Post by Stella Champeaux

“Clinical AI is a living tool, not a finished product.”🌳 Those were my closing words during my oral presentation at the European Society of Paediatric and Neonatal Intensive Care - ESPNIC 2026 International Congress in Malta last week. I am incredibly grateful to the Great Ormond Street Hospital DRIVE unit for funding me through the Clinical Informatics Research Programme and giving me the opportunity to attend and present this work. Over the last two years, I have been working with colleagues at GOSH and UCL on a simple but important question: 💡Why does AI fail to travel?💡 Our project evaluated a sepsis prediction model developed in the Children's Hospital of Philadelphia (CHOP) NICU and tested whether it could be transferred directly into a mixed ICU environment at GOSH. The findings were striking: 🔹 Performance deteriorated when the model was transferred directly between hospitals. 🔹 Local retraining and adaptation using GOSH data largely recovered performance. 🔹 Models appear to learn institutions as well as physiology, reflecting local patient populations, workflows, and clinical practice. These findings reinforce the importance of local evaluation and adaptation before deploying clinical AI tools in new healthcare settings. A special thank you to Mark Peters and Alex Brown for reviewing my presentation and providing invaluable feedback and guidance. It was a privilege to share this work and discuss the future of clinical AI with colleagues from across Europe. #ESPNIC2026 #ClinicalAI #DigitalHealth #AIGovernance