Use of Federated Learning for validating and updating privacy-preserving decentralized multi-study prognostic models in Traumatic Brain Injury

The article describes the implementation of a federated learning (FL) infrastructure for developing clinical prediction models without requiring sensitive data sharing between countries. The research team tested three IMPACT prognostic models to predict mortality and unfavorable outcomes six months after traumatic brain injury using data from two large studies: TRACK-TBI from the United States (441 participants) and CENTER-TBI from Europe and Israel (1,175 participants), totaling 1,616 participants. Both federated models achieved comparable results with AUC values between 0.77-0.88 and demonstrated better calibration and precision than original models. Federated learning proved feasible and efficient for developing and validating clinical prediction models with privacy protection, overcoming regulatory restrictions on data sharing. This infrastructure enables global collaboration and supports advances in data-intensive analytical methods including artificial intelligence.