Nonparametric Failure Time Analysis with Genomic Applications

• Activity Number: 233
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Cheng Cheng*+
• Companies: St. Jude Children's Research Hospital
• Keywords: Survival Analysis ; rank correlation ; GWAS ; proportional hazard ; permutation test ; nonparametric

Abstract

Genome-wide Association Study (GWAS) has become routine in cancer genomic translational research, which often requires genome-wide screening to identify ordinal genomic features that are associated with treatment outcome, for example, single nucleotide polymorphisms associated with time to relapse. The estimated coefficient of a hazard rate regression model (HRRM) is often used as the association test statistic. It will be demonstrated in this talk that in certain cases the HRRM approach is problematic. A robust, completely nonparametric alternative using rank correlation is then proposed. This method, called correlation profile test (CPT), consists of the correlation profile statistic and a very efficient hybrid permutation test combining permutation and asymptotic theory. Statistical performances are compared with several established methods, by a simulation study and analysis of real genomics data. It is shown that CPT performs much better than the HRRM approach in terms of maintaining the power and nominal significance level, especially in cases where the proportional hazard model does not hold.