All Times EDT
Keywords: Bayesian methods, Misclassification, attiributable risk, relative risk
Disclaimer: The contents of this presentation reflect the author's views and should not be construed to represent FDA's views or policies.
Estimation of the population attributable risk (PAR) has become an important goal in public health research, because they describe the proportion of disease cases that could be prevented if a set of exposures were eliminated from a target population as a result of some intervention. In epidemiologic studies, categorical covariates are often misclassified, leading to bias in PAR estimates. We present approximate and full Bayesian methods for point and interval estimation of the PAR in the presence of misclassification, by correcting for bias in both the exposure prevalences and relative risk estimates. We consider both main study/internal validation study and main study/external validation study designs, with and without the assumption of transportability of the exposure prevalences. We assessed the performance of these methods through a simulation study and applied these methods to estimate the PARs in the Health Professionals Follow-Up Study of dietary risk factors for colorectal cancer.