The abstract is read at the publisher; the summary is JClub's.
In a PRISMA-ScR systematic review of 192 PubMed and Web of Science papers, 511 ChatGPT application types, 690 concern items, and 368 usage factors were mapped with a 1.04% metadata error rate.
- Why it matters: LLM variability creates uncertainty for healthcare use, and the field needs timely, evidence-integrated views of what ChatGPT applications are being studied and what risks and techniques accompany them.
- What they did: The team used an adaptive, iterative LLM-based agentic AI with human-in-the-loop validation to extract metadata, classify content, and synthesize knowledge, then ran temporal/thematic analyses plus citation network and main-path analysis.
- The result: The human-agentic workflow supported efficient contextual synthesis across studies, enabling trend characterization of ChatGPT applications in healthcare and demonstrating expert-validated literature analysis for medical informatics.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.