Federated Causal Inference from Multi-Site Observational Data via Propensity Score Aggregation

The article presents a novel Federated Learning (FL) approach for estimating the Average Treatment Effect (ATE) from decentralized observational data without requiring centralized access to individual-level data. The method utilizes a federated weighted average of local propensity scores, calculated using Membership Weights (MW), to construct Federated Inverse Propensity Weighting (Fed-IPW) and Augmented IPW (Fed-AIPW) estimators. This approach outperforms traditional meta-analysis methods by effectively handling site-level heterogeneity in treatment assignment and covariate distributions, making it significant for practitioners needing to conduct causal inference across multiple data sites while adhering to privacy constraints.