This research proposed a Bayesian neural controlled differential equation (BNCDE) for estimating treatment effects over time in healthcare. Traditional methods are limited to point estimates, but the BNCDE provides meaningful posterior predictive distributions of the potential outcomes, thereby providing uncertainty estimates. This method models the time dimension through a system of neural controlled differential equations and neural stochastic differential equations, allowing for tractable variational Bayesian inference. This novel method is the first to provide uncertainty estimates of treatment effects in continuous time, offering significant potential for improving decision-making in healthcare.
Publication date: 26 Oct 2023
Project Page: https://arxiv.org/abs/2310.17463v1
Paper: https://arxiv.org/pdf/2310.17463
