Transforming medical device clinical trials with agentic AI: a Taiwan-specific framework integrating smart data, regulatory evidence, and trial optimization.

Abstract

Medical device clinical trials present distinctive regulatory and methodological challenges-including device-specific classification, engineering-based endpoints, operator-dependent performance, device-related safety considerations, and lifecycle evidence obligations-that differ from those of pharmaceutical trials. In our structured synthesis, we identified no published framework that explicitly integrates agentic AI into medical device clinical-trial workflows within Taiwan's regulatory context. Drawing on a structured synthesis of 39 core sources (29 scholarly publications and 10 regulatory or standards documents), supplemented by targeted contextual literature on AI governance and implementation, we developed a five-layer framework comprising Regulatory Decision, Evidence Requirement, Data Integration, Agentic AI Optimization, and Adaptive Feedback. The framework links bounded agentic AI functions to TFDA regulatory decision-making, device-specific evidence requirements, device-derived and real-world data, and lifecycle evidence generation. Cross-cutting governance considerations include algorithmic bias and representativeness, transparency and traceability, human oversight and accountability, cybersecurity and privacy, and model-drift and lifecycle change management. The framework is a contextual proposal for Taiwan rather than a validated or implemented system. We provide actionable recommendations for regulators, investigators, AI developers, and infrastructure planners. External assessment and prospective feasibility testing are required before implementation.