Development methodologies and methodological quality of artificial intelligence reporting guidelines in healthcare: A scoping review
Honghao Lai, Bo Tong, Karen Spruyt, Paweł Jemioło, Akira Kuriyama, Ivan D. Florez, Małgorzata Bała, Loai Albarqouni, Myeong Soo Lee, Jiayi Liu, Yimeng Li, Sibo Du, Long GeBackground:
The integration of artificial intelligence (AI) into healthcare has intensified the need for transparent reporting. Although numerous AI-specific reporting guidelines have emerged, their development methodologies have not been systematically mapped. This review aimed to systematically map the development methodologies of reporting guidelines for AI in healthcare and to describe their adherence to established frameworks and AI-specific innovations.
Methods:
We conducted a scoping review following the Levac framework and Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews. Five databases (PubMed, Web of Science, Embase, China National Knowledge Infrastructure, and Wanfang) were searched from inception to January 2026. We included publications that presented reporting guidelines, standards, or checklists specific to AI in healthcare. Two reviewers independently screened records and extracted data. The completeness of the development process reporting was assessed using an 18-item framework derived from a previously published methodology; AI-specific methodological features were identified through thematic analysis.
Results:
From 24,268 records, 442 full-text reports were assessed, and 106 publications representing 98 distinct guidelines were included. Over half (56.1%) were published in 2023–2025. Development was led predominantly by academic institutions (74.5%) with international collaboration in 66.3%. Adherence to established development steps was heterogeneous: evidence synthesis (62.2%) and consensus processes (54.1%) were relatively well-reported, whereas postpublication activities were poorly reported (e.g., 1.0% with an update plan). We identified 44 AI-specific methodological features in 21 guidelines (21.4%), including living or dynamic guideline models (n = 9) and real-world validation approaches.
Conclusions:
AI reporting guidelines in healthcare show diverse and often incomplete adoption of established development methodologies, with emerging innovations such as dynamic updates and real-world validation. This map can inform future guideline development and harmonization efforts.