500 – Beyond Randomized Studies: Nonrandomized, Single Arm, or Special Studies
Why Not Consensus Reading with Multiple Readers for Evaluating a New (Test) Device?
Bipasa Biswas
CDRH/FDA
Diagnostic devices, like imaging devices, often require an interface with a reader, who evaluates and interprets to provide a final diagnosis. The diagnosis is a qualitative assessment and thus the reader's responses are yes/no or present/absent. Examples include Colon Capsule Endoscopy (CCE) with cameras that image the colon to detect polyps. This presentation will discuss why a consensus reading of the subject device under evaluation, is not appropriate, when in practice only a single reader is involved in evaluating the images. Since both the subject device and the comparator are evaluated on the same subject, the design is paired and the readers, all read the same images, thus binary data from such a design is correlated. Data from correlated binary data were simulated to assess the impact of consensus readings of the test device output in evaluating the performance measures.