Using Counterfactuals to Account for Treatment Failures in Clinical Trials

During the course of a clinical trial, subjects may experience treatment failure. For ethical reasons, it is necessary to administer rescue medications for such subjects. However, the rescue medications may bias the set of response measurements. This bias is of particular concern if a subject has been randomized to the placebo group, and the rescue medications improve the subject’s condition. The standard approach to analyzing data from a clinical trial is to perform an intent-to-treat (ITT) analysis, wherein the data are analyzed according to treatment randomization. The Food and Drug Administration (FDA) requires an ITT analysis from any clinical trial sponsored by a pharmaceutical company, and the National Institutes of Health (NIH) requires an ITT analysis of any clinical trial that it sponsors. Supplementary analyses may be performed in addition to the intent-to-treat analysis to account for the effect of treatment failures and rescue medications. Some common approaches include the following: (1) eliminating all subjects who experience treatment failure, (2) eliminating data after the onset of treatment failure, and (3) imputing the last observation prior to treatment failure for all observations occurring after the treatment failure. Following the work of Imbens and Rubin (1997) for randomized experiments with noncompliance, Shaffer and Chinchilli (2001) proposed a likelihood-based counterfactual approach in which subjects who would not fail under either treatment assignment were compared with respect to the response variable. The objective of this paper is to develop a Bayesian counterfactual approach and compare the operating characteristics with those of the likelihood-based counterfactual approach. Section 2 reviews the model and assumptions in Shaffer and Chinchilli (2001). The Bayesian approach is outlined in Section 3. In Section 4, the results of the simulations to compare the two procedures are presented. Final remarks are given in Section 5. 2. Likelihood-based Counterfactual Approach