Fri, Feb 13, 2026, 08:00 – 09:00AM (PDT), 11:00 – 12:00 noon (EST)
Title:
Estimating Long-Term Treatment Effects by Integrating Randomized Trials and Real-World Data under Unmeasured Confounding, Study Heterogeneity, and Informative Dropout
Abstract:
Long-term treatment effects are central to regulatory and health technology assessment (HTA) decisions, yet are often not identifiable from randomized clinical trials alone due to limited follow-up, control-to-treatment crossover, and loss to follow-up. Real-world data can potentially supplement randomized evidence, but naïve integration is vulnerable to bias arising from unmeasured confounding, systematic differences between trial and routine care settings, and informative missingness.
We propose a semiparametric proximal difference-in-differences (Proximal DiD) framework to estimate long-term treatment effects in the original randomized trial population. The framework introduces two interpretable and realistic assumptions: conditional parallel trends given latent confounders, which accommodates time-varying effects of unmeasured disease severity while allowing persistent study environment differences, and missing at random given latent confounders, which permits informative dropout driven by latent health status. Identification is achieved through outcome and source bridge functions constructed from observable proxy variables, avoiding explicit modeling of unmeasured confounders.
We develop three estimators, namely proximal outcome regression, proximal inverse probability weighting, and a proximal doubly robust estimator that remains consistent if either the outcome or selection model is correctly specified and achieves semiparametric efficiency under joint correctness. Simulation studies demonstrate substantial bias reduction and robustness under realistic forms of model misspecification. An application to Alzheimer’s disease illustrates how the proposed method recovers clinically plausible long-term control trajectories in the presence of crossover and informative dropout, yielding materially different conclusions from existing approaches.
This work provides a transparent and practical framework for extending randomized evidence using real-world data in regulatory and HTA settings where long-term decisions must be made under imperfect information.
Speaker: Dr. Shu Yang
Shu Yang is a Professor of Statistics, Goodnight Early Career Innovator, and University Faculty Scholar at North Carolina State University. She received her Ph.D. in Applied Mathematics and Statistics from Iowa State University and completed her postdoctoral training at the Harvard T.H. Chan School of Public Health. Her research focuses on causal inference, real-world evidence, and data integration, particularly in the context of comparative effectiveness research in health studies. She also contributes extensively to methodological developments in missing data analysis and spatial statistics. Dr. Yang has served as Principal Investigator on multiple large-scale research grants from the NSF, NIH R01, and FDA U01. She has published over 115 peer-reviewed research articles and is a recent recipient of the Committee of Presidents of Statistical Societies (COPSS) Emerging Leader Award.
Website: https://shuyang.wordpress.ncsu.edu/
