Lay summary
Clinical trials usually tell us whether a treatment works “on average,” but they cannot show how each individual patient might respond. This is a problem because treatments that help some people can harm others. Machine learning, a type of artificial intelligence, offers a solution. It can analyse large trial datasets to detect complex patterns and predict how different patients will respond to treatment. We have pioneered this approach in intensive care, developing a machine learning model that estimates the individual benefit or harm of higher versus lower oxygen levels for patients on life support. In this programme, we will refine the model using data from the world’s largest oxygen trial and then test it in a new trial involving more than 24,000 patients. This will be the first major clinical trial to evaluate whether AI-guided treatment improves survival, setting an international benchmark for safe and personalised approaches to oxygen therapy and improving outcomes for critically ill patients.