Framework for artificial intelligence implementation research in healthcare: synthesizing current evidence on barriers and facilitators

There is a growing number of artificial intelligence (AI) models; however, few have made it into routine clinical practice to date. We describe the synthesis of existing knowledge on the barriers to AI implementation in healthcare settings into a consolidated theoretical Framework for Artificial Intelligence Implementation Research in Healthcare (FAIIR-H) to help guide future implementation evaluation and planning. We undertook a scoping literature review; elevant articles published between 01/2020 to 05/2024 were retrieved from Medline, Embase and Web of Science. Additional relevant frameworks were then identified through snowball sampling. Constructs and domains were consolidated across theories using affinity diagramming. The final consolidated framework included 12 domains and 63 constructs across five themes: Design and Development (data, model performance, design, regulation), Organization and Culture (resources, readiness, buy-in), and Deployment and Maintenance (integration, user-centredness, maintenance), with Quality and Safety, and Equity as separate, overarching themes. The findings of this work will help to guide the development of implementation plans tailored to current evidence on known barriers and enablers of adoption, and guide the systematic evaluation of the implementation of AI-driven technologies in healthcare settings.