Identifying and formulating appropriate clinical problems for clinical prediction models and AI decision support: the RIVU-AI framework

Despite rapid advances in clinical artificial intelligence (AI), most clinical prediction and decision support models fail to translate into routine care. Existing frameworks provide guidance on model development, evaluation and implementation but provide limited guidance on how to identify and formulate clinical problems that are most appropriate for prediction modelling or AI decision support. We propose that many failures may stem from this misalignment. We introduce the RIVU-AI framework, which evaluates clinical problems across three domains to identify those most likely to yield meaningful, implementable and equitable decision support: (1) Risk versus Impact, (2) Variability and Uncertainty and (3) Actionability within an Identifiable deployment setting. By shifting focus from how models are built to how clinical problems and intended use cases are formulated, RIVU-AI provides a practical framework that may improve the translation of prediction models and AI decision support into clinical care.