PREVALENCE ESTIMATION USING DIAGNOSTIC-TESTS WHEN THERE ARE MULTIPLE, CORRELATED DISEASE STATES IN THE SAME ANIMAL OR FARM

This paper contains a theoretical investigation of the effect of multiple testing on disease prevalence estimation. The material is based on the assumption that the cluster or unit of concern (for example, an animal or farm) may contain some diseased and some non-diseased elements (for example, quarters within cows or animals within farms). Cluster-level and clement-level prevalence are different. The similarity of disease states within clusters (measured by the intracluster correlation coefficient) affects the relation between cluster-level and element-level prevalence. This phenomenon, combined with imperfect test sensitivity and specificity, creates serious biases in the conventional estimators of cluster-level prevalence. The paper proposes a new estimator, called the 'empirical estimator', which exhibits much less theoretical bias than the conventional estimators. Theoretical variances and biases for the estimators are presented.