Comparing the Performance of Composite Endpoints and Responder Definitions with a Unified Modeling Framework for Multiple Endpoints Within Clinical Trials

• Activity Number: 584
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): David Andrae*+
• Companies: PPD, Inc.
• Keywords: Composite Endpoints ; Responder Definitions ; Clinical Trials ; Multiple Outcome Measures ; Joint Latent Variable Model ; Random Effects Specification

Abstract

Clinical trials for many medical indications employ multiple outcome measures to assess efficacy due to treatment effects. Examples include patient-reported symptom ratings or a set of clinical or pharmacodynamic criteria needing to be met. To operationalize these sorts of data parsimoniously, composites of individually-collected data, eg, summed scores or responder definitions, are formulated and analyzed as a surrogate endpoint. The construction of composites, while sometimes a beneficial data reduction technique, results in loss of information. A unified modeling framework for a set of clinical outcomes is proposed and compared to a composite of those same endpoints. The performance of the different modeling approaches is assessed via simulation and then compared directly on example clinical trial data. Implications for clinical interpretation and reporting of results are discussed.