A0691
Title: Model-free feature screening for high-throughput semi-competing risks data with FDR control
Authors: Chenlu Ke - Virginia Commonwealth University (United States) [presenting]
Abstract: Partial Identifying biomarkers that contribute to early detection and effective treatment of cancers is a vital yet ongoing research task, which is often characterized by high-throughput data generated in a massive and fast manner by omics technologies, along with complicated survival endpoints as cancer course often involves adverse events such as progression and recurrence. A new feature screening framework is proposed for high-throughput survival data subject to semi-competing risks. Compared with existing prototypes, the method does not require an estimation of the survival function and relaxes the common assumption of independent censoring. The sure screening property and the rank consistency property in the notion of sufficiency are established. A knockoff procedure is also developed for controlling false discoveries. The advantages of the proposed method are demonstrated by simulation studies and an application in discovering the prognostic significance of copy-number alterations in multiple Myeloma.