MODEL-BASED ESTIMATES OF POPULATION ATTRIBUTABLE RISKS FOR ORDINAL DATA

摘要

Threshold-specific population attributable risk measures are developed to account for ordinal disease classifications. A cumulative logit model is utilized to formulate the threshold-specific risks as functions of the underlying model parameters. Covariate-adjusted and overall attributable risk measures are proposed to quantify the impact of exposure to an ordinal risk factor on an ordinal disease classification, in the presence of a confounding variable. These methods are developed under prospective and cross-sectional sampling designs. The asymptotic dispersion matrices of the risk estimates are obtained using multivariate Taylor series expansions which incorporate the sampling variation of the estimated model parameters and the appropriate estimates of risk factor prevalences. These methods are illustrated within the context of a health examination data, investigating the potential influence of body mass index, adjusted for race, on the prevalence distribution of diastolic blood pressure among adult women in the U.S.