Evaluating of large language models in extracting clinical information
DOI:
https://doi.org/10.59681/2175-4411.v16.iEspecial.2024.1306Keywords:
Syndrome, Signs and Symptoms, Machine Learning, Natural Language ProcessingAbstract
Objective: investigate the effectiveness of large language models (LLMs) in named entity recognition (NER) in clinical notes in Brazilian Portuguese. Method: We evaluated the NER task in 30 clinical notes using the metrics and methods of precision, recall, and F-score. In the experiment conducted, we compared the performance of the models GPT-3.5, Gemini, Llama-3, and Sabiá-2 in extracting the entities "Signs or Symptoms," "Diseases or Syndromes," and "Negated Data." Results: We found that the Llama-3 model showed superior performance, especially in sensitivity, achieving an F-score of 0.538. GPT-3.5 demonstrated balanced performance, while Gemini showed higher precision but lower sensitivity. Conclusion: Our results indicate that the choice of model depends on the appropriate weighting of these factors concerning the individual requirements of each clinical application.
References
Yadav, P., Steinbach, M., Kumar, V., & Simon, G. (2018). Mining Electronic Health Records (EHRs). ACM Computing Surveys, 50(6), 1–40. doi:10.1145/3127881
Jensen, P. B., Jensen, L. J., & Brunak, S. (2012). Mining electronic health records: towards better research applications and clinical care. Nature Reviews Genetics, 13(6), 395–405. doi:10.1038/nrg3208
Assale, M., Dui, L. G., Cina, A., Seveso, A., & Cabitza, F. (2019). The Revival of the Notes Field: Leveraging the Unstructured Content in Electronic Health Records. Frontiers in Medicine, 6. doi:10.3389/fmed.2019.00066
Sun, Peng et al. "An overview of named entity recognition." 2018 International Conference on Asian Language Processing (IALP). IEEE, 2018. p. 273-278.
DA SILVA, Diego Pinheiro et al. "Exploring named entity recognition and relation extraction for ontology and medical records integration". Journal of Informatics in Medicine Unlocked vol. 43 (2023): 2352-9148. doi:10.1016/j.imu.2023.101381
Liu, Zhengliang, et al. "Deid-gpt: Zero-shot medical text de-identification by gpt-4." arXiv preprint arXiv:2303.11032 (2023).
Schneider, Elisa Terumi Rubel et al. "BioBERTpt: a portuguese neural language model for clinical Named Entity Recognition." Proceedings of the 3rd Clinical Natural Language Processing Workshop. 19 November 2020, 2020.
Schneider, E. T. R, et al., "CardioBERTpt: Transformer-based Models for Cardiology Language Representation in Portuguese," 2023 IEEE 36th International Symposium on Computer-Based Medical Systems (CBMS), L'Aquila, Italy, 2023, pp. 378-381, doi: 10.1109/CBMS58004.2023.00247.
Oliveira, L.E.S.e., Peters, A.C., da Silva, A.M.P. et al.. SemClinBr - a multi-institutional and multi-specialty semantically annotated corpus for Portuguese clinical NLP tasks. J Biomed Semantics. 2022;13(1):13. Published 2022 May 8. doi:10.1186/s13326-022-00269-1
https://openai.com/index/chatgpt/ [Internet]. San Francisco: OpenAI; c2024 [cited 2024 May 31]. Available from: https://openai.com/index/chatgpt/.
Apresentando o Gemini: nosso maior e mais hábil modelo de IA. [Internet]. California: Google; c2024 [cited 2024 May 31]. Available from: https://blog.google/intl/pt-br/novidades/tecnologia/apresentando-o-gemini-nosso-maior-e-mais-habil-modelo-de-ia/#mensagem-sundar.
https://llama.meta.com/llama3/ [Internet]. California: Meta; c2024 [cited 2024 May 31]. Available from: https://llama.meta.com/llama3/
https://www.maritaca.ai/sabia-2 Internet]. São Paulo: Maritaca AI; c2024 [cited 2024 May 31]. Available from: https://www.maritaca.ai/sabia-2
GE, Yao et al. "Few-shot learning for medical text: A review of advances, trends, and opportunities". Journal of Biomedical Informatics vol. 144 (2023): 1532-0464. doi: 10.1016/ j.jbi.2023.104458
Bird, S., Klein, E., & Loper, E. (2009). Natural language processing with Python: analyzing text with the natural language toolkit. " O’Reilly Media, Inc."
Downloads
Published
How to Cite
Issue
Section
License
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
Submission of a paper to Journal of Health Informatics is understood to imply that it is not being considered for publication elsewhere and that the author(s) permission to publish his/her (their) article(s) in this Journal implies the exclusive authorization of the publishers to deal with all issues concerning the copyright therein. Upon the submission of an article, authors will be asked to sign a Copyright Notice. Acceptance of the agreement will ensure the widest possible dissemination of information. An e-mail will be sent to the corresponding author confirming receipt of the manuscript and acceptance of the agreement.
