Machine learning prediction of glioblastoma patient one-year survival

Glioblastoma (GBM) is a grade IV astrocytoma formed primarily from cancerous astrocytes and sustained by intense angiogenesis. GBM often causes non-specific symptoms, creating difficulty for diagnosis. This study aimed to utilize machine learning techniques to provide an accurate one-year survival prognosis for GBM patients using clinical and genomic data from the Chinese Glioma Genome Atlas. Logistic regression (LR), support vector machines (SVM), random forest (RF), and ensemble models were used to identify and select predictors for GBM survival and to classify patients into those with an overall survival (OS) of less than one year and one year or greater. With regards to overall survival, a significant (p \u3c 0.05, n = 175) correlation was found with age (negative), radiation treatment (positive), and chemotherapy treatment (positive). IDH1 mutation and 1p19q codeletion showed insignificant correlation with OS in this dataset. This potentially implies that IDH1 mutation alone, although important in secondary GBM prognosis, is insignificant for primary GBM prognosis. 1p19q codeletion also appeared to be insignificant for primary GBM prognosis when considered alone. The ensemble model had the highest overall accuracy, achieving a mean AUC score of 0.644 and an F1 score of 0.799