Joint variable screening in accelerated failure time models

Variable screening has become increasingly popular as a method for analyzing high-dimensional survival data. Most existing variable screening methods for such data assess the importance of their variables using marginal models that relate the time-to-event outcome to each variable separately. This implies that the relevance of one variable is examined while other variables are excluded. Therefore, these methods exclude variables that only manifest their influence jointly, and may retain irrelevant variables that are correlated with relevant ones. To circumvent these difficulties, we propose a new approach to evaluate the joint variable importance in censored accelerated failure time models. We establish the sure screening properties of the proposed approach and demonstrate its effectiveness through simulation studies and a real-data application. Furthermore, we propose a novel procedure using stability selection for tuning.