 were in common among model containing genes of one subtype and OS related genes of other two subtypes. Besides, five genes' (RP11.229A12.2, RP11.62F24.2, C3orf67, RP11.275H4.1 and TBX3) function (protection or risk factor) opposed in different subtypes. More attention, non-coding RNA plays a vital important role in the prognosis of glioma (58.74%, 70.13% and 58.11% in the IDH-wildtype, IDH-mutation 1p/19q intact and IDH-mutation 1p/19q codeletion). What's more, after multivariate adjustment by Cox regression, prognostic value of the models maintained powerful and independent of clinicopathological features. In conclusion, the multi-gene-based classifiers add great prognostic value to the current stage system, and the RNAs and their related genes deserve further study for their mechanisms in influencing the OS and their clinical significance in improving the OS.