Digitizing the Pre-consultation Experience: Impacts and Design Recommendations

Paper Title

Digitizing the Pre-consultation Experience: Impacts and Design Recommendations

Publication Info

• Topic area: Use of LLM-powered conversational agents for clinical pre-consultation.
• Keywords: Pre-consultation, conversational agents, large language models, healthcare technology, patient-centered care, clinical workflows, medical summaries, information transfer, digital health, patient empowerment.

Background and Problem

• Problem / challenge: Traditional pre-consultation methods (e.g., surveys, intake nurses) are limited in scalability and engagement. LLM-powered agents can automate this process but risk producing redundant or irrelevant information, burdening physicians and amplifying errors in electronic health records (EHRs).
• Significance: Improving pre-consultation can enhance patient-physician communication, streamline clinical workflows, and empower patients to take a more active role in their care.
• Motivation and related work: Existing pre-consultation methods improve efficiency and communication but are constrained by survey fatigue and overburdened medical staff. LLMs have shown promise in gathering and summarizing patient information but lack sufficient focus on curating clinically useful summaries while supporting patient-centered care.

Solution

• Proposed approach: An LLM-powered pre-consultation agent with two components: (1) a chatbot interface for gathering patient information and (2) a summary interface for curating and presenting the information for review by both patients and physicians.
• Novelty:
  1. Integration of conversational agents with editable summaries to bridge the gap between patient narratives and clinical documentation.
  2. Design considerations for balancing patient empowerment with physician efficiency.
  3. Evaluation of patient experiences with a focus on transparency, support, and empowerment.
• Procedure and key techniques:
  • Chatbot interface engages patients in a conversational exchange to collect medical history and concerns.
  • Summary interface distills the conversation into two formats: a narrative summary in lay language and a structured table using medical terminology.
  • Patients review and edit the summary to ensure accuracy before it is shared with physicians.
  • GPT-4 was used for both the chatbot and summarization processes, with prompts tailored to ensure clarity, engagement, and follow-up on incomplete responses.

Results

• Concrete findings:
  • 86% of participants strongly agreed they would use the system for future appointments.
  • 86% found the summary easy to understand, and 71% believed it would help physicians understand their medical background.
  • Participants appreciated the ability to review and edit summaries, which enhanced their confidence and sense of control.
• Advantage over baselines:
  • Enabled richer information disclosure compared to static questionnaires.
  • Fostered a sense of being heard and supported, unlike rushed interactions with intake nurses.
  • Improved transparency by allowing patients to verify and edit their medical summaries.
• Experiments / evaluation:
  • Conducted a simulated lab study with 14 participants who interacted with the agent using hypothetical medical scenarios.
  • Participants completed pre- and post-interaction surveys, think-aloud protocols, and semi-structured interviews.
  • Analysis included coding transcripts and Likert-scale survey responses.
• Limitations and future work:
  • Study conducted in a non-clinical setting with hypothetical scenarios, limiting ecological validity.
  • Future work should explore real-world deployments, assess quantitative outcomes (e.g., consultation efficiency), and address accessibility concerns for diverse patient populations.

Summary

This study introduces an LLM-powered pre-consultation agent designed to gather patient information and generate editable summaries for physicians. The system enhances traditional pre-consultation by enabling richer patient disclosure, fostering transparency, and empowering patients to take an active role in their care. Evaluation with 14 participants demonstrated high satisfaction, with 86% expressing willingness to use the system in future appointments. While the agent shows promise for improving patient-physician communication and clinical efficiency, future work should explore real-world deployments and address design challenges such as balancing detail, language preferences, and accessibility.