Non-Inferiority Studies for Diagnostic Devices
*Lakshmi Vishnuvajjala, FDA\CDRH 

Keywords: Sensitivity, Specificity, predictive values, Likelihood raios

Non-inferiority studies are more complex for diagnostic devices as they usually involve two performance measures like sensitivity, specificity, or positive and negative predictive values or positive and negative likelihood ratios; if both measures of a pair are higher, or both lower, we can say one test is better, or worse than the other. If one is higher and one is lower, it is not always easy to determine which test is a better test. It depends not only on the magnitude of the difference, but also on the consequences of misclassification. For screening tests for example, this could also depend on the next step in the patient work-up. Area under the ROC curve is considered sometimes, but it can be problematic, as the consequences of false positives and false negatives can vary greatly not only between devices, but even for the same device, depending on the context in which it is used. We will consider some of these issues in determining non-inferiority of diagnostic tests.