Logistic Regression Under Independence of Genetic and Non-Genetic Covariates in a Case-Control Study
In a case-control study of a rare disease such as type 1 diabetes, covariate information is often collected on a genetic factor and a continuous attribute such as age. In some instances, it is reasonable to assume that the attribute and genetic factor occur independently in the population. Under this independence assumption, we develop maximum likelihood estimators of parameters in a logistic model of disease risk. Estimates are based on data from both patients and controls and may be obtained by fitting a polychotomous regression model of joint disease and genetic status. Our results extend previous log-linear approaches to imposing independence between a genetic factor and a categorical attribute, thereby avoiding potential loss of information from discretizing a continuous attribute. We apply the method to investigate the effects of age and a variant of the glutamate-cysteine ligase catalytic subunit on type 1 diabetes. The results are compared to those obtained from a standard logistic regression analysis, which does not make use of the independence assumption.
This type of interdisciplinary work is a hallmark of our program in Applied Statistics at Simon Fraser University. For more information, please contact Ji-Hyung Shin (firstname.lastname@example.org) or her supervisors Jinko Graham (email@example.com) or Brad McNeney