Name: Vance Berger

Firstname: Vance

Title: Mathematical Statistician

Institution: National Cancer Institute

Street: 6130 Executive Boulevard

City: Bethesda

Zip-Code: 20892-7354

Country: USA

Phone: (301) 435-5303

Fax: (301) 402-0816

Email: vb78c@nih.gov

Authors: VANCE W. BERGER (1,2) AND BONNIE J. LAFLEUR (3)

Title: Improving the Information Content of Clinical Trial Endpoints

Abstract: Because the severity of most diseases can be measured non-uniquely, different medical interventions, with different mechanisms of action, may be evaluated differently, even in the same patient population. Complicating this further is the fact that even for a given medical intervention, it may not be clear which endpoint, if any, will be most likely to show an intervention effect. For these and other reasons, clinical trials typically involve the evaluation of multiple safety and efficacy endpoints. As information accrues about diseases and patient populations, some endpoints may cease to be useful, but the trend would still likely be towards increasing numbers of potential endpoints. This trend would provide sponsors with increasing numbers of choices for the primary efficacy endpoint. If the endpoint selected as primary is not the optimal one for demonstrating the superiority of the experimental medical intervention, then a safe and effective medical intervention may mistakenly be found otherwise. On the other hand, the sponsor may find the endpoint that makes its case, and not study other endpoints that would have shown the experimental intervention to be inferior, in some way, to the control intervention. As such, the reliance of medical decisions on narrow primary endpoints can lead to inflation of both the Type I and Type II error rates. To address these concerns, we propose that all endpoints, especially the primary endpoint, be as informative as possible. This could be accomplished by combining some endpoints into composite endpoints. To avoid losing information in this transformation, we define the concept of information-preserving composite endpoints, and provide information concerning when this type of composite endpoint would be most useful. Specifically, we define the concept of joint fusibility of a set of endpoints, and note that this property confers upon the derived information-preserving composite endpoint the greatest amenability to statistical analysis. We also point out that using composite endpoints allows sponsors the most discretion in selecting their primary between-group statistical analysis. We illustrate these ideas with examples from a variety of therapeutic areas.