The abstract is read at the publisher; the summary is JClub's.
A structured clinical input-guided workflow generated Chinese discharge education drafts for 67 acute ischemic stroke patients, with expert interrater agreement (ICC 0.862–0.943) and higher ratings for most domains.
- Why it matters: LLMs may help with discharge education, but clinical deployment is difficult; this study addresses feasibility, clinician oversight, and whether outputs are rated favorably across key education domains.
- What they did: Using electronic medical record extraction mapped to a deidentified case report form, clinicians verified inputs, then GPT-4o, Grok-3, and DeepSeek-R1 produced drafts; two blinded senior neurologists rated 5 domains, with patient-centered evaluation in 50 patients.
- The result: Across risk factor control, rehabilitation guidance, follow-up planning, and health education, LLM drafts scored higher than physician-written instructions (Bonferroni-adjusted P <0.001), while medication management differences were not significant after correction; GPT-4o and Grok-3 showed higher empathy (Bonferroni-adjusted P <0.01) with no clear hallucinations noted.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.