Comparison of Six Statistics of Genetic Association Regarding Their Ability to Discriminate between Causal Variants and Genetically Linked Markers

Objectives: Genome-wide association (GWA) studies still rely on the common-disease common-variant hypothesis since the assumption is associated with increased power. In GWA studies, polymorphisms are genotyped and their association with disease is investigated. Most of the identified associations are indirect and reflect a shared inheritance of the genotyped markers and genetically linked causal variants. We have compared six statistics of genetic association regarding their ability to discriminate between markers and causal susceptibility variants, including a probability value (Pval) and a Bayes Factor (BF) based on logistic regression, and the attributable familial relative risk (FRR). Methods: We carried out a simulation-based sensitivity analysis to explore several conceivable scenarios. Theoretical results were illustrated by established causal associations with age-related macular degeneration and by using imputed data based on HapMap for a case-control study of breast cancer. Results: Our data indicate that a representation of genetic association by FRRs and BFs generally facilitates the distinction of causal variants. The FRR showed the best discriminative power under most investigated scenarios, but no single statistic outperformed the others in all situations. For example, rare moderate- to low-penetrance variants (allele frequency: 1%, dominant odds ratio: ≤2.0) seem to be best discriminated by BFs. Conclusions: Present results may help to fully utilize the data generated in association studies that take advantage of next generation sequencing and/or multiple imputation based on the 1000 Genomes Project.