Figure 1.
Causal graphs for ITE estimation. (a) a direct model ignoring treatment. (b) a naive model including treatment as a feature. (c) the proposed causal framework where patient status influences both treatment and outcome. To estimate the true effect of treatment, we must disentangle its influence from the learned representation of a patient’s status Refer to the image caption for details.The image shows three labelled panels a, b, and c that present causal model structures. Panel A is titled Direct model and shows x labelled Status pointing to y labelled I T E. Panel B is titled Naive treatment model and shows x labelled Status pointing to y labelled I T E, and t labelled Treatment also pointing to y. Panel C is titled Our proposal and shows x labelled Multimodal Status pointing to both t labelled Treatment and y labelled I T E. Arrows indicate the direction of relationships in each model.

Causal graphs for ITE estimation. (a) a direct model ignoring treatment. (b) a naive model including treatment as a feature. (c) the proposed causal framework where patient status influences both treatment and outcome. To estimate the true effect of treatment, we must disentangle its influence from the learned representation of a patient’s status

This Feature Is Available To Subscribers Only

or Create an Account

Close Modal
Close Modal