A1049
Title: Biased-corrected two-stage instrumental variable estimation in Mendelian randomization for binary outcomes
Authors: Chen-Hua Cho - National Tsing Hua University (Taiwan) [presenting]
An-Shun Tai - National Tsing Hua University (Taiwan)
Abstract: Mendelian randomization (MR) offers a powerful framework for causal inference under unmeasured confounding. While MR has been widely applied in epidemiological studies, its use with binary outcomes presents unique challenges. In particular, standard two-stage instrumental variable estimation for binary outcomes fails to converge to the true causal parameter, instead converging to a distorted value due to model misspecification and unmeasured confounding. Despite growing interest in this setting, existing methods remain limited in their ability to address this issue effectively. The aim is to propose a novel method tailored for binary outcome data within the MR framework, drawing on an analog of the two-stage least squares (2SLS) approach. The bias is formally identified due to unmeasured confounding and provides a more accurate and reliable biased-corrected estimation of the causal effect. The performance of the proposed method is evaluated through extensive simulation studies across a range of realistic scenarios, and it is benchmarked against existing approaches.
