Computing adjusted attributable fractions from two-phase case-control Data

Two-phase studies have proven to be a useful generalization of the standard case-control study: In the first phase study, disease status and partial or proxy information on covariates is known for a large number of patients. In a stratified sub-sample,the second phase study, complete and/or exact covariate
information is ascertained. Log odds ratios for the second phase covariates can be derived using both data sources. This talk develops maximum likelihood estimation of confounder-adjusted attributable fractions from data in a two-phase setup. An asymptotic variance formula is presented. The method is demonstrated with data from a recent occupational case-control study in the department of Lorraine, France. Our SAS-twophase package is used.