B1248
Title: Testing for association with survival in genome-wide analysis studies: Overcoming limitations and innovating approaches
Authors: Dominic Edelmann - German Cancer Research Center, Heidelberg (Germany) [presenting]
Abstract: The purpose is to critically examine the established methodologies for testing the association of single-nucleotide polymorphisms (SNPs) with survival, pointing out drawbacks of the current practice and offering innovative solutions to circumvent these shortcomings. It is shown that the Wald and Score statistics based on the Cox model cannot reliably control the type I error rate of 5E-8 commonly used in genome-wide analysis studies (GWAS). On the other hand, while likelihood ratio tests and Firth correction-based procedures are substantially more reliable, the runtime of these tests is prohibitively high. To address this challenge, a fast and precise testing procedure for GWAS is proposed based on prescreening via an extremely efficient version of the Score test, followed by a precise evaluation of the p-value for the screened subset of genes using the Likelihood ratio test. Alternatives for the multiplicative hazard models are further considered. To this end, a novel distance correlation-based test procedure is proposed for testing the association of SNPs and survival. Asymptotic properties are derived for this test. Moreover, it is shown that the testing procedure is the locally most powerful test for certain genomic models. Finally, an outlook on testing the association of SNP sets with survival is presented.
