AI-Enabled Clinical Decision Support for Standardized Nurse-Led Emergency Care Pathways: A Multisite Mixed-Methods Implementation Study

Abstract

Emergency departments face persistent challenges in triage consistency, delayed clinical decisions, fragmented workflows, and uneven adoption of artificial intelligence (AI) decision support, creating uncertainty about benefits for nurse-led care. This study examines how AI-enabled clinical decision support relates to standardized emergency nursing pathways and which organizational conditions influence implementation. Guided by the Consolidated Framework for Implementation Research, it uses a qualitative-led, cross-sectional documentary multiple-case synthesis of published quantitative, qualitative, and mixed-methods investigations, rather than an original quantitative hospital survey. The intended sample comprises 97 peer-reviewed publications, with documented emergency department and hospital cases as embedded comparisons; complete screening and extraction remain unverified. Key variables include predictive discrimination, triage agreement, care timeliness, protocol adherence, nursing acceptance, workflow compatibility, explainability, governance, and institutional readiness. Analysis employs structured evidence extraction, CFIR-guided thematic coding, descriptive comparison of reported metrics, cross-site appraisal, and provisional five-point ratings for seven hypotheses without pooled inference. In a source study of 135,470 emergency visits, machine learning achieved critical-care AUC 0.86 versus 0.74 for conventional triage, and hospitalization AUC 0.82 versus 0.69. A separate study of 1,491 assessments across seven departments found electronic triage agreement increased from 75.4% to 92.7%, while median assessment time increased from 312 to 347 seconds; this tool was not AI-based. Multisite AI validation using 468,167 encounters reported inpatient admission AUROC 0.873 and intensive-care AUROC 0.951. An implementation study interviewed eight rapid response nurses and seven emergency physicians, identifying three principal themes concerning trust, implementation practices, and workforce responsibilities in care. Provisional evidence ratings were 3/5 for hypotheses H1, H2, H3, H5, and H6, and 4/5 for nursing readiness (H4) and institutional variation (H7). Findings suggest meaningful predictive potential but insufficient direct proof of improved nurse-led adherence or patient safety. Responsible deployment requires nursing participation, independent clinical oversight, local validation, workflow adaptation, and ongoing safety auditing.