Multiple phenotype modeling in gene-mapping studies of quantitative traits: Power advantages

Genomewide searches for loci influencing complex human traits and diseases such as diabetes, hypertension, and obesity are often plagued by low power and interpretive difficulties. Attempts to remedy these difficulties have typically relied on, and have promoted the use of, novel subject-ascertainment schemes, larger sample sizes, a greater density of DNA markers, and more-sophisticated statistical modeling and analysis strategies. Many of these remedies can be costly to implement. We investigate the utility of a simple statistical model for the mapping of quantitative-trait loci that incorporates multiple phenotypic or diagnostic endpoints into a gene-mapping analysis. The approach considers finding a linear combination of multiple phenotypic values that maximizes the evidence for linkage to a locus. Our results suggest that substantial increases in the power to map loci can be obtained with the proposed technique, although the increase in power obtained is a function of the size and direction of the residual correlation among the phenotypes used in the analysis. Extensive simulation studies are described that justify these claims, for cases in which two phenotypic measures are analyzed. This approach can be easily extended to cover more-complex situations and may provide a basis for more insightful genetic-analysis paradigms.