Abstract

BACKGROUND: The established clinical criteria for gastric cancer prognosis are insufficient due to molecular heterogeneity. Therefore, constructing a robust prognostic model is essential to predict gastric cancer patient survival.

MATERIAL AND METHODS: A comprehensive method, which combined weighted gene co-expression network analysis (WGCNA) with elastic-net Cox regression, was utilized to identify prognostic long non-coding RNAs (lncRNAs) from Gene Expression Omnibus database for overall survival (OS) prediction. Methods using WGCNA or elastic-net Cox regression alone were treated as “contrast” methods. The univariate and multivariate Cox regression was used to identify independent prognostic clinical factors. We performed 3-year and 5-year area under the curve (AUC) of the time-dependent receiver operating characteristic comparison of 3 different methods in gene and clinical-gene models to explore the prediction ability of the comprehensive method. The optimal model identified in the training set were validated in the validation set. Biological information analysis for the optimal model was also explored.

RESULTS: The clinical-gene model containing 13 co-expression lncRNAs identified by the comprehensive method and 3 clinical factors including molecular subtype, recurrence status and operation type, was the found to be the optimal model in the study, with 0.832 and 0.830 for the 3-year and 5-year AUC in the training set, and 0.764 and 0.778 in the validation set, respectively. Biological information analysis suggested that lipid metabolism played an important role in the occurrence and development of gastric cancer.

CONCLUSIONS: We constructed a novel prognostic model containing 13 co-expression lncRNAs and 3 clinical factors for gastric cancer patients.

Keywords: Survival Analysis, Gene Expression Profiling, Prognosis, RNA, Long Noncoding, Area Under Curve, Biomarkers, Tumor, Databases, Genetic, Kaplan-Meier Estimate, ROC Curve, transcriptome