Combining expert information for the analysis of animal health screening tests

Richard Evans
College of Veterinary Medicine
Iowa State University
*Candidate for the Assistant or Associate Professor Position

Friday, February 3rd
10:00am
Moos 2-690
Minneapolis Campus

Abstract:

Losses attributable to porcine reproductive and respiratory syndrome (PRRS) costs the swine industry over $600 million each year, and controlling PRRS disease status in herds is an industry priority. Several screening tests have been proposed, but no reference test (i.e.; "gold standard") exits, so simple determination of test sensitivity and specificity is not possible. Hui and Walter (1980) developed a multi-test, multi-population model to circumvent the no gold standard problem, and Joseph (1995) and Johnson (2001) developed Bayesian approaches. The Bayesian methods require a content expert to provide distributions for disease prevalence and test sensitivity and specificity. The original Bayesian models permit information from a single content expert. However, most veterinarians are traditionally more comfortable with consensus approaches to decision making. We explore models for combining information (sensitivities, specificities and prevalence) provided by several swine veterinarians, and using them in a modified Joseph (1995) model for the analysis of a PRRS ELISA screening test. We also investigate the use of heavy-tail prior distributions to "automatically" down weight expert information that is in conflict with the data.

A social tea will be held at 9:30A.M. in A434 Mayo. All are Welcome.
For more details contact 612-624-4655 or see http://www.biostat.umn.edu/seminar_academic.html