Abstract
The importance of genetic determinants and risk factors of diseases has been consistently recognized in genetic epidemiology, which is one of the fastest growing areas in genomic medicine. Familial clustering is a common characteristic of genetic related phenotypes, providing vital insights into the etiology of diseases by establishing the relative contribution of genetic and environmental factors. The availability of family data has opened up new opportunities for studying genetic and environmental contributions to diverse diseases. Family data overcome the limitations of statistical power common in twin data analysis, but also enhance the breadth of genetic information.
The generalized linear mixed model has provided a central conceptual framework that allows estimation of the genetic and environmental contributions with adjustment for various epidemiological risk factors. However, estimation often requires high-dimensional integrals to integrate out the random effects and in the models that we considered this is general analytically intractable. Since we have to deal with large datasets with sparse binary outcomes, computation has been another stumbling block in the analysis of realistic models.
This thesis focuses on the analysis of population-based family data, for application in cancer, perinatal diseases and psychiatric disorders. We have closely investigated the marginal and hierarchical-likelihood approaches, and also considered ascertainment approaches for both binary traits and age-at-onset traits. We demonstrate that the newly developed methodologies for the analysis of family data are highly flexible and allow straightforward handling of covariates.
