A gene expression-based risk model reveals prognosis of gastric cancer

Background The prognosis of gastric cancer is difficult to determine, although clinical indicators provide valuable evidence. Methods In this study, using screened biomarkers of gastric cancer in combination with random forest variable hunting and multivariable Cox regression, a risk score model was developed to predict the survival of gastric cancer. Survival difference between high/low-risk groups were compared. The relationship between risk score and other clinicopathological indicators was evaluated. Gene set enrichment analysis (GSEA) was used to identify pathways associated with risk scores. Results The patients with high risk scores (median overall survival: 20.2 months, 95% CI [16.9-26.0] months) tend to exhibit early events compared with those with low risk scores (median survival: 70.0 months, 95% CI [46.9-101] months, p = 1.80e-5). Further validation was implemented in another three independent datasets (GSE15459, GSE26253, GSE62254). Correlation analyses between clinical observations and risk scores were performed, and the results indicated that the risk score was not significantly associated with gender, age and primary tumor size but was significantly associated with grade and tumor stage. In addition, the risk score was also not influenced by radiation therapy. Cox multivariate regression and three-year survival nomogram suggest that the risk score is an important indicator of gastric cancer prognosis. GSEA was used to identified KEGG pathways significantly associated with risk score, and signaling pathways involved in focal adhesion and the TGF-beta signaling pathway were identified. Conclusion The risk score model successfully predicted the survival of 1,294 gastric cancer samples from four independent datasets and is among the most important indicators in clinical clinicopathological information for the prognosis of gastric cancer. To our knowledge, it is the first report to predict the survival of gastric cancer using optimized expression panel.