The next Shaman:From hype to value in medical AI

Artificial intelligence (AI) is increasingly promoted as a means to address growing healthcare demand, workforce shortages, and rising costs, yet evidence on its economic value and acceptability in routine care remains fragmented and often methodologically limited. This doctoral thesis examines how medical AI can move from technological promise to sustainable real-world value through a Health Technology Assessment (HTA) approach that integrates economic evaluation with stakeholder perspectives and technological maturity. Conducted within the European AICCELERATE project, the research focuses on two complex care settings—Parkinson’s disease and pediatric palliative care—where continuous monitoring, coordination, and patient and caregiver involvement are central to care delivery. The thesis first evaluates the methodological quality, risk of bias, and technological maturity of health economic evaluations of medical AI. The systematic review shows that many evaluations are performed while technologies are still at relatively early stages of development, often omit relevant implementation and operational costs, and may overstate cost-effectiveness. Building on these findings, an early cost-effectiveness analysis of an AI-enabled remote monitoring system for Parkinson’s disease demonstrates how economic modelling can be used to identify key drivers of value and uncertainty during technology development. The thesis subsequently incorporates stakeholder perspectives through qualitative research and discrete choice experiments. Across Parkinson’s disease and pediatric palliative care, acceptability was shown to depend not only on expected clinical or economic benefits, but also on trust, clinician oversight, data use, impact on daily life, reassurance, coordination, and preservation of human judgment. Overall, the thesis argues that evaluation of medical AI should be stage-appropriate, multidimensional, and aligned with technological maturity. It introduces Technology Readiness Levels (TRLs) as a maturity framework for HTA, demonstrates their use in systematic review and early economic evaluation, and integrates stakeholder preference elicitation into early assessment. The resulting TRL-informed HTA framework provides a practical approach for generating evidence that is proportionate to the maturity of AI technologies and relevant to researchers, developers, policymakers, healthcare organizations, and HTA bodies. The central conclusion is that successful translation of medical AI from bench to bedside requires more than technical performance: it requires realistic economic evidence, meaningful stakeholder engagement, and evaluation strategies that support responsible, efficient, equitable, and patient-centred implementation.