Evaluation of candidate genes in family studies, generalized estimating equations and bootstrap approaches

grantor: University of TorontoIn family data, observations within the same family cannot be assumed to be independent. In genetic epidemiological studies, when the goal is to make inferences on the population average of certain covariate effects, one must not ignore the correlation in the data analysis. Two methods for analysing binary family data are evaluated. One method is the generalized estimating equations approach (GEEs). This is a semiparametric method that requires that the mean structure and a working correlation matrix are specified. The GEE is mainly used to model the relationship between covariates and the response, while treating the correlation as a nuisance parameter. The other method is a nonparametric bootstrap approach. A hierarchical bootstrap approach is proposed that involves two strategies to handle dependent data from nested structures, since the family structure can be viewed as such. A one-step logistic regression model is applied on bootstrap data ignoring the family structure, letting the bootstrap determine the variation. A simulation study is performed to compare the two methods using a multivariate logit-normal model to generate family data with a general correlation structure. An underlying multifactorial genetic model is assumed where affected status is determined by multiple candidate genes with both additive and dominant effects, an environmental factor, and a residual familial component that would exhibit the effect of unmeasured genes and/or environmental effects. The GEE and bootstrap methods are compared under different scenarios given in the simulation study. An application is made to a Genetic Analysis Workshop 9 data set.M.Sc