Predictive Model to Identify the Long Time Survivor in Patients with Glioblastoma: A Cohort Study Integrating Machine Learning Algorithms.
Journal:
Journal of molecular neuroscience : MN
PMID:
38662286
Abstract
We aimed to develop and validate a predictive model for identifying long-term survivors (LTS) among glioblastoma (GB) patients, defined as those with an overall survival (OS) of more than 3 years. A total of 293 GB patients from CGGA and 169 from TCGA database were assigned to training and validation cohort, respectively. The differences in expression of immune checkpoint genes (ICGs) and immune infiltration landscape were compared between LTS and short time survivor (STS) (OS<1.5 years). The differentially expressed genes (DEGs) and weighted gene co-expression network analysis (WGCNA) were used to identify the genes differentially expressed between LTS and STS. Three different machine learning algorithms were employed to select the predictive genes from the overlapping region of DEGs and WGCNA to construct the nomogram. The comparison between LTS and STS revealed that STS exhibited an immune-resistant status, with higher expression of ICGs (P<0.05) and greater infiltration of immune suppression cells compared to LTS (P<0.05). Four genes, namely, OSMR, FMOD, CXCL14, and TIMP1, were identified and incorporated into the nomogram, which possessed good potential in predicting LTS probability among GB patients both in the training (C-index, 0.791; 0.772-0.817) and validation cohort (C-index, 0.770; 0.751-0.806). STS was found to be more likely to exhibit an immune-cold phenotype. The identified predictive genes were used to construct the nomogram with potential to identify LTS among GB patients.