Adaptive Design and Analysis of Multi-Reader Multi-Case Studies

• Activity Number: 227 - Recent Advances in the Design and Analysis of Multi-Reader Imaging Studies
• Type: Topic Contributed
• Date/Time: Monday, July 29, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Weijie Chen* and Zhipeng Huang and Frank Samuelson and Lucas Tcheuko
• Companies: Food and Drug Administration and FDA/CDER and FDA/CDRH and FDA/CTP
• Keywords: adaptive designs; multi-reader multi-case studies; clinical trials; medical imaging

Abstract

The evaluation of medical imaging devices often involves multi-reader multi-case (MRMC) studies, in which both readers and cases are treated as random samples so that the conclusion can be generalized to both populations. Due to limited prior information, the sizing of such studies for a targeted statistical power is often inaccurate and it is desired to adaptively resize the study after an interim analysis. The challenge of adaptive sizing of MRMC studies lies in the fact that sizing involves both readers and cases and there is a complex correlation structure in MRMC data. We developed methods that allow resizing the study after an interim analysis while adjusting the critical value for hypothesis testing to control the type I error from inflation. Analytical results in the U Statistics framework have been derived under fully-crossed and split-plot designs using area under the ROC curve (AUC) or sensitivity/specificity as study endpoints. Our simulations show that the adaptive resizing can effectively adjust the power towards the target value while controlling the type I error rate under a variety of simulation conditions.