A Novel Two-Stage Approach to Identify Prognostic Biomarkers for Cancer Genetic Data with Survival Endpoints

• Activity Number: 696
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Zheng Li* and Ming Wang
• Companies: Penn State and Penn State
• Keywords: Time to event outcome ; P values ; Biomarker Selection ; Cox PH model ; Bootstrapping ; Multi-split

Abstract

Identification of prognostic biomarkers has gained substantial attention recently in developing personalized medicine for cancer patients. In particular, nowadays there are usually high-dimensional candidate biomarkers in discovery process because of enhanced development in proteomics and genomic technologies. Also, for such cancer genetic research, survival endpoints are very often to be analyzed. In this work, we adopt penalized techniques, i.e., LASSO and Elastic net, for variable selection in high-dimensional data analysis. However, those approaches add restriction to the models, thus, the traditional p-values are not satisfactory. We incorporate the idea from Meinshausen et al. (2009) to use multi-splitting algorithm and bootstrapping method for adjusted p-values under the framework of Cox Proportional Hazards model. A novel two-stage framework is proposed for selecting the prognostic biomarkers. Extensive simulations are conducted to evaluate its performance. It can be shown that our two-stage approach is less conservative compared with Meinshausen et al. (2009), and also the model selection accuracy is improved. Finally, we apply our method to a real breast cancer study.