THE ROLE OF ARTIFICIAL INTELLIGENCE IN EARLY SEPSIS DIAGNOSIS AND MANAGEMENT

Resumo

‌ Introduction: Sepsis is a clinical syndrome resulting from a dysregulated response to infections, carrying a high risk of organ failure and death if not treated promptly. The Sequential Organ Failure Assessment (SOFA) score is widely used to diagnose sepsis; however, it has limitations in both sensitivity and specificity. In this context, Machine Learning (ML), a branch of Artificial Intelligence (AI), stands out for its ability to learn from pre-existing data and generate useful predictions. In the case of sepsis, ML models show promise in predicting mortality, identifying high-risk patients, and guiding timely interventions. By analyzing large volumes of vital signs, laboratory tests, and medical histories, these models detect subtle patterns, enhancing the accuracy of predictions and supporting medical decision-making. Objective: Analyze the applications of AI in the early diagnosis and management of sepsis. Methodology: This is a simple narrative review based on data from PubMed (Public Medline) and the Virtual Health Library (BVS). The following Health Sciences Descriptors (DeCS) were used: “Machine Learning,” “Early Diagnosis,” “Artificial Intelligence,” “Sepsis,” and “Triage,” as well as their equivalents in Portuguese. Four original open-access articles, published in English and Portuguese between 2019 and 2024, and aligned with the study objective, were selected. Literature Review: Findings indicate that ML models can accelerate clinical decision-making and optimize resources in sepsis management¹. Their effectiveness depends on being developed by multidisciplinary teams and integrated into institutional sepsis protocols. Since no single biomarker identifies sepsis, combining multiple biomarkers within ML models improves early prediction². Training should include laboratory and clinical data — such as blood, inflammatory, coagulation, and organ function biomarkers — since they are linked to sepsis severity. The greater the diversity of variables, the higher the model’s accuracy². The use of multiple clinical variables in ML has proven more effective for early sepsis diagnosis than the SOFA score³ or conventional medical evaluation⁴, offering greater sensitivity and accuracy in predicting mortality, particularly in emergency settings³,⁴. Conclusion: In light of the evidence, studying AI applications in sepsis management is crucial to improving prognosis and increasing the likelihood of favorable outcomes. However, ML models do not replace the role of multidisciplinary teams. Instead, they should be employed as supportive tools that streamline clinical management and facilitate early treatment initiation, thereby contributing to reduced sepsis-related mortality.