Complex Heterogeneity in the Utility of a Surrogate Marker
Ouvert à tous, en présentiel.
University of Texas, Austin, USA.
For clinical trials that typically require lengthy or invasive follow-up procedures, the identification and evaluation of surrogate markers offers a promising path towards obtaining quicker and less burdensome results.
Many methods have been proposed to measure the utility of a surrogate marker in predicting the treatment effect of interest.
However, the existing literature largely focuses on statistical validation of a surrogate marker overall, and generally fails to identify when a surrogate marker may be useful for certain subpopulations and not for others. Understanding potential heterogeneity is critical when the intent is to replace the true outcome with the surrogate marker in future clinical trials.
To this end, we propose three methods for evaluating complex heterogeneity with respect to multiple baseline covariates: a fully parametric model, a semiparametric two-step model utilizing two-dimensional kernel smoothing, and a varying coefficient model.
We demonstrate the performance of our methods on simulated data as well as AIDS clinical trial data, and discuss the tradeoffs of using the different methods depending on the setting.