A0452
Title: Robust surrogate marker evaluation for survival outcomes: Leveraging instrumental variables in causal mediation analysis
Authors: An-Shun Tai - National Tsing Hua University (Taiwan) [presenting]
Abstract: Surrogate markers are commonly used in clinical and epidemiological studies to enable early assessment of treatment effects on long-term outcomes, particularly in survival analysis. However, traditional methods for surrogate evaluation can be misleading in the presence of unmeasured confounding, often resulting in the surrogate paradox, where the surrogate appears beneficial, but the treatment ultimately harms the true outcome. To address this issue, a robust causal mediation framework is proposed for surrogate marker evaluation that leverages instrumental variables to identify and estimate key causal effects, even when standard identification assumptions are violated. The method accommodates time-to-event outcomes and properly handles censoring, making it well-suited for survival data. Identification conditions are derived, a corresponding estimation procedure is developed, and the asymptotic properties of the proposed estimators are established. Extensive simulation studies demonstrate the robustness and efficiency of the approach compared to existing methods.
