Related Experiment Video
Updated: Jan 13, 2026

Quantitative Analysis of Cellular Composition in Advanced Atherosclerotic Lesions of Smooth Muscle Cell Lineage-Tracing Mice
Published on: February 20, 2019
Large Language Models in Patient Health Communication for Atherosclerotic Cardiovascular Disease: Pilot
Pengfei Li1, Yinfei Xu1, Xiang Liu2
1Department of Emergency and Critical Care Medicine, The Affiliated Suzhou Hospital of Nanjing Medical University, Suzhou Municipal Hospital, Gusu School, Nanjing Medical University, Suzhou, Jiangsu, China.
DeepSeek R1 demonstrated superior accuracy and completeness in answering patient questions about atherosclerotic cardiovascular disease (ASCVD) compared to ChatGPT-4o and Gemini. However, all large language models (LLMs) struggled with providing guideline-concordant treatment information.
Area of Science:
- Artificial Intelligence in Medicine
- Natural Language Processing for Health
- Digital Health Literacy
Background:
- Large language models (LLMs) show potential for improving public access to medical information on chronic diseases like atherosclerotic cardiovascular disease (ASCVD).
- The effectiveness of LLMs in multilingual patient-centered health communication, particularly for ASCVD, requires further investigation.
Purpose of the Study:
- To comparatively evaluate the performance of three advanced LLMs (DeepSeek R1, ChatGPT-4o, Gemini) in generating responses to ASCVD-related patient queries.
- To assess LLM performance in English and Chinese across accuracy, completeness, and comprehensibility for patient-facing ASCVD information.
Main Methods:
- A cross-sectional evaluation using 25 validated ASCVD questions across five domains (definitions, diagnosis, treatment, prevention, lifestyle).
- 750 responses generated by submitting each question five times to each LLM in English and Chinese under default settings.
- Independent scoring of responses by three blinded cardiologists using Likert scales, with rigorous randomization and consensus scoring.
Main Results:
- DeepSeek R1 achieved the highest "good response" rate (96%) in both languages, outperforming ChatGPT-4o (84%) and Gemini (48-68%).
- DeepSeek R1 showed superior median accuracy and completeness scores (P<.001).
- All models performed better on definitional and diagnostic questions than treatment/prevention topics; none reliably provided guideline-concordant treatment information.
Conclusions:
- DeepSeek R1 demonstrates strong potential for generating high-quality, patient-facing ASCVD information in multiple languages, enhancing digital health literacy.
- A critical limitation exists in providing guideline-concordant treatment information, suggesting LLM use for ASCVD should be restricted to lower-risk queries without expert oversight.
Related Concept Videos
Atherosclerosis III: Management
Atherosclerosis IV: Nursing Management
Coronary Artery Disease IV: Preventive Measures
Atherosclerosis II: Clinical Manifestations and Diagnostic Tests
Health Literacy
Atherosclerosis I: Introduction

