Evaluation of Biomarkers as Principal Surrogates for Time-to-Event Endpoints

• Activity Number: 122
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Erin E. Gabriel*+ and Peter B. Gilbert
• Companies: University of Washington and University of Washington
• Keywords: Surrogate endpoints ; Causal effects ; Time-to-event ; Vaccine efficacy trail

Abstract

Since 2002 when Frangakis and Rubin proposed a new definition of a surrogate endpoint based on causal effects, a principal surrogate, several methods have been developed for the evaluation and comparison of biomarkers as principal surrogates. To our knowledge only one such method has been developed for a time-to-event outcome, based on a structural Cox proportional hazards model. We introduce surrogate evaluation methods for a time-to-event endpoint using Weibull structural risk models, which allow for the evaluation of waning surrogate value in the presence or absence of treatment efficacy waning. The causal estimand of interest can be identified even with missing potential outcomes by standard causal inference assumptions plus additional assumptions described by Gilbert and Hudgens (08 Biometerics), together with augmented trial design features outlined in Follmann (06 Biometerics) of baseline predictors of the potential surrogate and placebo subject cross-over to active arm and measurement of the potential surrogate. The new methods are motivated by and illustrated with a simulation study of a proposed multi-vaccine arm placebo-controlled efficacy trial design.