The risks of relapse for lung adenocarcinoma patients were still higher than 30%, even after complete surgical resections in early stages. Although lots of prognosis studies using genome-wide profiling had been published, biological meaning and interactions among the prognostic genes were poorly understood. Therefore, we developed a novel method integrating gene set enrichment analysis and Cox-hazard regression model to investigate the relations between predefined gene sets and the survival outcome in lung cancer. The method was able to select gene sets associated with the survival outcome, clustering of the prognostic genes sets, and selection of a representative gene set from each cluster. Furthermore, kernel matrix was used to visualize the similarities between those representative gene sets. In addition to survival outcome, our method can also use other continuous variables to explore other biological interpretation concealed in the predefined gene sets.