JMIR Publications

ABSTRACT

Background:

Implementing artificial intelligence (AI) in healthcare involves more than technological capability because adoption is shaped by trust, clinical workflows, governance, organisational conditions, and patient safety. However, longitudinal evidence from professional healthcare discussions remains limited.

Objective:

This study aimed to identify longitudinal patterns and implementation challenges related to trustworthy AI in healthcare and to synthesise these findings into a systems-oriented conceptual framework.

Methods:

A mixed-methods text mining study was conducted using longitudinal discussion data from an international healthcare quality learning community collected between September 2024 and July 2026. After data screening and preprocessing, 799 discussion posts were analysed using temporal analysis, term frequency–inverse document frequency, document-level keyword co-occurrence analysis, latent Dirichlet allocation (LDA), and automated Systems Engineering Initiative for Patient Safety (SEIPS) classification. The computational findings were integrated with contextual interpretation to develop a corpus-informed conceptual framework.

Results:

The analysis identified sustained discussion activity across the observation period and recurring attention to AI, healthcare, data, trust, bias, and patient safety. A four-topic LDA model identified four latent thematic structures within the discussion corpus. SEIPS classification showed that Tools and Technologies accounted for 390 of 799 classified comments (48.81%), while Tasks accounted for 321 (40.18%); together, these dimensions represented 88.99% of classified comments. Person, Organization, and Internal Environment were less frequently represented. Across the integrated analyses, trust was closely associated with concerns regarding transparency, governance, privacy, workflow integration, clinical decision-making, and patient safety.

Conclusions:

AI implementation was discussed primarily in relation to technological characteristics and healthcare work processes, while human, organisational, and environmental considerations were less prominent in the automated lexical classification. These findings support a socio-technical interpretation of trustworthy AI implementation. The Trust-Centred Artificial Intelligence Implementation Framework provides an empirically informed conceptual synthesis of the patterns identified within the analysed discussion corpus. Further research is needed to test its transferability and validity across healthcare settings and implementation contexts. Clinical Trial: Not applicable.