Evaluating large language models for public patient education on levodopa therapy in Parkinson’s disease: accuracy, empathy, reliability, and readability

Large language model (LLM)-based chatbots are increasingly used for health information, but their performance in Parkinson’s disease (PD) patient education remains unclear. This study evaluated five widely used LLM-based chatbots in answering public questions about levodopa therapy. In this cross-sectional comparative study, ChatGPT, Gemini, Copilot, DeepSeek, and Doubao were assessed using a structured set of levodopa-related patient education questions. In the primary analysis, 265 responses were evaluated by five blinded raters for safety, accuracy, empathy, and information quality using predefined criteria, including DISCERN, EQIP, JAMA benchmark criteria, and the Global Quality Score. Readability was assessed using six established indices. An additional 265 responses were generated in an independent repeat-generation analysis using the same question set to assess response stability. Inter-rater agreement was high. Safe-response rates ranged from 86.8% to 94.3%. Accuracy differed significantly, with Copilot and DeepSeek showing the highest median scores, whereas empathy differences were modest and closely clustered. Doubao had the highest DISCERN and EQIP scores, while ChatGPT had the lowest information-quality scores. Gemini showed the best readability profile and Doubao the poorest. Repeat-generation analysis showed generally high stability, with 97.0% exact safety agreement and largely preserved model rankings. Current LLM-based chatbots generated generally informative English-language responses about levodopa therapy, but safety gaps, variable information quality, and poor readability remained. These findings reflect expert-rated output quality rather than demonstrated patient education effectiveness, and patient-centered validation is warranted.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-16
DOI
https://doi.org/10.1038/s41598-026-74929-y
Primary Topic
Artificial Intelligence in Healthcare and Education
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Evaluating large language models for public patient education on levodopa therapy in Parkinson’s disease: accuracy, empathy, reliability, and readability

琳晶 王, Luwen Zhu, Qiang Tang, Xin Zuo et al.
6 links
Scientific Reports
Artificial Intelligence in Healthcare and Education
article

Evaluating large language models for public patient education on levodopa therapy in Parkinson’s disease: accuracy, empathy, reliability, and readability

琳晶 王, Luwen Zhu, Qiang Tang, Xin Zuo, Changfeng Cheng, Chunying Wang, Huan Zuo
article en

Abstract

Large language model (LLM)-based chatbots are increasingly used for health information, but their performance in Parkinson’s disease (PD) patient education remains unclear. This study evaluated five widely used LLM-based chatbots in answering public questions about levodopa therapy. In this cross-sectional comparative study, ChatGPT, Gemini, Copilot, DeepSeek, and Doubao were assessed using a structured set of levodopa-related patient education questions. In the primary analysis, 265 responses were evaluated by five blinded raters for safety, accuracy, empathy, and information quality using predefined criteria, including DISCERN, EQIP, JAMA benchmark criteria, and the Global Quality Score. Readability was assessed using six established indices. An additional 265 responses were generated in an independent repeat-generation analysis using the same question set to assess response stability. Inter-rater agreement was high. Safe-response rates ranged from 86.8% to 94.3%. Accuracy differed significantly, with Copilot and DeepSeek showing the highest median scores, whereas empathy differences were modest and closely clustered. Doubao had the highest DISCERN and EQIP scores, while ChatGPT had the lowest information-quality scores. Gemini showed the best readability profile and Doubao the poorest. Repeat-generation analysis showed generally high stability, with 97.0% exact safety agreement and largely preserved model rankings. Current LLM-based chatbots generated generally informative English-language responses about levodopa therapy, but safety gaps, variable information quality, and poor readability remained. These findings reflect expert-rated output quality rather than demonstrated patient education effectiveness, and patient-centered validation is warranted.

Scientific ReportsVol. 16(1)
Heilongjiang University of Chinese Medicine (CN), First Affiliated Hospital of Heilongjiang University of Chinese Medicine (CN)
Harbin Science and Technology Bureau
Good health and well-being
Artificial Intelligence in Healthcare and Education
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon