Skip to content
Review

Artificial intelligenceâ€"assisted sepsis detection in adult intensive care units: a systematic review of clinical implementation and patient outcomes

2026 · International Journal of Medicine in Developing Countries · 0 citations

Abstract

Sepsis is a time-critical condition in which delayed recognition can adversely affect outcomes. Artificial intelligence (AI) and machine learning (ML) might facilitate earlier detection by integrating large amounts of routinely collected clinical data. This systematic review evaluated the performance and clinical utility of AI/ML approaches for sepsis prediction and early recognition. A systematic review was conducted to identify studies evaluating AI/ML-based models for sepsis detection, prediction, or risk stratification. Data were extracted regarding study characteristics, algorithms, predictors, diagnostic performance, comparison with conventional scoring systems, clinical implementation, validation, and patient outcomes. Eleven studies published between 2021 and 2026 were included. Studies used diverse approaches, including random forest, gradient boosting, and deep-learning models, across ICU, emergency department, and hospital settings. Most models demonstrated moderate-to-excellent discrimination, with reported AUCs ranging from 0.47 to 1.00. Several AI models outperformed conventional scores, including NEWS2, SOFA, qSOFA, SIRS, APACHE II, and BISAP. Some systems enabled earlier prediction, ranging from approximately 2â€"5.5 hours before antimicrobial administration to 24 hours before sepsis onset. Limited evidence suggested improved antibiotic timeliness and potentially reduced unnecessary antibiotic use; however, consistent benefits in mortality or length of stay were not demonstrated. External validation was limited, and concerns included class imbalance, alert fatigue, limited generalizability, and potential label leakage. AI/ML showed promising potential for early sepsis detection and risk stratification, frequently outperforming conventional screening tools. However, evidence for improved clinical outcomes remains insufficient. Prospective, multicenter, externally validated studies are needed before widespread clinical implementation.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.