Abstract:
|
Comparative diagnostic studies in which each patient has two tests conducted will generate correlated ROC curves. The traditional ROC comparison methods applied on the correlated data can result in incorrect statistical inference. Furthermore, to design and apply group sequential method in these comparative trials, we need to derive the theoretical variance-covariance structure and the joint distribution of sequential statistics. We first derive the theoretical covariance structure of sequential correlated ROCs' difference and further verify the findings through simulation studies. Based on the independent increments covariance structure that we have proved, we conduct group sequential studies for comparing ROC curves on both simulated and real data.
|