Feasibility of a Multimodal AI-Based Clinical Assessment Platform in Emergency Care: An Exploratory Pilot Study

Philipp Russ^1*Peter Michael Mross¹Gunter Kräling¹Fabian Lechner¹Mohamed Eldakar¹Nadine Schlicker¹Simon Bedenbender²Sylvain Brouwer¹Kirsten Zantvoort¹Andreas Jerrentrup³Ivica Grgic^1,2Martin Christian Hirsch¹

• ¹Institute for Artificial Intelligence in Medicine, University Hospital Giessen and Marburg, Marburg University, Marburg, Germany
• ²Department of Internal Medicine and Nephrology, University Hospital Giessen and Marburg, Marburg University, Marburg, Germany
• ³Department of Emergency Medicine, University Hospital Giessen and Marburg, Marburg University, Marburg, Germany

Background: Overcrowding in emergency departments (EDs) is a key challenge in modern healthcare, affecting not only patient and staff comfort but also mortality rates and quality of care. Artificial intelligence (AI) offers the potential to optimize ED workflows by automating processes such as triage, history-taking and documentation. To explore a potential approach to overcrowding, we developed a multimodal and modular AI-based platform that integrates these functions into a single system. This exploratory pilot study investigated the feasibility of implementing the platform, focusing particularly on usability and patient trust in the system. Methods: Ambulatory patients triaged as non-urgent at the Marburg University Hospital ED were recruited. After providing written consent, they underwent an AI-supported initial assessment, including vital sign monitoring, automated triage, suspected diagnosis and automatic report generation. Participants then completed validated questionnaires on usability, Trust in Automation (TiA), and a supplementary self-developed survey. Results: A total of 20 patients were enrolled (70% female, 30% male; mean age 45.1 years), with an average interaction time of 10.6 minutes. The majority (80%) reported feeling safe, satisfied, and willing to recommend the system, while areas for improvement were identified regarding patient inclusion in decision-making and the perceived quality of information received. Usability was rated as excellent, with a mean System Usability Scale (SUS) score of 90.6. Although familiarity with the system was low, trust-related measures assessed using the TiA questionnaire were generally high. Conclusion: This exploratory pilot study demonstrates the feasibility and user acceptance of a multimodal AI platform in an ED setting. The system achieved high patient satisfaction, excellent usability, and a generally high level of trust. While these findings are limited to feasibility and perception, they indicate that such systems could serve as a basis for multicenter studies that directly evaluate impacts on triage accuracy, patient engagement, and clinical efficiency.