Implementation Framework for AI Deployment at Scale in Healthcare Systems

Artificial Intelligence (AI) and its related digital health technologies are increasingly being used in the field of medicine. Despite promises of leading the future of personalised medicine and better clinical outcomes, implementation of AI-enhanced medical tools faces barriers for deployment at scale. We introduce a novel implementation framework that can facilitate digital health designers, developers, policymakers, and other stakeholders, including patient groups, to co-create and solve issues throughout the lifecycle of designing, developing, deploying, monitoring, and maintaining algorithmic models. It is aimed at large health systems that want to integrate multiple machine learning (ML) models with multiple modalities. This design thinking approach addresses increased clinical utility, moving beyond only improving model prediction. The framework combines a privacy-preserving approach with clinical parameters to establish a reward function for the optimisation problem of reinforcement learning to rank competing machine learning models. This enables the use of a reward function to implement explainable AI (xAI) methods, allowing clinical interpretability. It also incorporates governance mechanisms, such as version control for establishing audit trails and orchestration platforms to monitor and manage models within a health system. Overall, the proposed framework provides guidance to help users work towards human- centred AI design and develop AI-enhanced health-system solutions. It can help digital health designers ensure human-centric thinking when implementing highly automated systems.