Abstract
Regulatory agencies aim to ensure the safety and efficacy of medical products but often face legal and privacy concerns that hinder collaboration at the data level. In this paper, we propose federated learning as an innovative method to enhance data-centric collaboration among regulatory agencies by enabling collaborative training of machine learning models without the need for direct data sharing, thereby preserving privacy and overcoming legal hurdles. We illustrate how Swissmedic, the Swiss Agency for Therapeutic Products, together with its partner agencies, proposes to use federated learning to improve TRICIA, an AI tool for assessing incoming reports of serious incidents related to medical devices. This approach enables the development of robust, generalisable risk assessment models that can potentially improve current processes. A proof of concept was deployed and thoroughly tested during the 14th Global Summit on Regulatory Science using synthetic data with participants from Swissmedic, the U.S. Food and Drug Administration (FDA), and the Danish Medicines Agency (DKMA), with promising initial results. This innovation has the potential to serve as a roadmap for other regulators to adopt similar approaches to optimize their own regulatory processes, contributing to a more integrated and efficient regulatory environment worldwide.