Performance Evaluation of New Trial-Level Surrogacy Measures in Binary-Binary Endpoint Scenario in Clinical Trials

• Activity Number: 372
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Yiyi Chu* and Qian Shi and Daniel Sargent
• Companies: The University of Texas Health Science Center and Mayo Clinic and Mayo Clinic
• Keywords: surrogate endpoints ; clinical trials ; trial-level surrogacy ; Clayton Copula

Abstract

The coefficient of determination from weighted least squares regression (R2WLS) is one of the frequently used trial-level surrogacy (TLS) measures to assess the surrogacy of putative surrogate endpoints (S) for true endpoints (T). However, the estimation of R2WLS ignores the estimation errors of treatment effects (TEs) on S and T within study, and cannot address the question of significance prediction. Motivated by these concerns, two new TLS measures were proposed to account for TE errors (R2Z) and estimate significance prediction accuracy (R2AUC) based on the TE test statistics. The performances of these measures are evaluated and compared with R2WLS in large-scale simulation studies in the context of binary S and binary T. Clayton Copula model was used to generate correlated binary S and T on individual subjects within each trial. Our findings suggest that R2Z and R2WLS perform similarly; but R2AUC presents the largest estimation variability. The impacts of number of trials, sample size within trials, and range of the TEs were further evaluated.