Fusing trial and observational data improves treatment estimates
Prediction-Powered Data Fusion for Treatment Effect Estimation
Artificial IntelligenceMachine Learning
Summary
Randomized trials are reliable for measuring how treatments work but usually involve few people. Observational studies include many people but can have bias that confuses the results. The authors found a way to combine the small accurate trials with the large less reliable studies to get better estimates of how well treatments work, without making strong assumptions about the large studies. They created new methods that keep trial results unbiased while gaining precision from big data in observational studies. Their experiments show these methods work well for both overall and personalized treatment effects.
What this means in practice
- •For clinical data analysts: Combine small clinical trial and large observational data to improve treatment effect estimates without bias.
- •For healthcare data scientists: Develop personalized treatment recommendations by estimating individual treatment effects using fused trial and observational data.
Authors
Yonghan Jung, Shu Yang
Abstract
Randomized controlled trials (RCTs) identify treatment effects without confounding but are often small, whereas observational studies (OBS) are large but may be confounded. Many estimators combining a small RCT with a large OBS have been developed for the average treatment effect (ATE) and the conditional ATE (CATE). However, existing ATE estimators either make assumptions on the OBS or do not borrow enough power from them. The CATE has been studied less than the ATE. Existing CATE methods either assume the OBS are unconfounded, rely on a model of the confounding function, or accept bias in exchange for lower variance. We therefore propose a framework that, without special assumptions on the OBS, fuses the OBS and the RCT by preserving the unbiasedness of RCT-based estimation while borrowing power from the large OBS to boost precision. Applying this principle, we build an ATE estimator, AIPW-Fusion, with closed-form weights and confidence intervals, and two CATE learners, DR-Fusion and R-Fusion. Experiments corroborate our findings.