Online survival analysis software to assess the prognostic value of biomarkers using transcriptomic data in non-small-cell lung cancer

Abstract

In the last decade, optimized treatment for non-small cell lung cancer had lead to improved prognosis, but the overall survival is still very short. To further understand the molecular basis of the disease we have to identify biomarkers related to survival. Here we present the development of an online tool suitable for the real-time meta-analysis of published lung cancer microarray datasets to identify biomarkers related to survival. We searched the caBIG, GEO and TCGA repositories to identify samples with published gene expression data and survival information. Univariate and multivariate Cox regression analysis, Kaplan-Meier survival plot with hazard ratio and logrank P value are calculated and plotted in R. The complete analysis tool can be accessed online at: www.kmplot.com/lung. All together 1,715 samples of ten independent datasets were integrated into the system. As a demonstration, we used the tool to validate 21 previously published survival associated biomarkers. Of these, survival was best predicted by CDK1 (p<1E-16), CD24 (p<1E-16) and CADM1 (p = 7E-12) in adenocarcinomas and by CCNE1 (p = 2.3E-09) and VEGF (p = 3.3E-10) in all NSCLC patients. Additional genes significantly correlated to survival include RAD51, CDKN2A, OPN, EZH2, ANXA3, ADAM28 and ERCC1. In summary, we established an integrated database and an online tool capable of uni- and multivariate analysis for in silico validation of new biomarker candidates in non-small cell lung cancer.

Figure 1. Survival characteristics of the patients included in the database including histology of adenocarcinoma (adeno), squamous cell carcinoma (SCC) and large cell carcinoma (large), gender, stage (only with overall survival) and smoking history.

Figure 2. Validation of 29 previously published NSCLC biomarkers.

Meta-analysis of these genes and signatures in the respective sample cohort yielded CCNE1, CDC2 and CADM1 as the best performing individual genes (A–C) and the signature of Yamauchi et al. (D). A funnel plot depicting the hazard ratios (with confidence intervals) versus sample number for CDC2 and VEGF shows more reliable estimation with larger database sizes (E–F).