Assessment of prognostic prediction models for gastric cancer using genomic and transcriptomic profiles

Abstract

Gastric cancer is one of the leading causes of cancer deaths in the world. We performed an assessment of prognostic prediction model for gastric cancer with DNA, RNA level and clinical factors based on TCGA database. Systematic dimensional reduction strategy was adopted to screen the potential predictors at both DNA (3,076,323 SNPs) and RNA levels (459 cancer driver genes). Cox proportional hazards model and LASSO penalized regression were further used to harmonize the final predictive variables. We used the Receiver Operating Characteristic curve to evaluate the predictive ability of the model. Three genes (MKL1, TJP1 and WNT5A) and four SNPs (rs11209970, rs4377857, rs3892045 and rs2747537) along with clinical factors (age, gender and pathological stage) were incorporated into final prognostic models. For the model containing only the clinical predictors, the area under the curve (AUC) was 0.747. Next, we combined the clinical factors and DNA level factors, the AUC had a significant increase (AUC = 0.819) compared with the clinical model (P = 0.009). Further, we added the RNA level factors into the prediction model, the prediction ability raised to 0.837 with a P value of 2.82 × 10−3. The following time-dependent ROC revealed that the mean AUC was 0.799 for 6, 12, 18, 24, 30, 36, 42, 48, 54 and 60 months overall survival prediction. Our findings suggested that multi-omics information could increase the predictive performance of prognosis for gastric cancer. This model would provide clinicians an applicable tool to predict the prognosis of patients with gastric cancer.