Relationship of Different Surrogate Evaluation Frameworks to the Definition of a Valid Surrogate/Replacement Endpoint

• Activity Number: 219
• Type: Invited
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Peter B. Gilbert*
• Companies: Fred Hutchinson Cancer Research Center

Abstract

A goal of many biomedical research fields is identification of surrogate endpoints based on randomized efficacy trials. We have one randomized treatment (Z) for which two endpoints S and Y are both measured in each of the groups Z=0 and Z=1. S is an inexpensive endpoint measured shortly after randomization that is a candidate surrogate for the true clinical endpoint Y of interest. The primary objective of the trial is to learn about the treatment effect on Y, which is done by directly measuring P(y|Z). Where possible, future research would proceed by additional randomized trials that also directly assess the treatment effect on Y. Using the meaning surrogate = replacement, S is a valid surrogate for Y if in some sense measurement of P(s|Z) alone informs us about P(y|Z), without needing to collect data on Y. This talk discusses how different surrogate frameworks apply for this learning objective. This includes the result that principal surrogate based criteria generally do not imply a valid replacement endpoint, such that the chief application of the principal surrogate framework may be assessment of effect modification rather than of surrogates.