Deep Representation Learning for Individualized Treatment Effect Estimation using Electronic Health Records.

•To predict individualized treatment effects using electronic health records.•To formulate ITE estimation as a representation learning problem.•To learn the latent representations through multi-task deep learning.•Based on the learned representations, KNN is adopted to estimate the counterfactual outcomes in an interpretable manner.•Evaluation is conducted on a benchmark semi-simulated dataset and a real clinical dataset pertaining to heart failure.