Aplicação de modelos de linguagem natural para identificação de informações em registros clínicos

Abstract

This work proposes a solution to improve the quality, consistency, and traceability of clinical records in Emergency Care Units (UPAs) through Natural Language Processing (NLP) techniques and Large Language Models (LLMs). In high-demand environments, record-keeping errors - such as entering prescriptions in evolution fields - compromise data integrity, affecting both medical care and billing processes. The developed solution combines the spaCy library for data anonymization and preprocessing with GPT-5 and Gemini 2.5 Pro models, enabling accurate identification and classification of clinical information. An interactive comparative analysis software was implemented, allowing healthcare professionals to evaluate and compare responses from different LLMs. Preliminary results indicate that LLMs can extract relevant information with a high level of detail, especially GPT-5, demonstrating the potential of this approach to optimize patient care workflows and increase the reliability of medical data in UPAs.