GPT-MEDIC

AI holds great promise for improving patient care, but challenges related to data complexity have hindered its translation into clinical practice. Electronic health records in intensive care units are particularly difficult to leverage — they capture a complex interplay between the patient’s health and clinical decisions made in response.

GPT-MEDIC develops novel approaches to directly model this complexity. Funded by an ERC Starting Grant (€1.5M, 2026–2031), the project pioneers generative pre-trained models for clinical event streams, enabling flexible risk prediction across outcomes and hospitals. Training leverages federated infrastructure across European ICUs, covering up to 1 million patients with over 30 billion clinical events — the largest harmonised ICU dataset to date.

We work with collaborators at Amsterdam UMC, UCL, and the University of Cambridge.

Funding: ERC Starting Grant 2025