G-Net: A Deep Learning Approach to G-computation for Counterfactual Outcome Prediction Under Dynamic Treatment Regimes

TL;DR

G-Net is a deep learning framework designed to improve counterfactual outcome prediction under dynamic treatment regimes by capturing complex temporal and nonlinear dependencies in time series data.

Contribution

This paper introduces G-Net, a novel deep learning approach for G-computation that handles complex temporal data with minimal assumptions and estimates dynamic treatment effects.

Findings

G-Net effectively models complex temporal dependencies.

G-Net provides accurate estimates of time-varying treatment effects.

Evaluation on simulated cardiovascular data demonstrates its potential.

Abstract

Counterfactual prediction is a fundamental task in decision-making. G-computation is a method for estimating expected counterfactual outcomes under dynamic time-varying treatment strategies. Existing G-computation implementations have mostly employed classical regression models with limited capacity to capture complex temporal and nonlinear dependence structures. This paper introduces G-Net, a novel sequential deep learning framework for G-computation that can handle complex time series data while imposing minimal modeling assumptions and provide estimates of individual or population-level time varying treatment effects. We evaluate alternative G-Net implementations using realistically complex temporal simulated data obtained from CVSim, a mechanistic model of the cardiovascular system.