Related Experiment Video
Updated: Aug 6, 2026
06:38
The Stroke Preclinical Assessment Network Multi-Laboratory Model of Thromboembolic Stroke with Thrombolysis: TE-MCAo
Published on: December 19, 2025
Structured Clinical Input-Guided Large Language Model Workflow for Acute Ischemic Stroke Discharge Education: A
Juntao Yin1, Wan Wang2, Lijuan Wu1
1Department of Neurology (J.Y., L.W., Z.L., Weiwei Wang, G.L., L.T., X.Z.), Xingtai Central Hospital, China.
Stroke
|July 21, 2026
Summary
Large language models (LLMs) show promise for generating patient education materials for acute ischemic stroke survivors. This structured workflow produced high-quality, clinician-supervised discharge summaries, improving on physician-written instructions.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Healthcare
- Clinical Decision Support
Background:
- Large language models (LLMs) offer potential for patient education but face clinical implementation challenges.
- Evaluating LLM feasibility for generating discharge education drafts for acute ischemic stroke patients is crucial.
Purpose of the Study:
- To assess the preliminary feasibility of a structured, clinical input-guided LLM workflow for creating discharge education drafts for acute ischemic stroke patients.
- To compare LLM-generated drafts with physician-written instructions.
Main Methods:
- A workflow involving data extraction from electronic medical records, manual verification, and prompt-based LLM generation (GPT-4o, Grok-3, DeepSeek-R1) was used.
- Two neurologists evaluated LLM drafts and physician-written instructions across five domains.
- Patient-centered evaluations and interrater agreement (intraclass correlation coefficients) were assessed.
Main Results:
- LLM-generated drafts received higher expert ratings than physician-written instructions in risk factor control, rehabilitation, follow-up, and health education (P<0.001).
- GPT-4o and Grok-3 showed higher empathy ratings compared to physician-written instructions (P<0.01).
- No significant hallucinations were found, and LLM drafts had fewer unacceptable ratings.
Conclusions:
- A structured clinical input-guided LLM workflow is preliminarily feasible for generating clinician-supervised discharge education drafts for acute ischemic stroke.
- Further prospective studies are needed to evaluate workflow integration, usability, and impact on patient and clinical outcomes.