Development and Clinical Evaluation of a Large Language Model-Based System for Generating Patient-Friendly...

Area of Science:

• Artificial Intelligence in Medicine
• Medical Informatics
• Patient Communication

Background:

• Standard echocardiography reports use complex medical jargon, hindering patient comprehension.
• This complexity can increase patient anxiety before consultations.
• Large language models (LLMs) offer a solution by translating technical data into understandable narratives, including longitudinal comparisons.

Purpose of the Study:

• To develop and evaluate an LLM-based system for generating patient-friendly echocardiography reports.
• To assess the system's professional safety, impact on patient comprehension, and short-term anxiety levels.

Main Methods:

• A two-stage study involving retrospective report generation (n=60) and prospective clinical evaluation (n=100 patients, n=85 family members).
• LLMs integrated clinical data and serial echocardiographic data for report generation.
• Report quality was assessed by clinicians and an external LLM; patient outcomes (comprehension, anxiety) were measured using Likert scales and the STAI-6 inventory.

Main Results:

• LLM-generated reports demonstrated high professional quality with minimal hallucination events.
• Patients and family members rated the patient-friendly reports highly, indicating improved understanding and addressing of concerns.
• Anxiety levels decreased significantly after patients read the LLM-generated reports, particularly in older adults and outpatients.

Conclusions:

• The LLM-based patient-friendly echocardiography reporting system is feasible and shows preliminary clinical utility.
• The system enhances patient understanding of echocardiographic findings and reduces short-term anxiety.
• Further research is needed to evaluate long-term outcomes due to the nonrandomized design.