Systematic review finds healthcare AI publication governance remains fragmented

A systematic narrative review found that governance for artificial intelligence in healthcare publications has expanded from an emphasis on reporting model performance toward broader sociotechnical accountability, but the field remains fragmented and epistemically uneven. Researchers synthesized 100 publications, including 69 governance instruments and policies, and identified three persistent tensions affecting standards for AI-assisted scholarly publication.

The study searched PubMed, including MEDLINE, with 14 Boolean search strings covering 1 January 2018 through 19 May 2026. The database search was supplemented by a manual search of the EQUATOR Network guideline library. Investigators deduplicated the records and prioritized them using a relevance-scoring algorithm. Of 30,276 deduplicated records, 888 progressed to human screening, and 100 publications were included in the final review.

Among the included publications were 69 governance instruments and policies, which the researchers classified across four evidence tiers. The review also incorporated 31 contextual evidence sources. Together, these materials were organized into six thematic groups: predictive and diagnostic AI frameworks; sociotechnical and lifecycle governance; generative AI and large language model standards; specialty-specific extensions; surgical reporting updates; and editorial and publication policies.

Analysis of the literature identified four consecutive phases in governance development between 2018 and 2026. The first phase centered on transparency for predictive models, while the later trajectory broadened toward comprehensive sociotechnical accountability. The authors characterize this change as an evolution in the governance landscape rather than as an evaluation of any particular clinical AI system or a demonstration of improved patient outcomes.

The review showed that healthcare-publication governance now draws on instruments with different purposes and levels of authority. Predictive and diagnostic reporting standards address only part of that landscape, whereas sociotechnical and lifecycle approaches consider governance more comprehensively. Standards developed for generative AI and large language models form another category, alongside specialty-specific, surgical, editorial and publication-policy instruments that extend or operationalize governance for particular settings.

Three governance tensions emerged from the synthesis. The first is the trade-off between specificity and universal applicability: a framework can address a particular application in detail, but narrower standards may be less broadly transferable. The second is the gap between disclosure-focused standards and lifecycle governance, since requirements to document information do not by themselves amount to a comprehensive system of oversight across the technology lifecycle. These are structural features of the reviewed literature, not measured effects on publication quality or clinical practice.

The third tension is what the authors call epistemic conflation. This occurs when heterogeneous governance instruments are perceived as having equivalent authority because they are treated as interchangeable. The review’s classification into evidence tiers and six thematic groups highlights why instruments with different scopes and functions should not be assumed to provide the same form or degree of assurance. The findings indicate that the growing number of AI publishing standards has not yet created a unified governance system.

Based on its analysis, the review identifies harmonization across evidence tiers, adaptive updating mechanisms and lifecycle-oriented governance as priorities for advancing the field. These conclusions apply to the governance frameworks and reporting standards examined through the review’s stated search period, rather than to all AI activity in healthcare. The study is a narrative synthesis of policies, standards and contextual sources; it does not report an intervention trial, clinical performance estimate or direct effect on healthcare outcomes.

Subject of Research: Systematic review of AI governance frameworks and reporting standards in healthcare publications

Article Title: Transparency, accountability, and scholarly integrity: A systematic narrative review of artificial intelligence governance frameworks in healthcare publication

Article References: Silva, A., Canever, B., Lino, M., & Silva e Silva, V. (2026). Transparency, accountability, and scholarly integrity: A systematic narrative review of artificial intelligence governance frameworks in healthcare publication. PLOS Digital Health, 5(10), e0001783. https://doi.org/10.1371/journal.pdig.0001783

Image Credits: AI Generated

DOI: 10.1371/journal.pdig.0001783

Keywords: artificial intelligence, healthcare publishing, AI governance, reporting standards, scholarly integrity, generative AI, large language models, publication ethics, systematic review, sociotechnical accountability, Transparency, accountability