Abstract
Background/Objectives: Sepsis is a severe medical condition characterized by a dysregulated host response to infection, with potentially fatal outcomes, requiring early diagnosis and rapid intervention. The limitations of traditional sepsis identification methods, as well as the complexity of clinical data generated in intensive care, have driven increased interest in applying artificial intelligence in this field. The aim of this narrative review article is to analyze how artificial intelligence is being used in the diagnosis and prognosis of sepsis, to present the most relevant current models and algorithms, and to discuss the challenges and opportunities related to integrating these technologies into clinical practice. Methods: We conducted a structured literature search for this narrative review, covering studies published between 2016 and 2024 in databases such as PubMed/Medline, Scopus, Web of Science, IEEE Xplore, and Google Scholar. The review covered models based on machine learning (ML), deep neural networks (DNNs), Recurrent Neural Networks (RNNs), and clinical alert systems implemented in hospitals. The clinical data sources used, algorithms applied, system architectures, and performance outcomes are presented. Results: Numerous artificial intelligence models demonstrated superior performance compared to conventional clinical scores (qSOFA, SIRS), achieving AUC values above 0.90 in predicting sepsis and mortality. Systems such as Targeted Real-Time Early Warning System (TREWS) and InSight have been clinically validated and have significantly reduced the time to treatment initiation. However, challenges remain, such as a lack of model transparency, algorithmic bias, difficulties integrating into clinical workflows, and the absence of external validation in multicenter settings. Conclusions: Artificial intelligence has the potential to transform sepsis management through early diagnosis, risk stratification, and personalized treatment. A responsible, multidisciplinary approach is necessary, including rigorous clinical validation, enhanced interpretability, and training of healthcare personnel to effectively integrate these technologies into everyday practice.