AI Research as a Continuous Clinical Service

Advances in artificial intelligence (AI) have demonstrated potential in augmenting clinical decision making and expanding access to care, particularly in resource-limited settings.¹ Nevertheless, a persistent disconnect remains between reported research performance of AI systems and their effectiveness in real-world clinical practice. For example, although the COVID-19 pandemic precipitated a rapid increase in scientific publications, including numerous AI-based predictive models, systematic reviews have highlighted that many models, despite reporting high predictive performance, were inaccurate or had a high risk of bias.² This proliferation of inadequately validated models underscores the risk that impressive research model metrics may not translate into practical clinical applicability and emphasizes the need for rigorous validation and careful assessment of real-world utility before deployment.