EVALUATION OF AN ARTIFICIAL INTELLIGENCE MODEL FOR CLINICAL DATA EXTRACTION AND ABSTRACTION FROM ELECTRONIC HEALTH RECORDS.

Carolina de Souza Delage Faria; Thais Amanda Frank de Almeida Alves; Thaís Fontes de Magalhães; Nicole Maués Flexa de Oliveira; FLAVIA PANTOJA MACHADO; Mariana Storniolo Sanches Machado; Maria Julia Palacios Santos; Carla Ferreira Kikuchi Fernandes; José Maria Cordeiro Ruano; Marair Sartori

Título do Trabalho

Autores

Carolina de Souza Delage Faria
Thais Amanda Frank de Almeida Alves
Thaís Fontes de Magalhães
Nicole Maués Flexa de Oliveira
FLAVIA PANTOJA MACHADO
Mariana Storniolo Sanches Machado
Maria Julia Palacios Santos
Carla Ferreira Kikuchi Fernandes
José Maria Cordeiro Ruano
Marair Sartori

Modalidade

Resumo

Área temática

Inteligência Artificial e Saúde

Data de Publicação

21/08/2026

País da Publicação

Brasil

Idioma da Publicação

pt-BR

Página do Trabalho

https://www.even3.com.br/anais/xxvii-fiamc-world-congress-for-catholic-physicians-abstract-session-688844/1512121-evaluation-of-an-artificial-intelligence-model-for-clinical-data-extraction-and-abstraction-from-electronic-heal

ISBN

978-65-272-2684-0

Palavras-Chave

Artificial Intelligence, Electronic Health Records, Natural Language Processing, Data Abstraction, Clinical Research.

Resumo

Introduction: Electronic Health Records (EHRs) contain a large volume of clinical data distributed heterogeneously across structured fields and unstructured free-text entries, which limits their systematic use in research. Manual abstraction performed by specialists is time-consuming, costly, and susceptible to errors and interobserver variability. Artificial Intelligence (AI) technologies, particularly those based on Natural Language Processing (NLP), enable the automation of data extraction and structuring. However, their reliability depends on rigorous evaluation against a human reference standard (gold standard). Given the substantial volume of unstructured data, the high cost of manual abstraction, and the need for standardization, AI has the potential to reduce time, cost, and bias, provided it undergoes supervised validation. Objective: To evaluate the accuracy of an AI model in the extraction and tabulation of clinical data from electronic health records. The study aims to assess agreement between manual data abstraction and automated tabulation, refine the model through supervised external correction, and prepare it for large-scale application. Methods: This retrospective observational study evaluated an AI model applied to EHR data extraction. Two guidance manuals were developed to support the AI: (1) a descriptive manual detailing the organization of the medical record, including its structure and data location within the record; and (2) a database manual defining variables, category standardization, and data entry instructions. After parameterization, five medical records were randomly selected for initial testing. Cases were independently abstracted by the AI model and by previously trained healthcare professionals. Discrepancies were reviewed by an independent specialist, who established the reference standard (gold standard). Performance was assessed based on agreement with the established standard, and identified discordances were used for supervised refinement of the model. Results: Observed discrepancies were concentrated in five main categories: extraction of numerical data; semantic standardization of closed-ended responses; temporal interpretation of clinical events; comprehension of open-ended categories; and clinical inference not explicitly documented. Systematic analysis of these discrepancies enabled refinement of extraction rules and instructions, leading to iterative model improvement and greater standardization of automated abstraction. Conclusion: The AI model demonstrated potential to automate the extraction and abstraction of clinical data from Electronic Health Records, with the prospect of significantly reducing time and cost in the development of structured databases. However, its implementation in clinical research requires rigorous evaluation, specialized oversight, and continuous refinement to ensure accuracy and reproducibility. The proposed approach represents a feasible and scalable strategy for leveraging real-world data in clinical studies

Título do Evento: XXVII FIAMC World Congress for Catholic Physicians - ABSTRACT SESSION
Cidade do Evento: Brasília
Título dos Anais do Evento: Proceedings of the XXVII FIAMC World Congress of Catholic Physicians and VI Brazilian Congress of Catholic Physicians
Nome da Editora: Even3
Meio de Divulgação: Meio Digital

Como citar

FARIA, Carolina de Souza Delage et al.. EVALUATION OF AN ARTIFICIAL INTELLIGENCE MODEL FOR CLINICAL DATA EXTRACTION AND ABSTRACTION FROM ELECTRONIC HEALTH RECORDS... In: Proceedings of the XXVII FIAMC World Congress of Catholic Physicians and VI Brazilian Congress of Catholic Physicians. Anais...Brasília(DF) Casa Dom Luciano, 2026. Disponível em: https//www.even3.com.br/anais/xxvii-fiamc-world-congress-for-catholic-physicians-abstract-session-688844/1512121-EVALUATION-OF-AN-ARTIFICIAL-INTELLIGENCE-MODEL-FOR-CLINICAL-DATA-EXTRACTION-AND-ABSTRACTION-FROM-ELECTRONIC-HEAL. Acesso em: 11/09/2026