Recommended methods to compare the accuracy of two binary diagnostic tests subject to a paired design

摘要

This manuscript assesses different methods to compare the sensitivities and the specificities of two diagnostic tests. It studies the comparison conditioning on the disease status and testing each hypothesis test individually to an error , a global hypothesis test to simultaneously compare the parameters, and the individual comparison of the parameters along with the multiple comparison methods of Bonferroni and Holm. Simulation experiments were carried out to study the global type I errors and the global powers of the different methods, providing some general rules of application. When the prevalence is small and the sample size is small or moderate, it is recommended to compare the parameters individually along with the method of Bonferroni or Holm; in any other situation, it is recommended to compare the parameters simultaneously through a global test. The results were applied to an example of the diagnosis of coronary disease.