Guidelines for reporting artificial intelligence studies in medicines, pharmacotherapy, and pharmaceutical services: MedinAI development, validation and statement.

Abstract

Background: Artificial intelligence (AI) applications in medicines, pharmacotherapy, and pharmaceutical services are expanding, yet the lack of standardized reporting guidelines for scientific studies hinders transparency, comparability, and reproducibility in evidence-based healthcare decision-making.

Aim: To develop and validate comprehensive reporting guidelines for AI studies in these fields through expert consensus.

Method: Following the Guidance for Developers of Health Research Reporting Guidelines (Moher in PLoS Med, https://doi.org/10.1371/journal.pmed.1000217 , 2010), this study was conducted between May and October 2024 in two phases. Phase 1 involved drafting the initial guidelines through literature reviews and structured expert discussions by an internal committee. Phase 2 employed the Delphi method for validation and refinement. Twenty-six experts from nine countries, representing clinical pharmacy, pharmaceutical services, computer science, and AI, participated in the first round, with 21 completing the second round. Items were included if they received a median ≥ 7 on a 9-point evaluation scale, with ≥ 75% agreement defining publication consensus.

Results: The final MedinAI guidelines comprise 14 items and 78 sub-items across four domains: core aspects, ethical considerations in medication and pharmacotherapy, medicines as products, and services related to medicines and pharmacotherapy. All items achieved consensus (median = 8, with 95.2% agreement on publication readiness). MedinAI's items adapt to different AI development stages, and its structure operates in parallel with EQUATOR Network reporting guidelines for most study designs (CONSORT, STROBE, PRISMA, SPIRIT, etc.), ensuring versatility.

Conclusion: MedinAI provides validated reporting guidelines for AI studies in medicines, pharmacotherapy and pharmaceutical services, promoting transparency, comparability, reproducibility, responsible and ethical AI development for these fields.