Título: Predicting Circulatory Failure in ICU Patients with Deep Sequential Models
Autores: Gabriel Rolla Ferreira, Jose Gabriel V. de Souza, Laura Martins Froede, Lucas Pimenta Braga, Luísa Barros Ribeiro Andrade, Matheus Cândido Teixeira, Samuel L. V. Miranda, Anisio Mendes Lacerda, Gisele L. Pappa, Wagner Meira Jr & Alexandre Guimaraes de Almeida Barros
Resumo: Healthcare professionals working in intensive care units (ICUs) receive a large volume of measurements from numerous monitoring systems. However, due to human cognitive limitations, it becomes difficult to process all this information and promptly recognize early signs of patient deterioration. Critically ill patients are typically admitted to ICUs, where alarm systems are implemented for the early detection of circulatory failure — a condition associated with high mortality, especially among severely ill individuals. Current alarm systems, which rely on fixed thresholds, often generate a high number of false positives, contributing to alarm fatigue and potentially delaying appropriate clinical interventions. This study proposes the development of an early warning system based on machine learning (ML) techniques to integrate multivariate data and predict circulatory failure up to 8 hours in advance. The model was trained using the MIMIC-IV database. The dataset includes over 70,000 ICU admissions, and the exploratory analysis was conducted with a temporal resolution of 5 minutes, enabling fine-grained modeling of patient trajectories.
Palavras-chave: Intensive Care Unit; Circulatory Failure; Transformer; Deep Learning; Real-Time Prediction.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1191697
Artigo em PDF: CBIC_2025_paper1191697.pdf
Arquivo BibTeX:
CBIC_2025_1191697.bib