A0611
Title: Robust sparse reduced-rank regression for estimating treatment effects
Authors: Ryoma Hieda - Doshisha University (Japan)
Shintaro Yuki - Institute of Science Tokyo (Japan)
Kensuke Tanioka - Doshisha University (Japan) [presenting]
Hiroshi Yadohisa - Doshisha University (Japan)
Abstract: To estimate treatment effects on multiple outcomes, a method based on robust, sparse, reduced-rank regression is proposed. The method is designed to handle cases with highly correlated outcomes and the presence of outliers, which often degrade estimation accuracy. By decomposing the treatment effect matrix into low-rank components and introducing sparse and robust constraints, the proposed method achieves stable and interpretable estimation. This framework also allows subgroup identification through covariate selection via sparsity constraints. The simulations show that the proposed method outperforms existing approaches in terms of mean squared error and rank correlation, particularly when treatment effects lie in a low-dimensional subspace. For the application, the results are provided for real clinical trials. This approach contributes to the field of personalized medicine by offering a robust and interpretable framework for heterogeneous treatment effect estimation across multiple outcomes.
