STRATIFICATION IN NONPARAMETRIC ROC STUDIES

In many clinical experiments it is clear that external factors can affect the performance of diagnostic tests as these factors influence the distributions of separator variables. The authors investigate how stratification of cases and/or controls can be used in nonparametric receiver operating characteristic (ROC) studies, and develop statistical methods using stratification. The method can be used to simultaneously assess the ability of a diagnostic marker against several types of controls or cases. This allows a new method for the assessment of diagnostic tests.