Multiple reading procedures: the performance of diagnostic tests.

Abstract

This paper examines the effects on sensitivity and specificity of various multiple reading procedures. Multiple reading procedures entail the performance of a diagnostic test several times and consequent assignment of the subject to an 'affected' or 'unaffected' group. A Unanimity rule (that is, all tests must be positive) leads to the largest predictive value positive, but may have unacceptably low sensitivity. The rule which classifies the subject as positive based on the majority of the tests increases both the sensitivity and specificity of the individual test. Variability of subjects' sensitivity and specificity (that is, case of correct diagnosis on a test) affects the performance of these rules. We also study alternative procedures in which the final test is a better, but more expensive, test.