Feasibility of a Chatbot-Based Intervention to Support Case Management for Patients at Risk of Suicide in the Emergency Department: A Pilot Study
Background/Objectives: Emergency departments (EDs) are critical for suicide prevention, yet high workloads often hinder consistent case management. This pilot study assessed the clinical feasibility, usability, and acceptability of CAREBot, a rule-based chatbot designed to support ED-based case management for patients at risk of suicide, among patients and social workers. Methods: We employed an explanatory sequential mixed-methods design. Ten patients and three social workers used CAREBot and completed questionnaires assessing usability, perceived usefulness, and satisfaction. Patients’ psychological status was repeatedly assessed using the PHQ-9, GAD-7, and SBQ-R. Semi-structured interviews with social workers explored integration challenges and workflow impacts. Results: Patients rated ease of use (3.9/5) and information clarity (4.4/5) favorably, whereas emotional comfort was rated lower (2.5/5). Over 4 weeks, 191 URL access events were recorded among the 10 patients (mean, 19.1 per patient). Individual PHQ-9, GAD-7, and SBQ-R score trajectories varied across participants, with both increases and decreases observed. Social workers rated CAREBot favorably for quickly understanding patients’ risk level or condition (4.33/5) and ease of following the scenario flow (5.0/5). Qualitative findings suggested that CAREBot could support initial patient assessments through structured digital screening, although social workers raised concerns regarding risk-scoring accuracy and the management of multiple high-risk patients. Conclusions: CAREBot appears to be feasible and acceptable as an adjunct to ED-based suicide case management, although these preliminary findings require confirmation in larger controlled studies. A rule-based architecture was deliberately selected to provide predictable and controlled interactions in a high-risk environment. Future research should evaluate whether integrating large language models (LLMs) with rule-based safety mechanisms can improve conversational responsiveness and personalization while maintaining patient safety.
Authors
- Junsang Yoo (ORCID: https://orcid.org/0000-0003-2331-7644)
- Sejin Heo (ORCID: https://orcid.org/0000-0001-7387-4575)
- Hyunsu Choi (ORCID: https://orcid.org/0000-0002-5801-8882)
- Hansol Chang (ORCID: https://orcid.org/0000-0002-2624-0008)
- Ji Hyun An (ORCID: https://orcid.org/0000-0002-1628-9617)
- Minha Kim (ORCID: https://orcid.org/0000-0002-7178-0812)
- Won Chul Cha (ORCID: https://orcid.org/0000-0002-2778-2992)
- 전초원
- Sunjin Hwang (ORCID: https://orcid.org/0009-0009-5142-8334)
- Taerim Kim (ORCID: https://orcid.org/0000-0001-8340-9511)
Institutions
- Samsung Medical Center (KR)
- Sungkyunkwan University (KR)
Publication Details
- Journal
- Healthcare
- Published
- 2026-11-19
- DOI
- https://doi.org/10.3390/healthcare14223981
- Primary Topic
- Digital Mental Health Interventions
- Type
- article
- Field-Weighted Citation Impact
- 0.00