BEYOND ACCURACY: EXPLAINABILITY, WORKFLOW, AND GOVERNANCE IN THE CLINICAL IMPLEMENTATION OF ARTIFICIAL INTELLIGENCE

Abstract

Artificial intelligence (AI) is increasingly discussed as a transformative force in medicine, yet its practical value depends less on algorithmic promise alone than on whether tools can be integrated safely, intelligibly, and sustainably into clinical care. This review examines how free full-text PubMed literature describes the real-world implementation of AI in healthcare and identifies the technical, organizational, ethical, and social conditions that shape adoption. A structured narrative review was conducted using a targeted PubMed search with a free full-text filter, supplemented by focused searches on explainability, governance, workflow, bias, and implementation in clinical settings. The final core synthesis drew on 21 peer-reviewed articles published between 2019 and 2025, including reviews, qualitative studies, survey studies, consensus guidance, and quality-improvement evaluations. Across the literature, implementation emerged as a sociotechnical challenge rather than a purely technical one. Recurrent themes included the gap between proof-of-concept performance and routine use, the need for practical explainability, risks related to bias and patient safety, workflow and workforce adaptation, and the importance of lifecycle governance. Real-world studies suggest that AI can reduce documentation burden, improve structured data capture, and support triage or decision support, but adoption remains uneven and often limited by poor integration, uncertain accountability, insufficient training, and weak post-deployment monitoring. The review concludes that clinical AI should be evaluated not only by accuracy but also by its fit with human work, organizational routines, fairness, and governance.