This paper examines why national learning health systems repeatedly fail to translate local clinical data into sustained system-wide learning. Challenging the assumption that learning emerges through data aggregation alone, it argues that the core problem is architectural rather than technical. Building on insights from the Clinically-Grounded Systems (CGS) series, the paper outlines principles for designing health data platforms that scale from local clinical workflows to national learning without destabilising care delivery. It proposes a layered system architecture that preserves clinical flow, distributes cognitive load, embeds legitimate governance, and tolerates variability as a source of learning. As CGS Paper 5, this publication integrates prior analyses of system emergence, human-centered logistics, governance, and failure modes into a coherent framework for future learning health systems and AI-enabled health data platforms. It is intended for clinicians, health system leaders, policymakers, and designers involved in large-scale health data and digital infrastructure initiatives.
Anita Domargård (Thu,) studied this question.