Elucidativa: using language models to explain complementary exam results

• Luana Cruz Machado UFRJ
• Rafael Petri Pinheiro UFRJ
• Felipe Sahb Furtado Amparo

DOI: https://doi.org/10.5753/sbcas_estendido.2024.1955

Abstract

In Brazil, approximately two billion medical tests are conducted annually, many of which patients struggle to interpret, potentially leading to loss of follow-up and lack of awareness regarding their health status. Language models (LLMs), such as GPT-4, have been leveraged to interpret and elucidate medical tests, thereby fostering patient autonomy. This project utilizes optical character recognition (OCR) in conjunction with GPT-4 to extract and simplify medical reports, making information more accessible to patients. Despite being a proof of concept, qualitative studies are necessary to validate the accuracy of the explanations and assess whether patients retain information better after receiving simplified reports.

References

AGÊNCIA NACIONAL DE SAÚDE SUPLEMENTAR. Planos de saúde realizaram 1,8 bilhão de procedimentos em 2022.

CADAMURO, J. et al. Potentials and pitfalls of ChatGPT and natural-language artificial intelligence models for the understanding of laboratory medicine test results. An assessment by the European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) Working Group on Artificial Intelligence (WG-AI). Clinical chemistry and laboratory medicine, v. 61, n. 7, p. 1158–1166, 27 jun. 2023.

CAVALCANTE, G. H. O.; REIS, G. J. DOS. Avaliação do seguimento de lesões precursoras de câncer do colo do útero – uma revisão bibliográfica. Pesquisa e Ensino em Ciências Exatas e da Natureza, v. 5, 25 jul. 2021.

CHOW, J. C. L. et al. Developing an AI-assisted educational chatbot for radiotherapy using the IBM Watson Assistant platform. Healthcare (Basel, Switzerland), v. 11, n. 17, p. 2417, 29 ago. 2023.

MOKMIN, N. A. M.; IBRAHIM, N. A. The evaluation of chatbot as a tool for health literacy education among undergraduate students. Education and information technologies, v. 26, n. 5, p. 6033–6049, set. 2021.