Techniques to determine test protocol performance

This paper considers the problem of determining the performance of test protocols, i.e., combinations of diagnostic tests. Two techniques are developed which provide quantitative measures of performance. First, protocol performance can be calculated from clinical data. This technique is simple and straightforward. The primary limitation is that the necessary clinical data may not be available. A second technique is developed to predict protocol performance based on (1) protocol design, (2) individual test performance, and (3) test correlation. This second technique is applied to two types of audiological protocols: those to determine retrocochlear site-of-lesion, and those to identify hearing loss in newborns. Both techniques provide information essential to the design and selection of test protocols. The ability to predict performance is extremely important because clinical studies to evaluate protocols can take years to complete, or may be impractical.