Abstract
<title>Abstract</title> <p>Artificial intelligence (AI) increasingly mediates diagnosis, prognosis, triage, prescribing, referral, monitoring and escalation of care, yet the evaluation of clinical AI remains split between model performance, implementation evidence and high-level ethics. This integrative review reframes AI-mediated clinical decision support as an auditable knowledge-governance system rather than as an isolated algorithmic tool. We conducted an integrative synthesis structured with scoping-review procedures and a provenance-coded evidence map. An author-curated full-text corpus of 119 records was deduplicated and screened; 108 records were retained for ethical and information-systems mapping. Coding focused on autonomy, algorithmic justice, professional responsibility, institutional responsibility and human dignity. The synthesis shows that governance, fairness and accountability are more frequently operationalized than relational autonomy and dignity. Two exploratory quantitative lanes were feasible. A mortality-related surveillance lane showed a favorable but non-confirmatory pooled relative-effect signal under modified Hartung-Knapp uncertainty (0.68; 95% CI 0.24–1.93; k = 2). A downstream clinical-action lane suggested increased clinician activation after AI/CDSS exposure (1.57; 95% CI 1.01–2.44; k = 4), but heterogeneity was substantial and the result does not establish patient benefit, clinical utility or health-system value. The contribution is a reproducible knowledge-systems architecture that links data provenance, model mediation, clinical action, patient deliberation and institutional accountability while explicitly limiting translational claims.</p>