Variable selection for time-to-event data

Abstract

With the increasing availability of large scale biomedical and -omics data, researchers are offered with unprecedented opportunities to discover novel biomarkers for clinical outcomes. At the same time, they are also faced with great challenges to accurately identify important biomarkers from numerous candidates. Many novel statistical methodologies have been developed to tackle these challenges in the last couple of decades. When the clinical outcome is time-to-event data, special statistical methods are needed to analyze this type of data due to the presence of censoring. In this article, we review some of the most commonly used modern statistical methodologies for variable selection for time-to-event data. The reviewed methods are classified into three large categories: filter-test based method, penalized regression method, and machine learning method.