Advancing human-centred AI in emergency care: a multimethod evaluation of RapidX AI using the proliferate AI framework.

Abstract

Background:Artificial intelligence (AI) has revolutionised healthcare by enhancing diagnostic accuracy and supporting clinical decision-making, particularly in emergency departments (EDs). RAPIDx_AI is designed to assist ED clinicians in interpreting cardiac biomarkers for suspected myocardial infarction (MI). However, user-centred evaluations of its effectiveness in real-world settings remain limited. Objective(s):To evaluate the effectiveness of RAPIDx_AI using the PROLIFERATE_AI framework, focusing on constructs such as comprehension, emotional engagement, usability, barriers, and optimisation strategies. Method(s):This multimethod study involved 24 ED clinicians from 12 metropolitan and regional South Australian EDs, supported by a transdisciplinary expert team. Data collection utilised structured surveys based on Expert Knowledge Elicitation (EKE), Bayesian statistical analysis, and qualitative feedback. RAPIDx_AI was assessed across five human-centred constructs to identify usability challenges and opportunities for optimisation. Result(s):RAPIDx_AI demonstrated "Good Impact" across key constructs, with median comprehension and emotional engagement scores of 0.34 and 0.31, respectively. Registrars and advanced trainees achieved the highest comprehension (median 0.466) and preference scores (median 0.458), while residents and interns reported the lowest comprehension (median 0.198) and usage (median 0.078). Experienced clinicians (>10 years) showed stronger emotional engagement (median 0.391). Recommendations included enhancing the interface, automating data entry, and implementing gamified training for novice users. Conclusion(s):RAPIDx_AI shows strong potential to improve decision-making for cardiac biomarkers, particularly among experienced users. Addressing usability challenges through tailored training, workflow integration, and interface refinements is crucial for broader adoption. These findings provide actionable insights for clinicians and policymakers to refine AI tools for enhanced clinical outcomes and scalability.Copyright © 2025, The Authors. All rights reserved.