Evaluating the Utility of ROC Analysis in Blinded Evaluation of an Ongoing Clinical Trial
Daniel C. Bonzo, Amy Bian, Tosh Kameda
Biometrics and Data Management, Xenoport,Inc., Santa Clara, CA, United States

The importance of statistical monitoring in the ongoing evaluation of a clinical trial's risk-benefit to trial subjects and that of the Sponsor is well-recognized. When used appropriately, it can greatly impact the advancement or demise of a clinical development program. We take the view that blinded statistical monitoring is an appropriate tool to use to help reduce development efforts by identifying ineffective drugs when they are still in the exploratory phase of development. Likewise, it is an appropriate tool that can be used to trigger Phase 3 planning and development if it shows that the experimental drug is effective during its exploratory phase of development.

In this paper we propose an approach for use during a blinded evaluation of an experimental drug's risk-benefit which hinges on the identification of a useful surrogate for the actual treatment assignment. The method uses a treatment response indicator whose value can be determined by utilizing Receiver Operating Characteristic (ROC) analysis on the clinical trial's primary efficacy endpoint and using an appropriate global rating of change (GRC) scale as treatment response anchor.

The utility of the proposed approach was evaluated through limited simulation using various effect sizes of the primary efficacy endpoint and correlations between the GRC scale and the primary efficacy endpoint. Limited simulation performed to assess the performance of this method showed that as correlation increased the ability to predict a positive trial outcome also increased. In particular, the method started to have predictive value when the correlation reached the moderate range. However, the resulting surrogate treatment effect estimates were not as useful as they tended to be inflated with increasing inflation as correlation increased.

Keywords: receiver operatic characteristic (ROC) curve; global rating of change (GRC) scale; statistical monitoring; predictive profile

Biography: Daniel Bonzo is the Head of Biometrics and Data Management at XenoPort, Inc. which is located in Santa Clara, CA. His research interests include methods for analyzing mixed data and statistical methods application in clinical trials. Prior to his involvement in the pharmaceutical and biotechnology industries, he was Associate Professor at the School of Statistics of the University of the Philippines in Diliman. He also served as consultant in the Philippines' semiconductor and telecommunication industries. He obtained his PhD in 2003 from University of the Philippines in Diliman and did his post-doctoral research in 2005 at Rikkyo University in Tokyo.