External control arms (ECA) can inform the early clinical development of experimental drugs and, in specific cases, can be accepted by regulatory agencies as evidence of treatment effect. ECA methods, such as inverse probability of treatment weighting (IPTW), allow to tap into real-world data or historical clinical trials to build a control group for a single-arm trial. In this work, we leverage a recent privacy-enhancing technology called Federated Learning (FL) to scale real-world data access across multiple centers. The objective is to better power the ECA while limiting patient’s data exposure. To this end, we introduce a federated learning
IPTW method for time-to-event outcomes called FedECA. Using synthetic data, we show that FedECA is equivalent to IPTW on pooled data and that it outperforms its federated analytics counterpart relying on matching-adjusted indirect comparison (MAIC). Additionally, we study the impact of adding differential privacy on the accuracy of our method and discuss its applicability. We perform all our experiments using Substra, an open-source FL software with proven experience in privacy-sensitive contexts.