A carregar...
Publicação
Health & AI, the regulatory Interactions
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.29 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Tecnologia inovadora, interactiva e acessível, os modelos de grande linguagem (MLL) preenchem todos os requisitos. Esta revisão sistemática examina o seu comportamento no processo de tomada de decisões clínicas. Esta revisão visa identificar o seu desempenho, os principais factores, os desafios associados e a perspetiva da regulamentação dos dispositivos médicos; tudo isto é importante para a tomada de decisões clínicas informadas e consentidas. Uma pesquisa abrangente produziu 60 artigos através de três bases de dados eletrônicas PubMed, Scopus e Science Diret; Abrangendo artigos revisados por pares publicados entre novembro de 2022 e dezembro de 2023. Os resultados coletados indicam que quinze tipos diferentes de grandes modelos de linguagem foram usados para auxiliar o processo de decisão clínica, como encontrar diagnósticos, propor planos de tratamento, prever prognósticos e muito mais a partir de dados de saúde gerados pelo paciente em quatorze especialidades médicas e subespecialidades de cirurgia a psiquiatria; fatores-chave, como aprendizagem por transferência e técnicas de ajuste fino, estavam por trás de seu desempenho. Embora prometedoras, foram encontradas desvantagens como a falta de uma abordagem holística, a falta de contextualização, a priorização incorrecta das tarefas ou uma forte dependência dos dados introduzidos. Além disso, a restrição da intenção de utilização e a informação exclusiva sobre decisões médicas são alguns dos primeiros passos para os regulamentar como dispositivos médicos, sem deixar de considerar a possibilidade de enfrentar alucinações, fadiga de memória e preconceitos, entre muitos outros desafios. Esta análise apoiará os debates em curso sobre a inclusão plausível de ferramentas de inteligência artificial, como os modelos de linguagem de grande dimensão, no processo de tomada de decisões médicas e salientará o empenho das partes interessadas do sector da saúde, das autoridades reguladoras e dos beneficiários entre os profissionais de saúde e os doentes na promoção da sua utilização segura e eficiente, reconhecendo simultaneamente os seus limites.
Innovative, interactive, and approachable technology; large language models (LLMs) check all the boxes, this systematic review examines their behavior in the clinical decision-making process. This review aims to identify their performance, key drivers, challenges associated, and the perspective of medical device regulation; all important in making informed and consented clinical decisions. A comprehensive search yielded 60 articles through three electronic databases PubMed, Scopus, and Science Direct; Covering peer-reviewed articles published between November 2022 and December 2023. Collected findings indicate fifteen different types of large language models were used in assisting clinical decision processes such as finding diagnoses, proposing treatment plans, predicting prognosis, and more from patient-generated health data within fourteen medical specialties and subspecialties from surgery to psychiatry; key drivers such as transfer learning and fine-tuning techniques were behind their performance. Although promising, drawbacks such as lack of holistic approach, lack of contextualization, faulty task prioritization, or a severe dependence on inputs were met. Additionally restraining the intention of use, and solely informing on medical decisions are among the first steps to regulate them as medical devices while still considering confronting hallucinations, memory fatigue, and biases among many other challenges. This review will support ongoing discussions on the plausible inclusion of artificial intelligence tools like large language models in the medical decision-making process and highlight the engagement of healthcare stakeholders, regulatory authorities, and beneficiaries among healthcare professionals and patients in promoting their safe and efficient use while also recognizing their limits.
Innovative, interactive, and approachable technology; large language models (LLMs) check all the boxes, this systematic review examines their behavior in the clinical decision-making process. This review aims to identify their performance, key drivers, challenges associated, and the perspective of medical device regulation; all important in making informed and consented clinical decisions. A comprehensive search yielded 60 articles through three electronic databases PubMed, Scopus, and Science Direct; Covering peer-reviewed articles published between November 2022 and December 2023. Collected findings indicate fifteen different types of large language models were used in assisting clinical decision processes such as finding diagnoses, proposing treatment plans, predicting prognosis, and more from patient-generated health data within fourteen medical specialties and subspecialties from surgery to psychiatry; key drivers such as transfer learning and fine-tuning techniques were behind their performance. Although promising, drawbacks such as lack of holistic approach, lack of contextualization, faulty task prioritization, or a severe dependence on inputs were met. Additionally restraining the intention of use, and solely informing on medical decisions are among the first steps to regulate them as medical devices while still considering confronting hallucinations, memory fatigue, and biases among many other challenges. This review will support ongoing discussions on the plausible inclusion of artificial intelligence tools like large language models in the medical decision-making process and highlight the engagement of healthcare stakeholders, regulatory authorities, and beneficiaries among healthcare professionals and patients in promoting their safe and efficient use while also recognizing their limits.
Descrição
Tese de mestrado, Regulação e Avaliação do Medicamento e Produtos de Saúde, 2024, Universidade de Lisboa, Faculdade de Farmácia.
Palavras-chave
Artificial intelligence Large language model Medical expert system Generative AI Healthcare Medical device Teses de Mestrado -2024