Evaluating a continuous biomarker for infection by using observed disease status with covariate effects on disease

A continuous valued serology test has been developed for the infective agent causing bovine digital dermatitis. Since infection cannot be observed directly, disease status as indicated by foot lesions is used as an imperfect proxy to evaluate the test. The desire to specify covariate effects on disease rather than infection status leads to some difficulties in developing a coherent probability model linking the variables of interest. We resolve these issues by making some plausible conditional independence assumptions, and we use the resulting model to investigate the ability of the serology test to discriminate between infected and uninfected animals. Copyright (c) 2009 Royal Statistical Society.