[1Yin-A] Poster 1

[1Yin-A-03] Utility of Doubly Robust Estimation in Observational Data Analysis for Clinical ResearchCausal Effect Estimation in Observational Data with Confounding Due to Nonlinear Responses and Latent Variables

〇Yusuke Koyanagi¹ (1. TIS Inc.)

Keywords:

Causal inference,Doubly Robust Estimation,Clinical Research,Observable variable,Causal effect

In observational causal inference, doubly robust estimation combines propensity score and outcome regression models to mitigate treatment assignment bias. This study examines settings in which confounders satisfying the backdoor criterion are not fully observable and in which nonlinear physiological or biochemical responses are present. Through simulation studies, we evaluated the performance of causal inference methods under these conditions. The data-generating process assumed linear or nonlinear response functions and incorporated two causal scenarios: one in which observed variables directly affect both the outcome and treatment, and another in which latent variables directly affect the outcome and influence treatment through observed variables. In the analysis, latent variables were assumed to be unobservable, and models were constructed using only observed variables. The results show that under nonlinear relationships, outcome regression and propensity score models can be biased, whereas doubly robust estimation may attenuate such bias under certain conditions.