Mapping National Governance of AI for Health: Protocol for a Global Scoping Review

AI is rapidly transforming health systems, expanding from diagnostic imaging and predictive analytics to large language model-enabled clinical decision support. However, significant governance challenges persist, including algorithmic bias, privacy risks, limited transparency, and inequities in access. Despite the proliferation of national AI strategies, global governance remains fragmented, and systematic evidence on how national policies address ethical, regulatory, and implementation requirements is limited. No comprehensive synthesis currently maps national governance approaches against established frameworks or documents or accounts for implementation realities across diverse contexts.

This scoping review aims to (1) characterize national approaches to AI governance in health, (2) assess alignment with established governance frameworks, and (3) identify implementation challenges and enabling factors.

Following the Arksey and O'Malley framework and PRISMA-ScR (Preferred Reporting Items for Systematic Review and Meta-Analyses Extension for Scoping Reviews) guidelines, we searched 6 databases and key gray literature repositories for sources published between January 2015 and April 2025. Eligible documents include national-level policies, empirical analyses, and official reports on AI governance in health. Data extraction is guided by a framework integrating World Health Organization AI ethics and governance guidance and the strategic priorities of the Global Initiative on AI for Health across 4 dimensions-ethics, regulation, implementation, and operations. Descriptive mapping, governance principle coding, thematic synthesis, and subgroup analyses will be conducted.

Our systematic search across 6 electronic databases identified 21,278 records: 3409 (16.0%) from PubMed, 5691 (26.7%) from Embase, 3661 (17.2%) from Web of Science, 334 (1.6%) from Latin American and Caribbean Health Sciences Literature, 568 (2.7%) from the China National Knowledge Infrastructure, and 7615 (35.8%) from the WanFang Database. After removing 968 (4.6%) duplicates in EndNote (version V.21; Clarivate), 2 researchers independently screened 20,310 (95.5%) titles and abstracts. From 21,278 database records and 972 gray literature items, 149 (0.7%) sources met the inclusion criteria. Quality assessment and full data extraction will be finalized by June 2026.

This review protocol addresses a critical evidence gap by providing a comprehensive mapping of national AI governance policies in health against an established governance framework. The planned review will inform evidence-based, equitable, and context-specific governance frameworks essential for safe and trustworthy AI integration in health systems.