Binary Regression with Differentially Misclassified Response and Exposure Variables

• Activity Number: 42
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Li Tang*+ and Robert H. Lyles and Li Tang and Li Tang and Li Tang
• Companies: Emory University and Emory University and Emory University and Emory University and Emory University
• Keywords: Differential Misclassification ; Internal Validation ; Binary Regression ; HIV Epidemiology Research Study (HERS) ; 2 by 2 Table ; "Matrix" and "Inverse Matrix" Methods

Abstract

In epidemiologic studies, both exposure and response variables may be misclassified, and the potential threats to the validity of the analytic results are well-known. In existing literature, most discussion has been limited to nondifferential misclassification. Therefore, valid and accessible methods with which to handle differential misclassification are still in high demand. Here we begin with the common 2X2 table setting when both response and exposure are misclassified. We establish extensions of well-studied "matrix" and "inverse matrix" methods to facilitate differential misclassification adjustment in 2X2 tables. Then motivated by the HIV Epidemiology Research Study (HERS), we generalize the method to a standard binary regression setting by formulating a maximum likelihood framework that allows flexible modeling of misclassification in the response and exposure variables while adjusting for covariates. We place emphasis on the use of rich internal validation data available in the HERS in order to evaluate the differentiality in misclassification. The value of the method developed is demonstrated via simulation studies and through a detailed analysis of the motivating data.