Enhancing Patient-Centered Care: Examining the Design and Evaluation of Conversational Agents for Clinical Pre-Consultation

Abstract

Patients who feel heard, valued, and involved in their care consistently report greater satisfaction and better health outcomes. While patient-centered care is widely endorsed in theory, the realities of time-constrained appointments, physician burnout, and patient reticence often result in poor communication, misaligned expectations, and unaddressed concerns. Pre-consultation—the process of gathering patient information for physicians to review before the encounter—offers a promising solution. However, existing methods such as static intake questionnaires or clinician-led interviews are often too rigid or resource-intensive to scale effectively. My dissertation explores how conversational agents can bridge this gap by designing a pre-consultation chatbot capable of engaging naturally with patients and generating summaries for physicians. To inform its design, I first examined physician–patient communication strategies to identify key principles such as setting clear expectations, avoiding double-barreled questions, and using an open-to-closed-ended questioning structure with targeted follow-ups. Building on these insights, I developed a chatbot using a large language model (LLM) and deployed it in a medical clinic to gather real-world feedback on patient experiences. The findings showed that the chatbot was generally well-received, though its overly empathetic tone occasionally made patients feel uncomfortable.To evaluate the quality of information collected, I conducted an in-clinic study in which patients went through pre-consultation using one of three methods: a chatbot, a static questionnaire, or a medical resident. I found that both the chatbot and the resident elicited more informative responses, especially when follow-up questions were included. However, the chatbot remained inferior in its ability to recognize when further probing was needed, highlighting LLMs' limitations in contextual awareness. I then explored how information embedded in chatbot–patient dialogue could be summarized to help physicians prepare for consultations. Finally, I integrated these findings to develop a combined chatbot–summary system and examined how it influenced patients’ sense of agency in managing their healthcare. This research contributes to the design and evaluation of conversational agents for clinical care. Through empirical studies, real-world deployments, and user-centered approaches, it demonstrates that a pre-consultation chatbot–summary system can effectively gather and summarize information in a way that empowers patients and prepares physicians, ultimately enabling more practical patient-centered care.

Ähnliche Arbeiten