Overview (updated 02-20-18):
The Multi-Reader Sample Size Sample Size Program for Diagnostic Studies performs sample size computations for diagnostic reader-performance studies. Such studies are commonly used in radiology, where radiologists evaluate cases based on corresponding images resulting from an imaging modality (such as mammography for breast cancer) with respect to confidence of disease. In such studies a typical reader-performance measure is the area under the ROC curve. In the most common study design, the factorial design, each reader reads all cases under each modality (or test).
The program includes options for specifying five different study designs, nonequivalence and noninferiority hypotheses, and three different inference situations. The program works for any type of reader performance outcome. The program requires parameter estimates that are either conjectured or obtained from pilot data. Pilot data estimates can be obtained from analyses of data sets that use the methods proposed by Obuchowski & Rockette and by Dorfman, Berbaum and Metz, and updated by Hillis and colleagues. The software provides various combinations of the numbers of readers and cases needed for a specified power, or alternatively provides the power corresponding to specified numbers of readers and cases.
Questions? Contact Steve Hillis (firstname.lastname@example.org).