ESTIMATING THE PREVALENCE OF MASTITIS AND OTHER ORGAN-SPECIFIC DISEASES IN THE PRESENCE OF WITHIN-ANIMAL DISEASE CORRELATION AND DIAGNOSTIC-TEST CORRELATION

In estimating the prevalence of a disease such as mastitis, which can affect some or all organs in an animal, an epidemiologist may use a diagnostic test of known sensitivity and specificity to test all organs in each animal and designate the animal as disease-positive if at least one organ is test-positive. The epidemiologist is then faced with the problem of choosing a prevalence estimator that accounts both for the sensitivity and specificity of the test and also for the fact that multiple tests have been made on the same animal. This paper examines the effect of disease distribution and test characteristics, including sensitivity, specificity and correlation of test results, on the performance of four animal-level disease prevalence estimators. The crude estimator (theta(A)) is the proportion of animals that lest positive in at least one organ; the empirical estimator (theta(B)) is obtained by correcting the crude estimator using the test specificity and the number of times the test was performed on each animal; the corrected estimator (theta(C)) is obtained by correcting the crude estimator for the sensitivity and specificity of the test, but not for the fact that multiple tests were made on each animal; the distribution estimator (theta(D)) is obtained by substituting observed distribution parameters into the beta-binomial density formula. The paper extends the theory from a former paper (Donald, A., 1993, Prevalence estimation using diagnostic tests when there are multiple, correlated disease states in the same animal or farm. Prev. Vet. Med., 15: 125-145) and reports the results of a simulation study that compares the accuracy and precision of the four estimators. The empirical estimator (theta(B)), which exhibits the lowest bias, is recommended as the best estimator in most circumstances. The results of the investigation are applicable not only to mastitis prevalence estimation, but also to the estimation of disease that occurs in small clusters, such as organs on the same animal or animals within small litters.