A0598
Title: An unsupervised adaptive approach for causal inference in competing risks data using separable effects
Authors: Jih-Chang Yu - National Taipei University (Taiwan) [presenting]
Yen-Tsung Huang - Academia Sinica (Taiwan)
Po-Chun Liao - National Taipei University (Taiwan)
Abstract: Causal effects are investigated when the primary time to event outcome is unobserved due to truncation from competing risks. Using the separable effects framework, the total effect of exposure is decomposed into a direct effect on the primary outcome and an indirect effect through the competing event. A robust, unsupervised method is developed based on the generalized transformation model, which accommodates various data distributions with minimal parameterization. This approach adjusts for confounders as covariates, remains computationally efficient, and is suitable for limited sample sizes. It is shown that the proposed estimator is asymptotically consistent and weakly converges to a Gaussian process, enabling valid inference. Simulation studies demonstrate that it outperforms existing nonparametric methods by reducing bias from model misspecification and dependent censoring. Finally, applying the method to REVEAL data reveals that HBV infection directly increases the risk of liver cancer mortality, even in the presence of competing risks.
