Combination of TNM staging and pathway based risk score models in patients with gastric cancer.

Due to the complexity and heterogeneity of gastric cancer (GC) in individual patient, current staging system is inadequate for predicting outcome of GC. Comprehensive computational and bioinformatics approach may triumph for the prediction. In this study, GC patients were devided according to stage and treatment: curative surgery plus chemoradiotherapy in stage II, curative surgery plus chemoradiotherapy in stages III, and IV, unresectable metastatic gastric cancer. The training sets were downloaded from GEO datasets (GSE26253 and GSE14208). Gene set enrichment analysis (GSEA) was performed to explore enriched difference between recurrence and nonrecurrence. The core enrichment genes of enriched pathways significantly associated with recurrence or progression were identified using Cox proportional hazards analysis. Thereafter, the risk score models were externally validated in independent datasets-GSE15081 and The Cancer Genome Atlas (TCGA). We generated respective risk score models of patients in different stages and treatment. A five-gene signature comprising FARP1, SGCE, SGCA, LAMA4, and COL9A2 was strongly associated with recurrence of patients with curative surgery plus chemoradiotherapy in stage II. A six-gene signature consisting of SHH, NF1, AP4B1, COMP, MATN3, and CCL8 was correlated with recurrence of patients with curative surgery plus chemoradiotherapy in stages III and IV. And a four-gene signature composing of ABCC2, AHNAK2, RNF43, and GSPT2 was highly related to progression of patients with unresectable metastatic GC. Taking into consideration TNM stage and gene signature reflecting recurrence or progression, the risk score models significantly improved the accuracy in predicting outcome of GC.