Controlled Clinical Trials 24 (2003) 122–134 0197-2456/03/$—see front matter © 2003 Elsevier Science Inc. All rights reserved. doi:10.1016/S0197-2456(02)00321-5 ARTICLE IN PRESS Design and analysis of clinical trials with a bivariate failure time endpoint, with application to AIDS Clinical Trials Group Study A5142 A. Gregory DiRienzo, Ph.D.,* Victor DeGruttola, D.Sc. Department of Biostatistics, Harvard School of Public Health, Boston, Massachusetts, USA Abstract This paper investigates the use of a bivariate endpoint in clinical trials of antiretroviral drugs to treat infection with the HIV. While they provide the most reliable information about treatment effects, clinical events such AIDS or death are relatively rare among newly treated patients with HIV infection. As a result, surrogate endpoints, such as time to virological failure and time to regimen termination, are gener- ally used in place of clinical endpoints in studies that enroll patients without previous treatment expe- rience. Because no surrogate endpoint has been completely validated (i.e., shown to capture the full effect of treatment on clinical endpoint), it may be preferable to base treatment decisions on more than one surrogate. In this paper, we consider the bivariate endpoint of time to virological failure and time to regimen termination. Analysis of a bivariate endpoint requires appropriate statistical methods that maintain the experimentwise type I error rate when making simultaneous inference with respect to each endpoint. This article outlines such methods and compares their small sample properties using the simulation study that was executed for the purpose of designing a current AIDS clinical trial. © 2003 Elsevier Science Inc. All rights reserved. Keywords: Bonferroni correction; Log-rank test; Omnibus test; Power; Sequential testing; Type I error. Background and study design The choice of endpoint for studies of antiretroviral drug combinations in treatment-naive pa- tients is an area of current controversy [1]. Because of the low rate of clinical endpoints among such patients, investigators in these trials have relied on surrogate endpoints as a basis to compare treatments. These surrogates have included time to loss of virological control (virological failure) * Corresponding author: A. Gregory DiRienzo, Department of Biostatistics, Harvard School of Public Health, 655 Huntington Avenue, Boston, MA 02115. Tel.: +1-617-432-2831; fax: +1-617-432-2832. E-mail address: dirienzo@sdac.harvard.edu