Measurement Error and Misclassification in Covariates: Impacts on the Main and Interaction Effects and the Bayesian Adjustment

• Activity Number: 463
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Shahadut Hossain*+
• Companies: United Arab Emirates University
• Keywords: Measurement Error ; Misclassification ; Errors-in-variables ; Exposure Model

Abstract

Measurement error in quantitative factors or misclassifications in categorical factors or both are unavoidable in most epidemiological studies. If ignored, such measurement errors and/or misclassifications can seriously distort the estimated effects of the risk factors on the disease outcome of interest. Thus, adjustments for measurement errors and/or misclassifications are necessary to have accurate estimates of the effects of the suspected risk factors, and their interactions on the disease risks. In this talk, I will discuss a general Bayesian flexible parametric framework for eliminating the deleterious impacts of measurement errors and/or misclassification while estimating the main and the interaction effects in logistic regression model. The application of the proposed flexible method will be illustrated with reference to gene-environment interaction studies. Being a parametric framework, the method is unified and thus can be applied to any member of the generalized linear models, and to the non-linear regression models as well.