Abstract:

Family studies are an initial step in identifying diseases that may have genetic components. A commonly used model to analyze such studies is the quadratic exponential model [Zhao and Prentice, 1990], though the model has the drawback of only holding for one family size. In addition, available methods of analyzing studies of familial aggregation of disease are not able to handle complex ascertainment schemes where the probability of a family being sampled is function of the family history of disease. Lastly, methods of accounting for age in these studies are not always appropriate. All proposed methodology is applied to a large family study of cancer conducted by the National Cancer Institute-sponsored Cancer Genetics Network.

The first chapter proposes two approaches to utilizing the quadratic exponential model in the presence of varying family sizes. The model is assumed to hold for a particular size, and larger families are split into all possible subsets of that particular size. Alternatively, small families can be assumed to contain "missing relatives", and a likelihood-based approach is taken to account for this missing data.

The second chapter proposes an approach to accounting for complex ascertainment schemes and utilizes the bivariate quadratic exponential model. Whether an individual is responsible for bringing a family into the study is considered a random variable, as well as the disease outcome. This approach extends that of Tosteson, Rosner and Redline [1991] by allowing for sampling to be a function of the family's history of disease.

Thus far, disease status has been considered binary, that is one is affected or unaffected, though ignoring age can lead to potential misclassification of unaffected individuals. The last chapter extends the quadratic exponential model to account for age at disease onset or censoring.