Modeling gene-disease relationships in the presence of environmental effect modifiers: A methodological study

Genetic studies generate large amounts of information, usually as single nucleotide polymorphisms in candidate genes and their relationship to diseases. Analyzing such data becomes more challenging for multifactorial traits. Very limited statistical methods are available for such analyses since these diseases are considerably rare with very low incidence rates. In many instants also, persons with these usually congenital abnormalities may not survive beyond a relatively short time after birth or the diseased fetus may not be carried through birth.;Successful public health policy formulations like the prenatal folate diet supplementation program has in some instances controlled recurrences of Neural Tube Defects in child births among women with a previous birth instance of NTD. This in a limited way is also true for Cleft Lip/Palate and for Diabetes. This additional reduction in the incidence compounds the rarity of the disease making the use of traditional methods of statistical analysis not efficient. Penetrance, the risk of the disease among persons with the genetic predisposition then becomes an interesting issue to be pursued.;This study shows the utility of the zero-inflated Poisson (ZIP) regression model as a relevant approach to modeling disease penetrance. Using zero-inflated regression modeling of penetrance, the zero-inflated penetrance models are applied to simulated data.;The simulated data were first analyzed as a conventional ZIP regression model with covariate effects. This was compared with an EM based analysis. Since the EM does not readily provide the covariance structure of the maximum likelihood estimates, Bootstrap sampling methods and Importance Sampling techniques were explored in obtaining estimates of the standard deviations of the MLE estimates which could then facilitate the possible construction of confidence intervals for the EM estimates for the determination of statistical significance.;Subsequently, hierarchical modeling techniques were employed in a Bayesian context to estimate the model. WinBUGS and R2WinBUGS were used as efficient statistical tools of analysis. By assigning appropriate priors to the hierarchical model and incorporating information on the potential function of the covariates, full Bayesian and empirical Bayesian methods were used to estimate the zero-inflated penetrance model.;This study has shown that the zero-inflated model provides a convenient, easily interpretable model technique for modeling gene-disease relations. The corresponding Bayesian models allow much extension in the usefulness of hierarchical modeling of zero-inflated gene-disease data and provide tremendous flexibility in analysis.