Semiparametric estimation of major gene and random environmental effects for age of onset

Analysis of age of onset is a key factor in the segregation and linkage analysis of complex genetic traits, but is complicated by the censoring of unaffected individuals. Gamma frailty models have been used for age of onset in the biostatistics literature, but these models do not lend themselves to modeling the correlations due to genetic effects which segregate within a family. This dissertation proposes use of the multivariate relative risk regression models with random effects. More specifically, Cox model with latent major gene effects, and Cox model with latent major gene and family-specific random effects are developed. While the first model can only accommodate the dependence due to Mendelian segregation of a major gene, the second model can accommodate the dependence due to both gene segregation and shared unobserved environmental effects. It allows to test and to estimate the major gene effects in the presence of residual correlations. Incorporation of linked and biological markers information, and auxiliary genetic trait information into these proposed models is developed. Issues in study design and ascertainment correction are also discussed.; Generalized maximum likelihood methods are used for parameter estimation, using a Monte Carlo EM algorithm. Likelihood analysis of the proposed models is restricted by the difficulty in evaluating or maximizing the likelihood, especially when data are available for some of the members of an extended pedigree. Markov Chain Monte Carlo permits genotypic configurations to be realized from the posterior distribution given a current model and the observed data. Hence methods for likelihood analysis can be developed: Monte Carlo EM is used for estimation of the parameters and their variance-covariance matrix, and Monte Carlo likelihood ratios are used for hypotheses tests.; The proposed methods are applied to several simulated pedigree data sets for illustration. Data from the Alzheimer's disease research center in Seattle, and a case-control family study of breast cancer have been used as real data examples.; Multivariate survival data occurs frequently when family data are collected to detect familial aggregation and to identify genetic subtypes of diseases with a late age of onset. The proposed methods appear to be very useful in genetic epidemiologic studies where individual covariates and censored age of onset outcomes occur, especially when good estimates of effects at the individual level are important.