Federated Causal Inference from Multi-Site Observational Data via Propensity Score Aggregation
Researchers have developed a novel Federated Learning (FL) approach for causal inference from decentralized observational data. This method allows for the estimation of Average Treatment Effects (ATE) by exchanging aggregate statistics rather than individual-level data, addressing privacy and logistical constraints. The proposed technique utilizes a federated weighted average of local propensity scores, enabling flexible estimation with standard FL algorithms and improving data overlap compared to meta-analysis methods, especially when positivity assumptions are violated at the site level. AI
IMPACT Enables privacy-preserving causal inference from distributed datasets, potentially advancing research in fields with sensitive or siloed data.