Skip to main content
. 2021 Feb 16;28(6):1197–1206. doi: 10.1093/jamia/ocaa346

Figure 1.

Figure 1.

Schematic of the Deep Propensity Network using a Sparse Autoencoder (DPN-SA) framework. The DPN-SA module performs dimensionality reduction of the input through a latent variable space and then provides propensity scores to the Deep Counterfactual Network with Propensity Dropout (DCN-PD) that calculates the potential outcomes for treatment exposures vs controls.