On learning disentangled representations for individual treatment effect estimation

•To alleviate the problem of selection bias in individualized treatment effect estimation.•To disentangle outcome-oriented and treatment-oriented factors from input patient features by using representation learning.•To employ an individualized orthogonal regularization to eliminate confounders in the latent representations.•Evaluation is done using a semi-simulated dataset and a real-world datasetfor the task of individualized treatment effect estimation.