A Bayesian Approach to Handle Missing Data in Evaluation of Diagnostic Medical Devices

• Activity Number: 648
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Qin Li*+ and Gene Pennello
• Companies: FDA and FDA
• Keywords: Missing data ; diagnostic device ; medical device ; Bayesian

Abstract

When evaluating the accuracy of a diagnostic medical device, either compare to the reference standard or to a predicate device that is already on the market, missing data, including missing reference standard, and/or missing test result (either from the single test or from any of the two tests), are commonly encountered. It is well known that naive methods such as reporting completer diagnostic results will introduce bias to the diagnostic accuracy estimates, and more sophisticated statistical approaches are needed. In this talk, we introduce a Bayesian approach to take into account the missing data under missing at random assumption. This talk extends the examples in the draft FDA "Statistical Guidance on Reporting Results from Studies Evaluating Diagnostic Tests" and may serve as a small add on to the missing data scenarios covered by the National Research Council report "The Prevention and Treatment of Missing Data in Clinical Trials".