Preparing the Health Workforce for AI: Linking Health-System AI Adoption to Pre-Service and In-Service Training across the WHO European Region

Abstract

Artificial intelligence (AI) is being integrated into health systems at different speeds, raising questions about whether health-professional education is developing alongside technological implementation. This cross-sectional, country-level study examined associations between health-system AI characteristics and pre-service and in-service AI training across the WHO European Region using 2025 data from the World Health Organization Regional Office for Europe’s Artificial Intelligence for Health in the WHO European Region (AIRA) dataset. The overall dataset included 50 countries, with analytical sample sizes varying by indicator because of unknown or missing responses. Pre-service and in-service training status were examined in relation to three health-system measures: AI Application Implementation, AI Opportunity, and AI Adoption Barrier Burden. Composite indicators were derived from conceptually related AIRA items, with equal weighting within each index and “Don’t know” responses treated as unknown rather than negative. Descriptive statistics and Kruskal-Wallis tests were used, with Holm correction for multiple testing, Dunn-Holm post-hoc comparisons where appropriate, and epsilon-squared effect sizes. Established pre-service training was reported by 20% of countries and established in-service training by 24%. The clearest association was observed for in-service training: AI Application Implementation differed significantly across training levels after Holm correction (H(2)=10.21, adjusted p=.0182, ε²=.265), with established-training countries showing higher implementation scores than both the No and Under development groups. For pre-service training, implementation scores showed an ordered descriptive pattern, but the Holm-adjusted omnibus result did not meet the .05 threshold (adjusted p = .0501). AI Opportunity and AI Adoption Barrier Burden did not significantly distinguish training levels after correction. The findings indicate that established in-service AI training is associated with broader reported AI implementation at the health-system level, while the cross-sectional design does not permit causal inference.