Primary Submission Category: Bayesian Causal Inference
Data Fusion for Heterogeneous Treatment Effect Estimation with Multi-Task Gaussian Processes
Authors: Evangelos Dimitriou, Edwin Fong, Karla Diaz-Ordaz, Brieuc Lehmann,
Presenting Author: Evangelos Dimitriou*
In the pursuit of reliable causal predictions, balancing internal and external validity is crucial. Randomised Controlled Trials (RCTs), favoured for their internal validity due to randomisation, often encounter challenges in generalising findings due to strict eligibility criteria. Observational studies conversely, provide external validity advantages through larger sample sizes but compromise internal validity due to unmeasured confounding. Integrating RCTs with observational studies is a promising way to get the best of both worlds. In this context, we propose a novel Bayesian nonparametric approach leveraging multi-task Gaussian Processes to seamlessly integrate findings from both RCTs and observational studies. Our method treats potential outcomes from distinct data sources as tasks in a multi-task learning framework, modelling their relationships through a shared covariance function. This integration enables unbiased estimations within the target population, addressing both internal and external validity concerns. Our approach outperforms cutting-edge methods in point predictions across the covariate support of the target population. Additionally, it provides a quantifiable measure of uncertainty, offering a comprehensive understanding of the reliability of the estimated treatment effects. We demonstrate the performance of our approach through multiple simulation studies and a real world data application, showcasing its robust performance in diverse scenarios.