Assessment of Three-class Diagnostic Tests when Disease Verification is Subject to Selection

In the assessment of a three-class diagnostic test, the receiver operating characteristic surface has been one of the established tools when subsequent definitive disease verification can be obtained. However, omitting cases that have no definitive verification can seriously bias the assessment. In this paper, the assumption about the selection mechanism for verification is made in order to adjust for the bias. Through sets of numerical simulations with differential selection mechanisms, the properties of adjusted estimates are demonstrated. The proposed method is used to evaluate the accuracy of using Mini-Mental State Examination in diagnosing the three disease states of Alzheimer´s disease.