From Recommendation Correctness to Clinical Action: Transparency, Trust, and Verification in AI Decision Support

Open Access
Article
Conference Proceedings
Authors: Simranpaul Dhami, Dennis Vincenz, Kat Moran
Abstract

Artificial intelligence assisted clinical decision support systems are commonly evaluated using aggregate measures such as accuracy, sensitivity, and specificity. These measures do not establish whether a recommendation is dependable for an individual case. Clinical safety also depends on how interfaces communicate reliability and how clinicians interpret, verify, and use the output. This paper examines the pathway connecting AI recommendation correctness to clinical action through transparency, trust calibration, and verification. A qualitative, theory guided integrative framework approach was used to synthesize peer reviewed empirical and theoretical literature identified through Scopus, Web of Science Core Collection, and PubMed. The full multilevel safety framework provides context for a focused conceptual pathway linking recommendation correctness, interface communication, clinician reliance, verification, and action. The reviewed evidence suggests that correct recommendations can improve diagnostic performance, while incorrect recommendations can impair judgment when accepted without sufficient evaluation. Explanations, confidence displays, and uncertainty cues may help clinicians assess recommendations, but can also create false confidence or additional cognitive demands. Their value depends on whether they support interpretation and verification during clinical work. Clinician responses may include questioning, modifying, independently checking, or overriding an output. The proposed pathway identifies opportunities for interface design to support informed reliance and preserve clinician agency. It emphasizes communicating uncertainty and helping clinicians compare recommendations with clinical evidence before acting. These relationships remain conceptual and require empirical evaluation. Future studies should examine how reliability and uncertainty displays affect verification, reliance, diagnostic performance, and workload in realistic clinical workflows.

Keywords: Clinical Decision Support Systems, Recommendation Correctness, Interface Transparency, Trust Calibration, Automation Bias, Clinician Verification, Diagnostic Safety

DOI: 10.54941/ahfe1008358

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← Intra-Crew Dynamics of Multi-Unit Operations in a Nuclear Power Plant SimulatorAutonomous Hospitality Yachts: Designing Trustworthy Human-AI Experiences in Future Maritime Environments →
View all articles in Human Factors in Design, Engineering, and Computing →