| Numerous approaches for the linkage analysis of quantitative traits in humans have been described. The most commonly used methods fall into two camps: the use of variance components models with an assumption of multivariate normality, and the use of Haseman-Elston regression models or its various extensions. The variance components approach has the advantages of high power to detect the genetic loci (called quantitative trait loci, QTLs) contributing to the trait, the easy incorporation of environmental and other covariates, and the ability to consider extended human pedigrees. However, variance components methods suffer from inflated type I error rates in the case that the normal model is not correct. Haseman-Elston regression, and its various extensions, are generally robust to non-normality, but suffer from low power and are generally restricted to sibling pairs or sibships.; We describe a general framework, making use of generalized estimating equations (GEE), for quantitative trait linkage analysis in extended human pedigrees. The approach is shown to unify the variance components and Haseman-Elston regression methods, as these methods are each special cases of this general framework, corresponding to different choices for a working covariance matrix. This work has provided new insight regarding the relationship between these methods, and, most importantly, suggests important new quantitative trait linkage analysis methods which exhibit both the power of the variance components approach and the robustness of Haseman-Elston regression. In particular, we propose use of a working covariance matrix based on estimates of the higher moments of the phenotype distribution. We used computer simulations to investigate the power and robustness of numerous methods; the proposed higher moment approach is seen to perform best, in terms of both power and robustness.; In addition, we investigated the problem of estimating QTL location. Through computer simulation, we show that the standard approach to this problem, of estimating the QTL location as the position giving the highest test statistic, performs reasonably even when marker informativeness varies considerably. Further, 1.5 LOD support intervals, based on a test statistic that has been converted to the LOD scale, may be reasonably used to obtain an approximate 95% confidence interval for the location of a QTL. While the coverage of the 1.5 LOD support interval is not controlled at 95%, it is conservative, but not overly so. |
