Model-based Estimation of Population Attributable Risk under Cross-sectional Sampling

The covariate-adjusted population attributable risk (PAR) measures the proportionate reduction in disease prevalence in the target population when the putative risk factor is removed, after adjusting for covariate effects. This paper extends the model-based approach developed for retrospective and cohort studies to the cross-sectional sampling design. An appropriate logit linear model is utilized to estimate the covariate-adjusted attributable risk. The asymptotic variance of this complex ratio estimate is obtained using Taylor series expansions which incorporate the sampling variation of the estimated model parameters and the appropriate estimates of risk factor prevalence. These methods are illustrated with cardiovascular disease risk factor data from the second National Health and Nutrition Examination Survey (NHANES II).