Gene-gene interaction analysis for the survival phenotype based on the Cox model

Area of Science:

• Genetics
• Biostatistics
• Epidemiology

Background:

• Genome-wide association studies (GWAS) primarily focus on case-control designs, with limited methods for prospective cohort studies involving survival time.
• Existing methods like Surv-MDR, while addressing survival data, have computational drawbacks and lack covariate adjustment capabilities.

Purpose of the Study:

• To propose Cox-MDR, a novel approach extending generalized multifactor dimensionality reduction (GMDR) for survival phenotypes.
• To overcome the limitations of Surv-MDR by incorporating covariate adjustments and improving computational efficiency.

Main Methods:

• Cox-MDR utilizes martingale residuals to classify multi-level genotypes into high- and low-risk groups within a Cox regression framework.
• The method extends GMDR to survival analysis, accommodating discrete and quantitative covariates.

Main Results:

• Simulation studies demonstrated Cox-MDR's superior performance over Surv-MDR when adjusting for covariates, particularly at low minor allele frequencies (0.2).
• While Surv-MDR showed higher power at a minor allele frequency of 0.4, Cox-MDR and Cox regression models outperformed it with covariate adjustment.
• The study validated Cox-MDR's efficacy on real leukemia patient data for gene-gene interaction analysis with survival time.

Conclusions:

• Cox-MDR offers a computationally efficient and flexible method for detecting gene-gene interactions in survival data.
• The ability to adjust for covariates makes Cox-MDR a valuable tool for genetic association studies in prospective cohort designs.