Fig. 4: The results of literature search.
From: Streamlining evidence based clinical recommendations with large language models
a Workflow diagram of literature search. The search strategy typically includes both subject terms and free-text terms. The LLM first expands terms based on the clinical question and its decomposition, then determines their logical relationships to generate the final strategy. b Sensitivity of the literature search. Expert results for the 2020 EAN Dementia and 2021 ACR RA guidelines were obtained from Ovid MEDLINE searches provided in the original guidelines. In the 2024 KDIGO CKD, search strategies No.4 and No.5 are identical, as are strategies No.8 and No.9; their differences lie in the target literature being evaluated. Purple bars indicate the number of studies ultimately included. The Agentic method demonstrates comparable sensitivity (macro sensitivity: 0.5942) to expert searches (macro sensitivity: 0.5784). c Precision of the literature search. Bar charts show the total number of retrieved records, with exact numbers labeled for the method retrieving the most. The Agentic method achieves markedly higher precision than other approaches, particularly for 2020 EAN Dementia and 2024 KDIGO CKD.
