A mixed methods evaluation of artificial intelligence-enabled macula services

Abstract

Introduction Ophthalmology incurs an increasing number of NHS hospital outpatient appointments, more than any other specialty. Neovascular age-related macular degeneration (nAMD) makes the third largest contribution to ophthalmology appointments. A sight-threatening imbalance between demand and capacity for these macular appointments could be addressed by a well-validated artificial intelligence (AI) technology, yet to be prospectively applied in NHS research or practice. This thesis aims to explore the factors which limit clinical AI implementation and develop actionable solutions for the implementation of AI-enabled macula services in the NHS. Methods The thesis applies a pragmatist approach, draws on the disciplinary field of implementation science, and uses mixed methods. Qualitative evidence synthesis, qualitative interviewing, a retrospective diagnostic accuracy study and theoretically informed analyses are performed. Findings Five distinct stakeholder groups illuminate the interdependent factors that influence clinical AI implementation. AI-enabled macula services offer broad value recognised by most stakeholders who prioritise evidence that implementation will not lead to sight loss. A simulated AI-enabled medical device used a candidate AI technology to independently make nAMD treatment decisions with less undertreatment and less overtreatment than consultant-led-care. The AI-enabled intervention to operationalise this medical device should delegate treatment planning decisions away from ophthalmologists and partly apply freed resources to improve patient-clinician communication quality. Healthcare pathway analysis proposed AI use and training to optimise the safety, effectiveness, and fairness of AI-enabled macula services. Conclusions The novelty of clinical AI and limited connectivity between its stakeholders sustain the implementation gap observed generally and within macula services. The problem of mismatched demand and capacity in macula services is real for all key stakeholders and an AI solution appears able to offer value to each. NHS organisations are free to locally implement AI-enabled macula services and this thesis provides evidence to inform if and how they choose to proceed.