The comparison of diagnostic tests when one is based upon a discrete separator variable and the others are based upon continuous separator variables

摘要

A statistical method is developed for the comparison of diagnostic tests when one test is based upon a discrete separator variable and the other(s) is/are based upon a continuous separator variable(s). This is a situation in which present statistical methods are not always appropriate;The method so developed makes \u22point-wise\u22 comparisons of the sensitivities of the diagnostic tests after they have been set to have each of the specificities attainable by the one with the discrete separator variable. It is shown that, under very mild conditions, the vector of such comparisons is asymptotically multivariate normal. Monte Carlo studies show hypothesis testing procedures based upon this method have good power, but also tend to be anticonservative. This latter problem is seen to arise mainly from the required specificity estimation and to be alleviated with greater sample sizes. Several additional \u22fixes\u22 to this problem are also examined. Application of the method to two published data sets is presented.