The fairness paradox in public regulatory summaries of FDA-authorized artificial intelligence-enabled medical devices

Abstract

The U. S. Food and Drug Administration has published several types of public regulatory summaries of artificial intelligence-enabled medical devices, including 510(k) summaries, De Novo decision summaries, and premarket approval summaries of safety and effectiveness data. These documents are among the few regulatory texts routinely accessible to clinical institutions, patient advocates, educators, and researchers. Existing empirical audits have documented reporting deficits in areas such as demographic representation, model development details, and clinical validation data. This Perspective draws on these empirical findings and sociotechnical scholarship to formulate the saturation-gap paradox as a preliminary conceptual hypothesis requiring future empirical validation. It proposes that public summaries may be technically and procedurally detailed while providing only limited entry points for assessing fairness, clinical equity, and accountability. Rather than presenting a new systematic audit or inferring what FDA reviewers evaluated internally, this perspective focuses on the accessibility, structure, and accountability function of public disclosures. It considers three ways in which these disclosure genres may limit fairness scrutiny. These involve semantic abstraction, metricization and procedural governance, and accountability redistribution. Finally, building on this prioritized conceptual framework, this Perspective offers preliminary design propositions for disclosure reform and an illustrative TPACK-based educational outline as directions for future research, stakeholder consultation, and prospective development.