Statistical properties of Continuous Composite Outcomes: Implications for clinical trial design

Jesse D Troy; Ryan A Simmons

doi:10.1016/j.conctc.2020.100655

Statistical properties of Continuous Composite Outcomes: Implications for clinical trial design

Contemp Clin Trials Commun. 2020 Sep 24:20:100655. doi: 10.1016/j.conctc.2020.100655. eCollection 2020 Dec.

Authors

Jesse D Troy^{1

2}, Ryan A Simmons²

Affiliations

¹ The Marcus Center for Cellular Cures, Duke University School of Medicine, USA.
² Department of Biostatistics and Bioinformatics, Duke University School of Medicine, 424 Erwin Road Suite, 1102 Hock Plaza Box 2721, Durham, NC, 27710, USA.

Abstract

Statistical efficiency can be gained in clinical trials by using composites of time-to-event outcomes when the individual component outcomes have low event rates. However, the utility of continuous composite outcome measures is not as clear. Efficiency can be either gained or lost by using a continuous composite outcome measure depending on several factors, including the strength of correlation between the component outcomes and the size of the treatment effect on each component. In this article we review these concepts from the standpoint of planning a new trial. Statistical properties of composites formed from normally distributed continuous outcomes are discussed. An example dataset is used to demonstrate concepts and complete mathematical details are provided. Finally, a conceptual model for clinical trial design with continuous composites is proposed that could be used as a guide to evaluate the utility of a continuous composite outcome in a future trial based on existing knowledge in the therapeutic area.

Keywords: Clinical trials; Statistics; Treatment outcome.