Answering real-world clinical questions using large language model, retrieval-augmented generation, and agentic systems
Objective The practice of evidence-based medicine can be challenging when relevant data are lacking or difficult to contextualize for a specific patient. Large language models (LLMs) could potentially address both challenges by summarizing published literature or generating new studies using real-wo...
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
SAGE Publishing
2025-06-01
|
| Series: | Digital Health |
| Online Access: | https://doi.org/10.1177/20552076251348850 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850212271623503872 |
|---|---|
| author | Yen Sia Low Michael L Jackson Rebecca J Hyde Robert E Brown Neil M Sanghavi Julian D Baldwin C William Pike Jananee Muralidharan Gavin Hui Natasha Alexander Hadeel Hassan Rahul V Nene Morgan Pike Courtney J Pokrzywa Shivam Vedak Adam Paul Yan Dong-han Yao Amy R Zipursky Christina Dinh Philip Ballentine Dan C Derieg Vladimir Polony Rehan N Chawdry Jordan Davies Brigham B Hyde Nigam H Shah Saurabh Gombar |
| author_facet | Yen Sia Low Michael L Jackson Rebecca J Hyde Robert E Brown Neil M Sanghavi Julian D Baldwin C William Pike Jananee Muralidharan Gavin Hui Natasha Alexander Hadeel Hassan Rahul V Nene Morgan Pike Courtney J Pokrzywa Shivam Vedak Adam Paul Yan Dong-han Yao Amy R Zipursky Christina Dinh Philip Ballentine Dan C Derieg Vladimir Polony Rehan N Chawdry Jordan Davies Brigham B Hyde Nigam H Shah Saurabh Gombar |
| author_sort | Yen Sia Low |
| collection | DOAJ |
| description | Objective The practice of evidence-based medicine can be challenging when relevant data are lacking or difficult to contextualize for a specific patient. Large language models (LLMs) could potentially address both challenges by summarizing published literature or generating new studies using real-world data. Materials and Methods We submitted 50 clinical questions to five LLM-based systems: OpenEvidence, which uses an LLM for retrieval-augmented generation (RAG); ChatRWD, which uses an LLM as an interface to a data extraction and analysis pipeline; and three general-purpose LLMs (ChatGPT-4, Claude 3 Opus, Gemini 1.5 Pro). Nine independent physicians evaluated the answers for relevance, quality of supporting evidence, and actionability (i.e., sufficient to justify or change clinical practice). Results General-purpose LLMs rarely produced relevant, evidence-based answers (2–10% of questions). In contrast, RAG-based and agentic LLM systems, respectively, produced relevant, evidence-based answers for 24% (OpenEvidence) to 58% (ChatRWD) of questions. OpenEvidence produced actionable results for 48% of questions with existing evidence, compared to 37% for ChatRWD and <5% for the general-purpose LLMs. ChatRWD provided actionable results for 52% of questions that lacked existing literature compared to <10% for other LLMs. Discussion Special-purpose LLM systems greatly outperformed general-purpose LLMs in producing answers to clinical questions. Retrieval-augmented generation-based LLM (OpenEvidence) performed well when existing data were available, while only the agentic ChatRWD was able to provide actionable answers when preexisting studies were lacking. Conclusion Synergistic systems combining RAG-based evidence summarization and agentic generation of novel evidence could improve the availability of pertinent evidence for patient care. |
| format | Article |
| id | doaj-art-e32cc1f79b9645f3899c1190f2646bf0 |
| institution | OA Journals |
| issn | 2055-2076 |
| language | English |
| publishDate | 2025-06-01 |
| publisher | SAGE Publishing |
| record_format | Article |
| series | Digital Health |
| spelling | doaj-art-e32cc1f79b9645f3899c1190f2646bf02025-08-20T02:09:22ZengSAGE PublishingDigital Health2055-20762025-06-011110.1177/20552076251348850Answering real-world clinical questions using large language model, retrieval-augmented generation, and agentic systemsYen Sia Low0Michael L Jackson1Rebecca J Hyde2Robert E Brown3 Neil M Sanghavi4Julian D Baldwin5C William Pike6Jananee Muralidharan7Gavin Hui8Natasha Alexander9Hadeel Hassan10Rahul V Nene11Morgan Pike12Courtney J Pokrzywa13Shivam Vedak14Adam Paul Yan15Dong-han Yao16Amy R Zipursky17Christina Dinh18Philip Ballentine19Dan C Derieg20Vladimir Polony21Rehan N Chawdry22Jordan Davies23Brigham B Hyde24Nigam H Shah25Saurabh Gombar26 Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Department of Medicine, University of California, Los Angeles, CA, USA Department of Pediatrics, , Toronto, Ontario, Canada Program in Child Health Evaluative Sciences, Peter Gilgan Centre for Research and Learning, , Toronto, Ontario, Canada Department of Emergency Medicine, University of California, San Diego, CA, USA Department of Emergency Medicine, , Ann Arbor, MI, USA Department of Surgery, , New York, NY, USA Division of Clinical Informatics, , Stanford, CA, USA Department of Pediatrics, , Toronto, Ontario, Canada Department of Emergency Medicine, , Stanford, CA, USA Department of Pediatrics, , Toronto, Ontario, Canada Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Atropos Health, New York, NY, USA Division of Clinical Informatics, , Stanford, CA, USA Department of Pathology, , Stanford, CA, USAObjective The practice of evidence-based medicine can be challenging when relevant data are lacking or difficult to contextualize for a specific patient. Large language models (LLMs) could potentially address both challenges by summarizing published literature or generating new studies using real-world data. Materials and Methods We submitted 50 clinical questions to five LLM-based systems: OpenEvidence, which uses an LLM for retrieval-augmented generation (RAG); ChatRWD, which uses an LLM as an interface to a data extraction and analysis pipeline; and three general-purpose LLMs (ChatGPT-4, Claude 3 Opus, Gemini 1.5 Pro). Nine independent physicians evaluated the answers for relevance, quality of supporting evidence, and actionability (i.e., sufficient to justify or change clinical practice). Results General-purpose LLMs rarely produced relevant, evidence-based answers (2–10% of questions). In contrast, RAG-based and agentic LLM systems, respectively, produced relevant, evidence-based answers for 24% (OpenEvidence) to 58% (ChatRWD) of questions. OpenEvidence produced actionable results for 48% of questions with existing evidence, compared to 37% for ChatRWD and <5% for the general-purpose LLMs. ChatRWD provided actionable results for 52% of questions that lacked existing literature compared to <10% for other LLMs. Discussion Special-purpose LLM systems greatly outperformed general-purpose LLMs in producing answers to clinical questions. Retrieval-augmented generation-based LLM (OpenEvidence) performed well when existing data were available, while only the agentic ChatRWD was able to provide actionable answers when preexisting studies were lacking. Conclusion Synergistic systems combining RAG-based evidence summarization and agentic generation of novel evidence could improve the availability of pertinent evidence for patient care.https://doi.org/10.1177/20552076251348850 |
| spellingShingle | Yen Sia Low Michael L Jackson Rebecca J Hyde Robert E Brown Neil M Sanghavi Julian D Baldwin C William Pike Jananee Muralidharan Gavin Hui Natasha Alexander Hadeel Hassan Rahul V Nene Morgan Pike Courtney J Pokrzywa Shivam Vedak Adam Paul Yan Dong-han Yao Amy R Zipursky Christina Dinh Philip Ballentine Dan C Derieg Vladimir Polony Rehan N Chawdry Jordan Davies Brigham B Hyde Nigam H Shah Saurabh Gombar Answering real-world clinical questions using large language model, retrieval-augmented generation, and agentic systems Digital Health |
| title | Answering real-world clinical questions using large language model,
retrieval-augmented generation, and agentic systems |
| title_full | Answering real-world clinical questions using large language model,
retrieval-augmented generation, and agentic systems |
| title_fullStr | Answering real-world clinical questions using large language model,
retrieval-augmented generation, and agentic systems |
| title_full_unstemmed | Answering real-world clinical questions using large language model,
retrieval-augmented generation, and agentic systems |
| title_short | Answering real-world clinical questions using large language model,
retrieval-augmented generation, and agentic systems |
| title_sort | answering real world clinical questions using large language model retrieval augmented generation and agentic systems |
| url | https://doi.org/10.1177/20552076251348850 |
| work_keys_str_mv | AT yensialow answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT michaelljackson answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT rebeccajhyde answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT robertebrown answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT neilmsanghavi answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT juliandbaldwin answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT cwilliampike answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT jananeemuralidharan answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT gavinhui answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT natashaalexander answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT hadeelhassan answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT rahulvnene answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT morganpike answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT courtneyjpokrzywa answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT shivamvedak answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT adampaulyan answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT donghanyao answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT amyrzipursky answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT christinadinh answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT philipballentine answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT dancderieg answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT vladimirpolony answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT rehannchawdry answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT jordandavies answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT brighambhyde answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT nigamhshah answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems AT saurabhgombar answeringrealworldclinicalquestionsusinglargelanguagemodelretrievalaugmentedgenerationandagenticsystems |