Proposal Description

This invited session will highlight recent methodological advances in causal inference motivated by increasingly complex biomedical and health data. Modern studies often involve mediation mechanisms, survival outcomes, time-varying treatments, nonlinear and high-dimensional covariate structures, and growing interest in individualized treatment decisions. These challenges require new statistical tools that extend classical causal frameworks while preserving interpretability, valid inference, and meaningful uncertainty quantification.

The proposed session brings together four complementary talks at the frontier of this area. Xinyuan Song will present a GAN-based approach to causal mediation analysis, illustrating how modern generative modeling can enrich inference on causal pathways. Yi Li will discuss inference for the relative risk functional in deep nonparametric Cox models, advancing causal and survival analysis in flexible semiparametric settings. Liangyuan Hu will present a fully Bayesian structural nested failure time model for causal inference with time-varying confounding, addressing a central challenge in longitudinal observational studies where treatment decisions and patient trajectories evolve over time. Sijian Wang will introduce Bayesian-assisted conformal prediction of individual treatment effects, contributing new tools for uncertainty quantification in personalized decision-making.

Together, these talks reflect a unifying theme: the development of principled and flexible methods for drawing reliable causal conclusions from complex data settings where traditional methods may be inadequate. The session emphasizes several important frontiers in contemporary statistics, including causal mediation, survival analysis, longitudinal causal inference, Bayesian modeling, machine learning, and distribution-free prediction. Although the methodological themes are diverse, the talks are closely connected by their shared focus on modern causal questions arising in biomedical and health research.

This session is timely because causal inference continues to expand beyond classical low-dimensional settings into data-rich environments that demand methodological innovation. At the same time, there is increasing need for tools that not only estimate causal effects, but also characterize heterogeneity, quantify uncertainty, and support individualized decision-making. By bringing together experts working across these complementary areas, the session will provide a coherent overview of emerging directions in causal inference and their broader implications for statistics and data science.

The session will be of broad interest to researchers in statistics, biostatistics, epidemiology, and related fields who are developing or applying modern methods for causal inference in complex data environments.