Testing Secondary Endpoints in Group Sequential Trials
View Presentation View Presentation
Hsien-Ming James Hung, Division of Biometrics I, OB/OTS/CDER, FDA  *Hsien-Ming James Hung, Division of Biometrics I, OB/OTS/CDER, FDA 

Keywords: Group Sequential, Interim analysis, alpha spending function

Group sequential designs are increasingly employed to study clinical outcome endpoints, such as major adverse clinical events in cardiovascular trials. Such designs allow for stopping the trial early to conclude the treatment benefits of a test treatment are proven, determine the trial has no chance to show treatment benefits, or determine the test treatment is unsafe. Interim analyses are based primarily on the primary endpoint; thus, the primary endpoint is tested by repeated significance tests and handled using a pre-specified alpha spending function. In contrast, secondary endpoints are often tested after the primary endpoint achieves statistical significance. Hung, Wang and O’Neill (2007, Journal of Biopharmaceutical Statistics) report that if the secondary endpoint is tested at unadjusted alpha (usually at two-sided 0.05 level), the familywise type I error probability will be inflated and can nearly double the intended alpha level. Subsequently to this research, Glimm et al (2010, Statistics in Medicine) and Tamhane et al (2010, Biometrics) have confirmed the research finding and proposed that an alpha spending function be specified for testing the secondary endpoint and the optimal spending function is Pocock boundary. In this presentation, I shall revisit the issue of testing secondary endpoints from the design perspective and from the analysis perspective