Evaluating Candidate Surrogate Endpoints with Principal Stratification

Loading Map....

Date(s) - Dec 2, 2008
3:00 PM - 4:00 PM

Center for Community Health


Presented by: 

Corwin Zigler


Potential-outcomes methods for causal inference have been widely recognized for their applicability to settings susceptible to posttreatment selection bias in which variables measured post-randomization complicate estimation of treatment effects. One such setting with particular relevance to HIV/AIDS research is the evaluation of biomarkers or surrogate endpoints that can be used to predict treatment effects on difficult-to-obtain clinical outcomes. As these biomarkers are obtained post-randomization, the potential-outcomes framework of principal stratification has recently been applied to the evaluation of candidate surrogate measures. This talk will discuss the relevance of principal stratification to settings susceptible to posttreatment selection bias with particular focus on the definition and evaluation of valid surrogates. In order to determine the validity of estimating treatment effects on the basis of the surrogate, we quantify the extent to which a surrogate endpoint lies on the “causal pathway” between the treatment and the clinical outcome. We introduce the definition of a “principal surrogate” as well as quantities used to estimate surrogate value. As a motivating example, we use a simulated HIV field study that examines the reliability of a salivary biomarker in predicting intervention effects on stress-related outcomes. Using a Gibbs Sampler for computation, we illustrate a strategy for estimating the causal effect predictiveness (CEP) surface to characterize the surrogate value of a biomarker.


Corwin Zigler is a 3rd year Ph.D. student in the UCLA Department of Biostatistics. As a trainee on the department’s AIDS training grant, he has focused his research on causal inference methods relevant to social and behavioral HIV/AIDS applications. Under the direction of his advisor, Dr. Thomas Belin, he is preparing his dissertation on novel applications and computational strategies for principal stratification with particular focus in the context of biomarkers in HIV/AIDS research.