An Intelligent Model and Methodology for Predicting Length of Stay and Survival in a Critical Care Hospital Unit

• Published in: Informatics (Basel) 2024-06, Vol.11 (2), p.34
• Main Authors: Maldonado Belmonte, Enrique, Oton-Tortosa, Salvador, Gutierrez-Martinez, Jose-Maria, Castillo-Martinez, Ana
• Format: Article
• Language: English
• Subjects: architecture; Artificial intelligence; Big Data; Care and treatment; computing methodologies; Critical care; Data analysis; Data mining; Datasets; Decision making; Health care; Hospital utilization; interoperability; Length of stay; Machine learning; Methodology; Mortality; Patients; Scoring models; Survival; Systems integration; Validity
• ISSN: 2227-9709; 2227-9709
• DOI: 10.3390/informatics11020034

This paper describes the design and methodology for the development and validation of an intelligent model in the healthcare domain. The generated model relies on artificial intelligence techniques, aiming to predict the length of stay and survival rate of patients admitted to a critical care hospitalization unit with better results than predictive systems using scoring. The proposed methodology is based on the following stages: preliminary data analysis, analysis of the architecture and systems integration model, the big data model approach, information structure and process development, and the application of machine learning techniques. This investigation substantiates that automated machine learning models significantly surpass traditional prediction techniques for patient outcomes within critical care settings. Specifically, the machine learning-based model attained an F1 score of 0.351 for mortality forecast and 0.615 for length of stay, in contrast to the traditional scoring model’s F1 scores of 0.112 for mortality and 0.412 for length of stay. These results strongly support the advantages of integrating advanced computational techniques in critical healthcare environments. It is also shown that the use of integration architectures allows for improving the quality of the information by providing a data repository large enough to generate intelligent models. From a clinical point of view, obtaining more accurate results in the estimation of the ICU stay and survival offers the possibility of expanding the uses of the model to the identification and prioritization of patients who are candidates for admission to the ICU, as well as the management of patients with specific conditions.